PHYSMOD 2007

International Workshop
on Physical Modelling of Flow
and Dispersion Phenomena

August 29-31, 2007

Orlans, France

ISBN 2-913454-32-1
EAN 9782913454323
Presses universitaires dOrlans
Crdits photos : - Universit dOrlans / Communication / JSL

- David Hall

Proceedings of
International Workshop on Physical Modelling of
Flow and Dispersion Phenomena
August 29-31, 2007. Orlans, France

Scientific committee:
Dr. Sandrine Aubrun. L.M.E., University of Orlans, France.
Dr. Daniele Contini. Institute of Atmospheric Sciences and Climate, CNR, italy.
Dr. David Hall. Envirobods, Ltd. UK.
Prof. Jerry Havens. University of Arkansas, USA.
Dr. David Heist. U. S. Environmental Protection Agency, N Carolina, USA.
Prof. Zbynek Janour. Academy of Sciences, Prag, Czech Republic.
Prof. Tamas Lajos. Budapest University of Technology and Economics, Hungary.
Dr. Bernd Leitl. University of Hamburg, Germany.
Prof. Masaaki Ohba. Tokyo Polytechnic University, Japan.
Prof. Hideharu Makita. Toyohashi University of Technology, Japan.
Dr. Robert Meroney. Colorado State University, USA.
Prof. Alan Robins. University of Surrey, UK.
Dr. Eric Savory. University of Western Ontario, London, Canada.
Prof. Michael Schatzmann. Universiy of Hamburg, Germany.
Prof. Ted Stathopoulos. Concordia University, Montreal, Canada.
Prof. Jeroen Van Beeck. Von Karman Institute, Rhode-St.-Gense, Belgium.

Editor :
Dr. Sandrine Aubrun. L.M.E., University of Orlans, France.

Workshop organised by:

Universit dOrlans
PolytechOrlans
Laboratoire de Mcanique et dEnergtique

Special thanks for support to:

Universit dOrlans
PolytechOrlans
Laboratoire de Mcanique et dEnergtique
Rgion Centre
Ville dOrlans
Presses Universitaires dOrlans

Preface
On behalf of the Scientific Committee, it is my great pleasure to present to you the
proceedings of the international workshop on Physical Modelling of Flow and Dispersion
Phenomena, PHYSMOD 2007, held at Orlans, France, on August 29-31, 2007.
The workshop had 46 submissions and each of these was reviewed by two members of
the Scientific Committee. We have selected 32 as oral presentations and 14 as poster
presentations. Related full-length papers are gathered in these proceedings. I thank the
members of the Scientific Committee for their important contributions to the workshop,
for providing major input to the formulation of the workshop goals and their careful
reviews of submissions.
The objective of PHYSMOD is to bring together the community active in physical
modelling of flow and dispersion in wind tunnels or water channels. It is intended to
discuss, assess and report on the state-of-the art of experimental work in this field,
define directions of future research and encourage wider collaboration between
research institutes. It is intended to provide a wide platform for information exchange
and knowledge transfer, and participating institutions and laboratories are encouraged
to also bring their undergraduate and postgraduate students to present their work and to
incorporate them into the active fluid modelling community.
PHYSMOD 2007 concentrates on the physical modelling of flow and dispersion in the
natural environment, referring mainly to the following topics:
Heat and mass transfer due to atmospheric dispersion in urban areas (also
including heat island problems)
Unsteady properties of the dispersion process
Building effects on the flow characteristics in urban areas
Validation of numerical and analytical modelling methods
Improvement and validation of atmospheric flow and dispersion modelling
techniques
Quality assurance in physical modelling
Use of boundary layer modelling for wind technology

I wish you all a fruitful workshop and a nice stay in Orlans!

Dr. Sandrine Aubrun

CONTENTS

KEY-NOTE LECTURE
Compiling validation data sets from systematic wind tunnel measurements
requirements and pitfalls ........................................................................................................................ 1
B. Leitl, M. Schatzmann

DENSITY EFFECTS
Effect of Semi-circular Windbreak Array on the Heavy Gas Plume Dispersion in Urban Areas............ 7
B.S. Shiau, Y.C. Wu
Wind tunnel studies of LNG vapor dispersion from impoundments ..................................................... 15
J. Havens, T. Spicer, W. Sheppard
The Effect of Release Time on the Dispersion of a Fixed Inventory of Heavy Gas
A Wind Tunnel Model Study ................................................................................................................. 17
D.J. Hall, V, Kukadia, S. Walker, P. Tily, G.W. Marsland.

QUALITY ASSURANCE
Quality assurance of micro-scale meteorological models Action COST 732.................................... 23
M. Schatzmann, B. Leitl.
How comprehensive is comprehensive enough?
Model-specific reference data for the validation of micro-scale LES flow and dispersion models....... 27
F. Harms, B. Leitl, M. Schatzmann
How dense is dense enough?
Systematic evaluation of the spatial representativeness of flow measurements in urban areas......... 33
D. Repschies, B. Leitl, M. Schatzmann.
How close is close enough?
Sensitivity of wind tunnel results with respect to changing approach flow conditions.......................... 41
I. Herbst, B. Leitl, M. Schatzmann

WIND TECHNOLOGY
Properties of the far wake of a wind turbine in an atmospheric boundary layer .................................. 47
G. Espaa, S. Aubrun, P. Devinant, L. Laporte, E. Dupont.
Aerodynamic Design of the Princess Elizabeth Belgian Antarctic Research Station........................... 53
J. Sanz Rodrigo, C.Gorle, J. Van Beeck, P.Planquart.
Feasibility study of tests on scale models for the evaluation of the overpressures induced
by a passing train on adjacent structures............................................................................................. 59
L. Procino, G. Bartoli, C. Borri, A. Borsani.
Wind Tunnel Simulations of Pollution from Roadways......................................................................... 63
D. K. Heist, S. G. Perry, L. A. Brixey, G. E. Bowker.

URBAN FLOWS
How rough is rough?
Characterization of turbulent fluxes within and above an idealized urban roughness ......................... 69
M. Schultz, B. Leitl, M. Schatzmann.
Effect of roofshape on unsteady flow dynamics in street canyons....................................................... 75
J. Barlow and B. Leitl.
Study of Flow Fields in Asymmetric Step-Down Street Canyons......................................................... 79
B. Addepalli, E.R. Pardyjak
Spanwise variation of drag on roughness elements
in a nominally two-dimensional boundary layer.................................................................................... 87
P. Hayden, T. Mapurisa and A.G. Robins
CFD analyses of flow in stratified atmosphere ..................................................................................... 93
G. Kristf , N. Rcz,, M. Balogh

BUILDING EFFECTS ON PLUMES

Wind tunnel study of the concentration fields in a plume emission...................................................... 99
A.R. Wittwer, F. De Paoli, A.M. Loredo-Souza, E.B.Camano Schettini
Improved Building Dimension Inputs for AERMOD Modeling
of the Mirant Potomac River Generating Station................................................................................ 105
R. L. Petersen, J.J. Carter
Physical modeling of the downwash effect of rooftop structures ....................................................... 111
A. Gupta, P. Saathoff, T. Stathopoulos

DISPERSION IN URBAN AREAS

Dispersion from an area source in urban-like roughness................................................................... 113
F. Pascheke, J.F. Barlow and A. Robins
Dispersion of traffic exhausts in urban street canyons with tree plantings.
Experimental and numerical investigations ....................................................................................... 121
C. Gromke,J. Denev, B. Ruck
Flow and dispersion study in the simplified urban area ..................................................................... 129
H. Sedenkova, Z. Janour
Measurements on the Dispersion of Pollution in Urban Environment of Cubic Building Arrays
with Different Wind Directions ............................................................................................................ 135
B.S. Shiau, Y.S. Lin
Specifying exhaust systems that avoid fume reentry and adverse Health Effects............................. 143
R.L. Petersen, J.J. Carter

JETS AND PLUMES

Comparison of near-field behaviour of jets and plumes in a crosswind............................................. 149
E. Savory, M. Restorick, Z. Duan.
Comparison of experimental and modelled plume rise in stable environments................................. 157
D. Contini, A. Donateo, D. Cesari, A. G. Robins.
Validation of LES on plume dispersion
in the convective boundary layer capped by a temperature inversion ............................................... 165
T. Tamura, H. Nakayama, K. Ohta.

NEW TECHNOLOGICAL SOLUTIONS

Simulation of long time averaged concentration under actual meteorological conditions ................. 167
T. Hara, R. Ohba, K. Okabayashi, J. Yoneda, H. Nagai, T. Hayashi
A proposal for a new atmospheric boundary layer wind tunnel in the Netherlands ........................... 175
P. Builtjes, H. van Dop, B. Holtslag, H. Jonker
Active driving of a multi-fan wind tunnel ............................................................................................. 177
K. Sassa and H. Miyagi
Determination of Spatial Structure of Internal Gravity Wave
by Multi-channel Thermo-Anemometer Measurement....................................................................... 179
H. Makita and K. Ohba

POSTERS
Use of detection of coherent flow structures
for better understanding of 3D flow fields in urban environment........................................................ 185
T. Rgert, I. Goricsn, M. Balcz, K. Czder, T. Lajos
Air-quality and spatial planning .......................................................................................................... 191
FL.H. Vanweert,. J.I.J.H. van Rooij
Charateristics of the lowspeed wind tunnel of the LMF, France ...................................................... 197
L. Perret
Evaluation of pollution dispersion prediction using RANS
with turbulence models available in FLUENT 6.3............................................................................... 201
R. Izarra-Garcia, J. Franke
Flow and dispersion around tall buildings .......................................................................................... 203
T. Lawton, A. Robins
Quantifying the temporal representativeness of flow
and dispersion measurement in a complex urban area ..................................................................... 207
M. Rix, M. Schtzmann, B. Leitl.
How often is sufficient? A program for the statistical analysis
of puff dispersion in urban environments ........................................................................................... 213
R. Fischer, M. Schtzmann, B. Leitl

Issues concerning wind tunnel modelling of heavy gas dispersion

with focus on risk assessment............................................................................................................ 221
K. Bezpalcov, M. Ohba
Visualisation of dispersion processes in the surrounding of livestock buildings ................................ 225
K. von Bobrutzki, H.J. Mller
Wake development and interactions within an array of large wind turbines ...................................... 229
F. Pascheke and P.E. Hancock
Wind-Tunnel Modelling and Numerical Simulation Within Urban Intersection................................... 231
R. Kellnerov, M. F. Yassin, Z. Jaour
Adaptation of the Lucien Malavard
Wind Tunnel as an Atmospheric Boundary Layer Wind Tunnel......................................................... 237
S. Aubrun, G. Espaa, P. Devinant
Numerical Simulation of Room Air Motion, Physics and CFD Modeling............................................ 241
V. Esfahanian, E. Moallem

INDEX OF AUTHORS .............................................................................................243

i}i>}}-i

>i*6-i

i`
>}>i>``
>iy\VVV>`i`VV
*6>`}-ii`*6
/}>V*6

>i>}}-i

->>`*>Vi
>

>v>i>`
LVii
-v>Vi>}}

-i-i

>ii6iVi
*>iiivii

LAVISION GMBH
ANNA-VANDENHOECK-RING 19
37081 GOETTINGEN

TEL.: +49 (0)551 9004 0

EMAIL: INFO@LAVISION.COM
WWW.LAVISION.COM

TSI Europe

TSI - Leaders in Instrumentation

TSI Incorporated designs and manufactures precision instruments
used to measure flow, particulate, and other key parameters in
environments the world over.
TSI provides worldwide solutions for industry, governments,
research institutions, and universities, with applications ranging
from pure research to primary manufacturing. Innovative thinking
keeps TSI Incorporated a leader in the markets we serve.
We remain committed to helping customers solve their
measurement problems, and we have global technical talent,
resources, and commitment to find unique solutions for
difficult applications.

Call us:

TSI France
Europarc Bt D, 26 rue John Meynard Keynes, Technople de Chteau-Gombert, 13013 MARSEILLE
Tl : +33 (0)4.91.95.21.90 - tsifrance@tsi.com - Site internet : www.tsiinc.fr

For more information : www.quantel-laser.com

Imaging Systems
- Global flow
measurements
- Combustion
diagnostics
- Particle size
measurements

Fluid Dynamics

Optimisation of product
design and combustion

Aerodynamics and

- Air and gas flow

hydrodynamics
measurements
- Liquid flow measurements
- Particle size
measurements

Solid Mechanics

Optimisation of
materials and
components

- Strain & Stress

measurements
- Vibration analysis
- Non-destructive testing

Indoor Climate

- Thermal Comfort
- Air Quality
- Ventilation Efficiency

For improved
quality of life

Compiling validation data sets from systematic wind tunnel

measurements requirements and pitfalls
B Leitl, M Schatzmann

Center for Marine and Atmospheric Research, Environmental Wind Tunnel Laboratory
University of Hamburg,
Hamburg, Germany
bernd.leitl@zmaw.de
Abstract - Laboratory data are often used as a
reference for validating micro-scale flow and
dispersion models. In contrast to field data,
carefully generated test data sets compiled from
systematic wind tunnel tests provide a number of
advantages regarding the consistency and accuracy
of the test data, the completeness of a dataset or
concerning the documentation of the boundary
conditions of an experiment. Based on more than
10 years of experience in generating validation data
at the Environmental Wind Tunnel Laboratory
(EWTL), the paper is illustrating common pitfalls
and problems related to the compilation of
reference data sets.
From a wider, projectindependent perspective it is intended to define
minimum modeling quality requirements in order to
further promote boundary layer wind tunnels as a
source of reliable reference data for atmospheric
flow and dispersion modeling.
Key words physical modeling, atmospheric flow and
dispersion, quality assurance, modeling standard.

Introduction
Physical modeling of environmental flow and
dispersion phenomena is providing reliable and detailed
information on wind driven pollutant transport in built-up
areas for several decades. Whereas wind tunnel
modeling was used for almost any kind of environmental
flow and dispersion problem within the atmospheric
boundary layer in the past, nowadays a large fraction of
environmental flow and dispersion phenomena can be
modeled numerically with accuracy sufficient for many
practical applications. However, physical modeling in a
boundary layer is still the preferred choice if local-scale,
transient flow and dispersion phenomena in the range of
a few hundred meters and at time scales in the order of
minutes are a matter of particular interest. Based on
careful and diligent physical modeling and with the help
of state-of-the-art instrumentation, information and data
can be provided with a temporal and spatial resolution
which cannot yet be achieved by numerical modeling.
The high resolution of wind tunnel test data acquired
under controlled boundary conditions also enables
validation data to be compiled from systematic wind
tunnel tests. Based on systematic reference data, the
quality of different types of numerical models can be
evaluated and the 'fitness for purpose' for specific model
applications can be tested.
In general, the availability of test data and their
quality and accuracy significantly affect the outcome of
model validation procedures. In a strict sense, the
results of a model evaluation are as good or as
uncertain as the reference data used (Britter and
Schatzmann, 2006). Depending on the point of view

from which the validation problem is seen, different and

only partially overlapping data quality requirements can
be derived. Perhaps the most global set of data quality
requirements and constraints is developing from a strict
physical perspective but those requirements are most
likely not to be met entirely by field or laboratory data.
Seeing the problem from a numerical modeling point of
view also leads to very demanding data quality
requirements which are unlikely to be met entirely by
regular sources of data. From a physical modelers point
of view, the data to be delivered will satisfy what is 'state
of the art' in instrumentation and practically possible in a
wind tunnel, but not necessarily what is needed by
numerical modeling.
Until now, there is no clear definition of what should
be called 'validation data' and what quality criteria
should be met particularly by systematic test data
derived from wind tunnel tests. Consequently, various
test data sets have been used and sometimes even
misused for 'validating' atmospheric flow and dispersion
models. Since it is obvious, that reliable and qualityassured reference data are as important as a sound
validation strategy for testing accuracy and reliability of
atmospheric flow and dispersion models it is intended to
promote a more complete approach to the validation
data problem. Within the scope of the following paper
the basic requirements on validation data developed
and the methodology applied at the EWTL are outlined
and a methodology for compiling model- and
application-specific reference data is drafted.

1 Data Requirements and Constraints

In principal both, field data and laboratory data can
be used for testing the quality and accuracy of microscale flow and dispersion models.
Nevertheless,
laboratory
data
compiled
from
systematic
measurements in a boundary layer wind tunnel provide
a number of benefits regarding the evaluation of
numerical models and the most prominent general data
requirements are listed below.
Whereas field data in most cases represent very
complex flow and dispersion situations, in a wind tunnel
the degree of complexity can be chosen as needed and
reference data of different complexity can be
provided. Particularly the more simple configurations
such as individual structures or regular arrays of
obstacles enable a numerical code to be tested carefully
with respect to the physical behavior under well-defined
geometrical conditions. A unique feature of laboratory
test data sets certainly is, that even block-structured
'numerical' representations of a certain urban structure
can be tested in detail for fixed, well-documented
boundary conditions (Leitl et al, 2001).
A second major advantage of wind tunnel data to be
used for model validation purposes is the potential to

generate systematic test data sets. The ability of a

numerical model to replicate one particular test case
representing one particular configuration is not
necessarily documenting the models performance. A
difference between model results and reference data
might be acceptable, as long as the model is behaving
'physically correct'.
Testing the physically correct
behavior of a numerical model requires reference data
series with singular test parameters such as the mean
wind direction, the size of the model domain, the
location of emission sources or the type of sources to be
gradually varied in order to test the sensitivity of a model
to certain changes or uncertainties in model input. The
availability of systematic test data is mandatory for
testing the physical performance and the required type
of data can be provided with justifiable effort based
laboratory experiments only.
The completeness of reference data sets is
another point where wind tunnel test data can surpass
the quality of most of the field data sets. As it is nearly
impossible to completely measure and document the
physical and geometrical boundary conditions of a flow
and dispersion experiment in the field, laboratory
experiments provide at least a chance to measure all
driving boundary conditions required for a 'complete
documentation' of a data set. As a foundation for a
realistic evaluation reliable information on the
accuracy/uncertainty of the reference data is another
necessary requirement for a complete documentation
which can be provided not always for field experiments
but which is readily available from carefully prepared
wind tunnel tests.
As a further requirement, the representativeness of
validation data is listed here. For model evaluation, the
reference data must be of known statistical, temporal
and spatial representativeness. While it is again not a
serious problem to define the representativeness of
measured data within the scope of a diligently carried
out wind tunnel experiment it is more or less impossible
to document the representativeness of field data without
further assumptions or without the need of further input
from numerical or physical modeling. However, it must
be stated clearly, that defining and documenting the
'representativeness' of reference data requires
substantial efforts in addition to 'regular' experiments,
but without this information, data is of no value for model
validation.
Finally, the 'fitness for purpose' of validation data
sets must be mentioned. As long as the 'fitness for
purpose' is understood as providing test data for a
particular environmental flow and/or dispersion situation,
there is no real preference for data from field
experiments or laboratory data from a wind tunnel
experiment. However, if the fitness for purpose is
understood as the need for providing validation data
which fit the needs of a certain type of numerical model
to be validated, there are again several advantages in
using specifically generated laboratory data.
For
example, a RANS type code would require a statistically
representative mean flow and dispersion field to be used
as reference, which cannot be derived from field data at
all because the quasi-stationary conditions assumed in
a RANS-based model approach do not exist at full
scale. Similarly, the validation of LES-based numerical
codes is most likely to be based on statistic and
probabilistic measures of representative ensembles of
transient flow and dispersion data which require (longterm) data collection for quasi-stationary boundary
conditions as well.

Despite the advantages the use of carefully compiled

laboratory data has, the biggest constraint is of course
that wind tunnel data are model results only. As it is the
case for numerical models, the results from wind tunnel
modeling are significantly affected by the choice and the
quality of boundary conditions applied in the model and
the simplifications introduced for example regarding the
effects of thermal stratification or the spectral range of
turbulent eddies replicated in the wind tunnel.
Consequently, wind tunnel results do not automatically
provide reliable results for model validation or for upscaling to full-scale conditions when a model looks
'geometrically similar'. In fact, the scatter or uncertainty
in wind tunnel data can distort the physical results
entirely, when no state-of-the-art modeling rules are
applied and physical similarity is not sufficiently fulfilled.
On the other hand, results from independent wind tunnel
test campaigns or even from completely different wind
tunnel facilities are expected to match within the bound
of the measurement accuracy/repeatability of model
tests, if the same geometrical and physical boundary
conditions are realized.

2 Methodical Requirements
2.1 General Remarks

Up to now, there is no clear definition of what the

'state-of-the-art' in wind tunnel modeling is. Most of the
wind tunnel laboratories follow their own modeling
standard, which is driven mainly by the purpose of a
particular wind tunnel test. A comprehensive summary
of the similarity concepts of boundary layer wind tunnel
modeling is given in the 'Guideline for Fluid Modeling of
Atmospheric Diffusion' (Snyder, 1981). Although the
physical background as well as most of the
meteorological references is still appropriate, the main
focus of wind tunnel modeling has changed noticeably in
the past. Long-range pollutant transport from elevated
point sources, flow and dispersion in the upper part of
the atmospheric boundary layer (Ekman-layer) or flow
and dispersion phenomena under stratified atmospheric
conditions are nowadays modeled with sufficient
accuracy using numerical tools instead of a physical
model. A wind tunnel model remains the preferential tool
for modeling flow and dispersion phenomena in the
lower atmospheric boundary layer, where flow field is
dominated by the effects from the underlying roughness.
Accordingly, the modeling requirements must be
adjusted.
A more recent attempt to establish a standard for
environmental flow and dispersion modeling has been
made by the German Engineering Society VDI. The
'standard' defined in VDI 3783/12 (2000) was
established based on a guideline developed by the
WTG (Windtechnologische Gesellschaft) specifically for
building aerodynamics.
The modeling quality
requirements in general and the specification of target
values given in the guideline were based on a
interlaboratory comparison of flow and dispersion
measurements around a single cubic obstacle. As a
result, the uncertainty documented in evaluation/target
data cannot be counted as quality criteria. In contrast, it
is documenting the modeling uncertainty when wind
tunnel modeling is based on the VDI standard only.
Whereas an uncertainty of model results of more than
50% might be acceptable for some practical
applications, this order of magnitude expected as scatter
in model results is clearly unacceptable for reference
data to be used for model validation.

Own experience gained at the EWTL with different

models of exactly the same structure tested in different
wind tunnel facilities clearly shows that a repeatability of
model results better than 10% can be reached for
example for dimensionless concentrations modeled in a
complex street canyon (Liedtke et al, 1998 and
Schatzmann et al, 2006). It requires however a more
careful setup of the test, a much more strict evaluation
of the modeled boundary layer and a diligent selection
and use of the instrumentation applied.

2.2 Model Boundary Layer Evaluation

As stated above, the results from wind tunnel

measurements strongly depend on the physical
boundary conditions used in the model. Consequently, a
proper boundary layer modeling must be counted as a
minimum standard to be fulfilled and documented
completely to ensure a reasonable quality of wind tunnel
results.
However, keeping mean flow profiles,
turbulence quantities, turbulence length scales and
turbulent fluxes within the bounds of a conventionally
defined approach flow / exposure class does not
automatically ensure overlapping/matching model
results. Recent studies gave experimental prove, that
even within a complex urban geometry the model results
can scatter significantly if just the mean wind profile
approaching the model differs in the order of 10 20 %
(Herbst et al, 2007). Most often, it is hardly possible to
identify the uncertainties resulting from slightly
mismatching physical boundary conditions because the
approach flow is not completely documented from a
physical point of view. The documentation standards
currently applied in the EWTL must contain (with the
model in place in the wind tunnel) at least
measurements of
x representative, component-resolving lateral mean
wind profiles at different heights above ground,
documenting the lateral homogeneity of the
approach flow as well as the usable width of the
tunnel cross section
x at least three representative component-resolving
mean vertical approach flow profiles (one in the
center, two at the lateral edges of the model area)
with a sufficient number of measurement points in
order to derive mean wind profile parameters such
as roughness length and the height of the
logarithmic wind profile to be modeled
x several sufficiently long representative wind velocity
time series at appropriate heights above ground in
order to calculate turbulence spectra, turbulence
length scales, lateral and vertical wind direction
fluctuations as well as the time scales of wind
direction fluctuations for checking the 'physical
scale' of the modeled approach flow.
x at least one representative vertical momentum
flux/shear profile measured upwind of the model to
document that a sufficiently thick constant shear
layer is modeled and to provide experimental
evidence that a possible 'blockage effect' of the
model does not significantly affect the modeled
boundary layer flow a more simple but less
sensitive criterion would be a documentation of a
neglibigle longitudinal static pressure gradient
across the model area
If just one of the points listed is missing, it is
impossible to define from a physical point of view the
approach flow conditions applied or the quality of the

modeled boundary layer flow and the data compiled

under boundary conditions of unknown quality is not
qualified for model validation. On the other hand,
common pitfalls like the use of an improper model scale
with respect to the size of the tunnel cross section, an
inaccurate estimation of boundary layer parameters or
the effects of a pressure gradient in the test section of
the wind tunnel can be clearly visualized based on a
complete evaluation of the model boundary layer.

2.3 Physical Similarity under Test Conditions

and Repeatability

During wind tunnel tests, individual experiments are

usually carried out under more or less different 'ambient
conditions'. Strictly spoken, the efficient use of physical
similarity enables individual tests to be carried out under
conditions optimized with respect to the accuracy and
range of the applied wind tunnel instrumentation.
For example, for dispersion experiments the wind
speed is often adjusted depending on the distance of
the measurement point from the source in order to
maintain a sufficiently high concentration signal even for
large downwind distances.
In that case, quality
assurance of dispersion experiments requires the
bounds of physical similarity to be documented based
on systematic measurements at least for all the test
parameters changed in order to ensure that the data is
not collected 'at the edge' of physical similarity or the
operating range of a particular model. For the given
example, this requires repetitive measurements with the
wind speeds systematically changed for a near-field and
a far-field measurement location. Consistency of the
model results could be expected only if measurements
close to the release location do not show a significant
effect of the momentum ratio at the source exit and if
during the measurements carried out far from the source
the minimum wind speed is still sufficiently high. The
independence of concentration measurements form the
Reynolds number of the test and the initial release
conditions becomes critical particular for 'ground level
releases' from surface mounted sources and
measurements taken close to the ground or near walls.
In such cases we expect to see possible Reynolds
number effects or effects caused by the modeled
emission source first. On the other hand, near wall flow
and dispersion measurements in complex geometries is
what is studied most often in boundary layer wind
tunnels nowadays. Pre-testing physical similarity is also
crucial because once the data is transformed into nondimensional numbers for up-scaling to field conditions, it
is hardly possible to distinguish between different test
results and scatter due to mismatching similarity criteria
which would lead in the context of validation data to a
complete misinterpretation of the reference data set.
As a part of standard quality assurance it is also
necessary to carry out independent repetitions of
experiments in order to document the overall scatter
inherent in any experimental data set. Even in a
carefully carried out wind tunnel experiment it is most
often not the accuracy of the instrumentation which
defines accuracy of the results, but the repeatability of
the individual tests. Repeatability tests are necessary to
define the confidence interval of the measured data
needed for a correct interpretation of experimental
results. As long as no information on the repeatability of
the test results is provided, the quality of the test data
cannot be quantified and data cannot be used safely as
a reference for model validation.

2.4 Model Quality and Suitability of

Experimental Setup

In a wind tunnel experiment, not necessarily the

model with the highest geometrical detail or the 'most
realistic appearance' is giving the best or most realistic
answers. Designing and constructing 'aerodynamic
models' of urban structures also differs significantly from
common practice in aviation research, architectural
design or in the models used for investigations of
vehicle aerodynamics for instance. Surface roughness
and surface structure have to be considered as a major
factor influencing the results of near-ground or near-wall
flow and dispersion measurements. Unfortunately, there
are no common standards on how to built wind tunnel
models suitable for aerodynamic testing of urban
structures, except perhaps the local roughness
Reynolds number criteria to be considered, and it is
difficult to judge the 'fitness for purpose' of a model if the
test documentation is missing the information requested
in paragraph 2.3. A prominent example of how different
aerodynamic models might look like is the model
representation of trees in a wind tunnel. Whereas at the
EWTL a aerodynamic model of groups of trees was
developed for flow and dispersion measurements in and
above forest areas (Aubrun et al, 2004), other wind
tunnel laboratories still prefer the use of model trees
with a high degree of geometrical similarity.
Dispersion modeling also requires the emission
sources to be modeled properly in order to simulate a
particular dispersion process. To replicate for example
emissions from road traffic in a wind tunnel, an artificial
'line source' must be built into the model but the way line
sources are actually built normally differs. It is known,
that a well-defined release of emissions from any kind of
wind tunnel model source requires a sufficient pressure
loss across the entire release system to ensure the
source is working independently from the pressure
patterns introduced by obstacles/model buildings
surrounding the source area (Meroney et al, 1996;
Pascheke et al, 2003). If the release is not independent
from the pressure field above the source for example
changing the wind direction in the model will also
change the emission pattern and from concentration
measurements it is most likely not clear, whether a
change in the concentration pattern is caused by
changes in the flow field or whether it is resulting from a
different emission source behavior. A second problem
regarding emission source modeling is a possibly
unrealistic distortion of the mass balance or the local
flow field due to the release of large amounts of tracer.
Again, the street canyon case can be seen as a
prominent example, where the use of too high source
strength or the introduction of a significant momentum
due to jet-like releases can downgrade the accuracy and
reliability of the results significantly. The suitability of a
passive emission source must be documented at least
by giving experimental prove that the results from
nearby concentration measurements scaled with the
source strength give constant values over a wide range
of release rates.
There are several other issues regarding the
suitability of an experimental setup which cannot be
discussed in detail here. For instance, the need for
modeling traffic induced turbulence and how to model
car turbulence is still an open question. However, at this
point it was intended to make the clear statement, that
not just the (geometrical) wind tunnel model but the
entire experimental setup is defining the accuracy of
wind tunnel test results.

2.5 Wind Tunnel Instrumentation and

Experimental Facility

A few remarks on the instrumentation used for flow

and dispersion measurements shall be made here. Even
if it is assumed to be a matter of course, the
instrumentation utilized in a wind tunnel experiment
must be suitable for the type of measurements as well.
Again, there is no common sense, what exactly is
needed for a particular type of flow and dispersion
measurements and suitability is more often defined
based on the availability of instruments than on the
capabilities of measurement devices.
Just as an
example, the required reference wind speed
measurement at a location with sufficiently low
turbulence levels can still be carried out reliably by
means of a pneumatic probe and a calibrated pressure
transducer. In contrast, flow measurements within an
urban structure or next to obstacles will require LDA or
component-resolving PIV for reliable measurements.
Even a multi-component hotwire will not give proper
results because of the complexity and three
dimensionality of the flow field.
For dispersion measurements, most often FID
systems are used, giving sufficiently reliable and
accurate
measurement
results,
provided
that
instrumentation
is
used
properly.
Dispersion
measurements also require a sufficient monitoring
and/or control of the tracer release or emission source
strength as well.
In general, a properly compiled data report for
validation data must necessarily contain sufficient
information on the type of instrumentation used for a
specific measurement task. In addition, the accuracy of
the measurement devices as well as their calibration
status must be documented clearly in order to provide
confidence in the data to be used as a reference for
model validation.
Regarding the wind tunnel itself, there is no
fundamental restriction on what kind of wind tunnel can
be used. Even if specially built boundary layer wind
tunnels might be counted as the most efficient type of
facility, the same or sometimes even better quality of
physical modeling of environmental flows can be
reached in adapted conventional wind tunnel facilities.
The method of boundary layer modeling might, however,
differ. The common approach of using a boundary layer
specific set of turbulence generators and a specific floor
roughness is successfully applied particularly in wind
tunnels with a long test section or fetch upwind of the
model area. For smaller model scales, conventional
wind tunnels have been equipped with grid structures or
horizontal bars, enabling homogenous model boundary
layer flows to be generated within shorter test sections.
Of course, the size of the test section is clearly
defining the model scale to be realized in a specific wind
tunnel facility. From the experiences made in different
boundary layer wind tunnels at EWTL one can conclude
that in an approximately 1.5 m wide test section the
biggest model scale providing sufficient agreement with
full-scale conditions in the lower atmospheric boundary
layer is in the order of 1:500 to 1:400. A larger model
scale in the order of 1:200 or 1:300, as it might be
required for reaching sufficient spatial resolution during
local scale flow and dispersion measurements within
urban structures, already requires a wind tunnel facility
with an approximately 3 to 4 m wide test section when
urban type boundary layer flows have to be modeled.
When using a large scale geometrical model of an urban
structure in a narrow test section, physical similarity is

significantly limited because large scale low frequency

turbulent eddies are certainly not properly modeled at
the right spatial and temporal scale.

2.6 Post-Processing of Laboratory Data

Another set of problems and uncertainties of

laboratory based validation data sets is related to the
post processing and 'proper' scaling of wind tunnel
results. Usually, wind tunnel data are provided to the
'end user' in non-dimensional form, for example as C*concentration values. This enables the end user of the
data theoretically to scale the wind tunnel data
according to different/particular field conditions to be
used during a model validation procedure. In order to
re-scale data properly, it must be documented clearly,
how the raw measurement data was post-processed
and prepared and how the dimensionless values have
been calculated in order to ensure 'compatibility' of
different information sources. Quite often, different and
sometimes even wrong scaling is applied to wind tunnel
data because the data provider does not distinguish
between the type of the source (line/point/area) or
'universal scaling' does not apply in a particular case.
For example dispersion measurements in the near field
of a momentum-depended release such as an exhaust
jet should not be scaled using a C*-concept because it
implies that the results could be transferred to any
combination of wind speed and source strength, which is
not at all the case.
When data is handed out to the end user in nondimensional form, another basic requirement regarding
the documentation of the experiments is also to provide
the measurement uncertainty and repeatability in similar,
non-dimensional form. From this it becomes clear
instantly, that for example not only a very precise
concentration measurement is necessary in the case of
dispersion measurements.
The control and
measurement of the emission source strength and the
reference wind speed can contribute significantly to the
overall uncertainty of the wind tunnel results as well.
Just as another example, if the accuracy of the released
gas volume flow is measured with an accuracy of 20%
only, the minimum accuracy of the concentration
measurements cannot be better than 20%. In fact,
experience shows that most of the 'standard dispersion
measurements' suffer from a severe lack of accuracy in
the control and measurement of the 'initial and boundary
conditions' more than suffering from a lack of accuracy
in the concentration measurements itself. In some,
hopefully rare, cases an in depth analysis of 'standard
dispersion measurements' reveals that they cannot be
more precise than 50%. On the other hand, carefully
carried out dispersion measurements can reach an
accuracy of better than 5%.

3 Summary and Outlook

From the comments given above it is clear, that a
common standard for wind tunnel modeling of
atmospheric flow and dispersion phenomena is muchneeded.
Otherwise, the scatter found in similar
experimental results compiled in different wind tunnel
laboratories might be counted as a general limitation of
physical modeling and not as what it really is, an
uncertainty due to different boundary conditions and
different efforts spent for wind tunnel testing. Quality
assurance becomes even more important if results from
physical modeling are used as a reference for
development and testing of numerical flow and
dispersion models.

In addition to the already established technical

standards and common practices in the wind loads /
wind engineering community, a common practice
guideline for physical modeling of micro-scale
atmospheric flow and dispersion phenomena should be
developed which is sufficiently considering the particular
model quality requirements of dispersion modeling. In
order to further promote wind tunnel testing as a reliable
source of reference data, a quality standard must be
accepted among wind tunnel modelers, which is not
based on the 'least common denominator' only but
based on what is needed to assure reference quality of
data and based on what is technically possible with
'state-of-the-art' facilities and instrumentation.

References
Britter, R. and
Schatzmann, M. (Ed.) (2006).
"Background and justification document to support the
model evaluation guidance and protocol", COST
Action 732, Quality Assurance and Improvement of
Microscale Meteorological Models, ISBN 3-00-0183124.
Leitl, B.; Chauvet, C. and Schatzmann, M. (2001).
"Effects of Geometrical Simplification and Idealization
on the Accuracy of Microscale Dispersion Modeling"
Proc. 3rd Int. Conf. Urban Air Quality, March 19-23
2001, Loutraki, Greece
Snyder, W.H. (1981) "Guideline for Fluid Modeling of
Atmopsheric Diffusion" US EPA, Environmental
Sciences Research Laboratory, Research Triangle
Park, NC 27711
VDI 3783/12 (2000) "Environmental Meteorology
Physical modeling of flow and dispersion processes in
the atmospheric boundary layer, Application of wind
tunnels" Verein Deutsche Ingenieure VDI, Handbuch
Reinhaltung der Luft, Band 1B (in German & English)
Liedtke, J; Leitl, B.; Schatzmann, M. (1998) "Car
exhaust dispersion in a street canyon Wind tunnel
data for validating numerical dispersion models" Proc.
2nd East European Conference on Wind Engineering,
Prague, 7-11 September 1998, vol. 1, pp. 291-297
Schatzmann, M.; Bchlin, W.; Emeis, S.; Khlwein, J.;
Leitl, B.; Mller, W.J., Schfer, K.; Schlnzen, H.
(2006) "Development and validation of tools for the
implementation of european air quality policy in
Germany (Project VALIUM)" Atmos. Chem. Phys., vol.
6; pp. 3077-3083
Herbst, I.; Leitl, B.; Schatzmann, M. (2007) "How close
is close enough? Sensitivity of wind tunnel results
with respect to changing approach flow conditions"
Physmod2007, ibid.
Aubrun, S.; Leitl, B. (2004) "Development of an
improved physical modelling of a forest area in a wind
tunnel" Atmospheric Environment, vol 38, 2004, pp.
2797-2801
Meroney,R.N.;
Pavageau,
M.;
Rafailidis,
S.;
Schatzmann, M. (1996) "Study of line source
characteristics for 2-D physical modelling of pollutant
dispersion in street canyons". J. Wind Eng. Ind.
Aerodyn. 62 (1996), pp. 3756.
Pascheke, F.; Leitl, B.; Schatzmann, M. (2003)
"EVALUATION OF FIELD TRACER EXPERIMENTS
WITH RESPECT TO COMPLEX URBAN TYPE
BOUNDARY CONDITIONS". In Manfrida, Giampaolo
and
Contini,
Daniele,
Eds.
Proceedings
PHYSMOD2003: International Workshop on Physical
Modelling of Flow and Dispersion Phenomena, PratoItalia.

Effect of Semi-circular Windbreak Array on the Heavy Gas Plume

Dispersion in Urban Areas
Bao-Shi Shiau1,2,
1

Yi-Ching Wu2

Institute of Physics, Academia Sinica, Taipei 115, Taiwan

E-mail: bsshiau@gate.sinica.edu.tw

Department of Harbor and River Engineering, National Taiwan Ocean University, Keelung 202, Taiwan
E-mail: b0085@mail.ntou.edu.tw

Abstract - The wind tunnel experiments were

conducted to investigate the effect of
semi-circular windbreak array on the dispersion
characteristics of continuous spill of heavy gas
in urban areas. In the present sturdy, carbon
dioxide was adopted as the heavy gas, and
various initial densimetric Froude numbers
(Fr=16, 22 and 30), and different opening ratio
(P=50% 70% 85%) of the semi-circular
windbreak array cases were performed
experimentally. Experimental results show that
semi-circular windbreak array inhibits the
dispersion of the heavy gas plume and reduces
the plume concentration significantly. When the
opening ratio of the windbreak is 50%, the
concentration reduction downwind of the
windbreak is more effective than that of opening
ratio 70% and 85%. The horizontal dispersion
scales for different downstream distances of the
windbreak approach to the constant values at a
far downstream distance. The approaching to the
constant values of horizontal dispersion scales
is reached quickly as the windbreak opening
ratio decreases. The vertical dispersion scales
appear to close to the constant values at a far
downstream distance behind the windbreak for
different opening ratios of windbreak array.

modification in plume path theory. Duijm et al. [9] made a

review and on the evaluation protocol for heavy gas
dispersion models.
In order to reduce the spill of heavy gas impacts on the
air quality of environment, windbreak array is one of
effective methods and is commonly used to prevent the
dispersion of the heavy gas. The numerical model
developed by the Khan and Abbasi [8] did not account for
the windbreak shelter effect. Wind tunnel test made by
Donat and Schatzmann [5, 6] were also in absence of
windbreak in their experiments. In this study, we carried
out the measurements on the concentration distribution of
spilled heavy gas (carbon dioxide) in the turbulent
boundary layer flow with semi-circular windbreak array.
The urban types of neutral atmospheric turbulent boundary
layers were simulated as the approaching flows. Effects of
the semi-circular windbreak array on the heavy gas
dispersion under various densimetric Froude numbers
(Fr=16, 22 and 30), and different opening ratio (P=50%
70%85%) of the semi-circular windbreak array were
investigated.

2 Experimental Set-up
The experiments were carried out in the National
Taiwan Ocean Universitys Environmental Wind Tunnel.
The wind tunnel test section has a cross section of 2 m
wide by 1.4 m high, and 12.5 m long. The tunnel is an
open suction type and it contracts to the test section with
an area ratio of 4:1. The turbulence intensity of empty
tunnel in test section is less than 0.5 % at the mean
velocity of 5 m/s.
Four spires of 100 cm height and cubic elements (5 cm
x 5 cm x 5cm) are properly arranged as the roughness at
the entrance of test section to generate a thick turbulent
boundary layer which is used as the approaching flow.
An X-type hot-wire incorporating with the TSI IFA-300
constant temperature anemometer was employed to
measure the turbulent flow signals. Output of the analog
signals for turbulent flow was digitized at a rate of 4 K Hz
each channel through the 12 bit Analog-toDigital
converter. Since none of the analog signals containing
significant energy or noise above 1 K Hz, with the Nyquist
criteria, a digitizing rate of 2 K Hz was sufficient. The low
pass frequency for the analog signals is set as 1 K Hz in all
runs of the experiments.
Semi-circular windbreaks were arranged in array with
different opening ratios. The windbreak array was placed
four times of spill height (W=4hs) behind the spill source.
The spill height is the same as the windbreak height (hs=H).
Fig.1 is the schematic diagram of the arrangement of
windbreak array in the experiments.
Carbon dioxide (CO2) is chosen as tracer of heavy gas,

Keywords: windbreak, heavy gas plume, dispersion

1 Introduction
Many kinds of chemical gases storage tanks exist in
the industrial park. The chemical gases are almost toxic
and they are heavier than the air. These kinds of gases are
usually called heavy gases or dense gases. Accidents of
continuous spill of heavy gas from the storage tanks will be
dangerous and have a strong impact on the near
environment of air quality.
Studies on the heavy gas, such as Schatzmann [1, 2],
he indicated that the dispersion of the heavy gas is quite
different from that of the airborne pollutants. Britter [3] had
made a review on the atmospheric dispersion of heavy gas.
Zhu et al. [4] investigated experimentally on the flow
structure within a Dense Gas Plume. Robin et al. [5]
conducted the wind tunnel study on the dense gas
dispersion in a neutral boundary layer over a rough
surface. Donat and Schatzmann [6] had conducted
experiments on the one single-phase heavy gas jets
released in the turbulent boundary layer. Nielsen et al. [7]
made the field tests on the dispersion of pressure liquefied
ammonia. Khan and Abbasi [8] developed the numerical
model to simulate the heavy gas dispersion on the basis of

which is with a molecular weight of 44 (1.52 times

molecular weight of air). Dispersion experiments were
executed by continuously spill the carbon dioxide (CO2) of
heavy gas at a controlled flow rate from an elevated point
source. Sampling tracer of the heavy gas was carried out
for various spilling flow rates and windbreak opening ratios
for downstream distances away from source at different
downstream stations. For concentration measurements,
tracer gas sampling system was developed. It is
composed of 15 sampling tubes that arranged in a rake.
The 15 tubes attach to 15 air bags and suck the tracer gas
by pumps. In order to obtain sufficient tracer concentration
analysis, 5 minutes of sampling time was executed in
every downstream station. Mean concentrations of the
sampled tracer gas in air bags were obtained by using the
analyzer of Cole-Parmer carbon dioxide detector. The
detector has the concentration range of 0~100000 ppm
with a resolution of 10 ppm.

F1{x, y , z , g , hs , Ds , H ,U s , U s , P s ,U a ( hs ), n, Z ref ,

U a , Pa , P}

(1)

where coordinates : x is the downstream direction, y is the

lateral direction, and z is the vertical direction. hs: elevated
height of source; Ds: diameter of source, H: windbreak
height, P: opening ratio of windbreak array, Us: spill
velocity of heavy gas, U s :density of spilled heavy gas;
U a ( hs ) :cross mean wind velocity at the height hs; n:
exponent of power law for mean velocity profile of
approaching flow; Z ref :boundary layer thickness; U a :
ambient air density; P a :viscosity of the air; C is heavy gas
concentration at (x,y,z).
After the dimensional analysis, some dimensionless
groups were obtained, and equation (1) can be
transformed into the following dimensionless form as
shown in equation (2).
K

F2 {

x y z Us
, , ,
, n, P, Fr, Re s , Re}
H H H Ua

(2)

where ( hs H ): dimensionless source height; U s U a :

density ratio; where Fr
U s gDs ( U s  U a ) U s :
densimetric Froude number; Re s DsU s U s / P s : Reynolds
number in source; Re L0U 0 U a / P a : Reynolds number of
ambient flow. The dimensionless concentration K is scaled
as:

CQ
U s Ds H

(3)

As shown in equation 2, it indicates that the

dimensionless local concentration K in both of the model
and prototype is the same at the locations X/H, Y/H, Z/H,
provided that all the remaining dimensionless parameters
on the right-hand side of equation can be matched in the
wind tunnel experiments. In the present study, all the
parameters are set the same for model and prototype
except two parameters, such as: ambient Reynolds
number, Re and spill source Reynolds number, Res. The
ambient Reynolds numbers kept exceeding the critical
number (104) for ensuring flow becomes turbulent in the
experiments. A cross wire was placed at the exit of the
elevated source to trip the outflow. This is to ensure it
become turbulent flow.
The experiments carried out in the present study were
run for some important dimensionless parameters which
make sure of both safe experimental operating conditions
and a variation of the similarity parameters dominating the
heavy gas dispersion.
For non-Gaussian distributions associated with the
heavy gas dispersion behind the semi-circular windbreak
array, the parameters V y and V z can be viewed as the
standard deviation of the concentration distributions in
lateral and vertical directions, respectively. The
parameters are defined by,

hs

Fig.1 Schematic diagram of the arrangement of the

experiments

( y  y ) C ( x, y, z)dy
C ( x, y, z)dy
2

3 Similarity Analysis of Concentration

and Dispersion Scale Parameters

V 2y

From fluid flow modelling laws, some similarity

requirements have to be performed for transferring small
scales of wind tunnel model results to prototype scales.
The similarity laws are obtained by dimensional analysis.
The sketch of the continuously release of the heavy gas
and windbreak location is shown as in Fig.1. The local
heavy gas concentration downstream of the spill source
can be expressed as the following function:

(4)

yC ( x, y, z)dy
C ( x, y, z)dy

(5)

( z  z) C ( x, y, z)dz
C ( x, y, z)dz

(6)

where y

V z2


C ( x, y, z)dy

zC ( x, y , z )dy

where z

approaching flow is shown in Fig.4. Here u ' w' is the

Reynolds stress. As seen in the figure, a constant stress
layer exists near the ground of the boundary layer. This
agrees with the results of field observations reviewed by
Counihan [10].

(7)

4 Results and Discussion

4.1 Approaching Flow
The urban terrain type of neutral turbulent boundary
layer was generated as the approaching flow. Mean
velocity profile of the simulated turbulent boundary layer
flow is approximated by the power law shown as equation
(8).

U (Z )
U ref

Z n
)
Z ref

(8)

where U(Z) is the mean velocity at height of Z , U ref is

the free stream velocity, and Z ref is the boundary layer
thickness. In the present study, a rural terrain type of
neutral atmospheric boundary layer was simulated with a
model scale of 1/500. The free stream velocity is
U ref =3.5 m/s; and the boundary layer thickness, Z ref is
about 100 cm. The measured mean velocity profile is
shown in Fig. 2. Results indicate that the power exponent
n is 0.222. This value lies in the range of 0.21 to 0.40 as
proposed by Counihan [10] for the urban terrain type of
neural atmospheric boundary layer flow.

Fig.3 Turbulence intensity profile of approaching flow

n=0.222
Zref=100cm

U(Z)/Uref=(Z/Zref)n

0.8

X=10m,Y=1m
n=0.222

0.8

0.6
Z/Zref

0.6

0.4

Z/Zref
0.2

0.4

0.2

0.004
u'w'/Uref2

0.006

0.008

Fig.4 Reynolds stress profile of approaching flow

0
0.4

0.6

U(z)/Uref

0.8

4.2 Vertical Concentration

Fig.5 shows the vertical concentration profiles for
different opening ratios of windbreak array at various
downstream distances with Fr=22. Results indicate that
heavy gas concentration is higher for opening ratio 50 %
than that of 70 % and 85 % at the downstream station
x/H=11.667. It is also seen that the vertical concentration
profiles have no significant change as decreasing the
opening ratio of windbreak array at downstream distance
x/H>25. Others cases of Fr=16 and 30 have presented
similar results. In summary, the results show that
semi-circular windbreak array inhibits the dispersion of the
heavy gas plume and reduces the plume concentration
significantly. When the opening ratio of the windbreak is
50%, the concentration reduction downwind of the
windbreak is more effective than that of opening ratio 70%
and 85%.
The vertical concentration profiles for different initial

Fig.2 Mean velocity profile of approaching flow

The simulated longitudinal turbulence intensity profile
is shown in Fig. 3. The longitudinal turbulence intensity is
defined as:
Iu

0.002

u rms
U

(9)

Here urms is the root mean square of turbulent velocity

fluctuation, ,and U is the local mean velocity. It is seen that
the simulated longitudinal turbulence intensity close to the
wall is about 23%. Counihan [10] had indicated that the
longitudinal turbulence intensity close to the ground-level
in the urban terrain areas fell in the range of 20% to 35 %.
The Reynolds stress profile of the simulated

densimetric Froude numbers at various downstream

stations are shown in Fig.6. The opening ratio of
windbreak array is P=85 %, and spill height is hs=H. At the
downstream distance x/H<25, the concentrations become
higher when the initial densimetric Froude number
increases. The other cases of P=50% and 70 % also have
the general likeness.

X/H=25
P=50%
P=70%
P=85%

Z/H 2
4

X/H=8.333
P=50%
P=70%
P=85%

Z/H 2

40

90

140

190

240

X/H=41.667
P=50%

P=70%
P=85%
3

0
40

80

120

160

200

240

Z/H 2
4

X/H=11.667
P=50%
P=70%
P=85%

Z/H 2

40

90

140

190

240

X/H=58.333
P=50%

P=70%
P=85%
3
0
40

90

140

190

240

Z/H 2
4

X/H=16.667
P=50%
P=70%
P=85%

Z/H 2

40

90

140

190

240

Fig.5 Vertical concentration profiles for different opening

ratios of windbreak array at various downstream stations;
Fr=22

4.3 Ground level Concentration

The ground level concentration is an effective index of
how far the heavy gas reaches as it is spilled. In general,
the windbreak arrays have a shelter effect on the spread of
the heavy gas. The ground level concentration variations
along the downstream distances for different opening
ratios of windbreak arrays at Fr=16, hs=H are shown in
Fig.7. As comparing P=50 %, 70%, and 85%, it is found

0
40

90

140

190

240

10

that the ground level concentration firstly increases and

then decreases along the downstream distance for P=50
%. The other two cases of P=70 % and 85 % only exhibit
the decrease of ground level concentration along the
downstream distance.
Fig.8 is the ground level concentration variations along
the downstream distance for different initial densimetric
Froude numbers with P= 50 %.

X/H=25

Fr=30
Fr=22
Fr=16

Z/H 2
4

X/H=8.333

Fr=30
Fr=22
Fr=16

0
0

100

Z/H 2

200

300

X/H=41.667

Fr=30
Fr=22
Fr=16

0
0

50

100

150

200

250

300

Z/H 2

X/H=11.667

Fr=30
Fr=22
Fr=16

0
0

100

Z/H 2

200

300

X/H=58.333

Fr=30
Fr=22
Fr=16

0
0

100

200

300

Z/H 2

X/H=16.667

Fr=30
Fr=22
Fr=16

0
0

Z/H 2

100

200

300

Fig.6 Vertical concentration profiles for different initial

densimetric Froude number at various downstream
stations; P=85 %

4.4 Dispersion Length Scales in Vertical and

Horizontal Directions
The dispersion length scales are commonly used as
indication of the spread extent of the heavy gas as it is
spilled. The vertical dispersion length scale can be
calculated by applying the equations (6) and (7) for the
non-Gaussian concentration distribution of heavy gas

0
0

100

200

300

11

that horizontal length scales increase along the

downstream distance for all cases of opening ratios P=50
%, 70 %, and 85 %. For the case of 50 %, it has a more
shelter effect on the spread of spill heavy gas than that of
the cases of 70 % and 85 %. Therefore, the horizontal
length scale becomes longer.

dispersion over the semi-circular windbreak. The vertical

dispersion length scales for different opening ratios of
semi-circular windbreak array with Fr=16 are plotted in
Fig.9. Results show that the vertical dispersion length
scales increase as the downstream distance increases for
cases with opening ratio of windbreak P=50 %, 70 %, and
85 %. The length scales approach to a constant value at a
far downstream distance for all cases. The shelter effect
on the dispersion of heavy gas plume is better for P=50 %
than the other cases of 70 % and 85 %. So the vertical
length scale is smaller for the case P=50 % at the
locations not far from the windbreak (i.e. x/H<15).

0.5

0.48

200

Fr=16

0.46

P=50%
P=70%
P=85%

160

sigmaZ/H
0.44

Kground 120

0.42

Fr=16

P=50%
P=70%
P=85%

0.4

80

10

20

30

40

X/H

50

60

Fig.9 Vertical dispersion length scales for different opening

ratios of semi-circular windbreak array; Fr=16

40
0

10

20

30

X/H

40

50

60

Fig.7 Ground level concentration variations along the

downstream distances for different opening ratios of
windbreak arrays; Fr=16

2.8

300
2.6

P=50%

Fr=30
Fr=22
Fr=16

sigmaY/H
2.4

200

2.2

Kground

Fr=16

P=50%
P=70%
P=85%

100

10

20

30

X/H

40

50

60

Fig.10 Horizontal dispersion length scales for different

opening ratios of semi-circular windbreak array; Fr=16

0
0

10

20

30

X/H

40

50

60

5 Conclusion

Fig.8 Ground level concentration variations along the

downstream distances for different initial densimetric
Froude number; P=50 %

The wind tunnel experiments were conducted to

investigate the effect of semi-circular windbreak array on
the dispersion characteristics of continuous spill of heavy
gas in urban areas.
Experimental results show that semi-circular windbreak
array inhibits the dispersion of the heavy gas plume and
reduces the plume concentration significantly. When the
opening ratio of the windbreak is 50%, the concentration

By employing equations (4) and (5) to calculate will

yield the horizontal length scale. Fig.10 is the horizontal
dispersion length scales for different opening ratios of
semi-circular windbreak array with Fr=16. Results show

12

[5] Zhu, G., Arya, S.P., and Snyder, W.H., (1998), An

Experimental Study of the Flow Structure within a
Dense Gas Plume, Journal of Hazardous Materials,
Vol.62, pp. 161-186.
[6] Donat, J. And Schatzmann, M., (1999), Wind Tunnel
Experiments of Single-phase Heavy Gas Jets
Released under Various
Angles into Turbulent
Cross Flows, Journal of Wind Engineering and
Industrial Aerodynamics, Vol. 83, pp. 361-370.
[7] Nielsen, M., Ott, S., Jorgensen, H.E., Bengtsson, R.,
Nyren, K., Winter, S., Ride, D., and Jones, C.
(1997), Field Experiments with Dispersion of Pressure
Liquefied Ammonia, Journal of Hazardous Materials,
Vol. 56, pp. 59-105.
[8] Khan, F. and Abbasi, S.A., (2000), Modeling and
Simulation of Heavy Gas Dispersion on the Basis of
Modification in Plume Path Theory, Journal of
Hazardous Materials, Vol. 80, pp. 15-30.
[9] Duijm, N. J, Carssimo, B., Mercer, A., Bartholome, C.,
and Giesbrecht, H., (1997), Development and Test of
an Evaluation Protocol for Heavy Gas Dispersion
Models, Journal of Hazardous Materials, Vol. 56, pp.
273-285.
[10] Counihan, J., (1975), Adiabatic Atmospheric
Boundary Layers: A Review and Analysis of the Data
from
the
Period
1880-1972,
Atmospheric
Environment, Vol. 9, pp. 871-905.

reduction downwind of the windbreak is more effective

than that of opening ratio 70% and 85%. The horizontal
dispersion scales for different downstream distances of the
windbreak approach to the constant values at a far
downstream distance. The approaching to the constant
values of horizontal dispersion scales is reached quickly
as the windbreak opening ratio decreases. The vertical
dispersion scales appear to close to the constant values at
a far downstream distance behind the windbreak for
different opening ratios of windbreak array

6 References
[1] Schatzmann, M., (1993), Experiments with Heavy Gas
Jets in Laminar and Turbulent Cross-Flows,
Atmospheric Environment, Vol. 27A, pp. 1105-1116.
[2] Schatzmann, M., (1995), Accidental Release of Heavy
Gases in Urban Areas, Wind Climate in Cities,
Cermak, J.E., et al. (Eds.), pp. 555-574.
[3] Britter, R.E., (1989), Atmospheric Dispersion of Dense
Gases, Annual Review of Fluid Mechancis, Vol.21,
pp.317-344.
[4] Robins, A., Castro, I., Hayden, P., Steggel, N., Contini,
D., and Heist, D., (2001), A Wind Tunnel Study of
Dense Gas Dispersionin a Neutral Boundary Layer
over a Rough Surface, Atmospheric Environment,
Vol.35, pp. 2243-2252.

13

WIND TUNNEL STUDIES OF LNG VAPOR DISPERSION FROM

IMPOUNDMENTS
Jerry Havens, Tom Spicer, and Wendy Sheppard
Chemical Hazards Research Center
Ralph E. Martin Department of Chemical Engineering
University of Arkansas
Fayetteville, AR
Email : jhavens@uark.edu

Abstract - After briefly outlining the development of

the present regulatory requirements for LNG vapor
dispersion modeling in support of LNG terminal
siting applications in the United States, this paper
presents data from recent wind tunnel experiments
designed to determine experimentally, using carbon
dioxide gas with density similar to LNG vapor, the
effects upon dispersion of an impounding dike, with
or without an LNG tank centered inside the dike,
and compares these dispersion scenarios with that
which would result in the absence of either a tank or
dike. The results, while confirming an expected
reduction
in
downwind
exclusion
zone
requirements due to the presence of a dike and
tank, indicate that the reduction in these tests is
primarily due to the tank rather than the dike.

15

However, the reduction in exclusion zone distance

in this data set, due to the presence of a tank and
dike, appears to be masked by the atypically high
value of surface roughness used in the wind tunnel,
suggesting that the relative effect of the tank would
be even greater if more typical, smaller, site
roughness
were
present.
Surprisingly,
the
measured distance to the endpoint concentration
was shown to be greater when the dike (alone) is
present than when it is absent. This finding is
explained as resulting from the dikes restriction of
gravity spreading, which decreases the overall
planar area for vertical mixing of the vapor cloud,
thus increasing the downwind distance to the
endpoint concentration.

The Effect of Release T

Heavy Ga

D.J. Hall*, V. Kuk

Building Research Establ

Garston, Watford, He
WD25 9XX, U

djhall@envirobods.co.uk ; kukadia

Abstract The paper describe

experiments of releases of a fixe
dense gas, with release times vary
instantaneous to times long enou
continuous plume behaviour.
experiment more closely res
accidental release scenarios and
the most extensive gas clouds
intermediate release rates. The e
varying gas release Richardson n
this to be generally true, with grea
distances to specific gas cloud
distances occurring at inter
Richardson numbers and release
the instantaneous and near cont
rates.

Key words Dispersion, dens

inventory, effect of release rates.

Introduction

Risk assessment studies for t

accidental releases of toxic and flamm
are based either on direct experimen
frequently, on numerical models deriv
of data. Though there is a body
experimental field data of this sort, m
data is obtained using small scale
tunnels and water channels, provid
studies and specific site investigation
of these measurements cover two t
instantaneous (see, for exampl
Havens(1984) and Meroney and L
and continuous (see, for example
and Schatzmann(1991)).
Instant
models mostly use the sudden exp
volume of gas, initially contained in
sort, to the prevailing wind conditio
release is instantaneous, it is also
sense that it has no initial energ
Thorney Island type of field tria
Roebuck(1983)). Continuous releas
which the release time is long eno
plume to have reached a steady sta
no further development. In the case
air gas clouds this time can be c
equivalent of hours in large scale rele

The reality of most releas

somewhat different. Usually there is
a tank- or pipe-full of material und
example, and the variables in the rele
the form of release, its direction, d
and the time over which the releas
release time is the most critical

The Effect of Release Time on the Dispersion of a Fixed Inventory of

Heavy Gas A Wind Tunnel Model Study.
D.J. Hall*, V. Kukadia

S.Walker**, P. Tily, G.W. Marsland

Building Research Establishment,

Garston, Watford, Herts.
WD25 9XX, UK

* Now at Envirobods Ltd

** Now at Dstl Porton Down
Email: djhall@envirobods.co.uk

djhall@envirobods.co.uk ; kukadiav@bre.co.uk

parameter since the release rate of the discharge is in

inverse proportion to the time and in conventional
plumes the concentration is proportional to the
release rate. One might expect, therefore, that with
reducing release time the extent of the plume to a
given boundary would increase. However, as the
release time reduces the amount of energy in the
release increases and for relatively short release
times starts to produce some initial mixing of the gas
cloud, which in turn reduces concentrations in the
cloud. The balance between these effects implies
that the most extensive gas clouds (in terms of
distance to a given concentration) would occur at
some intermediate release time. Predicting this type
of behaviour is difficult as it falls between the two
common types of experiment previously described, so
that there is a lack of a satisfactory experimental data
base on this type of release. Interpolating between
the two extreme conditions, for which data does exist,
is unsatisfactory and at present unreliable.

Abstract The paper describes wind tunnel

experiments of releases of a fixed inventory pf
dense gas, with release times varying from nearly
instantaneous to times long enough to produce
continuous plume behaviour.
This type of
experiment more closely resembles many
accidental release scenarios and should lead to
the most extensive gas clouds occurring at
intermediate release rates. The experiments, for
varying gas release Richardson numbers, show
this to be generally true, with greatest downwind
distances to specific gas cloud concentrations
distances occurring at intermediate bulk
Richardson numbers and release times between
the instantaneous and near continuous release
rates.
Key words Dispersion, dense gases, fixed
inventory, effect of release rates.

The experiments described here address this

problem directly and cover the release of a fixed
inventory of gas over a range of release times. They
were some of the first to do this and used a small
scale wind tunnel model, at about 1/100 of the scale
of the Thorney Island releases. Only a summary of
the work is given here. Full details of the experiment
and its results can be found in Hall et al(1996).

Introduction
Risk assessment studies for the dispersion of
accidental releases of toxic and flammable gas clouds
are based either on direct experimental data or, more
frequently, on numerical models derived from this sort
of data. Though there is a body of large scale
experimental field data of this sort, most experimental
data is obtained using small scale models in wind
tunnels and water channels, providing both generic
studies and specific site investigations. The majority
of these measurements cover two types of release,
instantaneous (see, for example, Spicer and
Havens(1984) and Meroney and Lohmeyer(1983))
and continuous (see, for example, KoningLanglo
and Schatzmann(1991)).
Instantaneous release
models mostly use the sudden exposure of a fixed
volume of gas, initially contained in a tent of some
sort, to the prevailing wind conditions. Though the
release is instantaneous, it is also passive in the
sense that it has no initial energy, following the
Thorney Island type of field trial (Mcquaid and
Roebuck(1983)). Continuous releases are those for
which the release time is long enough for the gas
plume to have reached a steady state condition with
no further development. In the case of heavier-thanair gas clouds this time can be considerable, the
equivalent of hours in large scale releases.

1 Experimental Details and Scaling.

The experiments were carried out in the BRE
dispersion modelling wind tunnel, then at Warren
Spring Laboratory, which has a working section 22m
long, 4.3m wide and 1.5m high.
A simulated
boundary layer for a surface roughness, z0, of 0.4mm
was used in the experiments. The gas cloud, of a
fixed volume, V, of 2000 ml, of gas was released into
the wind tunnel over a variable time, T., discharged
from a flush opening in the tunnel floor of 100mm
diameter
(some
experiments
using
smaller
diameters). The gas discharge system (set below the
tunnel floor) is shown in Figure 1. It used a rolling
diaphragm cylinder in which the gases each side of
the piston are contained in leak proof rolling rubber
diaphragms, with the discharging gas contained in the
upper cylinder by a commercial pinch valve, which
permits filling the cylinder but allows a very free flow
of the discharging gas.
The gas cloud was
discharged by pressurising the lower side of the
cylinder and the discharge rate controlled by varying
this pressure and the use of constricting orifices.
Discharge times varied between 0.1 and 30s.
Concentration/time measurements were made along
the release centreline in the wind direction and both
laterally and vertically at two downstream distances,
of 150mm and 600mm. Ten repeat measurements
were made at each sampling point both to show the

The reality of most release scenarios is

somewhat different. Usually there is a fixed inventory,
a tank- or pipe-full of material under pressure for
example, and the variables in the release scenario are
the form of release, its direction, discharge velocity
and the time over which the release occurs. The
release time is the most critical

17

repeat variability in this type of release and to obtain

reliable mean concentration data.

shaded squares are for conditions with negligible

discharge momentum.

The characteristic length scale, L, of the experiment

was defined as,
L V1/ 3 .
The fixed discharge volume, of 2000 ml, gives a value
of L of 0.126m. The height for the reference wind
speed, U, was also taken as L. The reference wind
speed varied between 0.7 and 1.2 m s-1. The release
volume and source size matched several previous
experiments using instantaneous releases of the
Thorney Island type, so that a direct data comparison
with the earlier work was possible.

2 Results of Experiments and

Analysis
Figure 2 shows examples of ground level
concentration/time traces for low (Ri=0.5) and high
(Ri=5) Richardson numbers at a distance of 600mm
(X/L = 4.8) from the source. The left hand pair of
plots are for a short release time (Tr = 0.3s) and the
right hand pair are for a long release time (Tr = 10s)
Each plot shows the mean, 90%ile and 10%ile
concentration distribution derived from the ten repeat
measurements. All subsequent data in this paper are
of the mean values of the repeated measurements
only. The traces are all consistent with other heavy
gas release data. The short time high Richardson
number release data shows the characteristic sharp
initial peak in concentration due to the gravity current
front of the gas cloud from short term releases, while
the other traces show a more Gaussian type of
distribution, due either to longer release times or to
lower Richardson numbers.

Time scales, T, were non-dimensionalised with

respect to the characteristic length scale, L, and
reference wind speed U to provide the dimensionless
time,
UT
,
L
and the dimensionless release time,
UTr
L
where Tr was the release time of the gas cloud.
The release times, Tr, used, between 0.1 (the shortest
that could be readily managed) and
30s, produced dimensionless release times between
about 0.5 and 250. The shortest values of
UTr/L corresponded to the external airflow travelling
distances less than the source diameter during the
discharge, close to an instantaneous release, the
longest times produced a near continuous plume in
the downwind direction in most cases.

Figure 3 shows a complete set of ground level mean

concentration time traces for all the release
conditions, measured 600mm (X/L = 4.8) from the
source centre, so the relative effects of the
Richardson number and release time can be clearly
seen. Because of the initial release energy the
shortest release time gave the lowest concentration at
the ground, the highest concentrations occurring at
intermediate release times between 0.3s and 3s. It
can be seen from the traces that the release time
apparently had a greater effect than the Richardson
number on the gas cloud. The traces pass from those
characteristic of instantaneous releases at short
release times to those characteristic of plume
segments at the longest release time of 30s. The
release times for the highest concentrations in the gas
clouds varied with the Richardson number. For the
neutrally buoyant releases the highest concentration
in the gas cloud occurred for the 3s release time,
while at higher Richardson numbers it occurred at
time between 0.3s and 3s..

The relative stability of the gas release was defined by

the bulk Richardson number Ri,
'U L
Ri g
U U2
where, = g-a, where g is the density of the
released gas and a is the ambient density. A value of
Ri of zero represents a neutrally buoyant release and
a value of 10 a very stable gas cloud exhibiting strong
stability effects approaching those of still air
conditions. The values of Ri used here were, 0, 0.5, 1,
2, 5 and 10, obtained by varying both wind speed and
source gas density.

In practice the gas cloud concentrations shown in

Figure 3 are additionally dependent on whether the
release became elevated due to the discharge
momentum. For the neutrally buoyant releases this
was significant and the low concentrations were partly
due to elevation of the release. However, at the
highest Richardson number the gas cloud remained
largely on the ground and the low concentrations at
the shortest release times were mostly due to
enhanced initial mixing of the gas cloud.

The source discharge momentum, m, is of interest as

the shorter release times produced significant
momentum.
The
dimensionless
discharge
momentum, M, is defined as,
Ug 2
m
M
, where m
w A.
2 2
UL
Ua
w is the discharge velocity of the gas from the source
of area A. Note that this momentum is relative to the
ambient density, a, not absolute, though the
difference is small since the ambient density of air is
approximately 1.2kg m-3.

One of the main interests in dense gas cloud

dispersion is the downwind distance to a specific gas
cloud concentration. Figure 4 shows contour maps of
the dimensionless distances (X/L) to the maximum
and the mean gas cloud concentration at 1% and 10%
of the source concentration. Both maximum and
mean concentrations show similar behaviour. It can
be seen that the greatest distances for both
concentrations occurred in dense gas clouds of
relatively low Richardson number and intermediate
release times.

Table 1, shows the basic data for the experimental

test conditions as an array of Richardson numbers
and release times over which measurements were
made. Each square in these columns shows the
dimensionless release time in bold and the
dimensionless discharge momentum in italics. The

18

7.There is a complete archive of this data set

available for further investigations.

Some facets of the gas cloud behaviour can be seen

in Figure 5, which shows plots of ground level gas
cloud concentration with distance from the source for
all the release times at four Richardson numbers. In
general the release time has the greatest effect on
ground level concentration in neutrally buoyant
releases, and the least effect at the highest cloud
Richardson number. For the shortest release times
all the dense gas clouds show an inflection in the
reduction of concentration with distance, implying
some elevated part of the gas cloud contacting the
ground further downwind.
No confirmatory flow
visualisation was possible in these experiments, but
the fact that all the concentration distributions
otherwise show a falling concentration with increasing
distance suggest that a major part of the gas cloud
always remained in contact with the ground,
irrespective of the discharge momentum. There is
some evidence from the vertical concentration
distributions of the cloud that it partly bifurcated into
airborne and ground based components.

Acknowledgments
The work described here was jointly funded by the UK
Health and Safety Executive and the European
Commission as part of the FLADIS (CT90-0125)
project in the Major Technical Hazards Programme.
The experiments were carried out a the Warren
Spring Laboratory prior to its closure and the data
analysis and report completed at the Building
Research Establishment, to which two of the authors
and the dispersion wind tunnel transferred.

References
Hall D.J. Kukadia V., Walker S., Tily P, Marsland
G.W.(2006) The Effects of release time on the
dispersion of a fixed inventory of heavier-than-air
gas A wind tunnel model study. Building
Research Establishment, UK.
Report No.
CR149/96.
Hall D.J. and Waters R.A.(1989). Investigation of Two
Features of Continuously Released Heavy Gas
Plumes. Warren Spring Laboratory, UK. Report
No.LR707(PA).
Havens J.A., Spicer T.O.(1983). Gravity Spreading
and air Entrainment by Heavy Gases
Instantaneously Released Into a Calm
Atmosphere.
IUTAM
Symposium
on
Atmospheric Dispersion of Heavy Gases and
Small Particles, Delft, The Netherlands, August
29-September 2cnd. Springer-Verlag. ISBN 3540-13491-3.
Konig-Langlo G., and Schatzmann M.(1991). Wind
Tunnel Modelling of Heavy Gas Dispersion.
Atmospheric Environment, Vol. 25A, No. 7, pp.
1189-1198.
McQuaid J.D., Roebuck B.(1984). Large Scale Field
Trials on Dense Vapour Dispersion. Final Report
to Sponsors on Heavy gas Dispersion Trials at
Thorney Island, 1982-1984. Health and Safety
Executive, Sheffield, UK.
Meroney R.N., Lohmeyer A.(1983). Statistical
Characteristics of Instantaneous Dense Gas
Clouds Released in an Atmospheric Boundary
Boundary
Layer
Layer
Wind
Tunnel.
Meteorology. Vol. 28. pp. 1-22.

3 Discussion and Conclusions

1. The principle outlined at the start of this paper, that
the most extensive gas clouds from a
discharge of a fixed inventory occur at
release
times
intermediate
between
instantaneous and continuous, is borne out
by the experiments. Nearly all the measured
parameters
showed
this
type
of
characteristic.
2. The most extensive gas clouds occurred at
relatively short release times, corresponding
to values of the dimensionless release time
around 2-10, the time reducing with
increasing Richardson number.
3. The main cause of this effect was the initial mixing
of gas clouds generated by their release
energy when the release time was short,
which reduces the source concentration.
4. The overall gas cloud behaviour was the result of a
complex interaction between release rate as
a generator of initial mixing, release rate as a
regulator of concentration and the size of the
gas cloud, gas cloud stability (Richardson
number) source size and the details of the
release.
5. The basic release showed a mixture of omnidirectional and vertical discharge which
resulted at times in two-component plumes.
These could generate sudden increases in
concentration at the ground as the two parts
of the plume came together.
6. For longer release times, the effect of the discharge
momentum diminished and over about half
the of the measurements, the lower right half
of the conditions laid out in Table 1, it is
doubtful whether the details of the release,
other than the release time, were of any
particular importance to the resultant plume
dispersion.

19

Table 1. Operating conditions for experiments.

DImensionless release time, UTr/L, in bold
and dimensionless discharge momentum, M,
in italics.

Gas

g /a

/a

U
m s-1

Air

0.97

0.5

Argon

1.38

0.38

Argon

1.38

BCF/Air
50/50

5
10

Ri

Release Time, Tr (s)

0.1

0.3

10

30

0.77
3.4
#

2.31
0.38
B

7.7
0.034
#

23.1
0.0037
B

77
0.0003

231
0.00004

0.97

0.77
4.7

2.31
0.52

7.7
0.047

23.1
0.0052

77
0.0005

231
0.00005

0.38

0.69

0.55
9.12#

1.64
1.03B

5.47
0.091

16.4
0.010

54.7
0.0009

164
0.00001

3.37

2.37

1.207

0.96
7.3

2.87
0.82

9.6
0.073

28.7
0.0082

96
0.0007

287
0.00008

BCF

5.74

4.74

1.08

0.86
15.5

2.57
0.1.74

8.57
0.15

25.7
0.017

85.7
0.0016

257
0.00017

BCF

5.74

4.74

0.77

0.61
31#

1.83
3.42
-

6.1
0.31

18.3
0.034

61
0.0031

183
0.0003

Additional measurements made using

Reduced Diameter Source
#
Additional Measurements made using
Source Fitted with Raised Circular Cover
Shaded area represents approximate region where
source discharge momentum effects are small.

20

Figure 2. Examples of centreline ground level gas

cloud
concentration/time
traces
measurements.
600mm (X/L=0.=4.8)
downwind of the source.
Left side Tr = 0.3s, right side Tr = 10s.

Figure 1. Gas cloud discharge apparatus

Figure 3. Mean concentration/time traces for all

gas cloud release conditions, measured at
the ground 600mm (X/L=4.8) from the
source.

21

Figure 4. Contours of dimensionless distance (X/L) to 1% and 10% maximum (left) and mean (right) concentration
as a function of release Richardson number and release time.

Figure 5. Plots of maximum ground level gas cloud concentration against distance for four Richardson numbers
and all gas cloud release times.

22

Quality assurance of micro-scale meteorological models

Action COST 732
M. Schatzmann, B. Leitl

University of Hamburg, Meteorological Institute (ZMAW)

Hamburg, Germany
schatzmann@zmaw.de
Abstract A COST action has been launched which
is tasked to assure the quality of micro-scale
meteorological models that are applied for
predicting flow and transport processes in urban or
industrial environments. Main objective of the action
is to develop a coherent and structured model
evaluation procedure. Part of the procedure is the
provision of appropriate validation data. The talk
will focus on validation data requirements and the
role of boundary layer wind tunnels within this
context.
Key words quality assurance, obstacle resolving
numerical models, field and wind tunnel data, urban
canopy layer.

Introduction
The emergence of increasingly powerful computers
enabled the development of tools that have the potential
to predict flow and transport processes within the urban
canopy layer. These new tools are micro-scale
meteorological models of prognostic or diagnostic type.
Prognostic models are based on the Reynolds-averaged
Navier-Stokes (RANS) equations, whereas diagnostic
models are less sophisticated and ensure only the
conservation of mass. These two model types are
presently supplemented by even simpler engineering
tools. It is to be expected, however, that the latter will
sooner or later be replaced by Computational Fluid
Dynamics (CFD) tools such as RANS codes or the even
more complex Large Eddy Simulation (LES) models.
Micro-scale meteorological models are special in so
far as they are tailored to the needs of meteorologists.
They are adjusted to domain sizes of the order of
several decametres to a few kilometres (street canyons,
city quarters). They usually use boundary conditions
based on surface characteristics like land use,
roughness and displacement thickness and they may
contain modules that have the potential to simulate
chemical transformations, aerosol formation or other
important atmospheric physico-chemical processes.
Typically these models contain a substantial amount of
empirical knowledge, not only in the turbulent closure
schemes but also in the use of wall functions and in
other parameterisations.
Models play an important and often dominant role in
environmental assessment and urban climate studies
that are undertaken to investigate and quantify the
effects of human activity on air quality and the local
climate. Their increasing use is paralleled by a growing
awareness that the most of these models have never
been subject to rigorous evaluation. Consequently there
is often a lack of confidence in the modelled results.

23

Objective and Methodology

The main objective of action COST 732 is to improve
and assure the quality of micro-scale meteorological
models that are applied for predicting flow and transport
processes in urban or industrial environments. In
particular it is intended
x
to develop a coherent and structured quality
assurance procedure for this type of model that
gives clear guidance to developers and users as
to how to properly assure their quality and their
proper application,
x
to provide a systematically compiled set of
appropriate and sufficiently detailed data for
model validation work in a documented,
convenient and generally accessible form (www
data bank),
x
to invite from all participating states model
developers and users to apply the procedure and
to prove its serviceability,
x
to build a consensus within the community of
micro-scale model developers and users
regarding the usefulness of the procedure,
x
to stimulate a widespread application of the
procedure and the preparation of quality
assurance protocols which prove the fitness for
purpose of all micro-scale meteorological
models participating in this activity,
x
to contribute to the proper use of models by
disseminating information on the range of
applicability, the potential and the limitations of
such models,
x
to establish a consensus on best practises in
current model use and
x
to
give
recommendations
for
focussed
experimental programmes in order to improve
the data base.
It is to be expected that the very existence of a
widely accepted European standard for quality
assurance in the field of micro-scale meteorological
models in combination with the provision of suitable
validation data will significantly improve the culture
within which such models are developed and applied.
European model developers shall find step-by-step
guidance on how to demonstrate that their models are fit
for a particular purpose. Data sets (both flow and
concentration
data)
obtained
from
extensive
experiments will be made accessible and more widely
exploited. Relevant expertise available within the
member states will be brought together and combined to
develop a consensus for appropriate model use and
model improvement.

Results and Discussion

Model description: this should be a brief

description of the characteristics of the model,
the intended range of applicability, the theoretical
background on which the model development
was based, the software and hardware
requirements, etc.
x
Database description: a complete description of
the database that is to be employed for the
evaluation of the model, including the reasons
why this specific database was chosen. An
estimation of the data variability is required.
x
Scientific Evaluation: this is a description of the
equations employed to describe the physical and
chemical processes that the model has been
designed to include. If appropriate it should
justify the choice of the numerical modelling
procedures and it should clearly state the limits
with respect to the intended applications.
x
(Code) verification: this process is to verify that
the model produces results that are in
accordance with the actual physics and
mathematics that have been employed. This is to
identify, quantify and reduce errors in the
transcription of the mathematical model into a
computational model and the solution (analytical
or numerical) of the model.
x
Model validation: this is a structured comparison
of model predictions with experimental data and
is based on statistical analyses of selected
variables. It seeks to identify and quantify the
difference between the model predictions and the
evaluation datasets; it provides evidence as to
how well the model approximates to reality. A
quantification of the uncertainty of the model
predictions should be produced.
x
User-oriented assessment: is there a readable,
comprehensive documentation of the code
including technical description, user manual and
evaluation documentation? The range of
applicability of the model, the computing
requirements, installation procedures, and
troubleshooting advice should be available.
Five of the steps of the evaluation procedure
described above are relatively straightforward but the
model validation is complex and requires more attention.
Unfortunately this has led to the often-seen model
evaluation study that is no more than the validation step.
At the heart of the complexity of the model validation
process is the stochastic nature of atmospheric flows,
whether real or physically modelled. For example, and
prior to any comparison between mathematical model
and experimental results, the user or model evaluator
needs to address issues such as:
x
Which quantities should be compared?
x
At which point within the area of interest should
the comparison take place?
x
Should the comparison take place on a point-topoint basis or on an area averaged basis?
x
Should the compared quantities be averaged
over a specific period of time and if so what is the
time over which the averaging should take place?
x
Should the quantities be compared at the same
time or at different times?
The answers to these questions become clearer
when the purpose of the model is precisely stated. The
various metrics to be used need to be carefully selected
and agreed upon. Experience has shown that there may
be some generally expected values for these metrics for
x

The action started in July 2005 with a joint

ESF/COST 732 Exploratory Workshop on Quality
Assurance of Micro-Scale Meteorological Models in
Hamburg. About 45 scientists from Europe and the US
(the number of participants was limited in order to allow
ample discussions) attended the workshop. The
workshop proceedings (Schatzmann and Britter 2005)
contain a state of the art report on former quality
assurance initiatives in the field of micro-scale
meteorological models. These initiatives comprise the
'General Requirements for a Quality Assurance Project
Plan' by Borrego and Tchepel (1999), the 'Guidelines for
Model Developers' and the 'Model Evaluation Protocol'
which were worked out by the Model Evaluation Group
(MEG, 1994) under the CECs Major Industrial Hazards
Programme, the US-Environmental Protection Agencys
requirements for quality assurance of atmospheric
dispersion models (Irwin, 1998 and 1999) and the
experience
gathered within the initiative for
harmonization of atmospheric dispersion modelling for
regulatory purposes (Olesen, 1999 and subsequent
papers). Results from similar initiatives in related fields
were also taken into account, for example from the
investigations carried out within the Podbi-model intercomparison exercise (Lohmeyer et al., 2002), from the
FP5 project EMU (Hall 1997), the thematic network
QNET-CFD or COST Action C14 which dealt with the
industrial application of CFD codes for engineering
applications. Finally, the recommendations given by
national bodies, e.g., the Quality Assurance Guidelines
released by a task force of UKs 'Royal Meteorological
Society' (1995) and by Germanys 'VDI Commission on
Clean Air' (2002), were carefully considered. With
respect to data the considerations outlined in
Schatzmann et al (2002, 2003) were taken into account
and standards for validation data were defined which
can generally only be met by data sets based on a
combination of field and wind tunnel experiments.
Strategies for assuring the quality of a numerical
model can only be based on very generic scientific
principles such as the principle of falsification (K. R.
Popper, 1959). The decision about which particular
tests should be performed and which particular data
sets should be used for comparisons between model
results and observations can ultimately be only based
on a consensus built up within and by the scientific
community. The impact of COST 732 is dependent on
whether the quality assurance procedures suggested by
the Action are accepted by the community of model
developers and users or not. Therefore, the next logical
step was to draft a first version of the evaluation
procedure and its underlying motivation in order to
provide the basis for subsequent discussions within the
scientific community. This was done in form of two
related documents: A rather lengthy
x
Background and justification document to
support the model evaluation guidance and
protocol document (Britter, R., and Schatzmann,
M. 2007 a) and a much shorter
x
Model evaluation guidance and protocol
document (Britter, R., and Schatzmann, M. 2007
b).
The first document contains detailed explanations
concerning the general model evaluation philosophy and
the sequence of tasks that should be completed. These
tasks are

24

state of the art/science models when applied to

particular data sets subject to a specified protocol.
A special section is devoted to validation data
requirements. For the validation of micro-scale
meteorological models a suite of data sets with
increasing geometrical complexity is needed that allows
systematic testing of numerical codes. The data sets
must be complete, i.e. they must contain sufficient
information to set up a model run without further
assumptions concerning the model input parameters
and the uncertainty of the data must be known. It is
explained that the uncertainty of field data cannot easily
be quantified based on the results of field
measurements alone. It is not just the accuracy of the
instrumentation used for field measurements that
defines the reliability of field data. In addition, the
repeatability of field measurements for similar boundary
conditions as well as the spatial representativeness of
individual measurement locations with respect to a
particular flow and dispersion problem must be
evaluated and quantified with respect to the measured
quantities before corresponding data can be used safely
for model validation purposes. This is why COST 732
suggests validation data sets that always comprise
combinations of field and laboratory experiments. The
background document closes with a glossary of terms
since words like validation, verification, evaluation,
quality assurance etc. are not unambiguously defined
and used.
The second document (model evaluation guidance
and protocol document) is a condensed version of the
background document. It gives step-by-step guidance to
model developers and users on how to assure the
quality of a micro-scale meteorological model. The final
guidance and protocol document will come along with
recommendations for data sets that should be used
during the validation work. These data sets will be made
accessible in a unified format via a www data bank. In
practise the quality of model output depends not only on
the accuracy of the model itself and the model input.
Likewise important is the qualification of the person
running a model. Numerical simulation is a knowledgebased activity. Appropriate knowledge can be
transferred to users by recommendations concerning
the proper use of models. For obstacle resolving CFD
codes such recommendations are not straightforward.
COST 732 tried to respond that problem by drafting a
third document, the
x

Best practice guideline for the CFD simulation of

flows in the urban environment (Franke et al.,
2007a)

The recommendations given in the set of COST

documents are presently tested by the action itself. The
Mock-Up Urban Setting (MUST) data set which
comprises field and wind tunnel experiments from flow
and dispersion experiments carried out within and above
an urban building array made up by 256 ship containers
was selected and brought into a usable form. 11 groups
of numerical modellers (9 CFD and 2 non-CFD) started
to simulate the MUST case thereby following the
evaluation guideline. At a first workshop that took place
in Hamburg in January 2007 the results were presented
and compared and the differences were discussed.
Different evaluation metrics were tested and
recommendations for fair comparisons were given. It
followed another meeting in February in Brussels that
was mainly used to draw conclusions from the MUST
exercise, to discuss the next steps and to make

25

appropriate changes in the three before-mentioned

documents.

Future Work
Europe-wide discussion of the quality assurance
procedure, the use of specific data sets and the
recommendations given in the Best Practise Guideline
will lead to a harmonised and accepted approach. A
quality assurance activity will be launched and the
community of model developers and users will be invited
to apply the procedure to their models and to prepare
model evaluation protocols based on selected data sets.
This will be combined with a model inter-comparison
exercise within which several model developers and
users will simulate identical cases. Ideally this exercise
should comprise test cases for which the solution is not
known beforehand (blind testing). The intent is not to
pillory models that perform badly or to rank the models
in one way or the other. That only blocks the flow of
information and obstructs scientific exchange. The
differences in model results should be discussed and
the reasons for deviant model results should be
investigated. The strengths and weaknesses of
particular modules, parameterisations or closure
schemes will be determined. It is expected that
modellers will take this opportunity to test various
modules, develop common views about the most
appropriate set-up of micro-scale meteorological models
and, thereby, the quality standard of micro-scale
meteorological models and their application will
significantly improve. This leads to the expectation
expressed in the Introduction that the 'culture' within
which urban air pollution models are developed and
applied will be significantly improved.

References
Britter, R., and Schatzmann, M. (Eds.) (2007a):
Background and justification document to support
the model evaluation guidance and protocol
document. COST Office Brussels, ISBN 3-00018312-4.
Britter, R., and Schatzmann, M. (Eds.) (2007b): Model
evaluation guidance and protocol document. COST
Office Brussels, ISBN 3-00-018312-4.
Borrego, C., and Tchepel, O. (1999): General
Requirements for a Quality Assurance Project Plan.
Proceedings, 3rd SATURN Workshop, Aveiro,
Portugal.
Franke, J., Hellsten, A., Schlnzen, H., and Carissimo,
B. (Eds.) (2007a): Best Practice Guideline for the
CFD simulation of flows in the urban environment.
COST Office Brussels, ISBN 3-00-018312-4.
Hall, R.C. (Ed.) (1997): Evaluation of modelling
uncertainty - CFD modelling of nearfield
atmospheric dispersion. EU Project EV5V-CT940531, Final Report. WS Atkins Consultants Ltd.,
Woodcote Grove, Ashley Road, Epsom, Surrey
KT18 5BW, UK.
Irwin, J.S. (1998): Statistical Evaluation of Atmospheric
Dispersion Models. Proceedings, 5th Int. Conf. on
Harmonization within Atmospheric Dispersion
Modelling for Regulatory Purposes, Rhodes,
Greece.
Irwin, J.S. (1999): Effects of Concentration Fluctuations
on
Statistical
Evaluations
of
Centreline
Concentration
Estimates
by
Atmospheric
Dispersion Models. Proceedings, 6th Int. Conf. on

Harmonization within Atmospheric Dispersion

Modelling for Regulatory Purposes, Rouen, France.
Ketzel, M., Louka, P., Sahm, P., Guilloteau, E., Sini,
J.F., and Moussiopoulos, N. (2001): Inter-Comparison of Numerical Urban Dispersion Models.
rd
Proceedings, 3 Int. Conf. on Urban Air Quality,
Loutraki, Greece.
Lohmeyer, A., Mller, W.J., and Bchlin, W. (2002): A
comparison of street canyon concentration
predictions by different modellers. Final results from
the Podbiexercise. Atmospheric Environment, 36,
pp. 157-158.
MEG (1994): Model Evaluation Group: 'Guideline for
Model Developers' and 'Model Evaluation Protocol'.
European Community, DG XII, Major Technological
Hazards Programme, Brussels, Belgium.
Olesen, H.R. (1999): Model Evaluation Kit Recent
th
Developments. Proceedings, 6 Int. Conf. on
Harmonization within Atmospheric Dispersion
Modelling for Regulatory Purposes, Rouen, France.
Popper, K. R. (1959): The Logic of Scientific Discovery.
Basic Books, New York.
Royal Meteorological Society (1995): Atmospheric
Dispersion
Modelling.
Guidelines
on
the
Justification of Choice and Use of Models, and the
Communication and Reporting of Results. RMS,
104 Oxford Road, Reading, RG1 7LJ, UK.
Schatzmann, M., and Leitl, B. (2002): Validation and
application of obstacle resolving urban dispersion
models. Atmospheric Environment, 36, pp. 48114821.
Schatzmann, M., Grawe, D., Leitl, B., and Mller, W.J.
(2003): Data from an urban street monitoring station
and its application in model validation. Proceedings,
26th International Technical Meeting on Air Pollution
Modelling and its Application. Istanbul, Turkey, May
26-30.
Schatzmann, M., and Britter, R. (Eds.) (2005):
Proceedings from the International Workshop on
Quality assurance of microscale meteorological
models. European Science Foundation, ISBN 3-00018312-4.
Verein Deutscher Ingenieure (2002) Richtlinie 3783,
Blatt 9, VDI Guideline on Environmental
Meteorology - Prognostic microscale windfield
models - Evaluation for flow around buildings and
obstacles. Beuth Verlag Berlin.

26

How comprehensive is comprehensive enough? - Model-specific

reference data for the validation of micro-scale LES flow and dispersion
models.
F Harms, B Leitl, M Schatzmann
Meteorological Institute,
University of Hamburg,
Hamburg, Germany
Email: Frank.Harms@zmaw.de

Abstract The validation of time-accurate Large

Eddy Simulation (LES) models require wellcharacterized datasets with information adequate
for defining and evaluating unsteady simulations.
One of the main objectives of the study presented is
to identify and apply appropriate validation
procedures for such urban flow and chemical
transport (CT) simulation models. Based on two
extensive wind tunnel measurement campaigns, a
first model-specific dataset for the validation of
LES-based flow and dispersion models was
compiled. In the paper an outline of the ongoing
research project is given and selected results are
presented and discussed.
Key words validation of LES models, wind tunnel
experiments, flow and dispersion measurements, puff
dispersion

Introduction
The increasing possibility of an accidental or
deliberate release of chemical/biological/radiological
pollutants has caused detailed simulations of releases
and dispersion in complex geometries to attain a much
higher importance. For example, a credible and
validated computational fluid dynamics (CFD) based
plume prediction model of CT in urban areas can be
used in the licensing of new industrial plants, in safety
analysis studies for accidental releases of hazardous
materials in the chemical industry, or in the context of
crisis management after terrorist attacks in urban
environments.
Numerical simulations of dispersion in an urban
environment are of complex nature and need a
comprehensive computational effort. Hence simpler
models as, e.g., analytical (Gaussian) models,
diagnostic models (which use only the mass
conservation equation) or CFD-models with full
parameterization of turbulence, i.e., Reynolds-averaged
Navier-Stokes (RANS) codes were typically used in the
past for these complex tasks. Such models can
compute urban dispersion within a reasonable time, but
they are unable to capture the inherently unsteady
plume dynamics driven by urban geometry. Direct
numerical simulations (DNS) are simulations in
computational fluid dynamics in which the NavierStokes equations are numerically solved without any
turbulence model. These models are able to compute
transient flow dynamics, but they are prohibitively
expensive for most practical flows at moderate-to-high
Reynolds numbers, and especially so for urban CT
studies. LES constitutes an effective intermediate
approach between DNS and the RANS methods

27

(Sagaut 2004). Large Eddy Simulation (LES) is capable

of simulating flow features that cannot be handled with
RANS such as significant flow unsteadiness and
localized vortex shedding, and provides higher accuracy
than the industrial methods at lower cost. LES solutions
converge to the solutions of the Navier-Stokes
equations as resolution is increased, whereas RANS
generally do not. Because the larger-scale unsteady
features of the flow govern the unsteady plume
dynamics in urban geometries, the LES approximation
can capture some key features which the RANS
methods
and
the
various
Gaussian
plume
methodologies cannot. Nowadays increasing computer
power enabled the possibility to use LES models for
urban dispersion simulations.
Establishing the credibility of urban CFD solutions
has been one of the stumbling blocks to their
widespread use. Code validation using experiments
requires well-characterized datasets with information
adequate for defining and evaluating unsteady
simulations. Unfortunately, current full-scale field
studies do not fully provide this information. The
number of trials is limited and the data acquired is
typically too sparse and/or insufficient to properly
characterize
the
flows.
Further,
experimental
concentration variability cannot be measured from
these data.
Validation data for numerical models are not just any
experimental data; they must fulfill certain requirements
with respect to completeness, spatial and temporal
resolution,
accuracy,
representativeness
and
documentation of the measured results (Leitl, 2000). If
these requirements are not met, too many degrees of
freedom remain to set-up unique numerical model runs.
A wide variety of numerical results can be generated
with reasonable assumptions for the input data, with the
consequence that a solid conclusion concerning the
model quality cannot be reached.
To overcome these problems, datasets that match
the complexity of specific groups of models are needed.
For the validation of Large Eddy Simulation models for
urban applications, data are needed that comprise flow
and turbulence fields in combination with concentration
fields measured with high resolution in space and time
within the urban boundary layer. Field trials are
essential for validation but do not yet provide complete
data. In addition, specific data are required that test the
particular parameterizations the LES-type of model
applies. Under certain limiting conditions, such datasets
can be generated under carefully controlled conditions
in well-equipped boundary layer wind tunnels.
The present study is a joint research project of the
Meteorological Institute of the University of Hamburg
and the Naval Research Institute. Major goals of this
study are:

In order to provide all information needed for a

realistic comparison of field data with model results, two
sets of supplementary measurements were carried out
within the scope of three independent study projects.
One specific set of experiments focused the sensitivity
of wind tunnel results with respect to changing
approach flow conditions (Herbst et al, 2007) in order to
collect information on how close field data, experimental
results from the wind tunnel and numerical results are
expected to match, when the approach flow conditions
are not exactly the same. Another important question
regarding the comparison of field data with results from
wind tunnel measurements or numerical modelling is
related to the temporal representativeness of flow and
dispersion measurements within a complex urban area
(Rix et al, 2007). A specifically designed set of
experiments was intended to quantify the 'truncation
error' inherent in short-time averaged wind velocity and
concentration measurements as they are collected in
the field.
A third independent study project was intended to
provide a software tool for a efficient and systematic
analysis of the tremendous amount of puff dispersion
data collected within the project (Fischer et al, 2007).
Investigating the statistical representativeness of puff
dispersion data from field and laboratory experiments is
again crucial with respect to a comparison of model
results with the corresponding field data.
The analysis of the extensive amount wind tunnel
measurements is still in progress. The complete
reference database will be available approximately in
autumn 2007. Selected wind tunnel results will be
presented below.

To develop validation strategies and statistically

based figures of merit for urban models including
acute release scenarios
To create a benchmark quality dataset for the
validation of time-resolved urban CFD models such
as FAST3D-CD
To create a description for data acquisition and
reduction to adequately characterize turbulent
inflow and wind variability for urban CT models

1 Brief outline of the wind tunnel

experiments
Aim of the wind tunnel measurements is to create a
high quality reference dataset which is adequate to
validate micro-scale LES models. Therefore numerous
flow and concentration measurements in two extended
six month wind tunnel campaigns were carried out. A
wind tunnel model of the central business district of
Oklahoma City (OKC) was constructed for these
measurements. This city was selected to create a wind
tunnel dataset which can be combined with field
measurements. In Oklahoma City, the 'Joint Urban
2003' field experiment took place, which is one of the
most extensive field campaigns in urban areas including
flow and concentration measurements took place
(Allwine at al, 2004).
During the first wind tunnel campaign flow
measurements with a high temporal and spatial
resolution were carried out within the modeled area. A
2D Laser Doppler Anemometry (LDA) system provided
flow data at sampling rates of several hundred Hertz (up
to more than 1 kHz under favorable conditions),
resolving even small-scale turbulence in space and
time. More than 2500 individual flow measurements at
different locations were done to provide a sufficiently
high spatial density of measured data within the model
area. The second objective of the first wind tunnel
campaign was to characterize the inflow conditions
upstream of the Oklahoma model as precise as
possible. Accordingly, the development of the boundary
layer flow modeled within the test section of the wind
tunnel was systematically measured and analyzed and
the model boundary layer was documented by
comprehensive measurements upwind of the model.
The second wind tunnel campaign was focusing on
a systematic analysis of the dispersion of acute
releases in urban environments. Typical puff dispersion
phenomena are characterized by an immense variability
of individual releases and the scatter inherent in puff
dispersion measurements is expected to be even larger
for releases within a complex urban structure. From
that, the general problem arises that the scatter and the
variability of puff results cannot be quantified reliably
based on the limited amount of individual releases
typically available from field tests. In order to quantify
the uncertainty inherent in field data and to compile a
statistically representative amount of dispersion data,
systematic wind tunnel tests were carried out. Several
particular Joint Urban 2003 release configurations were
replicated in the wind tunnel for fixed meteorological
boundary conditions until a sufficient statistical
representativeness was reached. In addition, puff
dispersion data were collected for several series of
systematic wind tunnel tests, for example with gradually
changing wind directions or different release locations
at fixed wind conditions.

2 Experimental setup
The wind tunnel measurements were carried out in
the large atmospheric boundary layer wind tunnel
facility (WOTAN) at the Meteorological Institute of
Hamburg University. The 26 m long facility provides an
18 m long test section equipped with two turn tables
and an adjustable ceiling. The cross section of the
tunnel measures 4 m in width and 2.75 to 3.25 m in
height, depending on the position of the adjustable
ceiling. For precise probe positioning, the test section of
the tunnel is equipped with a computer controlled
traverse system with a positioning accuracy of better
than 0.1 mm on all 3 axes. The 1:300 scaled model of
the central business district (CBD) of Oklahoma City
was covering an area of approximately 1.6 km x 1.6 km
around the city centre in full-scale. Figure 1 shows a
downwind view of the wind tunnel model mounted on
the second turn table in the test section.

28

Figure 1. Model of Oklahoma City in the boundary layer

wind tunnel of the University of Hamburg

3 Wind tunnel measurements

measurements within a test series. The second LDA

system was measuring two components of the wind
vector simultaneously at measurement locations
surrounding the reference point. Depending on the
probe orientation the u-v or u-w component were
recorded at more than 100 different locations per cross
flow plane. From the recorded time series of both
synchronized LDA systems, spatial correlation
coefficients were calculated. Figure 2 shows the
correlation of the measured u component of both LDA
systems. The small dots indicate the measurement
locations of the second LDA system while the fixed LDA
system was measuring the u component of the wind
vector near the centre of this plot. As expected the
correlation coefficient is approaching a value of 1 when
both LDA systems are measuring at the same location.
In addition, it was found that the correlation coefficient
is decreasing with increasing distance in an almost
perfectly symmetric pattern, documenting the welldesigned approach flow conditions. For distances larger
than 150 meters in full scale no significant correlations
were measured.

3.1 Boundary layer modeling

In order to achieve a boundary layer flow similar to

the atmosphere, a model-specific arrangement of
turbulence generators and floor roughness developed.
In an iterative process the shape and arrangement of
the turbulence spires and the floor roughness elements
was varied until a sufficient agreement of the modeled
boundary layer with the conditions observed at full scale
upwind of Oklahoma City was reached. The proper
'physical' scale of the modeled boundary layer was
evaluated based on a comparison of integral length
scales, turbulence spectra and wind direction
fluctuations with those measured at full scale. For each
modification step, the complete set of boundary layer
parameters was measured. It was found that a careful
adjustment of the modeled boundary layer enables
even large scale turbulent wind fluctuations up to a time
scale of 45 minutes to be replicated properly in the wind
tunnel for the given model scale.

3.2 Approach flow characterization

The parameterization of turbulent inflow conditions

for LES models requires an as accurate as possible
characterization of the boundary conditions at the inlet
of the computational domain. Hence, one important
goal of the wind tunnel campaign was to adequately
describe the turbulent approach flow conditions. In
addition to measurements at vertical and lateral profiles
upwind of the model area the development of the model
boundary layer was documented by component
resolving measurements. At 17 measurement positions
along the center line of the wind tunnel vertical wind
profiles were measured. Each vertical profile consisted
of 20 individual measurement positions between 10 m
and 200 m height above ground at full scale. At each
location, 4 minutes long time series of the individual
components u, v and w of the wind vector was recorded
with a sampling rate of at least 500 Hz. From those
systematic measurements, a total of 660 time series is
available in the database to describe in detail the
development of the modeled boundary layer flow along
the wind tunnel test section.
Within a cross-flow plane upwind of the turn table
systematic, component-resolved correlated flow data
were collected. For these measurements two
synchronized LDA systems were used. The first LDA
system was measuring the u component of the wind
vector at a fixed reference position for all correlation

29

Figure 2. U against U correlations of the wind vector

within a cross flow plane upstream of the model area
Figure 3 shows the correlation of the u component
measured by the reference LDA system with the w
component measured by the second LDA. The contour
plot shows a significantly weaker correlation of the u-w
correlations compared to the u-u correlations. For both
LDA systems measuring at the same location a highest
correlation coefficient of about -0.3 is documented.

Figure 3. U against W correlations of the wind vector

within a cross flow plane upstream of the model area
In Figure 4 the u component measured at the
reference location is correlated with the v component

measured by the second LDA system. The correlation

coefficients calculated vary at a relatively small level
between 0.15 and -0.15. In contrast to the results
presented above, for this configuration the highest
correlations are not reached when both LDA systems
measure at the same location. The contour plot shows
areas of maximum positive and negative correlations for
distances of about 100 meters in full scale between the
two LDA systems.

releases the sources were equipped with solenoid

valves. Using ethane as tracer and Fast Flame
Ionization Detectors (Fast FIDs) for concentration
measurements, the transient concentration signal
resulting from the individual puff releases was recorded
with sufficiently high spatial and temporal resolution at
different locations. At each measurement position a
continuous puff time series of 30 minutes wind tunnel
time was recorded. Each time series contains at least
200 individual releases and the corresponding
concentration signals of puffs respectively.
The
measurement height of 6.7 mm in the wind tunnel was
kept constant for all puff dispersion measurements,
corresponding to a sampling height of 2 m above
ground at full scale. Figure 5 shows a typical cutout of a
recorded time series signal for three consecutive puffs.
The green line indicates the release periods of the
source.
EW TL@ ZM AW .D E U niversity of H amburg

200

c [ppm]

Figure 4. U against V correlations of the wind vector

within a cross flow plane upstream of the model area

100

3.3 Urban flow measurements

50

The main objective of the flow measurements within

the model area was to create a database of
measurements with a high temporal and spatial
resolution. Moreover, the measurements were expected
to cover the entire central business district of Oklahoma
City. More than 2500 individual flow measurements
were needed to provide the sufficiently large, areacovering
set
of
component-resolved
wind
measurements needed for model validation. The
measurement locations were arranged in a more or less
regular grid with a higher resolution near the city centre
and wider distances between the measurement points
at the outer edges of the domain. The data was
collected in heights ranging from 6 meter above ground
up to 260 meter above ground in full scale. At each of
the individual locations, a three minutes long time
series of the u, v and w component of the wind vector
was recorded. Almost four months of continuous wind
tunnel measurements were needed to compile a
comprehensive reference data set of component
resolved flow measurements which satisfies the specific
requirements for validation data.

610

615

620

wt time [s]

625

trigger signal [V]

150

630

Figure 5. Typical segment taken from a puff time series

signal, showing three consecutive puffs under identical
'mean' approach flow conditions
Due to the enormous variability of puff dispersion
data, as documented in Figure 5, it becomes clear that
a relatively large number of individual releases are
necessary to derive statistically representative results.
The analysis of these large puff ensembles includes
significant puff dispersion parameters such as puff
arrival time, dosage, peak time, flushing time, maximum
concentration level, calculation of a 'mean puff' for a
given sampling point. Furthermore, puff probability
analysis such as what is the probability of a certain
concentration to be reached at a given location at a
certain time after the release and analysis of the
representativeness of a limited number of puff releases
is applied to all data collected.
First of all the repeatability of a 200 puff ensemble
was investigated. For this purpose, the 30 minutes long
time series were repetitively measured and
subsequently analyzed several times. Figure 6 shows a
typical result of the repeatability of the mean arrival time
calculated. The mean arrival time denotes the time a
puff needs to travel from the source to the
measurement location on average. The repeatability of
the arrival time was found to be in the range of +/- 1%
for the ensemble sizes realized in the wind tunnel.
Similar repeatability of the results could be documented
for the peak time (the time when the maximum
concentration of the puff is reached) and the flushing
time (the time after the puff is washed-out completely at
the measurement location). Calculated dosage values
were found to be repeatable within a range of +/- 4%.

3.4 Urban dispersion measurements

As mentioned before, the second measurement

campaign was focusing on acute (puff) releases in an
urban environment. The objective was to create a
dataset which is appropriate for the validation of
emergency
response
dispersion models. Puff
dispersion in an urban area is driven by a transient flow
field. Even under identical mean approach flow
conditions the variability of puff dispersion is expected
to be very large. In order to quantify the variability
inherent in puff release data, hundreds of individual
puffs were released under exactly the same mean
boundary conditions for each release configuration.
For the puff measurements six special emission
sources were mounted at different locations in the
model area. The selected locations coincide with source
locations used in the Joint Urban 2003 field campaign.
To control precisely the duration of the short term

30

EWTL@ ZMAW.DE University of Hamburg

300

ontime : wt 0.288s
ontime : wt 0.493s
ontime : wt 0.689s
ontime : wt 0.991s
ontime : wt 1.998s

250
200

150
100

1.5

50
1

0.5

0.5

1.5

wt time [s]

2.5

Figure 8. Release duration test: measured mean dosage

of 215 puffs divided by the selected release duration
0

run Nr.

Figure 6. Repetition test: measured mean arrival time of

215 puffs for six different measurements under the same
conditions
Another set of experiments was intended to prove
that puff dispersion results can be scaled and
transferred to different mean wind speeds, release flow
rates and release durations. As an example, in Figure
7 the result of a systematic source flow rate test is
given. Five series of puff measurements with different
source flow rates were carried out at one location in this
case.

Finally, Figure 9 shows the results of a Reynoldsnumber or wind speed test. The puff measurements
were repeated at different approach wind speeds while
all other experimental parameters were kept constant.
The reference wind speed was measured in a height of
80 meters (full scale) upstream of the model area. The
test documented that the results of puff measurements
are scaleable also regarding to the mean wind speed.
EWTL@ ZMAW.DE University of Hamburg

200
180

dosage / reference wind speed

arrival time [s]

EWTL@ZMAW.DE University of Hamburg

dosage/ontime

To ensure a representative evaluation of the

documented repeatability of the results, the repeatability
tests were carried out for different release locations and
for different measurements locations at different
distances from the source as well.

160
140
120

EWTL@ZMAW.DE University of Hamburg

100

flow rate : wt 16.35 l/h

flow rate : wt 47.96 l/h
flow rate : wt 79.57 l/h
flow rate : wt 111.14 l/h
flow rate : wt 142.81 l/h

1.4

dosage / flow rate

1.2
1

0.8

60
40
20
0

0.6
0.4

reference wind speed @ 80m full scale

Figure 9. Reynolds test: measured mean dosage of

215 puffs divided by the selected wind speed

0.2
0

80

50

100

flow rate [l/h]

150

Figure 7. Flow rate test: measured mean dosage of 215

puffs divided by used flow rate
As it becomes clear from Figure 7, the measured
mean dosage divided by the source flow rate is
reaching a constant value for flow rates higher than
50l/h. The test shows that puff measurements are
scaleable regarding to the flow rate. Only for the lowest
flow rate the measured mean dosage is deviating from
the constant value, most likely because of an increased
uncertainty in the flow rate measurements. However, all
regular puff measurements were carried out with a
source flow rate much higher than 50 liter per hour.
In Figure 8 selected results from a systematic
release
duration
variation
are
plotted.
Five
measurements with release duration between 0.288
seconds and 2 seconds wind tunnel time were realized.
The figure shows the measured mean dosage divided
by the release duration (on-time of the emission source)
and indicates that the puff measurements can be
scaled to different release durations.

31

After a careful evaluation of the experimental setup

based on the results of the preliminary tests
documented above, a comprehensive puff dispersion
measurement program was started. The dispersion of
clouds released from 6 different source locations and
for 6 different wind directions was measured at dozens
of different measurement locations for each
configuration. By the time this abstract is published, the
measurements are still ongoing and the campaign will
be finished in summer 2007.

4 Conclusions
The validation of LES based flow and dispersion
models requires well-defined data sets with information
adequate for defining and evaluating unsteady
simulations. For these datasets flow and dispersion
measurements with a sufficiently high resolution in time
and space are essential.
To meet the validation data requirements,
comprehensive wind tunnel measurements were carried
out in the large boundary wind tunnel (Wotan) of the
Meteorological Institute at the University of Hamburg.

The database created contains more than 4000 time

series of approach flow, correlation and flow
measurements within the model area. The dispersion
measurements are still ongoing and will be finished in
summer 2007.
After a careful evaluation of the quality of all
measured wind tunnel data, a first model- and
application-specific reference data set for the validation
of LES-based urban flow and dispersion models will be
published.

Acknowledgements
The results presented above are compiled in a joint
research project of the Navy Research Laboratory and
the University of Hamburg, sponsored by DTRA.

References
Sagaut, P. (2004) Large Eddy Simulation for
Incompressible Flows, 2nd Edition, Springer.
Allwine, K. J., M. J. Leach, L. W. Stockham, J. S. Shinn,
R. P. Hosker, J. F. Bowers and J. C. Pace, (2004):
Overview of Joint Urban 2003 an atmospheric
dispersion study in Oklahoma City, Symp. on
Planning, Nowcasting and Forecasting in the Urban
Zone, January 11-15, Seattle, WA. Amer. Meteor.
Soc.. J7.1.
Herbst, I.; Leitl, B.; Schatzmann, M. (2007) How close is
close enough Sensitivity of wind tunnel results
with respect to changing approach flow conditions",
Physmod 2007, ibid.
Rix, M.; Schatzmann, M; Leitl, B. (2007) "How long is
long enough? - Quantifying the temporal
representativeness of flow and dispersion
measurements in a complex urban area", Physmod
2007, ibid.
Leitl, B. 2000 Validation Data for Microscale Dispersion
Modeling, EUROTRAC Newsletter, 22, 28-32

32

How dense is dense enough? Systematic evaluation of the spatial

representativeness of flow measurements in urban areas
D. Repschies, M. Schatzmann, B. Leitl
Meteorological Institute,
University of Hamburg,
Germany
Denise.Repschies@zmaw.de

Abstract For estimating the spatial representativeness of locally measured wind velocities in an
urban roughness structure, wind tunnel experiments were carried out in a dense array of idealized
obstacles. In a subunit of the cube field horizontal
and vertical planes were selected. At a large number
of points located in these planes time series of the
wind velocity vector were measured and subsequently analyzed. For several parameters characterizing the flow areal averages based on all values in
a plane were determined. Subsequently the number
of data points included in the averaging procedure
was systematically reduced by taking only every
second, third (and so on) data point into account.
The deviations were determined and used to define
a metric which quantifies the spatial representativeness of measurements based on a limited number
of data points only.
Key words wind tunnel experiments, cube fields, velocity time series, measurement of planes, reduction in
number of measuring points, spatial representativeness

Introduction
Pollutant transport within the urban canopy layer is
strongly affected by the individual geometrical structure
of the building array. Both mean velocity and turbulence
vary from point to point, local differences can be large. A
full documentation of the flow field within a city would
require measurements taken at numerous positions. To
obtain a complete picture of the flow field by field measurements is more or less impracticable and could not
be afforded. In reality, measurements are taken at selected points only, and the question arises how representative point-wise measurements are to characterize
the flow field within urban quarters or individual streets.
To obtain such an estimate wind tunnel experiments
were carried out in Hamburg Universitys large boundary

layer wind tunnel WOTAN (figure 1). For two idealized

urban geometries measurements were carried out in a
very dense spatial grid and a strategy was developed to
quantify the local variability of mean and turbulent flow
properties.

1 Methodology
The two obstacle arrays which were used in our investigations are shown in figures 2 and 3. Clearly, these
are not yet really urban geometries. Main advantage of
these idealized structures is, however, that measurements taken in a small subunit of the two cube arrays
can be expected to be representative for other similarly
chosen subunits as well and thus for the complete array.
To create the needed data base, measurements with
high spatial resolution in horizontal and vertical planes
were carried out. The horizontal planes were situated at
different heights and the vertical planes were chosen at
different horizontal positions within a subunit to get information about the local variability.
For each of the planes the areal average for a variety of flow properties (mean wind velocity components
u , v , w , the standard deviations u rms , v rms , w rms and the
mean turbulent fluxes u' v' , u' w' ) was calculated based
on data from all positions in the plane at which measurements were taken. Subsequently the ensemble of
data points incorporated in the averaging process was
reduced by systematically increasing the distance between points from which data were taken into account.
To obtain a quantitative indicator for the spatial (or more
exact: areal) variability of a flow property a metric was
defined (chapter 3) which, in a specific way, compares
the average over all data points with that over a subset
of points only.
In order to smooth the results and since also in a
laboratory experiment the number of measurements
which can be performed is limited, a plot software (Table

Curve 3D ) was
utilized
to
interpolate
the
effectively
measured
data
points (figure 4) to
an
even
larger
1
number of points
(figure 5) before the
variability indicator
0

2,625m

y/h

y/h [-]

14.625m

umean [m/s]
2.6
2.2
1.8
1.4
1
0.6
0.2
-0.2
-0.6
-1

-1

-2

x/h

Figure 1: Cube field (configuration 1) in the Environmental Wind Tunnel Laboratory (EWTL) with the used
-3
coordinate system and the field of view.
-2
-1
0
max. windspeed 20 m/s

field of view

33

blower (130kW) 1
x/h [-]

was determined. In these figures the values of the wind

velocity component u are displayed as a triangulated
colored area.
The influence of different canopy layer-structures on
the flow was investigated by varying the cube shapes,
whereas the distance between the cubes was kept constant. In a first series of experiments only the cubes
were located in the tunnel (figure 2). In the second setup roofs of different shape and orientation (flat roofs,
pitched roofs in flow direction and pitched roofs normal
to flow direction) were added to the cubes (figure 3).

found to be well below the threshold according

to VDI 3783/12 (1999).
The longitudinally pressure gradient was minimized with the help of the adjustable ceiling of
the tunnel.
The properties of the boundary layer upstream
from the intensive observation area were determined (for details see: Schultz, 2007).
Reynolds number independency tests were carried out by replicating some of the measurements with different wind velocities. The minimum wind speed used throughout the measurements was 6 m/s and secured a Reynolds
number well above the threshold.
Reproducibility tests were carried out throughout
the measurement campaign (for details see:
Repschies, 2006).

Figure 4: Horizontal plane with grid points at which

measurements were made ( top view).
Figure 5: Horizontal plane with grid points over which
the data were interpolated (top view).
Figure 2: Cube field without roofs in the boundary layer
wind tunnel. The vortex generators at the
1

y/h [-]

umean [m/s]
2.6
2.2
1.8
1.4
1
0.6
0.2
-0.2
-0.6
-1

-1

-2

-3
-2

-1

x/h [-]

tunnel entrance can be seen .

Figure 3: Cube field with roofs.

To guarantee the quality of measured data, the following standard investigations and tests were carried
out before data production really started:
The blockage of the wind tunnel cross section
due to the cube array was determined and

34

2 Experimental Setup
The idealized obstacle area was accomplished by
cubes of the length h = 125mm which were placed at a
distance of 1h to each other (for configuration 2 the
cube height was 1h as well but roofs were mounted on
top of some of the cubes which increased the total
height of those obstacles to 1.5 h). The whole cube field
consisted of 59 rows with 11 cubes per row. The first 39
rows served to generate a turbulent boundary layer,
whereas the last 12 rows were installed to prohibit any
disturbing influences from the downstream end of the
array.
Figure 2 shows the cube field of configuration 1,
whereas in figure 3 the cube field with added roofs is
displayed.
The area in which the intensive observations were
carried out is in the center of the block cube array in
Figure 2 (rows 46 to 48). The exact position of the measurement planes may be taken from figures 6 to 9.
Information on the distribution of roof shape patterns is
also given in these figures, the symbols are explained:
Figure 10.
The coordinate system (Cartesian) consistently used
throughout the study is as follows: The x-axis shows into
the direction of the wind, y is the lateral and z the vertical coordinate.
Using a 2d-Laser-Doppler-Anemometer, time series
of the velocity were recorded in horizontal planes at the
heights of 0.25h, 1h, 1.5h for configuration 1 and 0.64h,
1h, 1.5h and 2h for configuration 2, respectively (h =
cube height without roofs). The shapes of the measurement areas differ for the two configurations as can be
seen in figures 6 and 7 (darkened areas).
The vertical planes were positioned either at the
edges or in the center between two cubes and oriented
either in the x-z or y-z direction.

y/h
0

-1

-2

-1

x/h

Figure 7: Position of vertical planes in configuration 1.

y/h
0

-1

-2

-3

-2

-1

x/h

Fig. 8: Position of the vertical plane in configuration 2.

y/h
without roof

pitched roof normal

to flow direction

pitched roof in
flow direction

flat roof

Figure 9: Different roof shapes

x/h

The whole experimental program comprised measurements in 18 planes at altogether 2080 individual locations.

3 Data handling

Figure 6: Shaded area shows the sub-unit where the

measurements were made (configuration 1).

Due to experimental restraints the points at which

measurements were taken were not always evenly distributed over the measurement planes. To smooth gaps
out and to generate an array of values with uniformly
0.04h spacing (= 5mm) the real measurements were
homogenized by a plot software. Since the density of
points with real measurements was high (point distance
height-dependent: between 12.5 and 50mm), the danger
of producing spurious data was very low. All subsequent
analysees are based on processed data resulting from
the plot software.
If all data points of a given plane are included in the
averaging procedure, a reference mean value u r is calculated. In the next step, starting with the first data point
again, every second point in the x/h- and y/h- direction
(for horizontal planes) is neglected (velocity index: a for
area), subsequently every third point is taken out and so
on. The program used to determine the averages cycled

y/h

x/h

Figure 7: Shaded area shows the sub-unit where the

measurements were made (configuration 2)

35

Since the starting point for the analysis is chosen

more or less arbitrarily, the question arose how sensitive
the results may depend on the choice of a particular
starting point. If the second run starts with the second
data point, every second point in x/h- and y/h-direction is
left out again, but the attention is on other points now,
compared to the above mentioned case. The mean is
calculated from the other half of available points, thus it
is slightly different from the value calculated before. As a
consequence there are two mean values for the pointdistance of 0.08h.
The more the number of points is reduced and the
distance between data points increased the more possibilities exist to calculate areal averages for groups of
data points with identical spacing. Thus, the larger the
distance between data points considered, the more dots
exist for the same point-distance. This is shown again at
the example of u in figure 12:

20 times, thereby producing 20 averages u a based on

reduced numbers of data points with a maximum pointto-point-distance of 0.8h.
Figure 11 shows the results for the mean velocity u
versus the concerning point-distance for the lowest
measured plane of 0.25h in configuration 1. The absolute value of the percental deviation of an area mean u a
from the reference mean u r is given for every pointdistance in 0.04h-steps:
|delta u mean |=

100(u r  u a )
for u r > u a
| ur |

|delta u mean |=

100( u a  u r )
for u r < u a . (2)
| ur |

(1)

The dots are colored according to the number of points

included in the averaging process.

60 number of points

60 number of points

40

30

20

1060
250
120
100
70
60
30
24
20
14
12
10
8
6
5
4
3
2

|delta umean| [%]

|delta umean| [%]

50

50

40

30

20

1060
250
120
100
60
40
30
24
20
14
12
10
8
6
5
4
3
2

10
10

0
0
0

0.2

0.4

0.6

point-distance [h]

0.2

0.4

0.6

0.8

point-distance [h]
Figure 11: Results of a complete analysis of data points
at the example of u .

0.8

Figure 10: Areal averages acc. to Equ.(1) or (2) as a

function of data spacing for the mean
longitudinal velocity component u .

4 Results
Figure 12 showed that the area average for a selected flow property may be different for the same data
point density but a data point selection from a different
raster. To elucidate this fact even more, the total area of
the horizontal planes according to figure 6 is subdivided
into two sub-domains located in different parts of the
canopy (along-canyon and cross-canyon areas in figure
13).

Figure 12: Fragmentation of the horizontal measuring

area in two parts: along-canyon (left) and
cross-canyoun (right).
The question arises whether this result holds for the
horizontal mean wind velocity u only or for other flow
properties as well. When the vertical wind component

36

1800 number of points

1700
1200
1600
750
350
1500
250
1400
120
100
1300
80
60
1200
40
1100
30
24
1000
14
12
900
10
800
8
6
700
5
4
600
3
500
2
400
300
200
100
0
0
0.2
0.4

w is considered, the plots need to be re-scaled since

the plane averaged reference mean value for w is typi-

1800 number of points

1700
1200
1600
750
350
1500
250
1400
120
100
1300
80
60
1200
40
1100
30
24
1000
14
12
900
10
800
8
6
700
5
4
600
3
500
2
400
300
200
100
0
0
0.2
0.4

0.6

point-distance [h]

0.8

0.6

point-distance [h]

|delta wmean| [%]

|delta wmean| [%]

1800 number of points

1700
1200
1600
750
350
1500
250
1400
120
100
1300
80
60
1200
40
1100
30
24
1000
14
12
900
10
800
8
6
700
5
4
600
3
500
2
400
300
200
100
0
0
0.2
0.4

|delta wmean| [%]

cally a small number (figures 14 and 15). It becomes

evident that for w the spatial variability is highest in
planes at roof level z/h =1 (figure 14: right) whereas for
u the maximum was obtained at lower elevations.
Above the roofs (and configuration 1) the flow quickly
gains uniformity and |delta w mean | tends towards zero.

0.8

Figure 14: Results for the plane at 1.5h in the area

cross-canyon for the wind velocity w
(configuration 1).

0.6

point-distance [h]

0.8

Figure 13: Results for the plane at 0.25h (left) and 1h

(right) in the area cross-canyon for the wind
velocity w (configuration 1).

For the comparison between the areas along-canyon

and cross-canyon a different behavior of spatial variability metrics occurs particularly within the canopy layer.
The differences decrease with increasing height of the
plane and disappear above the roof level, except for
some exceptional cases.

60 number of points

40

30

20

50

|delta umean| [%]

50

|delta umean| [%]

When this concept is applied to our example, the

mean horizontal wind velocity u for the plane at z/h =
0.25 in configuration 1, figure 16 is obtained (crosscanyon case: left, along-canyon case: right). As expected, the figure shows a much larger increase of the
spatial variability metrics (|delta u mean |) with decreasing
data density for the cross-canyon case. The reason is
certainly the flow pattern which in the wake of a building
is more complex than in a street-parallel flow.
This explanation is supported by the results from
horizontal planes with higher elevation (figure 17: z=1h
and figure 18: z=1.5h). For the regular cube array (configuration 1) and z/h = 1.5 the flow is already so uniform
that in both sub-domains (along-canyon or crosscanyon) the metric is about zero. This means that u
wherever measured within this plane is representative
for the whole plane.

60 number of points

1060
250
120
100
60
40
30
24
20
14
12
10
8
6
5
4
3
2

30

20

10

10

0
0

0.2

0.4

0.6

point-distance [h]

0
0

0.8

0.2

0.4

0.6

point-distance [h]

0.8

Figure 15: Results for the plane at 0.25h in the area

cross-canyon (left) and along-canyon (right)
for the wind velocity u (configuration 1).
60 number of points

40

30

20

60 number of points

1060
250
120
100
60
40
30
24
20
14
12
10
8
6
5
4
3
2

50

|delta umean| [%]

|delta umean| [%]

50

0
0

40

30

20

1060
250
120
100
60
40
30
24
20
14
12
10
8
6
5
4
3
2

10

10

37

40

1060
250
120
100
60
40
30
24
20
14
12
10
8
6
5
4
3
2

0.2

0.4

0.6

point-distance [h]

0.8

0
0

0.2

0.4

0.6

point-distance [h]

0.8

Figure 16: Results for the plane at 1h in the area crosscanyon (left) and along-canyon (right) for
the wind velocity u (configuration 1).

30

20

10

40

30

20

0.2

0.4

0.6

point-distance [h]

0
0

0.8

0.2

0.4

0.6

point-distance [h]

40

30

20

50

|delta umean| [%]

|delta umean| [%]

50

10

30

20

30

20

10

0.2

0.4

0.6

point-distance [h]

0.8

0
0

0.2

0.4

0.6

point-distance [h]

0.8

Finally the variability of flow properties in vertical

planes is considered, again at the example of u and
configuration 1. The two planes which will be compared
with each other are shown in figure 21. The plane at
y/h=0 is in the wake, the plane at y/h=0.5 at the edge of
the cubes. Only measurement points below roof level
(z/h< 1) were taken into account.

1500
1060
250
120
100
60
40
30
24
20
14
12
10
8
6
5
4
3

10

0
0

40

40

1500
1060
250
120
100
60
40
30
24
20
14
12
10
8
6
5
4
3

Figure 19: Results for the plane at 1.5h in configuration

1 (left) and configuration 2 (right) for the wind
velocity u (cross-canyon).

0.8

60 number of points

1060
250
120
100
60
40
30
24
20
14
12
10
8
6
5
4
3
2

20

0
0

Figure 17: Results for the plane at 1.5h in the area

cross-canyon (left) and along-canyon (right)
for the wind velocity u (configuration 1).
60 number of points

30

50

10

10

0
0

40

60 number of points

1060
250
120
100
60
40
30
24
20
14
12
10
8
6
5
4
3
2

|delta umean| [%]

40

1060
250
120
100
60
40
30
24
20
14
12
10
8
6
5
4
3
2

50

|delta umean| [%]

|delta umean| [%]

60 number of points

1060
250
120
100
60
40
30
24
20
14
12
10
8
6
5
4
3
2

50

50

|delta umean| [%]

60 number of points

60 number of points

Figure 20: Positions of the two vertical planes in

configuration 1.

0.2

0.4

0.6

point-distance [h]

0.8

0.2

0.4

0.6

point-distance [h]

0.8

As figure 22 reveals, the plane in the centre of the

two cubes exhibits much more local variability as the
plane at the cube edge. This statement holds as well for
all other flow properties. The exception is the mean vertical velocity component w only, where observations
are reversed.

Figure 18: Results for the plane at 1h in configuration 1

(left) and configuration 2 (right) for the wind
velocity u (cross-canyon).

5 Conclusions

Comparing now results from configuration 1 and 2 with

each other one anticipates that the local variability
should be more pronounced for the latter case with its
mixture of roof shapes. As figures 19 and 20 show, this
expectation is fully corroborated. Since the horizontal
planes selected were somewhat larger for configuration
2 (see figure 7), the legend defining the number of data
points considered is of different scale.

An extensive study has been carried out to investigate the local variability of flow properties within and
above the urban canopy layer. From the numerous results obtained (Repschies, 2006) only a small selection
could be presented here.

38

300 number of points

200

150

250

|delta umean| [%]

|delta umean| [%]

250

300 number of points

500
375
100
80
40
30
20
15
12
8
6
3
2
1

100

50

0
0

200

150

500
375
100
80
40
30
20
15
12
8
6
3
2
1

100

50

0.2

0.4

0.6

point-distance [h]

0.8

0
0

0.2

0.4

0.6

0.8

point-distance [h]

Figure 21: Results for the vertical plane at y/h=0 (left)

and at y/h=0.5 (right) for the wind velocity u
within the canopy-layer.
The major finding is that the local variability is largest
in areas where the flow is heavily disturbed (wakes and
other flow separation zones). This statement is in line
with expectations and appears to be trivial. However, it
is first time that the degree of variability can now exactly
be quantified, at least for the configurations investigated.
A metric has been developed which allows the characterization of the local variability for the different flow
properties in a unified way. Starting from a reference
case with real good spatial resolution (0.04h) it can be
seen how the variability index for a selected plane increases when the spatial resolution successively decreases to 0.8h. As table 1 shows, the percental increase is different for individual flow properties, most
affected are the turbulent momentum fluxes.
Horizontal planes:

u rms

(30%)

v rms

(30%)

w rms

<u
(50%
)

< u' w'

(180%)

<v
(300%)

<w
(1700%)

< u' v'

(18700%)

(30%)

Vertical planes:

u rms

(30%)
v rms
(30%)

w rms

<u
(160%)

< u' w'

(240%)

<w
(490%)

< u' v'

(1500%)
<v
(1500%)

(30%)

Table 1: Maximum values of the variability index

according to equations 1 and 2 for horizontal
and vertical planes. Shown are the percental
increases for the variety of flow properties
investigated with respect to a resolution
decrease from 0.04h to 0.8h.

39

The results of this study are of practical relevance.

The European Air Quality Guideline 96/62/EU and its
daughter directives assume that a monitoring station in
an urban setting would be representative for an area of
about 200m surrounding the station. Although air quality parameters (concentrations) were not subject of our
investigations, it appears that this assumption is not well
founded. The scale of our experiments is between 1:100
and 1:250 which means that the 200m-assumption corresponds to a resolution of only about 0.8h.
With respect to numerical modeling of flow and dispersion in urban areas it is common practice to resolve
a street canyon by 5 grid cells in both directions of the
cross section. Transformed into our idealized urban area
this would correspond to a spatial resolution of 0.2h
which, at least for the more sensitive flow properties,
seems to be insufficient.

References
Repschies, D. (2006) "Untersuchungen zur Reprsentativitt lokaler Strmungsmessungen in einer
idealisierten
urbanen
Rauigkeitsstruktur"
(in
german), Univer-sity of Hamburg, Meteorological
Institute
Richtlinie 96/62/EG des Rates der EU vom 27.09.1996
ber die Beurteilung und Kontrolle der Luftqualitt.
Amtsblatt L296 vom 21.11.1996 (in german)
Schultz, M. (2006):
Fortschrittsbericht DFG 04-174: Systematische
Untersuchung der Grenzschichtentwicklung ber
Stdten, Hamburg (in german)
Schultz, M. (2007):
Systematic Investigation on the Urban Boundary
Layer project DFG 04-174, PhD-Thesis (under
preparation), Hamburg (in german)
VDI Richtlinie 3783, Blatt 12 (1999):
Umweltmeteorologie. Physikalische Modellierung
von Strmungs- und Ausbreitungsvorgngen in der
atmosphrischen Grenzschicht. VDI Handbuch
Reinhaltung Luft, Band 1b, Vorentwurf August 1999
(in german)
VDI Richtlinie 3783, Blatt 9 (2005):
Umweltmeteorologie. Prognostische mikroskalige
Windfeldmodelle - Evaluierung fr Gebude- und
Hindernisumstrmung. Available in English as VDI
Guideline on Environmental meteorology Prognostic microscale windfield models - Evaluation
for flow around buildings and obstacles.

How close is close enough? Sensitivity of wind tunnel results with

respect to changing approach flow conditions.
I Herbst, B Leitl, M Schatzmann

Environmental Wind Tunnel Laboratory, Meteorological Institute,

University of Hamburg, ZMAW,
Hamburg, Germany
ilona.herbst@zmaw.de
Abstract Modeling micro-scale atmospheric flow
and dispersion phenomena in a wind tunnel
requires the lower part of the atmopsheric boundary
layer to be replicated at a particular model scale.
However, in most cases the model boundary layer
flow will not be a precisely scaled copy of reality
and the possible differences between model and
reality will introduce an uncertainty when model
results and corresponding full-scale data are
compared. Based on a systematic wind tunnel
study, the sensitivity of wind tunnel results with
respect to changing approach flow conditions was
quantified. The paper is documenting how accurate
approach flow conditions should be modeled in a
wind tunnel in order to achieve a certain confidence
interval for flow measurements within a complex
urban geometry.
Key words boundary layer modelling, sensitivity of
model results to inflow conditions, urban canopy layer,
Oklahoma model

Introduction
Reliable modeling of flow and dispersion processes
within complex urban structures is one of the focal
points of environmental research. Besides expensive
and time consuming field measurements, physical and
numerical models are applied to investigate transient
dispersion phenomena of locally released pollutants in
urban areas.
From a mathematical point of view, flow and
dispersion processes in the atmospheric boundary layer
can be treated as a boundary value and initial value
problem.
Concerning the modeling of flow and
dispersion phenomena this means, that the model
results will certainly depend on the boundary conditions
applied. Most often, it is an open question, how sensitive
results from a particular boundary layer wind tunnel
experiment are with respect to a change in the modeled
approach flow conditions and how close model results
can represent the conditions expected at full scale.
Moreover, it is often impossible to define physically
precise enough the boundary conditions to be modeled.
Even in ideal situations, when mean approach flow
conditions can be derived from representative field data,
an uncertainty is introduced by averaging over physically
different situations within a certain class of
meteorological conditions and further uncertainty is
caused by the technical realization of a certain approach
flow condition in a wind tunnel.
Depending on the type of model investigated and
depending on the complexity of the model geometry, the
uncertainty inherent in the modeled boundary layer will
affect the reliability and representativeness of the model
results to a certain degree. Often it is argued, that

41

within a complex urban structure, the locally measured

flow and dispersion patterns are less sensitive to
changes in the outer boundary conditions because the
structure itself creates its own 'proper' wind field as long
as the model area is large enough. Since there is no
reliable experimental evidence for this claim, a
systematic study was carried out in the large boundary
layer wind tunnel facility "Wotan" at the Environmental
Wind Tunnel Laboratory EWTL. The goal of the
experiments was to analyze qualitatively and
quantitatively the sensitivity of model results for a very
complex urban structure.
In a second step this
information is intended to be used to evaluate the quality
of wind tunnel results based on a comparison with
corresponding field data.
The effect of different profile shapes of the mean
approach flow as well as the effect of varying turbulent
properties on the flow field measured within the modeled
area was investigated using a scaled model of the
central business district of Oklahoma City (OKC). The
model scale chosen was 1:300 and the entire model
was covering an area of approximately 1.6 km by 1.6 km
at full scale. Since there is no significant terrain
structure within the modeled area, topography was not
included in the wind tunnel model.
In a first series of measurements a boundary layer
was used which closely matched the conditions found
upwind of OKC during the field campaign of Joint Urban
2003 (Allwine, et al, 2004). Vertical profiles of mean
and turbulent properties were measured at several
locations within and above the business district.
Subsequently the measurements were repeated at
exactly the same locations with modified (false)
approach flow boundary conditions. The results were
compared and the sensitivity of the urban flow field to
varying approach flow conditions was quantified.

1 Boundary layer characterisation

For the given case, the 'proper approach flow
conditions' were defined based on an analysis of field
data available from the JU2003 experiment.
For
prevailing winds from the south, a number of field trails
were found for which near-neutral stratification in the
lower atmospheric boundary layer could be assumed.
The wind profile measurements from the field were
compared with the wind tunnel boundary layer prepared
for the OKC model in a previous project (Leitl et al, 2003
and 2005). Good agreement was found between the
approach flow conditions measured in the field and
modeled laboratory conditions. In the course of the
evaluation of the modeled boundary layer flows not only
the mean wind flow but also turbulence characteristics
and momentum flux measurements were compared
carefully.

In addition to the OKC approach flow conditions 5

different approach flow conditions were modeled by
means of different experimental setups upwind of the
model and with modifications of the tunnel cross section.
Figure 1 shows the vertical profiles of the timeaveraged mean velocity for all modelled boundary layers
measured upstream of the model area. The profiles are
scaled with reference data taken in a height of 220 m
above ground in full scale to detect deviations from the
reference approach flow within as well as above roof
level. The error bar attached to each individual
measurement point is documenting the overall
uncertainty of the measured data, calculated from
repetitive measurements.
x In order to analyse the influence of the pressure
gradient along the test section on the model results,
the variable wind tunnel ceiling was adjusted,
enabling an accelerated and decelerated boundary
layer to be modeled for nearly the same mean wind
conditions. The ceiling adjustment is mainly
affecting the modeled constant shear layer. One
configuration were measurements with a constant
size of the wind tunnel cross section. As is
becomes clear from Figure 1, the shape of the
mean wind velocity profiles hardly differs from the
reference approach flow.
x Using a setup with modified turbulence generators
and a different roughness upwind of the model, a
more rough approach flow condition was modeled.
A more round shaped mean wind profile is
achieved.

The modeled approach flow conditions clearly belong to

different roughness classes. According to the VDIguideline 3783/12 (2000), the OKC reference boundary
layer and the accelerated/decelerated configuration can
be classified as rough approach flow conditions. The
roughness length and power law exponent of the higher
turbulent boundary layer show values typical for flow
above a very rough surface. For the configuration
without boundary layer simulation, roughness length and
power law exponent indicate conditions typical for a wind
flow above a moderately rough surface when the
overall height of the model boundary layer is considered
properly.

reference BL
accelerated BL
decelerated BL
const. cross section
higher turbulence BL
no BL simulation

Zo [m]

0.20
0.35
0.43
0.31
1.53
0.03

0.18
0.19
0.20
0.18
0.26
0.13

Table 1. Roughness length Zo and power law exponent

for the different boundary layer
The turbulence properties of the approach flow are
more significantly changed by varying the approach flow
conditions than the mean wind profiles. Exemplarily, in
Figure 2 the vertical turbulent momentum fluxes are
plotted for the different test configurations.

x As a extreme case, no artificially thickened

boundary layer was modeled at all.
This
configuration represents a kind of an 'academic
case' where a relatively thin boundary layer
generated above the wind tunnel floor is
approaching the model. Accordingly, the wind
velocity doesnt increase at elevations above 60 m
height in full scale.

EWTL@ ZMAW.DE University of Hamburg

300

250

accelerated BL
decelerated BL
constant cross section
higher turbulence BL
no BL simulation
reference BL

Z fs [m]

200

150

100
EWTL@ZMAW.DE University of Hamburg

300

50

accelerated BL
decelerated BL
constant cross section
higher turbulence BL
no BL simulation
reference BL

250

0
-0.008

-0.004

U'W'mean/U ref [-]

-0.002

Figure 2. Profiles of the vertical turbulent momentum

fluxes for the different boundary layer

200

Zfs [m]

As expected, for the more round shaped wind profile

significantly higher vertical momentum fluxes were
measured with an average increase of the fluxes of
about 80 % compared to the reference case. The
decelerated approach flow generates approximately 30
% higher vertical fluxes than measured for the reference
case, whereas for the accelerated flow leads to slightly
lower momentum fluxes. Without boundary layer
modeling the vertical momentum fluxes are nearly zero
for heights above 80 m.

150

100

50

-0.006

0.2

0.4

0.6

U mean/U ref [-]

0.8

1.2

Figure 1. Vertical profiles of the averaged mean wind

velocity for the different boundary layer
The corresponding roughness length and power law
exponent of each boundary layer are listed in Table 1.

2 Results
In the subsequent part of the paper, selected results
for one exemplary measurement location are discussed.
In Figure 3, the measurement location in the center of a
narrow street canyon (Park Av.) is marked with the big
circle. The average building height at the measurement

42

EWTL@ ZMAW.DE University of Hamburg

300

accelerated BL
decelerated BL
constant cross section
higher turbulence BL
no BL simulation
reference BL

250

200

Zfs [m]

site is about 50 m and during the tests, the street

canyon was oriented perpendicular to the wind.
All measurement results are provided in a nondimensional form. For scaling, the reference wind
speed measured at 220 m height above ground was
used. In contrast to common scaling, in whereat often a
reference wind speed measured at roof height is
chosen, the approach used here enables the local
deviations from a reference case within the urban
roughness to be visualized more clearly.

150

100

50

0.2

0.4

0.6

0.8

U mean/U ref [-]

1.2

Figure 4. Vertical profiles of Umean in Park Avenue

2.1 Comparison of mean wind profiles

First, the mean wind profiles are compared with Uand W-component investigated seperately. The grey bar
attached to the vertical axis of the plots is indicating the
average building height around the measurement
location.
In general it can be stated that the location in Park
Avenue shows the lowest deviations from the reference
case of all measurement locations for the U-component
of the wind vector because of the high buildings
surrounding the measurement area (Figure 4).
For an accelerated/decelerated boundary layer flow
approaching the model, the difference in the measured
mean wind profile is in the order of 5% on average.
However, the highest deviations can be found right
above roof level, where differences in the mean wind
speed would have a major impact on pollutant transport
and dispersion. The wind profiles of the higher turbulent
flow and the flow without an explicitly modeled boundary
layer flow document a slightly higher similarity at this
location but local differences reach values of up to 12 %.
As documented in Figure 5, the vertical wind
component (W) is more sensitive with respect to
changes in the approach fllow than the horizontal
components. Nevertheless, the range of the deviations
from the reference case is similar to the deviations
observed in the approach flow and hardly influenced by
the building structure in this case. From Figure 5 it
becomes clear that the deviations from the reference
case are nearly the same for heights above 80 m for all
variations of the approaching boundary layer flow.

43

EWTL@ ZMAW.DE University of Hamburg

300

250

200

Zfs [m]

Figure 3. Measurement locations, discussed location

marked with big circle

A similarity in the shape of the vertical profiles can

be stated. The biggest shift of the profiles is, however,
observed for the approach flow with higher turbulence
and for the case without boundary layer simulation.
Surprisingly, the case without explicit boundary layer
modeling shows a better agreement with the conditions
of the reference case. The biggest differences are
documented for the higher turbulence boundary layer
flow, were locally measured values deviating up to a
factor of 3. From a modelers point of view this indicates
that the wrong approach flow conditions have a
significant impact on the quality of flow measurements
even within a very complex urban structure.

150

accelerated BL
decelerated BL
constant cross section
higher turbulence BL
no BL simulation
reference BL

100

50

-0.06

-0.03

0.03

W mean/U ref [-]

0.06

0.09

Figure 5. Vertical profiles of Wmean in Park Avenue

2.2 Comparison of turbulent properties

So far, only the averaged mean wind profiles were

discussed, but a complete evaluation requires
comparing turbulent properties as well. In general,
turbulent momentum fluxes are found to be more

sensitive with respect to changing boundary conditions.

On the other hand, momentum fluxes document
transient turbulent phenomena within the flow and they
are expected to have strong impact on dispersion within
the model area.
The vertical profiles of the horizontal momentum
fluxes measured for different approach flow conditions
are plotted in Figure 6. The profiles have nearly the
same shape and all profiles converge to the reference
profile above a height of 150 m (Figure 6).

EWTL@ ZMAW.DE University of Hamburg

300

250

Zfs [m]

200

150

EWTL@ ZMAW.DE University of Hamburg

300

accelerated BL
decelerated BL
higher turbulence BL
no BL simulation
reference BL

250

accelerated BL
decelerated BL
constant cross section
higher turbulence BL
no BL simulation
reference BL

100

50

200

Zfs [m]

0
-0.02

-0.016

150

-0.008

-0.004

U'W'mean/U ref2 [-]

Figure 7. Vertical profiles of the vertical momentum

fluxes in Park Avenue

100

0.002

0.004

0.006

U'V'mean/U ref2 [-]

0.008

0.01

Figure 6. Vertical profiles of the horizontal momentum

fluxes in Park Avenue
The biggest differences are observed slightly below
and near roof level. However, it should be kept in mind
that in about 70 m height the measurements show
strong gradients and a relatively high uncertainty of the
measured values is documented by the attached error
bars. With respect to the reference profile, the
accelerated and decelerated boundary layers generate
an average deviation in the order of about 40 %,
whereas the more turbulent approach flow is causing
differences of 60 % on average. The biggest deviation
in the measured horizontal flux profiles is observed for
the test case without an explicit atmospheric boundary
layer simulation, for which the differences reach a value
of 80 % on average.
In Figure 7 the vertical momentum fluxes measured
for the different approach flow conditions are given. At
roof level and in 180 m height above ground the highest
deviations are documented for the given measurement
location. The accelerated boundary layer shows the
highest differences with respect to the reference
configuration for heights right above roof level. For the
accelerated
and
decelerated
boundary
layers
approaching the model, average deviations between 10
% and 20 % were calculated. The more turbulent
approach flow is characterized by a deviation of about
40 % on average in the measured vertical momentum
fluxes. Without any explicit boundary layer modeling, the
differences between the reference case and the test
configuration reach values of 60% on average over the
entire profile.

Comparing the effect of changing boundary

conditions on the results of local flow measurements, it
can be stated that at roof level, a much bigger scatter
and bigger deviations are found in momentum flux
measurements than in the mean flow measurements.
Whereas the shape of the measured vertical mean wind
profiles are relatively similar for all test configurations,
the scatter in the shape of the measured flux profiles is
more pronounced.
Big deviations were found
particularly for the accelerated/decelerated model
boundary layers, indication that even small pressure
gradients across the model area will have a significant
effect on flow and dispersion modeling in a boundary
layer wind tunnel.
In order to document and compare the characteristic
scale of the turbulent eddies modeled, the integral
length scale of turbulent structures was calculated from
measured time series of the different boundary layers
(Figure 8). For the given measurement location in Park
Avenue the more turbulent approach flow produces the
highest observed deviation in measured turbulence
length scales of 112 m at a height of 110m.
EWTL@ZMAW.DE University of Hamburg

300

250

accelerated BL
decelerated BL
more turbulence BL
no BL simulation
reference BL

200

Zfs [m]

50

0
-0.002

-0.012

150

100

50

0 1
10

102

Lux [m]

Figure 8. Integral length scale in Park Avenue

44

103

3 Conclusions
The detailed analysis of locally measured data
reveals that proper boundary layer modelling is essential
for obtaining reliable model results.
However, the accuracy required for modeling proper
approach flow conditions certainly depends on the type
of problem to be investigated. If a more global flow
pattern over and around a complex urban roughness,
represented by horizontal mean velocity measurements,
is of particular interest, improper boundary layer
modelling can cause a minimum uncertainty of the
measured velocity components in the order of 10 % to
20 %.
If the local flow and dispersion phenomena at a
certain location are of particular interest, much more
care must be taken regarding a qualified experimental
setup providing proper approach flow conditions. As
shown in this paper, a highly dense built-up area can
even intensify the influence of varying approach flow
conditions on the locally observed turbulent properties.
Uncertainties up to a factor of three were found for the
example given here.
It is obvious that no atmospheric boundary layer
simulation at all produces the biggest errors. Hence, an
experimental setup without explicit boundary layer
modeling should never be used for investigations of
atmospheric flow and dispersion processes in urban
areas. Approach flow conditions with higher turbulence
levels than expected at full scale will have a significant
impact on the momentum fluxes and integral length
scales measured even at measurement locations below
roof level in a street canyon oriented perpendicular to
the mean wind direction. Turbulence characteristics
must be modeled and replicated in a wind tunnel
carefully in order to ensure a reasonable quality of
model results.
The longitudinal pressure gradient along the test
section in the wind tunnel and resulting accelerated or
decelerated model boundary layers approaching a
model was found to be an important factor regarding
uncertainty of the model results. Consequently, at least
the longitudinal pressure gradient across the model
needs to be checked and minimized carefully before any
model test is carried out. The most sensitive criterion
regarding accelerated/decelerated approach flow is,
however, a direct measurement of the vertical
momentum flux profile. Documenting a sufficiently thick
constant shear layer at the right momentum levels turns
out to be a necessary prerequisite for any reliable flow
and dispersion modeling in a boundary layer wind
tunnel.

References
Allwine, K. J., M. J. Leach, L. W. Stockham, J. S. Shinn,
R. P. Hosker, J. F. Bowers and J. C. Pace, (2004):
Overview of Joint Urban 2003 an atmospheric
dispersion study in Oklahoma City, Symp. on
Planning, Nowcasting and Forecasting in the Urban
Zone, January 11-15, Seattle, WA. Amer. Meteor.
Soc.. J7.1.
Leitl, B.; Pascheke, F.; Schatzmann, M. (2003)
"Generation of Wind Tunnel Data Sets in Supprt of
the Joint Urban 2003 Atmospheric Dispersion Study
Final Reort, Phase I", Meteorological Institute,
University of Hamburg, Germany
Leitl, B.; Schatzmann, M. (2005) "Generation of Wind
Tunnel Data Sets in Supprt of the Joint Urban 2003
Atmospheric Dispersion Study Final Reort, Phase

45

II", Meteorological Institute, University of Hamburg,

Germany
Pascheke , F., B. Leitl, M. Schatzmann (2004): Results
from recent observations in an Urban Boundary
Layer, COST 715 - Workshop on Urban Boundary
Boundary Layer Parameterisations, Office for Official
Publications of the European Communities, ISBN 92894-4143-7, pp. 73-83.
VDI (2000): Environmental Meteorology Physical
Modeling of Flow and Dispersion Processes in the
Atmospheric Boundary Layer, Application of Wind
Tunnels,
VDI
3783/12,
December
2000,
Kommission Reinhaltung der Luft im VDI und DIN,
Handbuch Reinhaltung der Luft, Vol. 1b.

Properties of the far wake of a wind turbine in an atmospheric boundary

layer
G. Espaa, S. Aubrun, P. Devinant

L. Laporte, E. Dupont

Laboratoire de Mcanique et dEnergtique,

Universit dOrlans, 8 rue Lonard de Vinci
F-45072 Orlans Cedex2, France
guillaume.espana@univ-orleans.fr

CEREA joint laboratory ENPC-EDF R&D,

6-8 avenue Blaise Pascal, Champs-sur-Marne
77455 Marne la Valle Cedex2, France
laurent.laporte-daube@edf.fr

Abstract This contribution is about wind turbines

modelling in a wind tunnel, taking into account the
atmospheric boundary layer. Wind turbines are
modelled using the concept of porous disc which can
replicate the wind velocity deficit and the shearproduced turbulence in the flow going through the
turbine. It is especially focussed on the influence of
the turbulence upstream of the turbine on the velocity
deficit and the additional produced turbulence.

1 Experimental set-up
Measurements were performed in the Malavard
wind tunnel of the Laboratoire de Mcanique et
dEnergtique (L.M.E), university of Orleans. It is a
close-circuit type with a test section of 2m high, 2m
wide and 5m long. However, as studies concerning
atmospheric flows (especially concerning the ABL)
require greater dimensions, we had to transform the
return section to get closer to atmospheric wind
tunnels conditions. Figure 2 underlines these
transformations. A sketch of the Malavard wind
tunnel is presented in figure 1 and a sketch of the
return section in figure 2. A convergent was placed
to control and homogenize the flow and a 16m long
and 5m wide wooden plate was constructed in the
return test section to model a flat terrain. It is
possible to see on this sketch that the return section
is divergent (angle 2.6) from the exit of the
convergent to approximately 12,5m. Beyond this
location, the section has a constant 5m width
(measurements were made to validate the
properties of the ABL in this constant part of the test
section).
Pieces of corner iron were placed on this flat terrain
to model the ground roughness, and turbulence
generators at the exit of the convergent allow the
establishment of the ABL. It was checked that
several ABL could be properly modelled at a
geometric scale of 1:400, from moderatly rough (low
grass, steppe) to very rough surfaces (centres of
European towns, suburbs).

Key words Wind turbine, atmospheric boundary layer,

wake, Laser Doppler Velocimetry.

Introduction
Wind farm efficiency naturally relies on
technological improvement of the wind turbine but it
also and especially relies on the resource
assessment of the upstream flow conditions. The
global resource assessment depends on the type of
terrain and on local meteorological conditions but it
could locally be modified when wind turbines
interact with each other. Indeed, the wind turbine
wake is characterised by a deficit of velocity and
production of turbulence. The estimation of flow field
modifications in the far wake of a wind turbine
located in the atmospheric boundary layer (ABL) is
not obvious. On one hand, field measurements are
rare, difficult to interpret and do not allow
parametrical study and on the other hand, numerical
simulations have still, in this topic, a high degree of
simplification. Physical modelling is therefore a
good tool to study this problem.
The goal of the present study is to model in a wind
tunnel, at a geometric scale of 1:400, a wind turbine
in a rough ABL. The first part of the project was to
modify the wind tunnel to allow the modelling of the
ABL. Then, a first set of measurements was needed
to validate the experimental device and for
comparisons with litterature concerning the wake
behind a single turbine. A third part focusses on a
parametric study and comparisons of the wakes of 2
wind turbines placed in 2 different turbulent flows.
Finally, in a future collaboration with CEREA (joint
laboratory between the French electricity provider
EDF and the Ecole Nationale des Ponts et
Chausses), comparisons will be made between
their numerical simulations and our physical
modelling to validate both methods. Thats why a
part of this paper is dedicated to the explanations of
the numerical method used by CEREA.

Return
section
Return
testtest
section
5 x 5 m
4u4m
Vmax = 22 m/s
Vmax = 12.5 m/s

Main test section

2u2 m 2
Vmax = 50 m/s

Figure 1 : Malavard wind tunnel

47

ABL (rougher however than for the Turbine1) which

characteristics could be obtained from Aubrun et al.
[4]. Measurements using Laser Doppler Velocimetry
have been compared with those from the first
turbine. The upstream mean wind velocity was also
10 m/s and the upstream turbulence intensity at hub
height was Iu 0 20% .

2 Numerical set-up
Numerical simulations of a horizontal axis wind
turbine wake could be achieved using different
methods (Vermeer et al. [11]). The porous disc
concept has been chosen to model the rotor and
source terms were added in the Navier-Stokes
equations to model the thrust and the rotation of the
blades in three dimensions. This approach was
chosen for different reasons. Firstly, the main
interest is the far wake region. Thus, there is no
need to reproduce the region directly behind the
turbine blades with accuracy. Secondly, the cost of
the simulation is very low compared to simulations
where the entire rotor is meshed. Thirdly, this
porous disc approach is not very different from the
physical modelling. Finally, using source terms
allows us to simulate the rotation of the blades and
as a result to simulate the rotation of the wake.
Mercure_Saturne is a three-dimensional CFD model
adapted to atmospheric flow from Code_Saturne,
CEREA general-purpose computational fluid
dynamic code for laminar and turbulent flows in
complex geometries. Mercure_Saturne solves the
RANS equations and the transport equations for
turbulent kinetic energy and dissipation. The
turbulence closure model used in the simulations is
k-
with
the
standard
parameterisation
(Archambeau et al. [1]). The discretization in space
is based on the unstructured finite volume
framework with collocated arrangement for all
variables. The time-marching scheme is based on a
prediction of velocity followed by a pressure
correction step. In all the numerical simulations, the
inflow wind profile is modelled in neutral conditions
by a logarithmic law. The inflow turbulent energy
and dissipation profiles are computed following the
Monin-Obhukov theory. The initial conditions and
the inflow boundary conditions are the same.
In order to validate this method, CEREA used the
wake measurements of a small wind turbine
(Rutland WG503) tested in the ENSAM Laboratoire
de Mcanique des Fluides wind tunnel (Rignault,
[7]). To illustrate the results of 3D numerical
simulations, the turbine wake is shown in figure 3
and figure 4. The simulation is in good agreement
with the measurements (figure 3). As it can be seen
in figure 4 the annular structure of the wake is well
defined. At 0.1D downstream of the disc, the hub
has a strong influence. The speed range is 0-10m/s.
The wake is very inhomogeneous.

Figure 2 : Transformations in the Malavard return section

The wind turbines were modelled at a geometric

scale of 1:400, according to the actuator disc
concept. The actuator disc extracts kinetic energy
(creating a velocity deficit downstream of the disc),
generates a spreading of the stream tube and an
appearance of shear-generated turbulence. A
porous disc produces exactly the same features.
This concept was in particular studied by Aubrun at
the L.M.E [2]. Using this concept, discs of 100mm
and 200mm diameter (D) (respectively 40m and
80m in full scale) made of metallic mesh (mesh size
= 3mm, wire diameter = 1mm) were used. With
these characteristics, the induction factor defined as

U disc
(where U disc is the mean velocity just
Uf
behind the rotor and U f the mean upstream
a 1

velocity) was a = 0.2 and the porosity level was

55%. It has been verified that, at a given velocity
deficit, the corresponding induction factor was in
agreement with thrust and power coefficients
(respectively CT
and Cp)
defined as
CT 4a (1  a ) and C P 4a (1  a ) (Aubrun [3]).
The porous discs were fixed to the floor with a
vertical cylinder of diameter 0.05D (standard wind
turbine mast diameters) at a hub height
corresponding to 1D.
The first part of the study deals with the first turbine
(Turbine1, D = 100mm) in a rough ABL (roughness
length z0 = 23cm in full scale and power law
exponent D = 0.19) and consisted in verifying the
properties of the velocity and turbulence fields by
comparison with litterature. The upstream mean
wind velocity was 10 m/s. The upstream turbulence
intensity
at
hub
height,
defined
as

H hub )
, where U RMS is the root
H hub )
mean scare velocity and U mean the mean velocity,
Iu 0

U RMS ( z
U mean ( z

was Iu 0

16% .

In the second part, the greatest turbine (Turbine2, D

= 200mm) was studied and also placed in a rough

48

At 4D, the wake structure is more homogenous with

wind speeds ranging from 6m/s to 10m/s. At 10D
the wind recovers 80% of its initial speed.

According to the VDI 3783/12, the power law

exponent for a rough surface, should be between
0.18 < D < 0.24, the roughness length between
0.1m < z0 < 0.5m and the displacement height d0 |
0.75h (h = height of the pieces of the roughness
elements). The values found for the modelled ABL
are:
D (20 to 180m) = 0.19
z0 (20 to 80m) = 0.23m
The agreement is satisfying. Turbulence spectra
and longitudinal integral length scale have also
been checked, reinforcing the validity of the
modelled ABL. Other types of ABL have been
modelled and corresponding graphs are available in
Aubrun et al. paper [4].

Figure 3 : Turbine wake for several locations downstream

of the simulated disc. (Numerical results from CEREA)

3.2 Results with Turbine1 (D=40m)

In this part, a 100mm diameter modelled wind
turbine was fixed to the floor in the ABL studied in
2.1. The upstream mean wind velocity was 10m/s
and the upstream turbulence intensity at hub height
was 16%.
Mean and fluctuating velocity were measured and
figure 5 shows vertical profiles of the longitudinal
component of the fluctuating velocity u (at several
locations downstream of the disc) divided by the
upstream mean velocity at hub height U0. Our plot is
compared to Smith & Taylors results [8]. Although
the second turbulence peak (lowest altitude) at
x=2D is smaller than for Smith & Taylor, the global
tendancy is respected.
3

Fi

2.5

Figure 4 : Annular structure of the wake for several

locations downstream of the simulated disc. (Numerical
results from CEREA)

x = 2D
x = 4D
x = 6D
x = 10D
x = - 1D

1.5

z/D

To fully validate this model, wake measurements

(wind speed and turbulence) on operational wind
turbines are needed. Unfortunately, such data are
very difficult to obtain. The comparison with the
small turbine Rutland is a first step toward validation
but the lack of turbulence data and the structural
differences between a small and a normal wind
turbine make the analysis difficult. The alternate
possibility is to compare results of numerical
simulation with results of physical modelling.

1
0.5
0
-0.5
-1
0.04

0.06

0.08

0.1

0.12

0.14

0.16

0.18

0.2

u / Uo

3 Results for validation of the

experimental device
In the following paragraphs, all the dimensions are
full scale dimensions.
3.1 Modelling of the ABL
Many references can be found in litterature which
provide requirements to properly replicate the
neutral ABL in wind tunnels (Snyder [9], VDI
3783/12 [10], ESDU [5]).

Figure 5 : Vertical turbulent velocity profile at several

downstream distances.
Top figure = measurements in Malavard wind tunnel

49

Bottom figure = Smith & Taylors measurements [8]

Figure 6 represents the velocity deficit for several

longitudinal positions downstream of the turbine.
This deficit is here, defined as:Ud=U(z)upstream Uw(z), and is divided by the upstream mean velocity
at hub height. Uw(z) is the mean velocity in the wake
measured at z and U(z)upstream is the upstream mean
velocity measured at z. This plot is compared to the
results of Vermeer, Sorensen and Crespo [11].
Despite a smaller value of our velocity deficit for 2D
downstream of the turbine, the global results are
already satisfying. Indeed, comparisons of the
measured velocity deficits at hub height and the
results from [11] are reported in the table 1. These

Figure 6 : Vertical distribution of velocity deficit at several

downstream distances.
Top figure = measurements in Malavard wind tunnel
Bottom figure = extracted from [11]

U d
in percents and

U
hub max

results represent

show a good agreement, especially for the far wake.

Malavard
wind tunnel
measurement
s
TNO
measurement
s

x = 2D
36.2%

x = 4D
18%

x = 6 / 7D
14%

51%

24%

12%

4 Results and analysis

In this paragraph, Turbine1 was replaced by
Turbine2, 2 times greater (D=200mm, 80m in full
scale). The upstream mean wind velocity was still
10m/s.
The main objective of the use of Turbine2 was to
make a comparison with the results of Turbine1.
Indeed, Turbine2 was located in a more turbulent
region of the flow (Tubulence intensity at hub height
| 20% for Turbine2 against 16% for Turbine1) and
this study could underline the effects of this change
of upstream flow conditions on mean and turbulent
velocity profiles.
On figures 7 and 8, vertical profiles of mean velocity
deficit are plotted for Turbine1 and Turbine2. This
deficit is defined as : Ud = U(z)upstream - Uw(z) (same
as figure 6).
On figures 9 and 10 are plotted vertical profiles of
non-dimensionalised root mean scare velocity

Table 1 : Comparison between the measured velocity

deficit at hub height and the results from [11]. (Deficits
expressed in percents of Uhub)

The main striking difference comes from the fact

that Vermeer, sorensen and Crespo [11] founded a
downshift of the maximum velocity deficit. It is
explained that this downshift could mainly be due to
the shear of the incoming flow and the presence of
the ground. This downshift is not seen on our
measurements. However, the characteristics of the
upstream flow for the results extracted from [11] are
not available. It is possible to think that this
downshift strongly rely on the type of ABL modelled.
A rougher ABL can create a stronger shear, which
modify the evolution of the wind turbine wake. In
this way, no precise conclusion can be given for this
difference. Future studies could try to underline this
phenomenon for better comprehension.

deficits defined as

measured at z and defined as : Iu

-0.5

0.1

0.2

0.3

0.4

0.5

z/D

Iu  Iu 0

U rms ( z )
and
U w ( z)

Iu0 is the upstream turbulence intensity at hub

height, defined in paragraph 1.

which represents, according to the Wind Energy

Handbook [12], the turbulence produced between
the upstream and the downstream flow, are plotted
on figure 11 and 12. Iu is the turbulence intensity

0.5

-1
-0.1

U hub

Finally, horizontal profiles of Iu 

x = 2D
x = 4D
x = 6D
x = 10D

U rms ( z )  U rms ( z ) upstream

0.6

Ud / Uhub

50

0.5

z/D

z/D

x = 2D (D=80m)
x = 4D (D=80m)
x = 2D (D=40m)
x = 4D (D=40m)

x = 2D (D=80m)
x = 4D (D=80m)
x = 2D (D=40m)
x = 4D (D=40m)

1.5

0.5

-0.5
-0.5

-1
-0.1

0.1

0.2

0.3

-1

0.4

[U(z)upstream - Uw(z)] / Uhub

-0.1

-0.05

0.05

[Urms(z) - Urms(z)upstream] / Uhub

Figure 7 : Vertical mean velocity deficit profiles for the

downstream locations x=2D and x=4D. Comparison
between Turbine1 (D=40m) and Turbine2 (D=80m)

Figure 9 : Vertical fluctuating velocity profiles for the

downstream locations x = 2D and x = 4D. Comparisons
between Turbine1 (D=40m) and Turbine2 (D=80m)

0.5

z/D

z/D

x = 6D (D=80m)
x = 10D (D=80m)
x = 6D (D=40m)
x = 10D (D=40m)

x = 6D (D=80m)
x = 10D (D=80m)
x = 6D (D=40m)
x = 10D (D=40m)

1.5

0.5

0
-0.5

-0.5

-1
-0.1

0.1

0.2

0.3

-1

0.4

-0.1

-0.05

0.05

[Urms(z) - Urms(z)upstream] / Uhub

[U(z)upstream - Uw(z)] / Uhub

Figure 8 : Vertical mean velocity deficit profiles for the

downstream locations x=6D and x=10D. Comparison
between Turbine1 (D=40m) and Turbine2 (D=80m)

Figure 10 : Vertical fluctuating velocity profiles for the

downstream locations x = 6D and x = 10D. Comparisons
between Turbine1 (D=40m) and Turbine2 (D=80m)

On figure 8, the velocity deficit of Turbine2, for

x=10D, is nearly null, showing that the wake at this
distance downstream of the turbine has already
disappeared. It is not the case for Turbine1 for
which the deficit is always superior than Turbine2
deficit. These results underline the fact that the
turbine wake is diffused faster for a turbine placed in
more turbulent atmospheric flow. The ratios of the
velocity deficit to the upstream mean velocity at hub
height are reported in table 2.

The peaks of fluctuating velocity at z = 0,5D (top of

the wind turbine) on figures 9 and 10 give another
indication of the influence of upstream turbulence.
Indeed, these peaks are always smaller for
Turbine2 reflecting the fact that the produced
turbulence is merged in the ambiant turbulence. In
Table
3
are
reported
the
values
of

x=2D
x=4D
x=6D
x=8D

D=40m
36.1%
17.9%
13.8%
6.4%

U rms ( z )  U rms ( z ) upstream

percents.

D=80m
33.8%
17.5%
7.2%
| 0%

U hub

x=2D
x=4D
x=6D
x=8D

Table 2 : Comparison between Turbine1 and Turbine2

mean velocity deficit at hub height. (Deficits expressed in
percents of Uhub)

at z / D = 0.5, in

D=40m
5.8%
4.8%
3.3%
2.5%

D=80m
4.8%
2.8%
1.1%
0.7%

Table 3 : Comparison between Turbine1 and Turbine2

fluctuating velocity deficit at the top of the turbine. (Deficits
expressed in percents of Uhub)

51

coefficients, any wind turbines operating point could

be reproduce.
With the prospect of the collaboration with CEREA,
we only have to know the type and the
characteristics of the wind turbine that is wanted to
be modelled and the comparisons of results should
validate our two modelling concepts.
Unfortunately, when this article has been written,
these comparisons were not finished yet. Interesting
results are expected to be found in the following
months.
Finally, field measurements would also be very
interesting to confirm the numerical simulations and
the wind tunnel measurements.

x = 2D (D=80m)
x = 4D (D=80m)
x = 2D (D=40m)
x = 4D (D=40m)

22
20
18

Iu+ (%)

16
14
12
10
8
6

-2

-1

y/D

Figure 11 : Horizontal profiles of the produced turbulence

for the downstream locations x =2D and x = 4D.
Comparison between Turbine1 (D=40m) and Turbine2
(D=80m)

References
[1]

F.Archambeau, N.Mehitoua, M.Sakiz (2004) Code

saturne a _nite volume code for the computation of
turbulent
incompressible
ows

industrial
applications. International Journal on Finite
Volumes, 1.

[2]

S.Aubrun (2005) Modelling wind turbine wakes with

a porosity concept. Proceedings of the Euromech
Colloquium 464b. Wind Energy pp.265-270. J.
Peinke, P. Schaumann, S. Barth (Eds.)

[3]

S.Aubrun (2007) Physical modelling of the far wake

from wind turbines. Application to wind turbine
interactions. Ewec2007, May 7-10, Milan, Italy.

[4]

S.Aubrun, G.Espaa, P.Devinant (2007) Adaptation

of the Lucien Malavard wind tunnel (L.M.E Orlans,
France) as an atmospheric boundary layer wind
tunnel. Physmod, August 29-31. Orlans, France.

[5]

ESDU 85020 (1985) Characteristics of atmospheric

turbulence near the ground

[6]

W.J.M.Rankine (1865) On the mechanical principles

of the action of propellers Trans. Inst. Naval
Architects, 6.

[7]

L.Rignault
(2002)
Etude
numrique
et
exprimentale du sillage d'un rotor olien. Technical
report, EDF.

[8]

D.Smith, G.J.Taylor (1991) Further analysis of

turbine wake development and interaction data.
Proceedings of the 13th BWEA Wind Energy
Conference. Swansea, UK, pp.325-331

[9]

W.H.Snyder (1981) Guideline for fluid modelling of

atmospheric diffusion. US Environmental Protection
Agency

22
x = 6D (D=80m)
x = 10D (D=80m)
x = 6D (D=40m)
x = 10D (D=40m)

20
18

Iu+ (%)

16
14
12
10
8
6

-2

-1

y/D

Figure 12 : Horizontal profiles of the produced turbulence

for the downstream locations x =6D and x = 10D.
Comparison between Turbine1 (D=40m) and Turbine2
(D=80m)

The two discs used for Turbine1 and Turbine2 have

the same characteristics (porosity = 55%, mesh size
= 3mm and wire diameter = 1mm, induction factor a
= 0.2). In this way, figures 11 and 12 give a sense of
universality of the produced turbulence measured.
Indeed, it reflects that the shear-produced
turbulence only relies on the type of disc and no
longer on its diameter. According to Aubrun [2],
when well non-dimensionalised, mean velocity and
turbulence intensity profiles seem to be independant
of the disc diameter. This result comes from
measurements on three discs with the same
porosity characteristics but with three different
diameters (100, 200 and 300mm) in a
homogeneous upstream flow. Nevertheless, this
result must also be proved in homogeneous flow,
for the type of disc used in this study.

[10] VDI 3783/12 (2000) Physical modelling of flow and

dispersion processes in the atmospheric boundary
layer Application of wind tunnels. Beuth Verlag,
Berlin

5 Conclusions
The validity and the feasability of the experimental
set-up is proved. By playing with the porosity level
of the porous disc, it has been shown that any
velocity deficit could be reproduced (Aubrun [2]). In
this way, as to a porosity level corresponds a
precise induction factor linked to power and thrust

[11] L.J.Vermeer, J.N.Sorensen, A.Crespo (2003) Wind

turbine wake aerodynamics. Progress in aerospace
science pp. 467-510
[12] Wind Energy Handbook. Burton et al. J.Wiley & sons
LTD.

52

Aerodynamic Design of Princess Elizabeth Belgian Antarctic Research

Station
J. Sanz Rodrigo, C. Gorle, J. van Beeck, P. Planquart

von Karman Institute for Fluid Dynamics

Rhode-Saint-Gense, Belgium
sanz@vki.ac.be, gorle@vki.ac.be, vanbeeck@vki.ac.be, planquart@vki.ac.be
Abstract - The new Belgian Research Station will be
constructed in the Sor Rondane Mountains,
Antarctica, during the austral summer 2007-2008. In
the conceptual design phase, wind tunnel modeling
using sand erosion technique allows efficient
evaluation of the snow erosion and deposition
around different building-block shapes. The best
performing concept is further analyzed in terms of
wind loading using a model instrumented with
pressure taps. The positioning of the main building
with respect to the ridge has a significant impact on
wind loading and snow erosion and deposition.
CFD modeling supports the detailed design.

such that it minimizes impact on the environment and on

the landscape during the construction, operation and
removal of the station.The station has a hybrid design,
with the main building above ground-level and anchored
onto snow-free rock area and a garage and storage
building constructed under the snow in the lee side of a
small granite ridge. The base is conceived as a
prototype to achieve the minimum environmental impact
and the highest efficiency in the use of energy
resources.

Key words Antarctica, snow, wind engineering, CFD

Introduction
The new Belgian Antarctic Research Station will be
constructed in the Sor Rondane Mountains (7157S
02320E, 1390m a.s.l.), during the austral summer
2007-2008. It is designed for optimal use by 12 people
with a surface area (living, technical, research, storage)
of 800 m. An extension, based on heated shelters, will
make it possible to accommodate another 8 to 18
people.

Figure 1: Impression of the Belgian Antarctic base

Consistent with the philosophy of the project,
managed by the International Polar Foundation (IPF),
the station design will make best use of the terrain
conditions for the integration of the buildings and will be

53

Figure 2: Sketch of building-ridge configuration

The system design of the station is based on
sustainable technology and high energy efficiency, with
full-year monitoring and remote sensing capabilities.
The facilities will use renewable energy as the primary
energy source, integrating a comprehensive energy
management regime, thereby minimizing the use of
fossil fuels. The station aims for zero emissions with
most of its energy covered by renewable sources. The
interested reader should refer to the CEE report
(BELSPO and IPF, 2007) for a broader description of
the scope of the base and its design particularities.
The efficiency of the building depends to a great
extend on its ability to cope with the extreme
environment in which it is immersed. The aerodynamic
design of the building is therefore one of the most
important conceptual design drivers. Wind tunnel testing
with sand erosion technique allows an efficient
evaluation of the snow and wind comfort for different
building block concepts and ridge integration
alternatives.
The best performing concept constitutes the
reference for the design optimization. A parametric
study is carried out to asses the impact of the building
positioning, orientation versus wind and elevation on
snow erosion/deposition and wind loading. The latter is
analyzed with a wind tunnel model instrumented with
pressure taps. Volumetric sand erosion tests are
conducted to study the integration of the under-snow
garage situated behind the main building. It will be
shown that the positioning of the main building with
respect to the ridge has a major impact in wind loading
and snow erosion and deposition.
The detailed design is conducted with a validated
CFD model and further erosion tests. The wind loading

successive steps of one minute duration reaching a

quasi-steady state on the erosion patterns. The velocity
at each step is increased by 0.5m/s, from 5 to 13m/s at
freestream level; 3) A picture is taken from the top at the
end of the end of each time step to record the
progression of the erosion patterns; 4) A contour
detection algorithm detects the contours and assigns
them the erosion velocity at which the picture was taken;
5) A gridding method is used to interpolate between the
contours to obtain the so called erosion speed contour
map.
Once the two erosion speed contour maps are
obtained, the ratio between both will reflect an
amplification factor, A, as a measure of the action of the
building on the flow.

on the final building geometry is assessed using a wind

tunnel model instrumented with pressure taps.
This project shows the advantages of introducing
Wind Engineering since the conceptual phase of the
building design process, giving the opportunity to
consider the most effective use of the environment.

1 Snow Erosion/Deposition Mapping

Using the Sand Erosion Technique
Wind tunnel modeling allows efficient support on the
conceptual design phase, when fundamental aspects
of the design, like building shape, orientation or
positioning, are considered.
The first concern from the aerodynamics point of
view is the ability of the building to cope with intense
snowdrift. Many Antarctic stations have had serious
problems with snow drifting, with eventual buried
structures that require costly maintenance to prevent
from inaccessibility and structural damage. To overcome
this problem, it is common practice to use elevated
buildings, which allow the wind to flow underneath
blowing the snow away from the building.
Snowdrift modeling in wind tunnel is a difficult task
due to the inherent impossibility to maintain similarity of
all the driving forces. It is also very time consuming (it
takes several hours to develop significant buildups). As
a result, it was not feasible to use this technique in the
very demanding conceptual design phase, as many
building configurations had to be tested in a very limited
amount of time. Instead, the sand erosion technique
proved to be a very efficient tool to evaluate the wind
conditions at ground level. It was used to point out those
areas of severe wind comfort and, at the same time,
identify snow accumulation and erosion regions.

U ref
Ub

(1)

where Uref is the erosion velocity at the building-free

case and Ub is the erosion velocity at the building case.
This is, A>1 implies an acceleration of the flow, A<1 a
deceleration of the flow and A=1 when the building does
not change the erosion conditions of the reference level.
Hence, those areas out of the influence of the building
flow deflection show an amplification factor close to 1.
The sand, as the snow, is eroded when the shear
stress of the wind acting on the ground exceeds a
certain threshold. This condition is satisfied, in the case
of the sand employed in this study, for a threshold
friction velocity of 0.23m/s, also a typical value for wind
hardened snow (Jaedicke, 2001). Those regions with
enough wind or turbulence to bring wind gusts to the
ground with higher friction velocities than the threshold
will detach the sand/snow particles from the ground and
transport them with the flow downstream. Therefore, the
contours obtained from the sand erosion patterns reflect
the positions at which the friction velocity is near the
threshold value, separating the erosion (friction velocity
higher than the threshold) from the deposition areas
(friction velocities below the threshold) for a given
velocity. This geometrical criterion was proposed by
Anno (1984) as the most important similitude in the
modelling of a snowdrift since the snowdrift would be
formed as a combination of erosion and deposition.
Of course, the sand erosion technique is a surface
visualization and does not provide information about the
development rate of the buildup height. It also cannot
predict the depth of snow erosion in those areas with
amplification factors higher than one. Therefore, the
comparisons between different building configurations is
based on the extension of the areas with A<1 as an
indicator of the efficiency of the building to accumulate
snow.
Figure 3 shows some typical images from erosion
tests. The left column presents the building-free test
which is used as a reference to compute the
amplification factor generated by the building on the
wind flow. The areas out of the influence of the building
present similar erosion evolution to the building-free
case. The small differences are due to the inherent lack
of repeatability of the manual sand spreading process.
Nevertheless, the reproducibility of the erosion tests is
quite good, with uncertainties on the erosion velocity
below 0.5m/s. At low erosion velocities, the first erosion
patterns appear near the corners and below the building,
where the turbulence and wind speed-up are the
highest. At high velocities, only the sand in the best
sheltered areas remains. These areas will have good
wind comfort but will also buildup snow.

Figure 3: Examples of surface sand erosion tests:

building-free test as reference (left), elevated building
oriented parallel (middle) and perpendicular (right) to the
incoming wind direction
The sand erosion technique is performed in two
steps: first without the building (reference case) and
then with the building included at the position and
orientation of interest. In both cases, the testing
procedure consist on: 1) Spreading a uniform sand layer
of ~3mm all over the model; 2) The wind tunnel is run in

54

Up to six different building geometries were tested

using the sand erosion technique. A trade-off between
aerodynamics, energy efficiency and compatibility with
the internal layout determined the best performing
concept, which is based on a one-storey square-based
building. This geometry constitures the reference for the
optimization to follow.
A parametric study was done on the reference
building to determine the impact that the building-ridge
integration would have on the snow erosion/deposition
and wind loading. The following parameters were tested:
orientation versus prevailing wind (0 or 45o incidence),
pillars height (elevation) and positioning on the ridge.
Figure 4 shows the amplification factor erosion maps for
four different integrations for the prevailing wind
direction from 101o: at reference position and 1m
clearance at ridge-top (top left map), at reference
position and 2m clearance (top right) and at 2.5m
backwards position with 1m (bottom left) and 2m
clearance (bottom right).

constitutes the setup for the volumetric sand erosion

tests. These tests are done, similar to the surface
erosion tests, at progressively increasing speeds but
now the time for each velocity step is extended to 5
minutes to account for the longer development of the
erosion patterns.
The sand bed is very erodable, compared with the
wind hardened snow and ice present at the site, so the
results cannot be directly related to the field conditions.
The volumetric sand erosion tests can be nevertheless
useful to qualitatively compare different building
configurations with each other.
The detailed design involves the adaptation of the
building envelop to other design features coming from
the internal layout or the system integration. The aim for
a modular base led to an octagonal building shape,
keeping all the corners with the same angles (Figure 5).
The accommodation of solar panels on the side walls
also required an inwards inclination of 30o for a more
efficient energy capture. The integration of some system
units in the technical room at the center of the building
required a second level that sticks out of the roof. This
add-on would be used also to give access to the top of
the station and to accommodate some more solar
panels.
Surface sand erosion tests (Figure 5) revealed that
the octogonal geometry was much more aggressive
than the reference building, with quite intense erosion
patterns.

Figure 4: Amplification factor erosion maps for reference

building
The effect of the building positioning across the ridge
appears to be an important parameter for snow
accumulation and erosion. A building with a larger area
on top of the ridge helps in the attenuation of the
vorticity generated under the building, leading to less
intense erosion on the sand/snow surface. The same
occurs when the clearance height is decreased: a larger
clearance develops more erosion.

2 Under-snow Garage Integration

Having the garage integrated next to the main
building provides evident advantages with respect to the
operation of the base. Aerodynamically speaking it also
serves as a platform to prevent form eroding too much
the area next to the main building, providing a more
stable ground to give access to the building from the
West.
To study the garage integration, an 80cm-wide
portion of the ridge model was removed from the lee
side at the location of the building. A vertical cut of 8cm,
following the rock profile as it was found on the site,
enables the accommodation of a 45x11x5 cm garage
model. The step height between the garage roof and the
ridge top was kept at around 1m in full scale. The empty
space left was filled up with sand plus an additional
centimetre to cover the garage entirely. This situation

55

Figure 5: Amplification factor contour map for octagonalbased building

By reducing the surface at the back of the bottom
face, with the larger chamfering of the corners, the
suction created at the bottom surface on the ridge top is
acting on a smaller area and the overall lift force is
increased. This V shape at the exit of the clearance also
enhances the vorticity which is the responsible for the
strong erosion streams just behind the building. To cope
with these problems, the back of the building was
modified back to its original shape, still leaving the
inclined walls for the solar panels.

(Comite Europen de Normalisation, 2005), which

deduce the design load from the mean pressure field.
The quasi-steady method assumes that the pressure
fluctuations on the building are fully correlated with the
velocity fluctuations, i.e. the maximum pressure will
appear at the maximum velocity peak. Then, the design
extreme pressure is calculated by multiplying the mean
pressure coefficient with the peak dynamic pressure
calculated using the extreme wind velocity.

Figure 6: Volumetric sand erosion tests for garage

integration study: reference (left), octogonal (middle)
and final (right) buildings.
Figure 6 shows the results of the voulmetric sand
erosion tests for three building iterations, all of them
situated in Ref-2.5m position and with 2m clearance at
ridge-top. It is remarkable the important effect of the
geometry at the back of the building on the erosion
patterns. The larger area on top of the ridge of the
reference and final buildings attenuates the trailing edge
vorticity decreasing the erosion very significantly with
respect to the octogonal shaped building.

'p50| AWS

Cp AWS

1
2
UU 50
| AWS
2

(2)

where and U50|AWS are the air density and extreme

velocity at the automatic weather station (AWS) instaled
at the site. The pressure coefficient CpAWS measured in
the wind tunnel, with respect to the AWS velocity, is
defined as:

3 Wind Loading from Instrumented

Wind Tunnel Models
A wind tunnel model of the reference building was
instrumented with pressure taps to study the wind
induced forces. The number of pressure taps was
limited (to 133) by the size of the evacuation tower
situated in the back of the building, through which all the
1mm tubes had to pass in order to reach four scanning
valves. Besides, the frequency response of the pressure
lines (10Hz) is too low to allow analysis on local peak
pressures. Again, a parametric study was conducted in
order to assess the wind loading sensitivity versus:
clearance height, across-ridge positioning, orientation,
wind direction and incoming boundary layer.
The final building geometry, obtained at the end of
the detailed design, was also instrumented with 159
pressure taps in order to assess the final wind loading
specifications. This time, the positioning of the pressure
taps was defined from CFD simulations allowing for
better characterization of the pressure gradients and
providing lower errors in the calculation of the global
forces. The uncertainty on the global forces due to tap
resolution is reduced from 30%, in the reference
building, to 3% in the final building model. The garage
roof was included in order to assess the dependency of
the building wind loading on the garage vertical
positioning, i.e. the gap beneath the building at the
clearance outlet.

Cp AWS

'p
1
2
U 0U AWS
2

(3)

where 0 and UAWS are the wind tunnel air density

and AWS velocity.
Figure 8 and Figure 9 show the pressure coefficient
interpolated from the pressure tap positions thorughout
all the building skin for the reference and final
geometries. Two wind directions are presented: 101o
o
(prevailing) and 135 (for extreme winds).

Figure 8: Pressure coefficient for reference (bottom raw)

and final (top raw) buildings at 101o and 135o wind
direction. TOP VIEW.

Figure 7: Building instrumented with pressure taps and 4

scanning valves
Uematsu and Isyumov (1999) state that, when the
characteristic dimension of the building is small
compared with the turbulence integral scale of the
incoming winds, the maximum load effects can be
evaluated using a quasi-steady approach. This is the
standard practice of many codes, like the Eurocode

Figure 9: Pressure coefficient for reference (bottom raw)

and final (top raw) buildings at 101o and 135o wind
direction. BOTTOM VIEW.

56

It is observed the typical high suction in the

separated-flow region at the leading edge of the top
surface. The pressure map at the top surface offers
roughly the same pressure field regardless of the
position or elevation of the building. The high suction
generated at the leading edge is slightly decreased at
higher elevation or with larger windward cantilever as
the approaching flow has less vertical velocity
component, or up-flow, due to the sloping terrain (the
incoming wind is more parallel to the building generating
a thinner flow separation at the leading edge).
The most interesting results happen in the bottom
surface. Here, it can be noticed a positive pressure from
the leading edge due to the terrain induced up-flow.
Then, the pressure decreases and turns to negative,
due to the speed-up generated by a convergent
clearance, as the flow approaches the ridge top.
The final building presents a larger tower on the
back. The tower constitutes an stagnation area
increasing the extension of the positive pressure in the
bottom surface. Besides, the acceleration of the flow at
the ridge top is lower than in the reference building
leading to lower suction and, therefore, higher lift forces.
In the other hand, a more aerodynamic shape in the
front reduces the streamwise pressure and the building
drag is reduced with respect to the reference case.
The overall forces acting on the building are
obtained by integration of the mean pressure
coefficients on the building facades. The force
coefficients along i axis, Fi, are normalized as follows:
CFi

Fi
1
2
UU AWS
Ai
2

(4)

where Ai is the frontal area for the drag coefficient

CFx and the base area for the lift coefficient CFz.
A sensitivity analysis on the building-ridge integration
is summarized in Figure 10 and Figure 11, showing the
effect of the clearance height on lift and drag coefficients
for different across-ridge positions in the reference
building (x=0m is the reference position, x=-3m and
x=+3m means that the building is positioned 3m
backwards and forwards respectively), compared with
the final building which is always situated in the back
position. The sensitivity to the incoming ABL on mean
pressures is also shown: ABL-smooth is Class I
(z0=0.0002m, 7% turbulence intensity), close to the
expected site conditions, and ABL-rough is Class IV
(z0=0.4m, 20% turbulence intensity), typical of urban
environment.

Figure 11: Drag coefficient on reference and final

buildings
The effect of the incoming ABL on the mean force
coefficients is not important on the lift force. In the other
hand, it is observed a 20-30% higher drag for the rough
ABL, due to a larger building wake produced by a much
higher turbulence level. The drag increases as the
building moves forward due to a larger building-ridge
wake (larger frontal area). The final geometry offers
40% reduction on drag due to a smoother aerodynamic
frontal shape.
The high positive lift force is significantly decreased
when the reference building is displaced backwards due
to the larger suction on the bottom surface. For the
same reason, when the clearance height is increased,
the lift force decreases. Therefore, the elevated building
is not only a good solution to cope with snow
accumulation but also a more convenient configuration,
from the structural point of view, in this particular
situation in which the building is installed in sloping
terrain. Above 2m the decrease in lift force is not very
significant, so it seems to be a good choice for the
pillars height, also providing comfortable accessibility
below the building in case of maintenance works. The
final building geometry offers higher lift coefficient due to
a lower suction under the building. Besides, the inclined
faces on the sides, most of them under suction, also
increase the lift.

Figure 12: Aerodynamic coefficients variability with wind

direction

Figure 10: Overall lift coefficient on reference and final

buildings

57

Regarding wind direction variability (Figure 12) it is

remarkable the low dependency of the lift coefficient with
the wind direction in the final building geometry.
Surprisingly, the drag coefficient is the lowest, not at

101 when the building has a parallel orientation to the

o
o
wind, but at 135 . At 101 the ridge is perpendicular to
the incoming wind and the frontal area offered by the
building-ridge ensemble is the largest, producing a
bigger wake and therefore higher drag. In the contrary, it
is when the building is align with the wind direction that
the highest speed-up is generated under the building
and the lowest lift is produced. The presence of the
tower increases the drag in the SE-S sector.

4 CFD Modeling
A CFD model was also implemented to support the
detailed design phase. The comercial solver Fluent 6.3
was used with steady RANS k- turbulence model.
Modified wall functions were used to account for the
roughness of the snow. A validation of the pressure
coefficient at the tap positions was performed at wind
tunnel scale on the reference building. In this case the
boundary conditions are defined by the wind tunnel walls
and the inlet velocity and turbulence profiles as
measured. Figure 13 shows that the discrepancy in the
pressure coefficients from the CFD and the wind tunnel
tests was less than 25% of the stagnation pressure
coefficient (this corresponds to a margin of 0.2 in Cp as
indicated by the thin lines in Figure 13) for 75% of the
measurement points.

Figure 14: Pressure Coefficient from CFD simulations

on final building geometry.

5 Conclusions
The design of an Antarctic building requires careful
consideration of the environmental conditions in which it
will be immersed. Wind and snow effects are important
parameters to be considered throughout the design
process.
In the case of the Princess Elizabeth station, the
positioning of the building on the top of a small ridge
requires carefull consideration of the flow under the
building in order to control the snow erosion and
accumulation and the wind loading.
The presented project is a good example of the
advantages of including wind engineering since the
beginning of the building design process, giving the
opportunity to consider the most effective use of the
environment.

Figure 13: Scattered plot of Cp measured vs simulated.

The pressure coefficients in the front and bottom
faces are well predicted, leading to good estimates of
the drag force (10% uncertainty). The pressure at the
top surface is reasonably well predicted. In the bottom
surface, the CFD model predicts higher suction, leading
to lower lift forces on the building (30% uncertainty).
The main differences in the numerical model are
atributed to the difficulties in reproducing the flow
conditions under the building. The correct speed-up will
be obtained when the building wake is correctly
modeled. The building wake is very dependent upon the
building positioning as it is deduced from the high
sensitivity of the lift coefficient.
The final building geometry was also simulated
leading to similar validation results as in the reference
building. The pressure coefficient on the skin is
presented in Figure 14.
The CFD models are mainly used for complementing
the wind tunnel tests in those areas with low tap
resolution. It also helps in optimizing the tap positioning
for a more accurate assessment of the overall wind
loading from the wind tunnel measurements.

References
Anno, Y. (1984). Requirements for Modeling of Snow
Drift, Cold Regions Science and Technology, Vol 8,
pp 241-252.
Belgian Science Policy and International Polar
Foundation (2007). Construction and Operation of
the New Belgian Research Station, Dronning Maud
Land,
Antarctica.
Final
Comprehensive
Environmental Evaluation (CEE).
Comite Europen de Normalisation (2005). Eurocode 1:
Actions sur les structures Partie 1-4: Actions
gnrales Actions du vent. EN 1991-1-4.
Uematsu Y. and Isymov N. (1999). Wind Pressures
Acting on Low-rise Buildings. Journal of Wind
Engineering and Industrial Aerodynamics Vol 82, pp
1-25.

58

Feasibility study of tests on scale models for the evaluation of the

overpressures induced by a passing train on adjacent structures
L. Procino, G. Bartoli, C. Borri

A. Borsani

CRIACIV,
University of Florence,
Florence, Italy
lorenzo.procino@pin.unifi.it
gbartoli@dicea.unifi.it
cborri@dicea.unifi.it

Department of Energetics S. Stecco,

University of Florence,
Florence, Italy
alessandra.borsani@pin.unifi.it

Abstract - The high speed trains in Europe develop

rapidly, which creates a need to investigate the
accompanying issues in detail. One of main
problems are wind loads on structures induced by a
passing train. The pressure fluctuations on the
structures induced by the running train are
amplified on noise-barriers, as well as on all those
structures which are particularly sensitive to these
kind of actions (e.g. deformable or brittle
structures). Some of full-scale measurement data
are already available. However these data refer only
to specific cases and can not identify the whole
nature of the problem. Therefore, the evaluation of
wind loads on structure is not always taken into
consideration neither by Specifications nor by
National and International Codes, and it is advisable
to carry out model tests. The main purpose of this
study is to try to determine an optimal experimental
set-up,
which
would
provide
a
correct
representation of full-scale problems in model.
Firstly, a known full-scale case was compared with
a model at the scale of 1:50, and the results were
found to be in good agreement. Consequently, the
same technique was adopted for further
experiments on the model of a new high speed
train-station in Reggio Emilia, Italy. The model
consists of a train moving on a track with a speed of
over 22 m/s. During the experimental campaign
measurement of induced over-pressures have been
performed.

x use of the optimal setup for the high speed railway

station of Reggio Emilia.

Keywords Pressure measurements, moving train,

high speed railway.

In order to validate the measuring system the

available results of in situ measurements were
analyzed.

1 Scale considerations
Model tests are often performed in order to
understand
aerodinamical
problems;
but
the
transferability of the results to full-scale is not always
certain.
In this study a 1:50 scale test simulates a real case
in which the train is moving at a velocity of
approximately 83 m/s. In order to maintain the Reynolds
number similarity the model train should be 4150 km/h
fast, which is not possible from the point of view of the
reliability of the test. Such a high velocity is not
desiderate also because it is necessary to remain under
the critical Mach number (Ma<0.3), which allows us to
consider the air as an uncompressible fluid.
The solution adopted is to reach a train velocity
which is high enough to induce a significant
overpressure. It should be also possible to measure the
chosen velocity with the available instruments. Changing
the shape of the train with a sort of technical roughness
allowed us to influence the flow structure to some
amount. The best experimental set-up was chosen by
comparing the obtained results with in situ
measurement.

2 Comparison between full-scale and

model case

2.1 Description of the measurement

Introduction
The purpose of this study is to determine the
overpressures on the roof and on the noise-barriers at
the new high-speed railway station of Reggio Emilia
(Italy), which are induced by the running through of a
train.
The following procedure has been accepted:
x gathering of available full-scale data,
x manufacturing of a model at the scale of 1:50
related to one of the real cases described in the
preceeding point (from now on called 'case 1') ,
x manufacturing of a lauching system which
simulates the moving of the train,
x calibration of the measuring system varying the
speed and the shape of the train,

59

Data referred to pressure measurements obtained on

barriers adjacent to the platforms during the passing of a
high speed train through the station . The pressure is
nondimensionalized as follows:

cp

pbarrier  p air

U air 2 vtrain 2

The train velocity measured with a radar is about 83

m/s.

3.8 cm

6 cm
7 cm

3.8 cm

24 cm
Figure 5. Model 3

Figure 1. Pressure measurement in situ

6 cm
4.8 cm
35 cm
Figure 6. Model 4

Figure 2. Scaled model

The preceeding case was reproduced at the scale of
1:50. The pressure taps collocated in the same position
of the ones in the real case have been connected to a
Setra Systems 239 transducer with range of measure 6
mms H2O. On this model a rail and an elastic propulsion
system was built, which allows to launch the model of a
train up to around 26 m/s. The velocity of the train was
measured by positioning two lasers to the mutual
distance of 2 m and simultaneously recording their
signals. The train velocity was detemined from the
distance between the lasers and from the time which is
needed for the train to pass by them.
Measurements were performed by varying the speed
in the interval between the 10 and the 26 m/s. Also the
shape of the train was varied in order to approach the
similarity with the full-scale.
The adopted shapes of trains are presented in
Figures 3, 4, 5 and 6.

Figure 7. Train - model n.1

The best results were obtained by using the model
n.4. Figure 8 shows the overpressure generated by the
different train models compared with full-scale.
Therefore, the experimental setup applied on the model
n.4 was used also for further experiments on the highspeed station of Reggio Emilia.

2.2 Results

The obtained results show as the passing through of

a train produces an overpressure, followed by a
depression, which is a typical for this phenomenon. The
time history of a pressure tap record is presented in
Figure 8.

3.8 cm
20 cm

3.8 cm

Figure 3. Model 1

Train passage

3.8 cm
3.8 cm
24 cm

Figure 8. Time history of a pressure tap during the

train passge, X axis=time Y axis= Cp

Figure 4. Model 2

60

Cp max
0.16

cp [-]

0.14
real case

0.1

model n.1
model n.3

0.06

model n.4

0.04
0.02
0
40

60

80
v [km/h]

100

2
Noise barrier

3
4

The procedure of measuring the train speed and the

system which simulates the moving of the train is the
same as in the preceeding case.

model n. 2

0.08

Figure 11. Sketch of the pressure taps position (1 on

the steel structure, 2 and 3 on the glasses, 4 on the
noise barrier)

0.18

0.12

Steel structure

glass

It seems that the cp values do not depend

significantly on the train speed.
The values vary instead with the different shapes of
train. The shape that optimally reproduces the results of
full-scale measurements is the n.4. Therefore it will be
used during the test on the Reggio Emilia station. The
maximal cps values obtained with different train models
are presnted in Figure 9, together with the reference
value obtained in the full-scale test.

3.2 Results

120

Figure 9. Cp max for differents model shapes, the

outlined line is the full-scale value

3 Case of interest
Once the optimal experimental setup was
determined (case 1), the test for Reggio Emilia train
station were carried out.

3.1 Description of the model

The structure of the station is made of 13 steel portal

that repeat in a total lenght of 400 m in full-scale. The
steel portals have a mutual distance of about 100 cm
and between them there are glass walls.

The overpressure was evaluated in the positions

shown in Figure 11. Position n.1 refers to a pressure tap
positioned under the portal (pos 1a) and on the side of
the portal (pos 1b). On the glasses the taps are in the
inferior part (position 2 and 3) while on the barrier it is on
the side toward the rail. The results for the various
positions is shown in Figure 12, 13, 14 and 15.

3.2.1

Position 1a

test

cp_max1

cp_min1

v [m/s]

v [km/h]

a2

0.058

-0.071

11.308

40.708

a3

0.044

-0.038

18.208

65.549

a4

0.054

-0.022

20.082

72.295

a5

0.045

-0.029

17.133

61.678

g2
0.039
-0.036
18.000
Table 1. Cp measured in case 1a

64.800

0.015

0.01

0.005

-0.005

-0.01

Figure 10. Reggio Emilia station

-0.015

The main interest was to evaluate the pressures

produced by a running train on the noise barriers and on
the roof of steel and glass. The pressure taps have been
positioned in the points marked in the Figure 11.

Train passage
1.4

1.5

1.6

1.7

1.8

1.9

2.1

Figure 12. Time history of a pressure measurement

during the train passage, X axis=time Y axis= Cp
It can be seen that the passage of the train
generates a small overpressure followed by a
depression. However the entity of the variation is
comparable to the sensibility of the pressure transducer.

3.2.2
test
d2

61

Position 1b
cp_max1 cp_min1
0.036

-0.034

v [m/s]

v [km/h]

19.955

71.837

e2
0.047
-0.030
18.329
Table 2. Cp measured in case 1b

65.985

0.3
0.25
0.2
0.15
0.1

0.03

0.05

0.02
0

0.01

-0.05
-0.1

Train passage

-0.15

-0.01

-0.2

0.5

1.5

2.5

3.5

4.5

-0.02

-0.03

1.3

1.4

1.5

1.6

1.7

1.8

1.9

2.1

Figure 13 Time history of a pressure measurement

during the train passage, X axis=time Y axis= Cp
The overpressure caused by the train passage is
very small.

3.2.3

Position 2 and 3

test

v [m/s]

v [km/h]

h2
0.034
-0.038
19.865
Table 3. Cp measured in case 2

71.514

cp_max1 cp_min1

0.03

0.02

0.01

-0.01

-0.02

-0.03

Train passage
2

2.1

2.2

2.3

2.4

2.5

2.6

2.7

Figure 14 Time history of a pressure measurement

during the train passage, X axis=time Y axis= Cp
The overpressure generated by the train on the
glasses was not registered.

3.2.4
test

Position 4
v [m/s]

v [km/h]

18.798

67.673

f3
0.315
-0.239
22.205
Table 4. Cp measured in case 4

79.940

f2

Figure 15 Time history of a pressure measurement

during the train passage, X axis=time Y axis= Cp

Train passage

cp_max1 cp_min1
0.293

-0.153

The passage of the train can be clearly seen from

the second 1.7 to 1.8 (Figure 15). It has the typical
course of an overpressure followed by depression.

4 Conclusions
The obtained results show that the pressure effects
generated by the passing of trains are very low both on
the superior part of the steel structure and on the
glasses of the roof. In these positions the measured
values are near the value of the sensibility of the
pressure transducer. On the barrier instead the
overpressure is appreciable. The results of the adopted
measuring system are acceptable and it is possible to
validate these results with the results of full-scale
measurements. In order to have a good agreement with
the reference values it was necessary to modify the
shape of the train model. The maintenance of the
geometric staircase can cause errors because the
Reynolds number similarity was not obtained.
This study is to be considerd as a first attempt to
resolve this kind of problems. The aspects regarding the
correct scaling of the this phenomena are to be further
investigated.

Acknowledgments
All the test are realized in the Boundary Layer Wind
Tunnel from CRIACIV (University of Florence, Italy).

References
Niemann H. J. and Hlscher N., (2003) "Messung der
aerodynamischen
Einwirkungen
aus
der
Zugvorbeifahrt am Trogbauwerk Ittembach"
Macciacchera I. U. and Ruck B., Pressure Fluctuations
Induced by Road Vehicles in Ambient Air A Model
Study-.
Shaw C. T., Garry K.P. and Gress T., (2000) Using
singular system analysis to characterise the flow in
the wake of a model passenger vehicle. Journal of
Wind Engineering and Industrial Aerodynamics, 85
(2000) 1-30.
Kanda I., Uehara K., Yamao Y., Yoshikawa Y. and
Morikawa T., (2006) A wind-tunnel study on
exhaust gas dispersion from road vehicles- Part I:
Velocity and concentration fields behind single
vehicles. Journal of Wind Engineering and
Industrial Aerodynamics, 94 (2006) 639-658.

62

Wind Tunnel Simulations of Pollution from Roadways

D K Heist*, S G Perry*, L A Brixey**, G E Bowkerg

*Atmospheric Sciences Modeling Division, ARL/NOAA, MD-81, USEPA, RTP, NC 27711, USA
**Alion Science and Technology, P.O.Box 12313, RTP, NC 27709, USA
g
Atmospheric Modeling Division, NERL/EPA, MD-81, USEPA, RTP, NC 27711, USA
heist.david@epa.gov

Abstract - A wind tunnel study has been

conducted to examine the influence of roadway
configurations and nearby structures on the flow
and dispersion of traffic-related pollutants within
a few hundred meters of the roadway.
Experiments involved smoke visualization and
tracer concentration measurements from a
modeled roadway as well as flow characterization.
A finite length source was used in the
experiments, but by employing lateral shifts in the
measured results and superposition, results from
an infinite line source were constructed.
Measurements of ground-level concentrations
showed a remarkable similarity among the cases
with sound barrier and depressed roadway
configurations. The data suggest that simple
adjustments to the flat terrain concentration
distributions could be used to describe that of
the more complex cases.
Key words Near-road pollution, line sources.

Introduction
Studies have shown that long-term exposure to
traffic-related pollutants is an important risk factor for
certain respiratory problems and for mortality (e.g.,
Nitta et al., 1993, Finkelstein et al., 2004). In the U.S.
there is increasing concern specifically for the many
people that live, work and attend school in close
proximity to major roadways. Many applied dispersion
models in use currently were developed for simplified
roadway scenarios that do not include the complex
geometries often found surrounding urban highways.
Hosker et al. (2003) found that guidance for the
application of such models in these situation has not
been adequate. In particular, they note that problems
can be anticipated in applying Gaussian dispersion
models in low wind situations, in areas with complex
highway configurations, and at urban intersections.
While a number of wind tunnel studies (e.g.,
Hayden et al. 2002, Kastner-Klein and Plate, 1999)
have examined the effects of urban street canyons
and intersections, there is a need to examine the
influence of roadway configurations and nearby
structures where urban buildings are not dominating
the overall flow. Therefore, the current wind-tunnel
study has been designed to consider these influences
on the flow and dispersion of traffic-related pollutants
within a few hundred meters of the roadway. This
paper reports on the five selected configurations
shown in Figure 1 (all with a model six-lane, divided
highway at a 1:150 scale). They included flat terrain
with no structures, flat terrain with sound barriers and
depressed roadways with vertical or sloping walls.
Experiments involved smoke visualization and tracer
concentration measurements from a uniform line-type
source as well as flow characterization.

1 Experiment
Experiments were conducted in the US EPAs
Fluid Modeling Facility meteorological wind tunnel.
The test section is 370 cm wide, 210 cm high, and
1830 cm long. The air speed in the test section was
fixed at 4.7 m/s at a height of 165 cm.
To simulate the traffic along a six lane highway, a
280 cm long and 24 cm wide roadway was installed in
the wind-tunnel with the roadway perpendicular to the
wind direction. Six parallel line-source segments were
placed in the center (laterally) of the roadway oriented
parallel to the axis of the highway.
Each line
contained approximately 30

Figure 1. Elevation view showing cross-sections

through the various roadway configurations studied.
Flow is from left to right.

63

The above equation was fit to the data (varying d

and z0) until the computed value of u* matched the
value found from the turbulent shear stress
measurements. The fit was performed between z =
150 and 400 mm, in the region of constant shear. The
best fit parameters were found to be z0 = 5.2 mm, d =
54 mm, and u* = 0.3 m/s. At a scale of 1:150, this
corresponds to z0 = 0.78m and d = 8.1m.
The tracer used in this study is high-purity ethane
(C2H6; CP grade; minimum purity 99.5 mole percent),
which, with a molecular weight of 30, is only slightly
heavier than air. In combination with the high
turbulence level at the release point and a total
release rate of only 1500 cc/min, this tracer may be
regarded as neutrally buoyant.
All samples were drawn through Rosemount
model 400A hydrocarbon analyzers (flame ionization
detectors). The output signals from the analyzers
were digitized at the rate of 20 Hz (each unit) for 120
seconds and were processed on a personal
computer.
The corrected concentrations (with background
subtracted) were normalized to give the nondimensionalized concentration, F= CUr/(Q/LxLy),
where C is the corrected concentration (a fraction by
volume), Ur is the reference wind speed (equal to
-1
2.46m.s , measured at a full-scale equivalent height
of 30m), Q is the volumetric effluent rate (1500 cc/min
of ethane), Lx is the alongwind dimension of the
roadway segment (12cm) and Ly is the lateral length
of the source segment (24cm).

Figure 2. Source construction. Three plates of

aluminum make up the box: the bottom plate, with
two holes to connect to the source gas, the middle
plate, hollowed out to form the perimeter of the box,
and the top plate with six lines of small holes forming
the emission lines. The box measures 24 cm x 48
cm with a depth of approximately 3 cm.

small holes (approximately 0.1 cm diameter) spaced

1.5 cm laterally on center with the holes in subsequent
lines staggered to provide a near continuous release
laterally (see Figures 2 and 3 for source construction
and geometry). This source road segment is 24 cm in
the along-wind direction and 48 cm in the lateral or
crosswind direction.
To enhance the near-road
turbulence, 0.6x0.6x1.2 cm blocks were placed
approximately 0.1cm upwind of each release hole.
The origin of the coordinate system for the study is
at the center of the roadway model at the tunnel floor,
with x positive in the streamwise direction, y along the
axis of the roadway, and z vertically upward. Laser
Doppler Velocimetry (LDV), was used to characterize
the boundary layer.
The boundary layer used in this experiment was
designed to simulate flow in an urban area. The
tunnel ceiling was adjusted along the length of the test
section to compensate for blockage effects and allow
for a non-accelerating freestream flow. The standard
logarithmic velocity profile was used to assess the
roughness length (z0), displacement height (d) and
friction velocity (u*):

U
u*

2 Methods
After measuring the pattern of downwind
concentration from the finite line segment, the infinite
line source results can be constructed by
superimposing the results from the finite line source
employing a lateral shift in the source location, as
follows:

C x , z 

C x, y  iL , z 

i f

fls

where C(x,z) is the predicted concentration from an

infinite line source and Cfls is the measured
concentration based on a finite line source of width,
Ly. An example of this calculation is shown in Figure
4, where a lateral profile of the measured ground-level
concentrations at x=20H is shown for the case where
a sound barrier is installed upwind of the roadway.
The measured ground-level concentrations are shown
as open diamonds. The filled diamonds represent
those measurements shifted laterally by multiples of
the source width (1Ly and 2Ly). The red line is the
summation of the five profiles shown and
demonstrates that there is a central portion which is
uniform in the y-direction, as if it was measured from
an infinitely wide source.

1 z d

ln
N z 0

where N was taken to be 0.4.

Figure 3. Source hole spacing in the top plate.

Six lines of approximately 30 holes each form the
source. All dimensions are in millimeters.

64

10

10

-1

10

-2

10

-3

10

-4

the same three cases examined in Figure 5. The

enhanced mixing of the sound barrier and depressed
road cases is evident in these figures as well,
especially close to the source. Moving downstream
the spread of plumes begin to look the same, as can
be seen in the vertical concentration profiles shown in
Figure 7. The profiles are shown at x = 4H and 40H,
corresponding to 24m and 240m downstream of the
middle of the source, full-scale. At 4H, there are
significant differences between the flat terrain case
and others, but by 40H those differences have
diminished. Although, even at 40H, the flat terrain
case shows approximately 25% higher concentration
at ground level than the other cases.

-40

-20

Y/H

20

40

Flat terrain

Figure 4. Superposition of results from a finite linesegment source to create the effect of an infinite line
source. Open diamonds are measured results; filled
diamonds represent measured results shifted laterally
by multiples of the source width; red line is the
summation of all five profiles.

4
2
0

3 Results

65

-12

-6

Y/H

12

Sound barrier (6m tall) upwind of roadway

Z/H

4
2
0

-12

-6

Y/H

12

Depressed road (6m), 90 degree walls

Z/H

Before examining the infinite line source

concentration patterns, it is instructive to view the
measured plume as it emerges from the edge of the
roadway from its finite width source. Figure 5 shows
the plume (flow is toward the viewer) for three of the
cases studied: the flat terrain case with no sound
barriers, the flat terrain case with a sound barrier at
the upwind edge of the road, and a case with the
roadway depressed 6m (full-scale) below the
surrounding terrain and vertical walls connecting the
roadway to the surrounding terrain.
The
concentration contours show that the addition of the
sound barrier has the expected effect of enhancing
the mixing of the plume both vertically and laterally
compared to the flat terrain case. The depressed
roadway case also enhances the mixing relative to the
flat terrain case, but to a lesser degree than the sound
barrier case. Although the vertical extent of the plume
is not as great in the depressed road case (relative to
the sound barrier case), the lateral mixing appears to
be comparable.
After superposition of the results as described
above, one can view slices in the X-Z plane to
examine the behavior of the plume that would
disperse from an infinitely long source. Figure 6
shows these cross-sectional views of the plume for

Chi
50
20
10
5
2
1
0.5
0.2
0.1

Z/H

Non-dimensional concentration

10

2
0

-12

-6

Y/H

12

Figure 5. Concentration contours from finite linesegment sources measured at the downwind edge of
the roadway for three cases. The flow in these
figures is toward the viewer.

Flat terrain

Z/H

10

0
0

10

X/H

20

30

40

Sound barrier (6m tall) upwind of roadway

Z/H

10

0
0

10

X/H

20

30

40

Depressed road (6m), 90 degree walls

Z/H

10

0
0

10

X/H

20

30

40

Chi
100
50
20
10
5
2
1
0.5
0.2
0.1

Chi
100
50
20
10
5
2
1
0.5
0.2
0.1

Chi
100
50
20
10
5
2
1
0.5
0.2
0.1

Figure 6. Concentration contours for an infinite line source. Flow is from left to right.

4 Discussion
The departures from flat terrain, though
significantly different from each other geometrically,
produce similar results, especially at ground level and
downwind of the immediate vicinity of the source.
This can be seen in the Figures 6 and 7, but also in
an analysis of the plume heights and the ground-level
concentrations downwind of the source.
The height of the plume can be characterized by
the plume centroid, z , defined:
f

zCdz
0
f

Cdz
0

where the upper limit of the integral in practice is

the point where the concentration reaches zero.
Figure 8 shows that, beyond approximately x = 10H

Figure 7. Vertical concentration profiles.

66

(60m, full scale) the plume heights are remarkably

similar for all cases except the flat terrain case. Close
to the source, differences between the cases are
more pronounced, with the deepest plume being the
case with sound barriers on both sides of the
roadway.

Figure 10. Inverse concentration as a function of

downwind distance. Legend as in Figure 8.
Figures 11 and 12 show the effect of shifting the
flat terrain results upwind a distance of 7H. The
agreement between all of the profiles with the shift is
very good. It should be noted that this is a limited
range of scenarios and does not include any variation
in wind direction from flow perpendicular to the road.
Further research is needed to understand how
variations in barrier height and placement, roadway
depth, roadway width and wind direction may affect
the utility of this finding.

Figure 8. Plume centroid heights versus downwind

distance.
Of particular interest for the study of exposure to
pollution related to traffic are the ground-level
concentrations downwind of a roadway. Figure 9
shows that the ground-level concentration patterns
are very similar for all cases except the flat terrain
case. The inverse of the concentration is plotted in
Figure 10 as a function of downwind distance.
Following the analysis method of Briggs et al. (2001),
the ground-level concentrations were expected to vary
-1
roughly with x . Interestingly, except immediately
downwind of the roadway structures, the data suggest
that a simple upwind shift in the flat-terrain results
may be used to describe the variation in concentration
in the other cases.

Figure 11. Concentration as a function of

downwind distance, showing the effect of effectively
shifting the source location for the flat case a distance
of 7H upwind. Legend as in Figure 8.

Figure 9. Concentration as a function of downwind

distance. Legend as in Figure 8.
Figure 12. Inverse concentration as a function of
downwind distance, showing the effect of shifting the
source location for the flat case a distance of 7H
upwind. Legend as in Figure 8.

67

5 Summary
These measurements demonstrate the use of a
finite, line-segment source to simulate the effect of an
infinite line source in modeling emissions from
various roadway configurations. The configurations
studied include flat terrain, flat terrain with sound
barriers upwind of the roadway, flat terrain with sound
barriers on both sides of the roadway, depressed
roadway with vertical walls and depressed roadways
with sloping walls.
Concentration measurements show that soundbarriers and depressed roadways produce remarkably
similar ground-level concentration patterns downwind,
which are significantly lower than flat terrain results.
The concentration patterns suggest that an effective
shift in the flat terrain concentration results a distance
of 7H upwind would describe the measurement results
well for the barrier and depressed roadways cases.
The utility of this approach needs to be investigated
with addition roadway configurations and wind
directions.

Acknowledgments
The authors would like to thank Richard Baldauf,
Vlad Isakov, and Tom Pierce for their advice and
support.

Disclaimer
The research presented here was performed
under the Memorandum of Understanding between
the US Environmental Protection Agency (EPA) and
the US Department of Commerces National Oceanic
and Atmospheric Administration (NOAA) and under
agreement number DW13921548. This work
constitutes a contribution to the NOAA Air Quality
Program. Although it has been reviewed by EPA and
NOAA and approved for publication, it does not
necessarily reflect their policies or views. Mention of
trade names or commercial products does not
constitute endorsement or recommendation for use.

68

References
Briggs, G.A., Britter, R.E., Hanna, S.G., Havens, J.A.,
Robins, A.G. and Snyder, W.H. (2001) Dense
gas vertical diffusion over rough surfaces: results
of
wind-tunnel
studies,
Atmospheric
Environment, vol. 35, pp. 2265-2284.
Finkelstein MM, Jerrett M, Sears MR. (2004) Traffic
related air pollution and mortality rate
advancement periods. Am J Epidemiol. Vol. 160,
no. 2, pp.173-177.
Hayden, R.E., Kirk, W.D., Succi, G.P., Witherow, T.
and Bouderba, I.
(2002) Modifications of
Highway Air Pollution Models for Complex
Geometries Volume II: Wind Tunnel Test
Program US Department of Transportation,
Federal Highway Administration Report, FHWARD-02-037.
Hosker, R.P, Jr., Rao, K.S., Gunter, R.L., Nappo, C.J.
and Meyers, T.P. (2003)
Issues affecting
dispersion near highways: light winds, intra-urban
dispersion, vehicle wakes, and the ROADWAY-2
dispersion model, US Department of Commerce,
NOAA Technical Memorandum OAR ARL-247.
Kastner-Klein, P. and Plate, E.J. (1999) Wind-tunnel
study of concentration fields in street canyons.
Atmospheric Environment 33, 3973-3979.
Nitta, H., Sato, T., Nakai, S., Maeda, K., Aoki, S. and
Ono, M., (1993) Respiratory health associated
with exposure to automobile exhaust I. Results of
cross-sectional studies in 1979, 1982, and 1983.
Archives of Environmental Health 48, pp. 5358.

How rough is rough? Characterization of turbulent fluxes within and

above an idealized urban roughness
M. Schultz, B. Leitl, M. Schatzmann
Meteorological Institut,
Hamburg, Germany
Merike.schultz@zmaw.de

Abstract Results of an extensive experimental

study will be presented that was caried out at the
Environmental Wind Tunnel Laboratory (EWTL) at
Hamburg University. Measurements were made
between and above arrays of closely packed
roughness elements. Three cube configurations
with and without added roofs were investigated.
The measurements focused on vertical momentum
fluxes which were subsequently used to determine
the vertical extent of the roughness sublayer (RS)
and inertial sublayer (IS) of the urban boundary
layer.

shows the basic experimental set-up in the wind tunnel,

Figure 2 the roughness configurations.
Within this study three different configurations of
cube arrays were investigated. The packing density
(plan area of roughness elements divided by the total
surface area) was kept constant at p = 0.25 for all three
configurations. Flow measurements were carried out
with an optical LDA fibre probe of size 85 mm (Dantec) with a focal length of 800 mm. A reference wind
speed was measured with a pitot tube at 2 m height
above ground mounted at the probe positioning system.

Key words roughness sublayer, vertical momentum

flux profiles, scaling

Introduction
The urban surface layer possesses a pronounced
layer, the roughness sublayer (RS), which is highly influenced by the local canopy and characterized by a
strong heterogeneity of the flow field. E.g. Cheng and
Castro (2002) showed in a wind tunnel study, that especially above very rough surfaces the vertical extent of
the RS is not negligible. It covers a big portion of the
lower part of the surface layer, usually on the expanse
of the inertial sublayer (IS). Furthermore, for specific
roughness it can be questioned if there exists an IS at
all. The logarithmic wind profile is not valid within the
RS. This makes it difficult to determine for example the
roughness length z0 as a parameter describing surface
characteristics. To replicate the effect of the urban
roughness, e.g. in mesoscale numerical models,
roughness parameterizations are needed. However, to
improve an appropriate roughness parameterization
detailed information of the flow field within the canopy
and the overlying RS and IS is required.

1 Experimental Set up
This study presents first results of an extensive
measuring campaign which focused on the determination of vertical momentum fluxes in and directly above
an idealized urban roughness. LDA Flow Measurements were carried out at the large boundary layer wind
tunnel Wotan. The wind tunnel has a total length of 25
m with a test section which is 4 m wide and 2.75 m high
and contains a flow establishment section of about 18
m length. The wind tunnel is equipped with an adjustable ceiling allowing 0.5 m height extension of the test
section.
All together 59 rows of cubes were positioned in the
wind tunnel, covering the whole floor from the vortex
generators at the entrance into the flow establishment
section to the suction fan at the end of the testsection.
An intensive observation area where most of the measurements were made was located after row 43 in order
to ensure a sufficiently long upstream fetch. Figure 1

69

Figure 1: sketch of the experimental set up in the Wotan wind tunnel at the EWTL of Hamburg University.

1.1 Detailed description of the roughness

The first configuration (top of Figure 2) consists of

an aligned arrangement of sharp edged wooden cubes
of height hc = 125 mm. The packing density of the array
is p = 0.25 which corresponds to an aspect ratio w/hc =
1. For cubes, the frontal area density (area of roughness facing the wind in relation to the total surface area)
f is equal to p, i.e. f = p = 0.25.
In the second configuration complexity was increased by adding roofs of different shape (flat and
pitched roofs) on every second cube of the first array in
order to reach a more heterogeneous surface (centre of
Figure 2). The extension of the added (flat and pitched)
roofs from bottom to roof ridge is half of the cube height
(62.5 mm) giving a new total cube-with-roof height of hr
= 187.5 mm. An average cube height of the array can
than be defined to be ha = 156.25 mm. The alignment of
the pitched roofs varied alternately between gable facing the approach flow (0) and rotated by 90. The
packing density p was kept constant, but adding roofs
results in a decreased aspect ratio w/ha = 0.8 and an
increased frontal area density f = 0.32.

Figure 2: Left hand side Pictures of the three investigated configurations in the wind tunnel. Right hand
side: Profile measurement locations for the three investigated types of roughness. Top: aligned without
roofs (config 1), centre: aligned with roofs (config 2) and bottom staggered with roofs (config 3). The
coordinate system for all three configurations is defined as follows: X=0 between rows 46 and 47, Y=0 at
the wind tunnel centerline and Z=0 at the ground floor, posive into the upward direction.

70

The analysis of the measurements is still in progress

and not yet completed. Therefore only a small part of
results can be presented within this paper.
The main objective of this paper is the investigation
of the roughness sublayer (RS). In literature many findings about the vertical extent of the RS can be found
and may be summarized with a height range of 2-5
times the average building height (ha) (as, e.g., proposed by Raupach et al. 1991). Within this paper the
definition of the heights of RS and IS are adopted from
Cheng and Castro, 2002 (i.e. the height of the RS
equals the converging height of several profiles measured above the roughness and the IS is estimated from
the range above the RS where the variation of the averaged vertical momentum flux profile is <5%). The
method proposed by Cheng and Castro (2002) is feasible and gives good results. Other definitions of the vertical extent of the RS like e.g. the height where the
mean wind speed profile deviates from the log law do
not lead to more accurate results. As can be seen in the
upper plot of Figure 3 all profiles above the aligned
cube array without roofs converge roughly at 1.5 to 1.6
hc. The error bars indicate the scatter band. Cheng and
Castro reported for cube arrays with the same packing
density a slightly higher RS-height of 1.8 hc. Considering the inaccurate definition of a convergence height,
this is fairly good agreement. The IS can be estimated
between 2 and 2.5 hc (2.3 hc) given the above definition.
The bottom plot of Figure 3 shows all flux profiles obtained from the aligned cube array with added roofs
(configuration 2). The height Z above ground in Figure 3
is scaled with the average height ha. The extension of
the RS above the aligned cube array with rooftops is
increased to 2 ha while the IS does not exceed 2.4 ha.
No differences of the vertical extensions of the RS and
IS have been observed between the aligned and staggered arrangement (therefore a plot of the staggered
arrangement is not shown here).
Another dominant feature that can be observed in
the top plot of Figure 3 is a peak of turbulent fluxes at
cube height hc. This peak occurs at flux profiles measured behind the obstacle and is indicating the pronounced shear generated by the obstacles. A similar
peak can be observed at the bottom plot of Figure 3 for
the cube array with added roofs. The maximum peak
value of the normalized Reynolds stress above this heterogeneous cube field (configuration 2) has almost
doubled and is located at 1.3 ha above ground. Observed peaks in Reynolds stress flux profiles are also
reported from field studies, e.g. Rotach (1993) measured above Zrich (Switzerland) an increase of turbulent fluxes with height up to 2.1 ha. Similar findings are
reported by Oikawa and Meng (1995) from measurements at Sapporo (Japan), their finding is an increasing
shear stress up to 1.5 ha with a subsequent decrease of
shear stress at higher elevations. Feigenwinter (1999)
observed a peak stress at 2.1 ha above Basel (Switzerland) and Louka et al., 2000

71

cube array: aligned without roofs

3.5
3
2.5

Z / hc

2 Experimental Results and discussion

1.5
1
0.5
0
-0.25 -0.2 -0.15 -0.1 -0.05

U'W' / U ref2

0.05

0.1

0.05

0.1

cube array: aligned with roofs

3.5
3
2.5

Z / ha

In the third configuration a staggered arrangement

of the cube array similar to configuration 2 was chosen
to access a more realistic urban roughness. The packing density p and frontal area density f are same as
for configuration 2.
The roof patterns are indicated at the right hand side
in Figure 2 as well as the measurement locations of the
accomplished vertical flux profiles.

1.5
1
0.5
0
-0.25 -0.2 -0.15 -0.1 -0.05

U'W'/U ref2 [-]

Figure 3: Measured vertical flux profiles at all indicated

measurement locations of Figure 2, for the aligned cube
arrays; top: without roofs (configuration 1), bottom: with
added roofs (configuration 2). Errorbars indicate scatter
band.
found a peak at roof level for their more idealized field
location of industrial buildings. Measurements behind
obstacles within a cube array carried out in a water tunnel by Mac Donald (2002), showed a peak at cube
height too. This leads to the conclusion that the peak of
Reynolds stress is directly linked to flow separation at
the upper edge of the obstacles. Observed peaks
above roof level depend on the defined averaged height
of locations with varying obstacle heights. The magni-

tude of the turbulent shear stress peak depends on the

roof shape, as can be seen in Figure 4, where profiles
of configuration 2 measured behind different roof types
are compared. Figure 4 shows that the pitched roof 90
(blue triangles) generates the largest peak of shear
stress. Similar results were found in wind tunnel studies
of 2D street canyons with varying roof shapes by
Rafailidis (1997) or Kastner-Klein (2004). Rafailidis,
found increased turbulence intensities for pitched roof
arrangements and Kastner-Klein et al. showed that
pitched roofs produce more turbulent kinetic energy
(TKE) than flat roofs which influences even the pattern
of the street canyon flow. The shear stress profile obtained behind the cube without roof (circles in Figure 4)
shows first the peak at cube height accompanied by a
slight decrease of the turbulent fluxes.

4
3.5

the obstacle. Merely the peak value is slightly higher for

the heterogeneous roughness indicating a larger
roughness effect. Above 1.5 hpeak the profiles scatter
more but all lie within the scatter band, so that no further conclusions can be drawn. In contrast to this, profiles measured at crossings (bottom of Figure 5) show
differences at roof (or peak) level. Profiles measured in
and above the cube array without roofs are constant
with height, i.e. they are not affected anymore by the
obstacles. For the configurations with roofs the aspect
ratio is decreased and profiles obtained at crossings
are affected by turbulence created by the roofs. Profiles
from the staggered array are influenced by the previous
obstacle. However, only minor differences can be observed above 1.5 hpeak. It seems that either the roughness is not rough enough to produce strong differences
in the vertical flux profiles or the flow may be not in
equilibrium despite of the long fetch.
Therefore, in another series of measurements it was
investigated to which degree the flow in the RS and IS
is in equilibrium with the underlying roughness structure. This was done by using different sets of spires
(vortex generators) at the wind tunnel entrance. If the
flow in the RS and IS above the intensive measurement
area would not change, it could be concluded that the
flow is entirely dominated by the underlying roughness
structure (which covered the whole wind tunnel floor).
During the measurement campaign three different types
of approach flow were created by using a set of seven
small spires of triangle shape, a set of three large
spires and no spires at all.

cube array: aligned with added roofs

profile 2 pitched roof 0
profile 9 cube
profile 16 flat roof
profile 23 pitched roof 90

Z / ha

2.5
2

1.5

3.5

0.5

scaling with peak height

config1: profile 4 cube
config2: profile 16 flat roof
config3: profile 10 flat roof

U'W'/U ref2 [-]

0.05

0.1

2.5

Z / hpeak

0
-0.25 -0.2 -0.15 -0.1 -0.05

Figure 4:vertical turbulent flux profiles measured

behind cubes with varying roof shape within and above
the aligned array with roofs (configuration 2).

1.5

The question remains, how to scale data accomplished over various surfaces (or cube fields) in order to
make them comparable to each other. Basis information
is needed which involves the characteristics of the individual roughness. Within this paper it is suggested to
use the peak height hpeak (which is roughly the height of
the surrounding dominating rooftops) as a basis of scaling the data. The reference wind speed Uref is scaled to
the wind speed at peak height hpeak. Adopting the proposed scaling to Figure 3 would lead to almost identical
expressions describing the extension of the RS for all
three investigated types of roughness, namely hRS = 1.5
hpeak (not shown as a figure). The vertical extension of
the IS is slightly higher for the homogenous roughness
(roughly 2.3 hpeak = 2.3 hc, compare above) than for
heterogeneous cube fields with roofs added. Figure 5
presents measured profiles scaled with hpeak behind the
cube (homogeneous array of configuration 1) and the
profiles measured behind a cube with flat roof of the
other two configurations. The profiles show surprisingly
similar shapes and collapse well especially between 1
hpeak and 1.5 hpeak, where they are directly influenced by

1
0.5
0
-0.25 -0.2 -0.15 -0.1 -0.05

U'W'/U ref2 [-]

Figure 5a.

72

0.05

0.1

4
3.5

scaling with peak height

config1: profile 11 crossing

config2: profile 21 crossing
config3: profile 12

cube array: aligned without roofs

profile 4, no spires
profile 4, 3 large spires
profile 4, 7 small spires

Z / hpeak

Z / hpeak

2.5
2

4
3

1.5
2

0.5
0
-0.25 -0.2 -0.15 -0.1 -0.05

U'W'/U ref2 [-]

0.05

0
-0.15

0.1

Figure 5b
7

Figure 5: Comparison of vertical flux profiles scaled with

peak height hpeak for the three investigated
configurations. Top: behind an obstacle of similar
shape, bottom: locations at crossings. Errorbars
indicate the scatter band.

3 Conclusions
First results of a measurement campaign with three
different types of roughness were presented. For selected profiles measured behind similar types of obstacles only minor differences can be found between a
surface consisting of obstacles with equal height and a
surface consisting of obstacles with heterogeneous
height. Clearly, more measurement locations and spatially averaged profiles have to be considered to complete the investigation.

73

-0.05

U'W'/U ref2 [-]

0.05

0.1

0.05

0.1

cube array: staggered with roofs

profile 10, no spires

profile 10, 7 small spires
profile 10, 3 large spires

Z / hpeak

Figure 6 shows measurements at the same location

with the three different approach flows. The profiles
start to depart beyond the scatter band at a height Z =
4-5 hpeak. The range of influence of the cube field without added roof shapes (configuration 1) can therefore
be estimated up to the height level 4 hpeak (= 4 hc). The
peak value of shear stress is amplified by use of the
large, broad spires, which produce a high turbulent flow.
This tendency can be observed as well at the flow
above the roughness but is not distinct enough through
the scatter band. It indicates that the homogeneous
cube array appears like a relatively smooth surface.
The staggered cube arrangement with added roof
shapes (configuration 3) seems to be less vulnerable to
different approach flows. The peak value stays constant
and the profiles follow the same shape up to the highest
level 4 hpeak (= 4.8 ha) within the scatter band. Nevertheless a trend of departing profiles can be observed at 2.6
hpeak (= 3.2 ha). These findings support the assumption
that the flow is in equilibrium with the underlying roughness at least up to the height of 2 hpeak, the highest level
where an IS can be found. An IS only forms in equilibrium flows.

-0.1

4
3
2
1

0
-0.15

-0.1

-0.05

U'W'/U ref2 [-]

Figure 6: Measurements carried out under varying

approach flow conditions for the aligned cube array
without roofs (top)and the staggered array with roofs
(bottom). Errorbars indicate the scatter band.
The next step of data analysis will concentrate on the
determination of appropriate values for basic roughness
parameters as, e.g. the roughness length z0 or the displacement thickness d0, possibly expressed as a function of morphometric properties (p and f). Finally, an
attempt will be made to extent the analysis to the shear
stress u* as well.

Acknowledgments
The authors gratefully acknowledge financial support from the DFG (Deutsche Forschungs Gemeinschaft) in the frame of the project DFG 04-174 Systematic Investigation of the Urban Boundary Layer

References:
Cheng, H. & Castro, I. P. 2002. Near wall flow over
urban-like roughness. Boundary Layer Meteorology,
104, 229259.
Feigenwinter, C., Vogt, R. & Parlow, E. 1999. Vertical
structure of selected turbulence characteristics
above an urban canopy. Theor. Appl. Climatol., 62,
5163.
Kastner-Klein, P., Berkowicz, R. & Britter, R. 2004. The
influence of street architecture on flow and
dispersion in street canyons. Meteorology and
Atmospheric Physics, 87, 121131.
Louka, P., Belcher, S. E. & Harrison, R. G. 2000.
Coupling between air flow in streets and the welldeveloped boundary layer aloft. Atmospheric
Environment, 34, 26132621.
MacDonald, R. W., Carter Schofield, S. & Slawson, P.
R. 2002. Physical modelling of urban roughness
using arrays of regular roughness elements. Water,
Air and Soil Pollution: Focus, 2, 541554.
Oikawa, S. & Meng, Y. 1995. Turbulence characteristics
and organized Motion in a suburban roughness
sublayer. Boundary Layer Meteorology, 74, 289
312.
Rafailidis, S. 1997. Influence of building areal density
and roof shape on the wind characteristics above a
town. Boundary Layer Meteorology, 85, 255271.
Raupach, M. R., Antonia, R. A. & Rajagopalan, S. 1991.
Rough-wall turbulent boundary layers. Appl. Mech.
Rev., 44 (1), 125.
Rotach, M. W. 1993a. Turbulence close to a rough
urban surface part I: Reynolds stress. Boundary
Layer Meteorology, 65, 128.

74

Effect of roofshape on unsteady flow dynamics in street canyons

J F Barlow

B Leitl

Department of Meteorology,
University of Reading,
Reading, UK
j.f.barlow@reading.ac.uk

Meteorologisches Institut,
University of Hamburg,
Hamburg, Germany
bernd.leitl@zmaw.de

Abstract - Ventilation of pollution from street level to

the boundary layer above is dependent on the flow
dynamics around buildings close to the source. As
instabilities in the shear layer at roof level are
responsible for intermittent mixing within streets,
sensitivity of the shear layer to roof shape is
hypothesised to significantly influence ventilation of
street canyon air. This study presents preliminary
analysis of wind tunnel simulations of mean and
unsteady flow dynamics for both flat and high pitched
roof street canyons at height to width ratios of
H/W=0.6 and 1. Mean flow patterns are consistent with
previous work in that they show an upward shift of
the recirculation centre for the pitched roof case. For
H/W=1, the high pitched roof street canyon exhibited
smaller skewness values in vertical velocity near roof
level, indicating less intermittent turbulent venting
than the flat roof case for the same aspect ratio.

Pascheke and Barlow (2005) used naphthalene

sublimation to derive mass transfer coefficients for street
canyons with flat, or low and high pitched roofs, given a
street level area source. Vertical mass transfer reached a
maximum for height to width ratio H/W=1 in the case of
flat roofs, whilst for high pitched roofs it occurred at
H/W=0.6 with a sharp drop as aspect ratio was increased
above this value. Mass transfer for the street canyon with
high pitched roofs was 88% of the flat roofed street for
H/W=0.6, dropping to 64% for H/W=1. In order to explain
these results, the present study was conducted to
investigate the change in flow dynamics due to changing
roof-shape and aspect ratio. Flow visualisation was used
to enhance understanding of the unsteady flow dynamics
in addition to measurement of the flow using Laser
Doppler Anemometry (LDA).

1 Experimental Set-up

Key words urban meteorology, urban morphology,

shear layer, intermittency, ventilation, street canyon.

Experiments were performed using the Blasius wind

tunnel at the University of Hamburg. The wind tunnel test
section was 1.5m wide, 1m deep and 4m long. A
boundary layer was generated upstream of the street
canyons. A combination of vortex generators and 325mm
fetch of roughness elements (Lego) were used to
generate the boundary layer. The roughness element
section ended 525mm upstream of the model street
canyons. Model scale of the boundary layer was
estimated to be 1:400. There was a logarithmic windprofile
up to a height of G ~ 300mm, which scaled up to
approximately 120m fullscale.

Introduction
Ventilation of pollution from street level to the
boundary layer above is dependent on the flow dynamics
around buildings close to the source. The conceptual
model of street canyon flow has been formed from classic
experiments highlighting the mean flow patterns in a flat
roofed street canyon and their dependence on aspect
ratio. However, due to high turbulence intensities in the
roughness sublayer, the turbulent component of the flow is
just as important as the mean flow in dispersing pollution.
Several authors have noted the intermittent nature of
roughness sublayer turbulence, and have studied the role
of the shear layer at the top of the urban canopy in
generating intermittent, coherent turbulent structures
which penetrate downwards, causing mixing of air within
the streets (Christen, 2005; Coceal et al., 2007). It is
hypothesised here that shear layer dynamics are sensitive
to roof shape and that local turbulent flow statistics will be
significantly altered, as well as overall mean flow patterns.

Two different street canyon aspect ratios were used

(H/W=0.6 and 1), using three different roof shapes (flat,
low pitched (26.6) and high pitched (45)). The street
geometry is shown in Table 1 note that height is
measured to the top of the roof, not to the gable. The
street canyons were placed perpendicular to the flow
direction, spanning the entire width of the wind tunnel.
Flow was held constant with reference velocity UG ~6 ms-1,
giving Reynolds number based on roof height ReH ~ 6600
for flat roofed street canyons, and 10000 for the high
pitched roof. Flow visualisation was performed for a
source in the first street canyon in an array of four, and
the third street canyon, to qualitatively ascertain
differences in flow dynamics close to a step change in
roughness. Windspeed was reduced to approximately 2
ms-1 to allow sufficient temporal resolution of dynamics by
the video camera, albeit compromising slightly on ReH, i.e.
flow near the canyon floor was not necessarily fully
turbulent. Laser Doppler Anemometry (LDA) was used to
measure the wind velocity near the top of the street
canyon. The flow measurements were limited to a
minimum height for each geometry (see Table 1), which
did not permit analysis of flow near street level, but
allowed focus to be made on roof level flow.

Wind tunnel street canyon studies have previously

shown that a change in roof-shape between flat and
pitched roofs causes a significant change in tracer
concentrations within the street (Rafailidis, 1997; KastnerKlein and Plate, 1999). Kastner-Klein et al (2004)
reviewed several studies including their own and
concluded that the dynamics within the street were
changed in the following ways: an in-street vortex does
not form in the presence of pitched roofs, thus inhibiting
exchange of air vertically; flow near the street is weak or
reversed in direction; and a recirculation forms instead at
roof level. Numerical simulation by Theodoridis and
Moussiopoulos (2000) presents a similar view of mean
flow patterns.

75

Roof shape

Roof angle
(degrees)

Flat
0
Low pitch
26.6
High pitch
45
Table 1. Dimensions of
pitched roofs.

Height
(mm)

Lowest LDA
measurement
mm
(ratio to H)

63
33 (0.52H)
78.7
49 (0.62H)
94.5
67 (0.71H)
street canyons for different

2 Flow visualisation
Flow within the first street canyon in the row for each
setup is presented in Figure 1 a to d. Visualisation was
achieved by placing a line source of smoke on the ground
along the canyon axis, equidistant between the buildings.
Laser light was reflected from a cylindrical mirror located
above the canyon, causing a longitudinal cross section of
the flow to be visible at the canyon centre, i.e. the (x,z)
plane at y=0. Snapshots of the flow are shown, which
highlight more intermittent turbulent structures ventilating
smoke with the air above. Wind vectors measured by LDA
are overlain on the instantaneous flow visualisation in
each case. Note that the flow is from right to left, and the
upward tilt of the vectors on the right hand side are due to
the displacement of the flow over the first building row.

Figure 1c. Flow visualisation for flat roof, H/W=1

Figure 1d. Flow visualisation for high pitch roof, H/W=1

Comparing Figures 1a and 1b, for the same aspect
ratio H/W=0.6 classically thought of as being in the wake
interference regime the centre of the mean flow
recirculation is displaced upwards in the high pitched roof
case. Although flow vectors for Figure 1b do not extend
down to the ground, visualisation showed a weak anticlockwise recirculation was present, although intermittent.
In comparing Figures 1c and 1d, the mean recirculation is
also displaced upwards for H/W=1, but flow at street level
was reversed for the high pitched roof case, i.e. the
smoke appeared to be predominantly transported to the
left side of the street, rather than the right. This suggests
that the transition to vertically stacked counter-rotating
vortices in a street canyon is not just a function of street
aspect ratio H/W (as suggested by Sini et al 1996, for flat
roof numerical simulations), but perhaps should also
include roof aspect ratio. In all cases, flow visualisation
showed intermittent acceleration of the recirculation, and
coherent ejections of smoke into the air above.

Figure 1a. Flow visualisation for flat roof, H/W=0.6. Flow

from right to left, yellow vectors indicate LDA
measurements of mean velocities. Snapshot of smoke
illuminated by laser light sheet, showing cross section of
street canyon flow at y=0.

3 Flow measurements
As the flow visualisation suggested intermittent vertical
ventilation of smoke into the air above, differences in flow
statistics between street canyon geometries are now to be
investigated. Preliminary results presented here focus on
Skewness in the vertical wind component, Skw, as it is
indicative of intermittent vertical motions caused by
coherent turbulent structures. For reference, the results of
Christen (2005) showed that profiles measured to one
side of a full scale street showed negative values within

Figure 1b. Flow visualisation for high pitch roof, H/W=0.6.

76

the street; a peak just below roof height; and a switch to

positive values some height above the street.

Figure 2a shows Skw values for the H/W=0.6 case for

the flat roof (cf. Figure 1a) the x axis is x/H, the x
position relative to the centre of the street, x = x x0,
scaled by roof height. Blue colours indicate
measurements within the street (i.e. z<H) and show
positive values on the leeward side of the street (x/H < 0).
This indicates that the mean flow pattern shows a
consistent updraught with intermittent, larger, upward
gusts. On the windward side, the pattern is reversed,
although Skw values are smaller in magnitude. These
distributions are consistent with the occasional
accelerations of the recirculation observed earlier. For z/H
= 0.9 to 1.0H a change occurs, with large negative values
of Skw at roof level across the street, particularly on
leeward side, close to the point of separation. These
strong downdraughts may be associated with downward
movement of the shear layer itself. By z = 1.3H, Skw is
positive and by z = 1.6H it approaches a constant value
across the street.

0
1.5

0.5

z=33mm 0.5H
z=40mm 0.6H
z=47mm 0.7H
z=54mm 0.8H
z=61mm 0.9H
z=73mm 1.1H
z=83mm 1.2H
z=98mm 1.5H
z=113mm 1.7H

Figure 2c. Skewness in vertical wind component, Skw,

for flat roof canyon, H/W=1.
1

0
Skw

-0.5

-1

z=92mm 1.0H
z=107mm 1.1H

-1

-1.5

Figure 2d. Skewness in vertical wind component, Skw, for

high pitched roof canyon, H/W=1.

0.5

-0.5

-1

In Figure 2b, H/W=0.6 for the high pitched roof case,

the patterns are similar at all heights, with larger negative
values of Skw seen near roof level on the leeward side of
the street. In Figure 2c, H/W=1 for the flat roof case, a
similar pattern is seen to Figure 2a, except that positive
skewness is seen at a slightly lower height, at 1.2H. In
Figure 2d the most notable difference is seen: the large
negative values of Skw near roof height on the leeward
side of the street are not observed. Without occasional
strong, downward injections of cleaner air from above, this
may explain the accumulation of smoke seen in Figure 1d
-1.5
near the leeward building.
Skw

-0.5

4 Preliminary Conclusions

z=87mm 0.9H
z=107mm 1.1H

-0.5

x'/H

0.5

z=97mm 1.0H

-0.5

z=122mm 1.3H

-1

z=77mm 0.8H

z=77mm 0.8H

-0.5

Figure 2a. Skewness in vertical wind component, Skw,

for flat roof canyon, H/W=0.6.

1
z=67mm 0.7H

z=67mm 0.7H

-1.5

-1.5
x'/H

1.5

0.5

It has been shown thus far that mean flow patterns for
street canyons of aspect ratios H/W = 0.6 and 1 with
either flat or high pitched roofs show some consistency
with previous results, viz.: the centre of recirculation is
moved upwards for the pitched roof case. Flow was still
observed at ground level, even for the H/W=1,high pitched
roof case, where it was seen to reverse weakly in the flow
visualisation, indicating a weak counter-rotating vortex.
This is different to Kastner-Klein et al (2004) who
observed no in-street vortex for the pitched roof case,
although their aspect ratio of H/W=1.3, with otherwise

-1

z=122mm 1.3H
-1.5
x'/H

Figure 2b. Skewness in vertical wind component,

Skw, for high pitched roof canyon, H/W=0.6.

77

-1

Skw

0.5

0
z=33mm 0.5H
z=40mm 0.6H
z=47mm 0.7H
z=54mm 0.9H
z=61mm 1.0H
z=73mm 1.2H
z=83mm 1.3H
z=98mm 1.6H
z=113mm 1.8H

-1.5

-1.5
x'/H

0.5

-1

-1

0.5

-0.5
-0.5

1.5

Skw

0.5

Institute of Meteorology, Climatology and Remote

Sensing, University of Basel.
Coceal, O., Dobre, A. and Thomas, G. (2007) "Unsteady
dynamics and organised structures from DNS over a
building canopy", in press
Kastner-Klein, P. and Plate, E. J. (1999). Wind-tunnel
study of concentration fields in street canyons,
Atmos. Environ., vol. 33, pp. 3973-3979.
Kastner-Klein, P., Berkowicz, R. and Britter, R. (2004).
The influence of street architecture on flow and
dispersion in street canyons, Meteorol. Atmos. Phys.,
vol. 87, pp. 121-131.
Pascheke, F and Barlow, J.F. (2005) Wind tunnel
modelling of scalar ventilation from idealised urban
surfaces, Proceedings of PHYSMOD 2005, 24-26
August, London, Canada.
Rafailidis, S. (1997). Influence of building areal density
and roof shape on the wind characteristics above a
town , Boundary-Layer Meteorol., vol. 85, pp. 255271.
Sini, J-F., Anquetin, S. and Mestayer, P.G. (1996)
Pollutant dispersion and thermal effects in urban
street canyons, Atmos. Environ., vol. 30, 2659-2677
Theodoridis, G. and Moussiopoulos, N. (2000). Influence
of building density and roof shape on the wind and
dispersion characteristics in an urban area: A
numerical study, Environ. Monit. Assess., vol. 65, p.
407-415

similar experimental conditions (ReH, upstream boundary

layer), could have led to more inhibited flow at street level.
An initial analysis of skewness of vertical velocity, as
an indicator of sweeps associated with shear layer
generation of coherent turbulent structures (Christen,
2005), showed a difference in flow dynamics at roof level
between the pitched and flat roof cases for H/W=1,
namely smaller skewness with a high pitched roof. This
may indicate a decrease in vertical exchange due to
unsteady, intermittent turbulent structures which still act in
the flat roof case and maintain overall mass transfer with
the air above. Further analysis of the data is required to
fully characterise the unsteadiness of the flow, and
differences in shear layer dynamics between the cases
presented.

Acknowledgments
JB acknowledges the generous use of facilities at the
University of Hamburg granted by M. Schatzmann and
support and advice from B. Leitl.

References
Christen, A. (2005) Atmospheric turbulence and surface
energy exchange in urban environments, PhD thesis,
.

78

Study of Flow Fields in Asymmetric Step-Down Street Canyons

Bhagirath Addepalli1,2

Eric R. Pardyjak2

D-3, Systems Engineering and Intergration Group

Los Alamos National Laboratory
Los Alamos, U.S.A
connoisseur_b@yahoo.com
Abstract The objective of the present work is to
better understand the physics associated with the
flow in complex street canyons. Specifically, an
Asymmetric Step-Down Street Canyon in which
the height of the downwind building (Hd) is less than
the height of the upwind building (Hu) is considered.
For the present study, the height of the upwind
building (Hu) was kept constant at 120mm and the
height of the downwind building was increased in
increments of ~ 0.08 H to span the range:
u

0.08Hd/Hu1. The footprints of the buildings were

square with a characteristic dimension W = 32mm
for all experiments. This provided a constant height
to width ratio for the upwind building of Hu/W=3.75
which was designed to correspond to a relatively tall
and slender building. Measurements were taken
along the center plane (X-Z plane) of the model
buildings for a normal incident flow. These
experiments were conducted for four different
canyon widths (S/W = 2.5, 2, 1.5, 1) to investigate the
effect of the seperation distance between the
buildings on the flow structure in the canyon and
turbulence statistics. In the present paper, the
streamtraces of the mean velocity field in the canyon
for all the above mentioned configurations have
been presented and the flow field for the canyon
width S / W = 2.5 has been analyzed. The results
indicate that the mean flow structure in the canyon
is highly dependent on the the seperation distances
(S/W) and the ratio of Hd / Hu.
Key words Step-Down Street Canyons, Flow
Structure, Vortex Cores, Saddle Points, Flow Topology

1. Introduction

Department of Mechanical Engineering

University of Utah
Salt Lake City, U.S.A
pardyjak@mech.utah.edu
Santiago, 2005), building alignment and geometry (Baik,
2000; Jiang, 2007; Addepalli & Pardyjak, 2005). For the
present work, the building geometry of the downwind
building has been varied, while the building footprint, street
canyon width and upwind building height remain constant.
The effect of these parameters on the flow patterns and
turbulence statistics in the canyon have been investigated.

2. Experimental Method
2.1 Flow Conditions and Setup
The experiments were performed at the Environemental
Fluid Dynamics Laboratory (EFD) at the University of Utah
at an atmospheric pressure of 640mm of Hg, temperature of
21C in a 7.9 m long boundary layer wind tunnel facility
having a working cross section of 0.61m 0.91 m. Lego
sheets with circular dimples of height 2mm lined the floor of
the tunnel and were used to produce rough walled turbulent
flow. The experiments were run at a free stream velocity of ~
7.2 m/s with a corresponding boundary layer depth () and
power law exponent () at the measurement location of /H ~
2 and ~ 0.205.

2.2 Canyon Models and Configurations

The model buildings were constructed using Legos with
square footprints. The plan area of the buildings was
maintained constant at LW = 3232 mm2. Two sets of
experiments were performed. The first experiment
investigated the flow field around an isolated tall building of
height H = 120mm. The second experiment investigated the
effect of adding a single downstream building of varying
heights in the wake of the tall building. For thse second set,
the upstream building height was kept constant at H =
u

120mm (/H ~ 2) and the height of the downwind building

was increased in increments of ~ 0.08 H to span the range:
u

The study of flow fields in modeled urban areas is

motivated by a number of environmental concerns such as air
quality, contaminant releases and pedestrian comfort. A vast
body of literature exists that characterizes flow over simple
building structures such as isolated buldings and idealized
street canyons. Tall buildings are a prominent feature of urban
downtowns, yet there is very little literature that describes the
physics that govern flow fields around tall buildings and their
interactions with other buildings. The aim of this paper is to
better understand the flow around tall buildings and their
interactions with other buildings and to contribute to the
existing literature on the subject.
Variation in height between the upwind and downwind
buildings of a street canyon significantly alters the flow
compared to a classical street canyon with buildings of
uniform height. The flow structure depends on a number of
parameters including the shear and turbulence in the incoming
flow (Davies, 1979), the seperation distance (Ohba, 1998;

79

0.08Hd/Hu1. This resulted in 12 cases for a given street

canyon width (S). These experiments were performed for four
different street canyon widths (S/W = 2.5, 2, 1.5, 1). This
resulted in a total of 48 cases for the second set of
experiments.

2.3 Notation
The following notations will be used to address the
different canyon configurations and the various parameters
used to understand the flow physics. Some of these
parameters are shown in Figure 1.
H
Hd
Hu
L
PQ
S
UH

Height of the isloted tall building = 120mm

Height of the downwind building
Height of the upwind building = 120mm
Along-wind length of the building = 32mm
Ground Shear Layer
Street canyon width
Velocity at Building Height

W
X
XP
XS
XV
ZS
ZSt1
ZSt2
ZV

Cross-wind width of the building = 32mm

distance of a given profile from the upwind building
distance of the ground shear layer from the upwind
building
Saddle point distance from the upwind building
Vortex core distance from the upwind building
Saddle point distance from the ground
First stagnation point height on the downwind building
2nd stagnation point height on the downwind building
Vortex core distance from the ground

turbulence fields. The spatial resolution of the of the final data

sets obtained was ~ 2.35mm.

3. Results and Discussion

3.1 Isolated Tall Building

The following example explains the notation used to

describe the various canyon configurations. S = 2.5W; (Hd /
Hu) = 0.88, would represent a street canyon configuration
with the buildings separated by S = 2.5W, and the ratio of the
heights of the downwind and upwind building being (Hd / Hu)
= 0.88.

Vortex
Core

Saddle
Point
Q

Hu

Ground
Shear Layer

ZV ZS

P ZSt1

Figure 2: Figure showing the streamtaces in the near-wake

of an isolated tall building of square cross-section with an
height to width ratio of 3.75
Figure 2 depicts the streamtraces in the wake of a tall
isolated building constructed using Tecplot. The prominent
features of the flow in the near wake are: 1) a primary vortex
below the rooftop downstream of the leeward face with a core
located at XV / H ~ 0.185 and ZV / H ~ 0.943 2) a saddle point
located at XS / H ~ 0.558 and ZS / H ~ 0.628 3) a shear layer
(PQ) emanating from the surface at XP / W ~ 1.0 4) a
counter-rotating vortex below the primary vortex.

Hd

XP
XV
XS

Figure 3 illustrates the physics behind the wake structure of a

tall building.

S
Figure 1: Figure showing the definition of the various
parameters used in the paper

2.4 Measurement Technique

2D PIV was used as the measurement technique. The
measurements were taken along the center plane (symmetry
plane; X-Z plane) of the model buildings. The flow was
seeded with olive oil particles generated using Laskin
Nozzles. The aerosols were illuminated with a 532nm
wavelength laser sheet generated using a 50mJ NewWave
Research (Fremont, CA) Solo PIV III Nd-Yag Laser. The
thickness of the laser sheet at the area of interest was ~
1.5mm. A 4.0MP (20482048pixels) CCD camera
manufactured by TSI Inc. (Shoreview, MN) having a frame
rate of 17fps was used in conjunction with a frame grabber for
image acquisition. A LASERPULSE synchronizer was used
to control the timing between the laser pulses and the camera
shutter open time through a PC desktop computer. Analysis of
the acquired image pairs was done using TSI INSIGHT6
analysis software. FFT based cross-correlation analysis was
performed on the conditioned image pairs by dividing them
into 6464 pixel interrogation regions. One thousand image
pairs were considered for computing the average velocity and

80

Y
X

Figure 3: Figure showing the flow around a tall building

The addition of a small downwind building only slighlty

modifies the flow field such that it is still dominated by the
upwind buildings wake. A close inspection of the streamline
patterns suggests that as the height of the downwind building
is increased from 0.08Hu to Hu, the flow transitions from a
wake dominated regime to a deep canyon skimming flow
regime. The deep canyon skimming flow regime can be
thought of as a skimming flow regime for tall buildings and is
charecterized by the presence of a counter-rotating vortex pair
in the canyon as shown previously (e.g. Baik et al., 2000).
The transition between regimes was found to occur when
Hd/Hu > 0.88 and is charecterized by the conversion of the
saddle point in the wake of the upwind building to a
stagnation point on the windward face of the downwind
building (see Fig. 5). For Hd/Hu > 0.88, the flow transitions
into skimming flow regime for deep canyons with the
formation of a primary and secondary vortex.

Figure 4: Figure showing an instantaneous snapshot of the

flow field in the near wake of a tall building
Davies et al. (1979) first demonstrated the presence of a
saddle point in the near wake of a tall building of height to
width ratio of 6. They called it an elevated free stagnation
point. Snyder and Lawson (1994) determined that the saddle
point appeared when the height of a square foot-print building
was increased from 1W to 2W. Ohba (1998) also detected an
elevated free stagnation point in the wake of a tall building.
Understanding the formation of this saddle is pivotal to
understanding the physics that govern the flow in the wake.
This saddle point can be understood to be a result of the
increased flow around the sides of the building compared to
that over the rooftop. As the height of a building is increased,
the flow begins to transition from flow over a 3D bluff body
immersed in a boundary layer to flow over a 2D bluff body.
This results in a large low pressure region being created
below the rooftop near the leeward face of the building. Since
the bulk of the flow goes around the sides and very little goes
over the top, the flow going over the sides moves towards the
low pressure region. At a certain height in the wake, the
horizontal and vertical motions come into equilibrium
resulting in the formation of a saddle point. Thus, above the
saddle point we find a primary vortex rotating in the
clockwise direction and below it we find a secondary vortex
rotating in the counter-clockwise direction. The secondary
vortex is highly unsteady and is not apparent in the avergaed
flow field (Fig. 2). But examining an instantaneous snapshot
of the flow as shown in Fig. 4 illustrates the unsteady nature
of the secondary vortex. The distance of the shear layer
emanating from the surface from the leeward face of the
building (XP) determines the extent to which the flow over the
sides peneterates into the wake before moving towards the
low perssure region near the roof.

Figure 6 further shows the change in the flow regime in

the canyon for Hd/Hu > 0.88. Figure 7 shows the change in the
height of the vortex core from the ground with increasing
downwind building heights. It can be seen from Figure 6 that
as the height of the downwind building is increased, the
vortex core moves towards the rooftop of the upwind
building. This behavior can be attributed to the change in the
pressure field due to the addition of downwind buildings of
different heights. But the change in the elevation is only ~ 2%
and hence can be assumed to be constant for dispersion
modeling work.

Wake Dominated
Regime

S
k
i
m
m
i
n
g
F
l
o
w

Figure 6: Dimensionless streamwise locations of the vortex

core & saddle point.

3.2 Step-Down Canyon S/W = 2.5

Figure 5 shows the effect of adding a downwind building
of different heights on the wake of the tall upwind building.
The width of the street canyon formed by the addition of the
downwind building was kept constant at S/W = 2.5 for all the
cases.
The downstream location of the origin of the ground
shear layer for an isolated tall building was found to be at
XP/W~1.

81

Figure 7: Dimensionless vortex core height as a function of

street canyon building ratio

Streamline Patterns for Street Canyon Width S = 2.5W

Hd / Hu= 0.08

Hd / Hu= 0.16

Hd / Hu= 0.61

Hd / Hu= 0.69

Hd / Hu= 0.27

Hd / Hu= 0.80

Hd / Hu= 0.35

Hd / Hu= 0.43

Hd / Hu= 0.53

Hd / Hu= 0.88

Hd / Hu= 0.96

Hd / Hu=1.00

Hd / Hu= 0.35

Hd / Hu= 0.43

Hd / Hu= 0.53

Hd / Hu= 0.88

Hd / Hu= 0.96

Hd / Hu= 1.00

Streamline Patterns for Street Canyon Width S = 2W

Hd / Hu= 0.08

Hd / Hu= 0.16

Hd / Hu= 0.61

Hd / Hu= 0.69

Hd / Hu= 0.27

Hd / Hu= 0.80

Streamline Patterns for Street Canyon Width S = 1.5W

Hd / Hu= 0.08

Hd / Hu= 0.16

Hd / Hu= 0.61

Hd / Hu= 0.69

Hd / Hu= 0.27

Hd / Hu= 0.80

Hd / Hu= 0.35

Hd / Hu= 0.43

Hd / Hu= 0.53

Hd / Hu= 0.88

Hd / Hu= 0.96

Hd / Hu= 1.00

Figure 5: Streamline patterns for the various cases considered

82

Streamline Patterns for Street Canyon Width S = 1W

Hd / Hu= 0.08

Hd / Hu= 0.16

Hd / Hu= 0.61

Hd / Hu= 0.69

Hd / Hu= 0.27

Hd / Hu= 0.80

Hd / Hu= 0.35

Hd / Hu= 0.43

Hd / Hu= 0.53

Hd / Hu= 0.88

Hd / Hu= 0.96

Hd / Hu= 1.00

Figure 5: Streamline patterns for the various cases considered contd

highly three dimensional nature of the flow.

Figure 8 : Saddle point height as a function of street

canyon building height ratio

A close inspection of Figs. 6 & 9 shows that the vortex

core and the ground shear layer move towards the downwind
building with increasing downwind building heights until Hd /
Hu 0.88. This behavior is somewhat counter-intuitive as one
might expect the vortex and the ground shear layer to move
towards the upwind building due to the adverse pressure
gradient induced by the downwind building. However, the
downwind building also forces the velocity near the wall to be
parallel to the wall. Also, there is no substantial change in the
height of the stagnation point on the downwind building (ZSt1
/H~0.17) with increasing Hd. It remains almost constant at
ZSt1 ~ . Above the stagnation point, the flow moves parallel to
the widward face of the downwind building.
Figures 6 & 8 show the change in the location of the
saddle point with increasing downwind building height. It can
be seen that the saddle point adjusts to the pressure field
created by the downwind building by moving towards the
upwind building until Hd / Hu ~ 0.69. Between 0.69 Hd / Hu
0.88, the saddle point moves towards the downwind
building. At Hd / Hu ~ 0.88, it appears that a vortex core, a
saddle point and a second stagnation point (ZSt2) exist
simultaneously on the downwind building. The physics
leading to this flow structure are not well understood and at
this point it can only be conjectured that this is a result of the

83

Figure 9: Ground shear layer location as a function the street

canyon height ratio

Figure 10: Dimensionless saddle point height as a function of

street canyon building height ratio
Figure 10 shows the height of the saddle point nondimensioanlized by Hd for various Hd / Hu values plotted on a
log-log plot. From the linear fit shown on the log-log axis, it

can be inferred that the height of the saddle point decreases as

a power law with increasing Hd/Hu. A fit of the data suggest
the vortex core behaves like ZS/Hd = a(Hd / Hu)b, where a =
0.86 and b = -0.88.
3.2.1 Mean and Turbulent Statistics at the Center of the
Canyon
Figures 11 17 show the mean and turbulent statistics
plotted at the center of the canyon (X/S = 0.5). Figure 11
shows mean streamwise velocity profiles (Um) nondimensionalized by the velocity at the top of the street canyon
(UH) for all of the test cases including the isolated building.
From the figure it can be seen that the primary vortex extends
from the top of the canyon to Z / H ~ 0.65, which is
approximately the height at which the saddle point was
detected for these cases. It reaches a maximum value at Z / H
~ 0.6. This is due to the jetting effect brought about by the
interaction of the primary and secondary vortices. The region
between 0 Z / H 0.6 is the region containing the
secondary vortex and is charecterized by a change in the sign
of the velocity from negative to positive.
The mean vertical velocity profiles (Wm) have been
plotted in Fig. 12. It is observed that as the height of the
downwind building is increased, the profiles shift to the right
indicating that an increase in Hd results in weaker vertical
velocities in the top half of the canyon and stronger vertical
velocities in the bottom half.
Figure 13 shows the mean Reynolds stress profile plotted
at the center of the canyon. As expected, the peak values of
Reynolds stress occur near the top of the canyon at the
height of the roof - and decrease to very small values in the
lower half of the canyon. Positive values of Reynolds Stress
are found in the lower half of the canyon. The form of the
profile in the lower region is unusual considering the fact that
these profiles lie in the region of a secondary vortex with
positive vertical velocities and negative horizontal velocities.
This behavior can be hypothesized to be due to the
momentum transfer from the lower leg of the street canyon jet
(with strong negative velocities). To further investigate this
unique pattern, the gradients of the mean streamwise velocity
(Um) were plotted. The basis for undertsanding the behavior
of the Reynolds stress profile based on the velocity gradient
can be illustrated with a simple eddy viscosity model, namely,

-uw = T U
Z
Figure 14 shows the vertical gradient of the mean
streamwise velocity (Um) plotted at the center of the canyon.
It is evident from the figure that the changes in the sign of the
Reynolds stress track the changes in the sign of the velocity
gradient with a near zero value of the Reynolds stress where
the vertical shear is zero.
Figure 15 shows profiles of mean turbulent kinetic
energy (TKE) at the center of the canyon. The maximum TKE
occurs in the shear layer at the top of the canyon and
decreases to a local minimum where the velocity gradient
goes to zero. Below the saddle point, the TKE begins to
increase reaching a local maximum at Z / H ~ 0.3. The
second peak in the TKE profile is associated with strong
mean shear in the lower 40% of the canyon.
Figure 16 corresponds to the mean normalized vorticity
profile at the center of the canyon. The peak values of
vorticity occur at the top of the canyon (in the rooftop
seperation shear layer) and decrease into the canyon. At

84

Z/H~0.6, the profile changes sign indicating the transition

from the region of primary vortex to secondary vortex.
Figure 17 illustrates how the various flow structures in
the wake of a tall building adapt to the addition of a
downwind building. The cavity of a tall building can be
charectarized by a vortex core, saddle point and the ground
shear layer. The relative movements of these entities have
been plotted for varying downwind building heights. The
cavity adapts to the addition of a downwind building while
preserving the basic bow-shaped structure until Hd/Hu~0.69.
Beyond this ratio, the ground shear layer merges into the
downwind building.

4. Conclusions
For Step-down street canyons formed from square
footprint buildings with Hu/W = 3.75 and S/W 2.5, the flow
structure in the canyon is quite complex. In these situations
where the upwind building is relatively tall and slender, the
flow field is strong function of the ratio of downwind and
upwind building heights (Hd/Hu). It was observed that there
exist two distinct flow regimes in the canyon. For Hd / Hu
0.88, the flow patterns in the canyon come under the wake
dominated regime and for Hd / Hu > 0.88, the flow structure
is in the skimming flow regime.

Acknowledgments
This work was funded through an LDRD project by Dr.
Robert E. Ecke, CNLS Director - Los Alamos National
Laboratory. The financial support is gratefully acknowledged.
We would like to thank Dr. Xavier Tricoche, Research
Assistant Professor, Scientific Computing & Imaging Institute
University of Utah for helping us with the critical points
detection code. We would also like to thank Dr. Inanc
Senocak & Dr. Michael J. Brown at the Los Alamos National
Laboratory for their valuable inputs during the course of this
research.

References
Addepalli, B. and Pardyjak, E.R., (2005). 2D PIV
measurements between a pair of model buildings with
varying geometries, 58th Annual Meeting of the Division
of Fluid Dynamics, Chicago,IL, USA.
Baik, J.J., Park, R.S., Chun, H.Y. and Kim, J.J., (2000). A
laboratory model of urban street-canyon flows, Journal
of Applied Meteorology,vol 39, pp. 15921600.
Davies, M.E., Quincey, V.G. and Tindall, S.J., (1979). The
near wake of a tall building block in uniform and
turbulent flows, Journal of Wind Engineering
Proceedings of the Fifth International Conference, Fort
Collins,Colorado,USA, pp. 289 - 298.
Jiang, Y., Liu, H., Sang, J. and Zhang, B (2007). Numerical
and experimental studies on flow and pollutant
dispersion in urban street canyons, Advances in
Atmospheric Sciences, vol. 24, no. 1, pp. 111 - 125.
Ohba, M., (1998). Experimental Study of effects of
seperation distance of twin high-rise tower models on
gaseous diffusion behind the downwind tower model,
Journal of Wind Engineering and Industrial
Aerodynamics, vol. 77-78, pp. 555 - 566.
Santiago, J.L. and Martin, F., (2005). Modelling the air flow
in symmetric and asymmetric street canyonsl, Int J.
Environment and Pollution, vol. 25, no. 1/2/3/4, pp. 145
- 154.

Figure 11: Mean streamwise velocity profile

Figure 15: Mean TKE profile

Figure 12: Mean vertical velocity profile

Figure 16: Mean spanwise vorticity profile

Vortex
Core

Saddle
Point

Origin of the
Ground Shear Layer
Figure 13: Mean Reynolds stress profile

Figure 17: Cavity adaptation to the addition

of a downwind building

Figure 14: Gradient of the Mean streamwise velocity

85

Spanwise variation of drag on roughness elements in a nominally twodimensional boundary layer

P Hayden, T Mapurisa and A G Robins
EnFlo School of Engineering,
University of Surrey,
Guildford, UK
p.hayden@surrey.ac.uk
Abstract The link between spanwise variations in
the flow and the drag in nominally two-dimensional
boundary layers over uniform roughness arrays has
been investigated. The drag on an individual
roughness element is shown to be related directly
to the local flow speed above it. However, it was
also found that small departures from uniformity in
array geometry could mask this relationship. This
led to a further study of the sensitivity of the drag
on an element to small changes in its height relative
to the average array height. Significant changes in
surface drag were found to result even when the
sheltering of downwind elements was accounted
for. It is believed that these observations explain
some of the scatter in compilations of the measured
properties of roughness arrays.

drag is proportional to the square of the speed, a

streamwise velocity variation of 5% could well lead to a
variation of 10% in the cube pressure drag. The aim of
the present study is to investigate the correlation
between the spanwise variation in mean streamwise
velocity and the drag measured using a pressure tapped
cube. In conjuction with this study, additional drag
measurements were made to determine their sensitivity
to variation in element heights within a array.

Key words Rougness arrays, surface drag, boundary

layers, spanwise flow variations.

Introduction
Recent studies of boundary layer development over
rougness arrays, reported by Reynolds et al (2006),
indicate the presence of spanwise variations in the
streamwise velocity field having amplitudes greater than
5% in the mean and greater than 10% in the
turbulence quantities. The initial wavelength of these
spanwise variations was set by the width of the
repeating unit of the roughness, but as the boundary
layer developed downstream this wavelength was
observed to double. For the 10mm cube array this
doubling occured 3 times over the 3.5m fetch of
roughness that was studied, resulting in wavelengths of
8 times that of the repeating unit of roughness.
Figure 1 shows some results from this work at 0.23m
into an array of 20mm obstacles. Lateral profiles of the
streamwise velocity are plotted throughout the
boundary layer (here 56.4mm deep), showing that the
spanwise periodicity extended to all heights. The peakto-peak amplitude of the variation decayed with
increasing fetch, but was still clearly present at the
furthest downstream station studied, 3.2m. There is an
associated periodicity in the shear stress and surface
drag, which has implications for the specification of the
aerodynamic properties of the underlying surface.
The surface drag of an array of cubes can be
inferred from velocity profiles measured above the array
but a more direct measurement can be obtained from
pressure measurements on the individual cubes; e.g. as
described by Cheng and Castro, 2002. The assumption
was made that the pressure drag on each cube at a
given downstream location was approximatly the same.
However, the measurements discussed above suggest
that significant lateral variations are likely. Given that the

87

Figure 1. Variation of streamwise velocity with spanwise

location at a number of heights in the boundary layer.
Data taken at x=230mm with a boundary depth of
56.4mm over the 20mm staggered array.

1 Experimental setup
The experiments were conducted in the EnFlo "A"
wind tunnel, which has a test section 4.5m long, 0.6m
high and 0.9m wide. The floor of the wind tunnel was
covered with 9mm baseboards onto which the 20mm
wooden cube array was fixed with a packing density of
25%. A section of the array is shown in Figure 2.
At the leading edge of the array a 175mm long by
9mm high full width ramp was used to bring the flow up
to the level of the base of the cubes from the tunnel
floor. The measurements presented in this paper were
made 150 cube heights downstream from the first row of
cubes (i.e. at x = 3.0m). To improve the quality of the
cube roughness array around the measurement
location, a section of 11 rows of wooden cubes was
replaced with aluminium cubes screwed to an
th
aluminimum base plate. The 9
row (looking
downstream) on the aluminium plate had holes bored
out underneath the central 13 cubes so that the
pressure tapped cube (shown in Figure 3) could be
inserted in any of these locations. Individual aluminum
cubes were used to fill the remaining holes.

Figure 2. Part of the 20mm staggered cube roughness

array having a packing density of 25%.

by the tunnel speed control loop to maintain consistent

operating conditions during the experimentation.
A single hot wire probe, operated by a Newcastle
constant temperature anemometer, was used to
measure the velocity profiles over the cube arrays. The
hot wire was calibrated in the freestream flow using the
velocity obtained from a nearby pitot static tube.
Pressure drag measurements were made using the
pressure tapped cube connected to a pair of 24 port
scanning valves. The output from each valve was
connected to the the +ve input of a Furness Controls
FCO12 micromanometer (pressure range 0-2mmH20)
which had its ve input connected to the static pressure
tapping from the Pitot static tube located in the
freestream. This enabled two pressure tappings to be
sampled simultaneously for 1.5 minutes at 100Hz, and
by stepping through 21 ports of the scanning valves all
42 pressure tappings could be measured in just over
half an hour.

Figure 3. The pressure tapped cube having 21 pressure

tapping poinbts in each of two opposite faces; first used
by Cheng and Castro 2002
The system described above allowed the pressure
drag on an individual cube to be measured at 13
different spanwise locations. Detailed measurement of
the heights of the aluminium cubes were made using a
LVDT (Linear Variable Differential Transformer) height
gauge mounted on the wind tunnel traverse system.
This information was used to guide adjustment of the
plate mounting to achieve a cube height variability over
this section of the array of 0.06mm which is 0.3% of
the cube height.
In what follws, (x, y, z) are the stream-wise, lateral
and vertical coordinates, with x= 0 at the leading edge
of the array, y = 0 on the centre-line and z = 0 being the
level of the top surface of the baseboard to which the
roughness elements were located.
The free stream velocity in the wind tunnel was
obtained by measuring the dynamic pressure from a
Pitot static tube connected to a Furness controls FCO12
micromanometer. This reference velocity was also used

Figure 4. The locations of the pressure tapping on the

front and back faces of the pressure tapped cube.
The locations of the pressure tappings shown in
Figure 4 are symetrical about the cube centre.
Measurements were made with one of the pressure
tapped faces as the upstream face and the cube was
then rotated by hand through 180 degrees and the
measurements repeated. In this way, the pressure
difference between the front and back faces could be
determined at all 42 locations. Care was taken when
rotating the cube to ensure that its height always
matched that of the surrounding cube array.
The data acquisition was controlled by the EnFlo
Software created at the University of Surrey using

88

LabVIEW supplied by National Instruments. This

comprehensive acquisition package was able to operate
the traverse, control the tunnel speed, move the
scanning valve, acquire signals from the A/D convertor
and process the results in a variety of ways.

in measured (pressure) drag coefficient. Here the

coefficient has been based on U25 measured on the
centre-line (y = 0). The two sets of data are very well
correlated and, further, the range of speeds is about
10% and the range of drag coefficients 20%, confirming
our expectations.

2 Results and discussion

2.1 Cube pressure distribution
The form of the pressure distribution on the front
face of a cube in an array differs greatly from that of an
isolated cube because of sheltering from the
surrounding array. Sheltering reduces the overall flow
speed to which the cube is exposed and displaces the
region of highest pressure towards roof level. Pressure
on the rear face is relatively constant and close to the
local static value. Figure 5 shows the measured
distribution of the pressure difference coefficient (for a
cube in the staggered array illustrated in Figure 1),
where the coefficient, Cp, is defined by:

Cp =

Pfront " Prear

1

2
2 #U 25

where U25 is the mean speed at a height of 25 mm

(i.e. 5 mm above the array), used as a local reference
speed. The great majority of the pressure force derives
from the upper section of the front face. The difference
between the front and rear pressures can be integrated
over the frontal area to give the drag, D, and a drag
coefficient, Cd, defined as:

Cd =

D
2
2 "U 25 A

Figure 6. (top) Streamwise velocity variation with

spanwise location taken at 1.25 HRef above the row
where the drag coefficient was measured. (bottom)
The expected relationship between the local drag
coefficient plotted in Figure 6 and the local velocity at a
height of 25mm is:

" U 25 (y) % 2
Cd (y) = Cd (y = 0)#
&
$ U 25 (y = 0) '

where A is the frontal area, H .

This relationship is shown to hold in Figure 7, where

the constant of the linear fit, 0.27, is Cd(y=0). The data
fit is not perfect, as this would imply full similarity of the
velocity profiles at every location examined. Clearly,
there are likely to be local variations in the profile shape
that will be responsbile for at least some of the observed
scatter.

Figure 5. Mapping of the distribution of pressure

difference coefficient over the frontal area of the cube.
As previously noted, the span-wise speed variations
in the boundary layer are expected to induce span-wise
variations in the drag on the cubes within the array.
Figure 6 shows the lateral profile of mean streamwise
velocity at a height of 25mm at the location of the
pressure tapped cube and the corresponding variation

89

Figure 7. Correlation between the measured drag

coefficient and the local velocity at 1.25Href.

2.2 Cube height sensitivity

With the link between the velocity non-uniformities
and the cube pressure drag established, interest was
turned to the sensitivity to variations in the height of
elements in an array. This interest is linked directly to
the question of spanwise varaibility as when this was
first investigated it was found to be masked by the
effects of imperfect array height uniformity. That was
why the special aluminium array section was installed
and why considerable attention was focussed on its
levelling.
Firstly the height of a pressure tapped cuboid was
increased or decreased relative to the cube height; a
height range of 15% was examined. This was achieved
by either sinking the pressure tapped cube into the base
plate or placing packing beneath it to increase the
overall element height. There were no pressure
tappings in the packing material and only the pressure
field on the tapped cube was actually measured. The
resulting drag is therefore somewhat too small. The
error is likely to be small, as can be illustrated with the
data in Figure 5. For this distribution, slight over 10% of
the total drag is derived from the lower 15% of the cube
face.
Figure 8 plots the drag coefficeint of the cuboid
relative to that of the cube as a function of the height
ratio, H/H ref; H ref being the cube height. The ratio would
be constant at 1.0 if there was no dependency of Cd on
H. However, the sensitivity is large, as might be
expected since more of the front face is exposed to the
oncoming flow as the cuboid height is increased. The
results reveal about a 45% increase in Cd for a 15%
increase in height; the decrease in Cd though is only
about 25% for a 15% reduction in cuboid height. The
curvature of the Cd-H relationship is important as it is
consitent with an increase in overall array drag with
increasing variability in element height, even though the
mean height remains equal to the cube height.

Figure 8. Sensitivity of the drag coefficent to the height

of the pressure tapped cube relative to the surrounding
array .(Href = 20mm)

measured as the height of elements directly upstream

(the only ones that were found to matter) was varied.
The results are plotted in Figure 9 as the ratio of the
drag coefficient to that for a cube in a uniform array as
a function of the height ratio, H/H ref, of the nearest or
the second nearest upwind cuboid (i.e. 4 Href or 8 Href
upwind). When the upwind cube was removed, the
drag coefficient is seen to have increased by about
35%. The drag coefficient rises a little as the height of
the upwind cuboid is increased from zero before
decreasing rapidly to give a coefficint ratio of unity
when H/Href = 1. The minimum drag coefficient ratio in
this case is about 0.6 when H/Href is about 1.5 and the
asymptotic limit of about 0.95 is attained once H/Href
exceeds 3.

Figure 9. Sensitivity of the drag coefficent to height

variations of an upstream cube at the same spanwise
location but for different distances upwind of the
pressure tapped cube.

The overall effect is illustrated in Figure 10 where

the increment in drag coefficient, Cd, relative to a
cube in a uniform array is plotted as a function of
cuboid number for the case H/Href = 1.1; i.e. cuboid 1
is 10% taller than the array elements and cubes 2, 3, 4,
and 5 are directly downwind from it at distances of 4, 8,
12 and 16Href. Clearly, some of the excess drag on the
taller cuboid is offset by the reduction in drag on the
sheltered cubes. The larger frontal area of the taller
cuboid has to be aken into account when analysing the
net outcome and the overall effect turns out to be an
increase in total drag of about 12% relative to the drag
on a cube in the uniform array; i.e there is a significant
increase in drag even when the sheltering of the
downwind cubes is taken into account. The equivalent
result for the case of a reduced height cuboid is not yet
available. However, the fact that the drag on a nonuniform height array is greater than that on a unifrom
array of the same average height indicates that it cannot
cancel the result for the taller cuboid.

Figure 8 does not tell the full story as although, for

example, a relatively tall cuboid experiences greater
drag, the array elements downwind of it are sheltered
and experience less drag. The net effect on the array is
the difference in these two changes and is not
immediately obvious. Some experimenst were therefore
preformed in which the drag of a cube in the array was

90

Figure 10. The behaviour of the individual cube drag

coefficients when the height of the first cube is
increased by 10%.

3 Conclusions
A direct link was proved between the pressure drag
on elements in an array of cubes and the lateral
variation in the velocity field over them. The drag
coefficient based on the local flow speed was effectively
constant. This was only apparent once the array had
been carefully adjusted to be level. It is likely that the
effect would be masked in many arrays used in wind
tunnel research by small variations in the level of the
array elements. The sensitivity of the pressure drag
coefficient to element height perturbations was shown to
be large. It was also non-linear, in that the increase in
drag on tall elements was greater than the decrease on
the correspondingly short elements. In an overall sense,
the increased drag on a tall element is off-set to some
degree by the sheltering of donwind elements. The net
effect in the case illustrated was that something like a
half of the drag on the tall element was cancelled.
The span-wise flow and drag variations and the
sensitivity of local drag to element height variations
probably explain a significant part of the scatter in
compilations of the measured properties of roughness
arrays.

Acknowledgments
We are grateful to Allan Wells for his technical
support in this project and for consultation from Tom
Lawton. Particular thanks needs also go to Florent
Champet who worked during the summer of 2005 taking
prelimiary pressure drag measurments without the
benefit of a high precision array.

References
Reynolds, R.T. Hayden, P. Castro, I.P. and Robins A.G.
(2007) Spanwise variations in nominally twodimensional
rough-wall
boundary
layers.
Experiments in Fluids, 42(2):311320.
Cheng, H. and Castro, I.P. (2002) Near wall flow over
urban-like roughness. Bound. Lay. Met., 104:229
259.

91

CFD analyses of flow in stratified atmosphere

G. Kristf, N. Rcz, M. Balogh
Dept. of Fluid Mechanics, Mechanical Engineering, Budapest University of Technology and Economics,
Budapest, Hungary
kristof@ara.bme.hu, racz@ara.bme.hu, baloghm@ara.bme.hu
Abstract - General purpose CFD solvers are
frequently used in small scale urban pollution
dispersion simulations involving no extensive
vertical flow. Vertical flow, however, plays an
important role in many atmospheric phenomena
such as in formation of urban heat island induced
country breeze, which has a great significance from
the point of view of ventilation of big cities.
Stratification effects must be taken into account in
such simulation studies. A general purpose CFD
software has been adapted to atmospheric flow
problems in order to take advantage of high
flexibility in geometrical description and meshing.
Non-hydrostatic formulation of the governing
equations has been employed together with
realizable k-e turbulence model. Compressibility
and thermal stratification effects were taken into
account by utilizing a novel system of
transformations to the field variables and by adding
the consequential source terms to the standard set
of transport equations. This approach offers further
possibilities for the development of the underlying
physical model and can be used in many application
area. An automated fast running pre-processor
software has been developed to help the creation of
model geometry meshing and specification of
boundary conditions. In this paper two validation
studies are presented: for thermal convection and
for internal gravity wave phenomena. Simulation
results are in line with the results of laboratory
scale experiments. Further validation is necessary
against full scale meteorological observations.
Key words non-hydrostatic model, CFD, urban heat
island, gravity waves, atmospheric flow

Introduction
More and more detailed numerical description of the
atmospheric flows is possible due to the development of
the computer technology. There are several advantages
of the usage of CFD solvers for computing atmospheric
flows in micro and meso-scale. High flexibility in mesh
refinement and grid structure allow more accurate
description of the relief, which has an impact on forming
both the aboveground and the upper streams, therefore
the computed flow field could be highly detailed. Many
type of available turbulence models, application of
effective numerical techniques, easy implementation of
model refinements, and the possibility of parallel
computing are all useful features for researchers of
micro meteorology, climatology and environmental
technology. Most CFD solvers have compressible flow
option, which means, that the air density can be a
function of local pressure and temperature through the
ideal gas law. This option, however, does not allow
effective simulation of buoyancy driven atmospheric

93

flow due to the severe limitation on the magnitude of

time steps. In those cases when density depends only
on temperature, ideal gas law with constant pressure,
or Boussinesq approximation with constant thermal
expansion coefficient can be used. Potential
temperature formulation allows the application of
incompressible gas models. As the potential
temperature in standard atmosphere is a linear function
of vertical coordinate, it leads to a parabolic vertical
pressure profile and thus causing several order of
magnitude difference between horizontal and vertical
components of the pressure gradient vector, which can
give rise to numerical instabilities in pressure based
CFD solvers.
In meteorological models of local atmospheric
circulations some variations of boundary layer
equations with hydrostatic approximation to the vertical
component of momentum equation are commonly used.
Numerical solution is usually obtained by purpose
developed solvers [1], [2], [3]. If micro-scale details of
the flow field, such as transport of pollutants in urban
street canyons, are investigated then meso-scale
models with nested CFD sub-models are usually
employed. The drawback of this method is the
interpolation of variables between grid interfaces with
different resolution which is the source of numerical
errors and model uncertainties [4]. Our purpose is to
reformulate the mathematical model in a commonly
used CFD solver in order for micro- and meso-scale
flows in one common system to be analyzed, in such a
way, micro structures could be studied by a simple local
grid refinement process.

Mathematical transformation
The mathematical relation between the absolute
physical quantities and the field variables used in the
CFD solver are described by a set of transformation.
The transformation formulae are based on the standard
ICAO [5] temperature and pressure profiles (Eq.1 and
2) and an approximate exponential function for the
density profile which allows the simplification of some of
the transformation expression (Eq.3).
T = T0 z

(1)
g

T z R

p = p0 0

T0

(2)

= 0 e z

(3)

z denotes the vertical coordinate,

= 0.65 C / 100 m ,
T0 = 288.15 K ,
p0 = 1.01325 10 5 Pa ,

0 = 1.225 kg / m 3 ,

g /(R ) = 5.2553 ,
= 10 4 m 1 .

~ 2
2
p
0
Sw = 0 g z T~ T0 02 1
2

Note that, the above density profile approximates

the standard profile within 0.4% error bound if z < 4000
m, but the error increasing rapidly at higher altitudes.
We employ a transformation defined by (Eq.4-8)

T = T~ T0 + T
p=

~
~+p
p + p = e z p
0

= ~ 0 +
z=

w=

Ln (1 z~)

0 ~ ~ z
w =we

~ = 0 0 (T~ T0 )

(6)

(7)

(8)

(9)

in which, w is the vertical component of the velocity

vector, T, p, , z, w denotes absolute physical values
~, ~, z~, w
~ are the transformed variables used in
and T~, p
the simulation software. Unsteady variable density
forms of continuity, momentum, and energy equations
are solved by the simulation system with Boussinesq's
approximation (Eq.9) to the density, in which, is the
cubic thermal heat expansion coefficient of air. In the
simulation system ~
is used only for evaluating the
buoyancy force in the vertical component of the
momentum equation, in every other place 0 is used
instead. Turbulent transport is modeled by the
Realizable k- turbulence model with full buoyancy
effects (included in transport equations of both k and ).
Exact form of the governing equations can be referred
from CFD literature and software manual therefore it is
not cited here. Eq.7 and Eq.8 can be derived from the
identity of stored mass in a dz high layer of air (Eq.10)
and assuming the identity of vertical mass flux in real
and transformed system (Eq.11)

w = ~ w~ 0w~

(11)

(12)

(13)

Eq.12 allows the application of arbitrary equilibrium

density profile, Eq.13 is valid only with density profile
given by Eq.3.
The energy equation and the turbulent transport
equations also need to be corrected by additional
volume sources which are proportional to the difference
between the magnitude of the equilibrium temperature
gradient and the dry adiabatic temperature gradient
= 0.976 C / 100 m.
A user defined source term in the energy equation
expresses the amount of heat introduced into unit
volume of fluid in unit time therefore it has the
3
dimension of W/m . Temperature of vertical air flow
follows the dry adiabatic gradient (). The
transformation applied to the temperature field variable
(Eq.4) already includes a temperature gradient .
Amount of heat corresponding to a temperature
gradient - must be taken out from the vertical flow,
therefore the necessary source term in the energy
equation is:

(5)

(10)

)(

2
~(1 z~)1
Sw = 0 g z T~ T0 (1 z~) 1 + p

(4)

dz = ~ dz~ 0 dz~

~ ( )
ST = c p ~ w

(14)

In k- models there is a production term for the

turbulent kinetic energy which is calculated
proportionally to the vertical component of the gradient
of temperature T~ . This term represents the effect of
buoyancy on turbulence and it should be zero if the
thermal stratification is neutral, that is, Eq.15 is fulfilled
by the absolute temperature. Temperature gradient can
be expressed by the gradient of equilibrium temperature
profile plus the gradient of transformed temperature
(see. Eq.16).

T
z

(15)

stable

~
T T

=
z z

(16)

From Eq.15 and Eq.16 a source term described by

Eq.17 for the turbulent kinetic energy can be derived, in
which t denotes the turbulent viscosity and Pr is the
turbulent Prandtl number. Note that Sk is negative
consequently the introduction of Sk into the model
causes a damping effect to the turbulent intensity.
Similar correction must be applied to the transport
equation of (Eq.18).

With this transformation the vertical extent of the

atmosphere is "compressed" below a well defined limit
(1/ ) therefore z~ cannot have values higher then 1/
(see Eq.7). Conditions described in Eq.10 and Eq.11
ensures good approximation to the continuity and to the
horizontal components of the momentum equation.
Only the vertical component of pressure gradient needs
correction. By assuming hydrostatic equilibrium at high
altitudes, which is valid in atmospheric flows, it is
possible to derive a source term for the vertical
momentum equation which compensates for the vertical
component of the pressure gradient (see Eq.12. and
Eq.13.).

Sk = g

S = C1 C3

t
( )
Pr

(17)

g t ( )
k
Pr

(18)

Here k is the turbulent kinetic energy, is the

turbulent kinetic energy dissipation, C1 and C3 can be
referred from the CFD literature or software

94

documentation. Effect of Coriolis force can be taken into

account ether by using rotating frame of reference
option or by some further user defined source terms.

Preprocessing
As it is known simulation of boundary layer flows,
such as atmospheric phenomena in focus of our
investigation, are very sensitive for grid quality and for
grid resolution in vertical direction. Optimal computing
cost can be achieved only by applying structured or
layered type unstructured grids. Although these type of
grids can be generated by general purpose meshing
software, it is a rather complex and time consuming
work, therefore it was necessary to develop an
automated fast running preprocessor utility involving the
following functions:
1. Extraction of surface points of arbitrary terrain area
from the online USGS elevation database of 1 arc
second (~ 30 m) resolution.
2. Transformation of co-ordinates from spherical to a
local projection Cartesian system employing UTM
(Universal Transverse Mercator) and EOV (Uniform
National Projection system) projections (Bcsatyai,
2005). Deformation of the transformed domain is
negligible in case of horizontal extents smaller than
100 kilometers which is the case in simulation
problems of our interest.
3. Extension of natural relief by a ramp zone around
the boundaries, as shown from Fig.1., for easier
definition of inlet velocity profile.

Figure 1. Surface model of Pilis (Hungary) extended

with a surrounding ramp zone
4. Selection of vertical extent, number of layers and
parameter of vertical grid refinement. Vertical grid
structure can be seen in Fig.2.
5. Generation of structured grid of even horizontal
resolution or optional import of layered unstructured
grid containing arbitrary horizontal refinements.
6. Mesh export in ANSYS-FLUENT 6.x format.
7. Creation of customizable wind profiles and export in
ANSYS-FLUENT Profile Format for easy definition
of inlet boundary conditions.

95

Figure 2. Vertical grid structure

Validation with laboratory scale

thermal convection
Reproducible measurement data are available
mostly from laboratory experiments. Since laboratory
scale experiments on atmospheric flow are themselves
based on approximate mathematical description only
partial validation of the above mathematical model is
possible on the basis of these data. As a first step
experimental studies involving thermal convection in
stably stratified medium are looked for.
Urban heat island circulation and also its interaction
with see breeze are investigated by Cenedese and
Monti [7] in water tank with stable thermal stratification
created by heated and cooled heat exchanger plates
positioned at the top and the bottom of the tank.
Thermal convection is generated by a thin layer D=100
mm diameter circular electric heater plate placed on the
bottom of the tank. Velocity field has been recorded by
PIV device at high spatial resolution in the horizontal
symmetry plane along with temperature profile
measurements above the heater plate.
Comparative CFD simulation has been carried out in
FLUENT 6.2 simulation system after implementing
source terms (Eq.13, 14, 17, 18) in the form of User
Defined Function (UDF). A drawback of laboratory scale
modeling is that the value of Reynolds number is
several orders of magnitude lower than that of
characterizing atmospheric scale flow. Due to the very
weak turbulence k- model and the corresponding
source terms (Eq.17-18) had to be switched off and
Large Eddy Simulation (LES) method has been used
instead. Variable density effects are not taken into
account in the laboratory experiments, therefore source
term (Eq.13) included in the vertical component of the
momentum equation had to be switched off too, and so
are the corresponding parts (Eq.5 and 7) of the
transformation system. This way we have got a
simplified version of the above mathematical model
which still does have the capacity to describe low
Reynolds number flow phenomena occurring in
stratified liquid with the help of the most important
volume source (Eq.14) included into the energy
equation.
Results of the LES simulation in comparison of
experimental data are plotted in Fig.3-7. x is horizontal z
is vertical coordinate zi denotes the mixing length, u is
horizontal w is vertical velocity component U and W are
velocity scales, temperature scale is dTm=(Taxis-Ta)z=0 in
which Ta is the unperturbed (initial) temperature profile.
Simulation results agrees well with experimental data
both for velocity and temperature field.

Figure 3. Computed flow field in the symmetry plane

(time interval of averaging is 300 s)

Figure 7. Dimensionless profiles of vertical velocity

components, symbols: measured data, solid line: LES
result

Validation with laboratory scale gravity

waves
In order to test our models ability to capture buoacy
and unsteady inertial effects internal gravity waves
generated by symmetric and asymmetric obstacles
were examined. Obstacle shape is characterized by a
Gaussian function, symmetric and asymmetric shapes
are used. Because of the larger number of simulation
setups we restrict the discussion of results to a
representative case (illustrated in Fig.8.) where an
obstacle with steep upstream edge and a gentle
leeward slope was investigated.
Measurements were carried out by Gyre and
Jnosi [8] in a narrow plexi glass tank with a size of
2.4 m * 0.4 m * 0.0087 m filled with linearly stratified salt
water towing the obstacles in the bottom of the tank.
The range of parameters, obstacle height h = 2-4 cm,
towing velocity U = 1-15 cm/s and Brunt-Visl
frequency N = 1.09-1.55 s-1 corresponds to an
atmospheric flow up to a level of 5-10 km for an
obstacle height of 600 m and wind speed of 10-70 m/s.
Reynolds number however cannot be kept in the
experiment because of the working fluid, it varies
2
3
8
9
between 10 and 10 (in full scale flow around 10 -10 ).
Due to the low Reynolds number range turbulence
models cannot be applied laminar approach was used
instead.
Two dimensional flow was assumed in the
simulation. Structured grid of 65 * 320 elements was
used. Grid structure surrounding the obstacle is
illustrated in Fig.8. Constant inlet flow speed and
moving bottom wall has been applied simulating the
experimental situation in co-moving frame of reference.
According to measurements 10-15 buoyancy period
was enough for a quasi-steady wave pattern to be build
up.

Figure 4. Dimensionless profiles of the temperature

perturbation along the axis of the heater plate, LES
results

Figure 5. Dimensionless profiles of the temperature

perturbation along the axis of the heater plate
measured data

Figure 8. Grid structure around the obstacle

Figure 6. Dimensionless profiles of horizontal velocity

components, symbols: measured data, solid line: LES
result

Typical streamlines for a high Froude number

(U/Nh = 1.4 where U is the towing velocity, N denotes
the buoyancy frequency, h is the obstacle height) and
a low Froude number (U/Nh = 0.3) scenario are
illustrated in Fig. 9. As can be seen in the lower part of
the picture wave breaking can occur at low wind speed

96

causing a so called rotor (vortex of horizontal rotation

axis) behind the obstacle. Good qualitative agreement
has been found between computed streamlines and
camera images taken during the experiments.

Figure 9: Computed streamlines. Heavy solid lines

indicate wave fronts at U/Nh = 1.4 (top) and
U/Nh = 0.3 (bottom) non-dimensional flow velocity.
Wave length and amplitude information have been
extracted from the computed flow patterns and than
compared with corresponding experimental data (see
Fig.10).

with some additional source terms to the governing

equations which are implemented in the form of User
Defined Functions in ANSYS-FLUENT simulation
system.
An automated fast running pre-processor software
has been developed to help the creation of model
geometry meshing and specification of boundary
conditions.
Model has been validated with laboratory
experiments for thermal convection and for internal
gravity waves. Flow pattern in both cases successfully
reproduced. Quantitative agreement has been found
between simulation results and experimental data such
as temperature profiles, velocity profiles and wave
parameters. Further validations are necessary with full
scale atmospheric measurements.
The presented approach may be used in future with
benefits in climatology, micro-meteorology, and also in
environmental science for analyzing phenomena:
Gravity waves behind high mountain;
Cloud (such as cumulonimbus) formation;
Flow around high mountain;
Urban heat island (UHI) and urban ventilation;
See breeze;
Valley breeze.
Plumes of cooling towers, chimneys or volcanoes;
Energy potential of wind farms;
Forest-fire or town-fire.

Acknowledgement
Authors of this paper whish to express their thanks
for the support of OTKA T049573 project and to
Hungarian National Office for Research and Technology
for support of NKFP 3A/088/2004 project.

References

Figure 10. Normalized averaged wave length (uper)

and amplitude (lower) as a function of non-dimensional
horizontal flow velocity
Wave amplitudes obtained both from laboratory and
numerical experiments are in line with linear theory. In
most cases wave amplitudes predicted by the
simulation model agree well with experimental data. In
those cases when steep leeward profile are examined,
depending on the velocity of the undisturbed flow,
separation bubbles can appear behind the obstacle
which obviously modifies the shape and size of lee
waves, therefore in these cases, accurate simulation of
boundary layer separation is very important. Surface
roughness and 3D effect such as effect of the end walls
can be significant in these cases.

Conclusions
General purpose CFD software has been adapted to
atmospheric simulation problems involving stratification
and
compressibility
effects.
Modifications
are
formulated in a novel system of transformation together

97

Yoshikado, H. 1992. Numerical study of the daytime

urban effect and its interaction with sea breeze. J.
Appl. Meteorol. 31, 1146-1163
Lu, J., Arya, S.P., Snyder, W.H. and Lawson Jr, R.E.,
1997. A laboratory study of the urban heat island in
calm and stably stratified environment. Part II.
Velocity field. J. Appl. Meteorol. 36, 1392-1402
Kurbatskii, A.F., 2001. Computational modeling of
turbulent penetrative convection above the urban
heat island in a stably stratified environment. J.
Appl. Meteorol., 40, 1748-1761
Sarma A., Ahmad N., Bacon D., Boybeyi Z., Dunn T.,
Hall M. and Lee P., 1999. Application of Adaptive
Grid Refinement to Plume Modeling, Air Pollution
VII, WIT Press, Southampton, 59-68.
Manual of the ICAO Standard Atmosphere / Doc 7488,
1993.
Bcsatyai, L.,
Magyarorszgi vetletek, Nyugat Magyarorszgi Egyetem, Szaktuds Kiad
Cenedese, A. and Monti, P., 2003. Interaction between
an Urban Heat Island and a Sea-Breeze Flow. A
Laboratory Study, J. Appl. Meteorol. 42, 15691583.
Gyre, B. and Jnosi, I.M., 2003. Stratified flow over
asymmetric and double bell-shaped obstacles.
Dynamics of Atmospheres and Oceans 37, 155170.

Wind tunnel study of the concentration fields in a plume emission

A. R. Wittwer

F. de Paoli

Depto. Ingeniera,
Universidad del Nordeste,
Resistncia, Argentina
a_wittwer@yahoo.es

Depto. Engenharia Civil

Universidade Federal do Rio Grande do Sul,
Porto Alegre, Brazil
fdepaoli@gmail.com

A. M. Loredo-Souza

E. B. Camano Schettini

Depto. Engenharia Civil

Universidade Federal do Rio Grande do Sul,
Porto Alegre, Brazil
acir@ufrgs.br

Instituto de Pesquisas Hidrulicas,

Universidade Federal do Rio Grande do Sul,
Porto Alegre, Brazil
bcamano@iph.ufrgs.br

Abstract The
concentration
fields
in
the
proximities of a local gas emission source are
experimentally analyzed in several combinations of
wind incidences and source emissions. The
concentration measurements were performed by an
aspirating probe, in a boundary layer wind tunnel.
The analysis has included the determination of the
concentrations mean values as well as the intensity
of the fluctuations. The source model represents a
local gas emission which is dispersed in a turbulent
boundary layer in neutral stability. Several different
conditions are determined by the plume buoyancy,
the emission velocity and incident flow wind speed.
The model scales were 1:200 and 1:400, according
to the simulated atmospheric boundary layer factor
scale. Pure helium as well as an air-helium mixture
are used for the emissions. The approaching flow
represents a neutral atmospheric boundary layer.
Three configurations were tested: an isolated
chimney in a homogeneous terrain, the same
chimney with one bluff body in close proximity, and
the chimney surrounded by a non homogeneous
urban terrain. The experimental mean concentration
values are compared with Gaussian profiles. The
plume vertical dispersion values are obtained and
compared with the Briggs curves. The profiles of
the
concentration
fluctuation
intensity are
compared with the experimental results obtained
from other authors. Finally, the dilution factor is
analyzed with regard to the empirical curves of the
minimum dilution.
Key words plume, gas emission, physical model,
wind tunnel, aspirating probe.

Introduction
The study of dispersion and pollutant concentration
levels discharged in the atmosphere has become a
fundamental issue due to the new environmental
demands. Nowadays numerous computational works
related with dispersion phenomena are being
developed. Usually these studies must be validated with
experimental results. The high costs of field
experimentation lead to laboratory reduced models
studies. In this context, the boundary layer wind tunnel
becomes an important tool. However, it is necessary
that the main characteristics of the atmospheric
boundary layer and of the dispersion processes be
reproduced. Up to now there are no publications

99

regarding experimental studies of atmospheric

dispersion performed in South America using reduced
models, despite the great urban concentrations and the
registerd atmospheric contamination problems. The
development of a toll to allow the evaluation of these
problems would be a great contribution to the
improvement of the region environmental conditions.
To reproduce dispersion phenomena in wind
tunnels, besides the requirements of simulation of the
ABL, similarity requirements are established to the
modeling of the plume emission behaviour. These
requirements, according to Isyumov & Tanaka [1980],
may be summarized in the following way: source
geometric similitude, Froude number similarity, density
and velocity ratio similarities as well as similarity of the
source Reynolds number.
The practical difficulties related to the exact
simulation of the emission process makes approximate
solutions an acceptable alternative. Cermak & Takeda
[1985] propose, besides the geometric similarity, the
equality of the relation between the emission velocity
and the local wind velocity, and the equality of
densimetric Froude number. The similarity criteria may
be modified according to the zone to be simulated, and
this may be the the region closest to the source, in
which the own chimney structure and the discharge
cause modifications in the flow field, or the furthest
zone, where these effects are not observed.
In this work the dispersion process of an emission
plume is studied through a reduced model in a
boundary layer wind tunnel. The source model
represents a light punctual gas emission that disperses
in a neutrally stable turbulent boundary layer. O modelo
da fonte representa uma emisso de gs pontual leve
que se dispersa em uma camada limite turbulenta em
estabilidade neutra. Distinct conditions are considered
which are determined by the degree of the plume
buoyancy, the exit emission velocity and the
approaching flow velocity, as well as two distinct
configurations for the emission surroundings. These
configurations of the urban surroundings are
represented by a simple building located windward and
leeward to the emission, allowing the analysis of the
modifications which are produced in the dispersion
process in comparison with the case of na isolated
source.

1 Experimental Design
The basic requirement of a wind tunnel dispersion
study is the physical simulation of the atmospheric flow.

These tests were performed at Prof. Joaquim

Blessmann closed returne boundary layer wind tunnel
of UFRGS [Blessmann, 1982].
Roughness elements mixture devices were used to
reproduce a neutrally stable boundary layer
[Blessmann, 1982]. The scale of the simulation is
approximately 1:350.
Figure 1 shows the nondimensional profiles obtained
with the higher velocity flow and two cases with low
velocity flow. In the high velocity, at the reference
position, the mean velocity is approximately 35 m/s,
while that for the low velocities the values are 0.85 and
1.91 m/s, respectively. A power law velocity profile with
exponent 0.23 is fit to the experimental values.

[(0 a )gw 0D0 ]/ aU0 3 . The plume buoyancy is also

600

known as the Richardson number.

In table 1, the values of the velocities w0 and U0
correspond to the reduced model values.

Alta velocidade
Uref=0.96 m/s

450

The wind velocity in the wind tunnel scale varied

between 0,96 and 3,44 m/s to allow the modification of
the plume characteristic parameters. Two cases were
considered: the isolated chimney in a homogeneous
terrain and the chimney with a single vicinity defined by
a prismatic building.
The plume characteristic conditions are determined
by the nondimensional parameters indicated below.
Tables 1 and 2 present values of the parameters
considered in the plume characterization. The adopted
relations and nondimensional parameters are the
ratiobetween the emission and the approaching flow
velocities,
w0/U0,
the
emission
momentum,
2
2
the
plume
buoyancy:
0 w 0 / a U0 ,

Uref=3.44 m/s
Perfil 0.23

z [mm]

Table 1. Conditions of plume simulation.

300

150

0
0,00

0,50

1,00

1,50

U/Uref

Alta velocidade
Uref=0.96 m/s
Uref=3.44 m/s

w0 [m/s]

U0 [m/s]

A
B
C
D
E
F
G
H

He
He
He
He - Ar
He
He
He
He - Ar

0,56
1,26
0,95
0,75
0,56
0,95
0,56
1,45

0,85
1,91
0,85
0,85
3,04
3,04
1,91
1,91

Condition

w0
U0

0 w0 2
aU 0 2

[( 0 a )gw0 D0 ]

A
B
C
D
E
F
G
H

0,66
0,66
1,11
0,88
0,18
0,31
0,29
0,76

0,060
0,060
0,171
0,278
0,005
0,013
0,012
0,145

-0,154
-0,031
-0,260
-0,154
-0,003
-0,006
-0,014
-0,031

Logartmica (Alta
velocidade)

y = -257,28Ln(x) + 827,38

z [mm]

Emission

Table 2. Characteristic parameters of the simulation.

Nondimensional parameters

600

450

Condition

300

150

0
0

10

20

30

Intensidade local de turbulncia

Figure 1. Nondimensional mean wind velocity and

turbulence intensity profiles for several wind tunnel
velocities.
The corresponding values of turbulence intensity are
also presented in Figure 1. As in the case of the mean
velocities, the configuration of the profiles is similar, but
in this case the values of the turbulence intensity are
greater for Uref = 1.91 m/s, presenting deviations for
Uref = 0.85 m/s. Due to the fact that the local
turbulence intensity are presented, the variation is
mainly the product of the relative decrease of the mean
velocity.
The emission source model was built with a circular
tube of 20 mm diameter that could represent, according
to the adopted scale, a 4m or 8m diameter chimney and
a variable height. The gas used in the emission is pure
helium.

100

aU 0 3

With regard to the incident wind and the chimney

surroundings, four configurations were considered in
the tests. Configuration I corresponds to the isolated
chimney (Figure 2), and the configurations II and III
(Figure 3) refer to the cases of a building located
windward and leeward of the emission, respectively. In
the figures D is the chimney diameter and x indicates
the distance from the measurement point.
In configuration I, the height of the chimney is
H = 250 mm. For configurations II and III, the building
height B is 270 mm, while H varies from 250 to 270 mm
according to the test. The separation between the
chimney and the building model in both cases is
80 mm.

is the flow emission, UH is the wind velocity

corresponding to the emission height and z is the
vertical coordinate measured from the wind tunnel floor.

2 Results
Instantaneous concentrations were measured in
vertical profiles located in several distances x/H,
measured from the emission. The concentration mean
and rms values were obtained from Reynolds
decomposition, for each point.
Figure 4 presents the vertical profiles of the
concentration coefficient K and the IC for condition A
and configuration I (no building), related to the positions
x/H = 0,33, 0,66 and 1,00. Due to buoyancy effects, the
profiles present an asymmetry, tending to deviate the
plume upwards. Similar behaviour is observed in
Figure 5, related to condition B and to the same
configuration I, for the three leeward positions.
2

Figure 2. Isolated emission Configuration I

x/H=0.315 (a)
x/H=0.630 (a)
x/H=0.315 (b)
x/H=0.630 (b)
x/H=1.260 (b)

z/H

1.5

0.5
0.0

1.0

2.0

3.0

Figure 3. Single building located windward and leeward

from the emission - Configurations II and III.

K=

CUHH2 ,
Q0

Ic = c ,
C

x/H=0,33
x/H=0,66

1.4

x/H=1,00

z/H

For the study of the plume dispersion process, the

concentration field was evaluated leeward from the
emission source. The measurements were performed
with a hot-wire anemometer with an aspiring probe. This
probe is composed by the hot wire and a 0,3 mm
internal diameter ceramic tube, connected to a vacuum
pump, allowing the measurement of instantaneous
concentrations. The capilar tube, together with the
aspiration, causes a sonic blokedge and the wire
becomes a probe sensitive to the variations of the fluid
properties (in this case, measures the variations of the
gas concentration), independently of the external
velocity field. In each point, a one minute sample was
taken, at a sampling frequency of 1024 Hz.
The results obtained in the tests are presented as
concentration coefficient (K) and intensity of the
concentration fluctuations (Ic) profiles, where

1.6

1.2

(1)
0.8
0.1

10

100

Ic

(2)

C and c are the mean concentration and the

standard deviation of the fluctuations, respectively, Q0

101

Figure 4. Concentration profiles K and Ic, condition

A, configuration I, for x/H = 0,33, 0,66, 1,00.

1.6

2
x/H=0.315 (a)
x/H=0.630 (a)
x/H=0.315 (b)

1.4

x/H=0.630 (b)
x/H=0.60

x/H=1.260 (b)

x/H=1.20
x/H=1.80

1.5

z/H

z/H

1.2

1
1

0.8

0.6

0.5
0.0

2.5

5.0

7.5

0.0

1.0

2.0

3.0

1.6

2
x/H=0.60
x/H=1.20
x/H=1.80

1.4
x/H=0.315 (a)
x/H=0.630 (a)

1.5

x/H=0.315 (b)
x/H=0.630 (b)

1.2

z/H

z/H

x/H=1.260 (b)

1
1

0.8

0.6

0.5
0.1

10

100

0.1

Ic

10

100

Ic

Figure 5. Concentration profiles K and Ic, condition B,

configuration I, for x/H = 0,60, 1,20, 1,80.

Figure 6. Concentration profiles K and

Configuration II, (a) Condition A, (b) Condition B.

From the comparison of these two situations is

possible to infer that, even keeping the same emission
velocity and momentum parameters, the behavior of the
plumes is not the same for the two cases since the
Richardson numbers Ri are different. As Ri diminishes,
the plume width decreases.
Figure 6 shows the vertical profiles corresponding to
configuration II, for two positions x/H and condition A
(a in figure 6), and in three positions x/H and condition
B (b in figure 6). The smaller absolute values of K with
respect to configuration I indicate a greater dilution
effect in the concentrations caused by flow conditions
imposed by the presence of the building. The effect is
also shown with the increasing in the vertical dispersion
z. In the leeward position, the level of the
concentration fluctuations is much lower.

102

I c,

The profiles corresponding to configuration III are

indicated in Figure 7. In the cases x/H = 0,54 (a) and
1,080 (a), the chimney height matches the building
height and the profiles are gaussian. In the case
x/H = 0,54 (b), the ratio of the heights chimney/building
is 0,96 and in the lower part (close to the building roof)
a distortion is produced in comparison with the
Gaussian behaviour. These three cases correspond to
condition A in which the inertial effects are smaller, the
effective plume height (H + z) increases allowing the
escape, almost totally, of the plume from the wake
region. In cases x/H = 0,54 (c) and (d), the ratio of the
heights chimney/building is 1,00 and 0,93, respectively.
The inertial effects are bigger (condition B) diminishing
the plume effective height and therefore increases the
concentration level in the lower part (close to the roof).
The intensities of the concentration fluctuations have a
great drop in this region of the profile.

x/H=0.33 (inf)
x/H=0.33 (sup)
4

x/H=0.66 (inf)

x/H=0.540 (a)

x/H=0.66 (sup)

x/H=0.540 (b)
3

x/H=0.540 (c)

(z-z 0)/

1.6

x/H=1.00 (inf)

x/H=1.080 (a)
x/H=0.540 (d)

x/H=1.00 (sup)
Gauss

z/H

0
0.00

1.2

0.25

0.50

0.75

1.00

1.25

C/C0
5
x/H=0.60 (inf)
x/H=0.60 (sup)
4

x/H=1.20 (inf)
x/H=1.20 (sup)

0.8

x/H=1.80 (inf)
2.0

4.0

6.0

x/H=1.80 (sup)

(z-z 0 )/

0.0

K
2

Gauss

x/H=0.540 (a)
x/H=0.540 (b)
x/H=1.080 (a)

x/H=0.540 (c)
x/H=0.540 (d)

0
0.00

0.25

0.50

0.75

1.00

1.25

C/C0

1.6

z/H

Figure 8. Gaussian profile and mean concentration

experimental values, configuration I, conditions A (top)
and B (bottom).

1.2

0.8
0.1

10

100

Ic

Figure 7. Concentration profiles K and Ic,

Configuration III, condition A (x/H = 0,540a,
x/H = 0,540b, x/H = 1,080a) and B (x/H = 0,540c, d)
The experimental values were fit to the Gaussian
expression:
2
(3)
C( z) / C 0 = exp[ ( z z 0 ) 2 / 2 z ] ,
where C0 is the maximum value of the concentration
(central position of the plume), making it possible to
determine the vertical dispersion parameter z. Figure
8 shows the vertical nondimensional concentration
profiles corresponding to three leeward emission
positions, considering conditions A and B for
configuration I. Due to the profiles asymmetry, the
representation is made considering one value of the
parameter z for the upper region and another one for
the lower region. The fitness of the values to the
Gaussian profile is acceptable, but the quality
diminishes in the positions far from the emission
source.

103

In Figure 9 the values of the parameter z are

presented plotted against the distance x/H, for
conditions A and B, and configuration I. The values
taken to the atmospheric scale are comparable to those
from the PTG curves in the initial part, considering
atmospheric stability between neutral and slightly
unstable. The comparison of the experimental values
with the expressions of Briggs given by Zanetti [1990]
and Baechlin [1992], for the condition of neutral
atmospheric stability (D) in urban and rural terrains, is
indicated in the figure.
Making an extrapolation, the experimental values
have a good agreement with the results obtained by
Robins [2001] in wind tunnel tests, always considering
the approaching flow case that better fit to these tests.
In the proximities of a gas source emission the
dispersion process may be analyzed from the dilution
factor defined by the expression Dn = C0/C. the values
obtained in these experiments are compared with a
prediction model of the minimum dilution based on the
work of Chen & Wilson [1988]. This model considers
the diffusion in a turbulent flow of great scale where the
dispersion of the plume is proportional to the distance
from the emission. The minimum dilution, considering
the plume dispersion linear, varies with the square of
the distance and may be obtained from expressions
dependent of the relation between momentum M and
empirical constants. This kind of expression is used for
the control and prevision of the gas emissions [Saathoff
et al., 1998].

10000

0.1

z/H

1000
Cond. A (inf)
Cond. A (sup)

0.01

Dn=C0/Cmax

Cond. B (inf)
Cond. B (sup)
Urbano D
D [Briggs]
Urban
Rural D [Briggs]
0.001
0.10

1.00

Cond. E
Cond. H

100

Cond. F

10.00

x/H

Figure 9. Comparison of the experimental values of the

plume vertical dispersion in configuration I with the
expressions of Briggs for atmospheric stability (D)
[Zanetti, 1990].

Configurao I I
Configuration

10

Configurao IIII
Configuration
Configurao IIIIII
Configuration
Configurao IVIV
Configuration

In Figure 10 the experimental values and the

minimum dilution curves are presented for momentum
ratio M = 0.245 (conditions A and B) and M = 0.414
(condition C). The experimental values were always
larger than the minimum dilution. Configuration I, which
presents the closest values to Dmin, exhibits a defined
tendency. In the beginning of the plume, for conditions
E and F, the dilution increases, a result from the smaller
momentum ratio (Table 1). At a measurement distance
x/H = 1.80, for condition H, the dilution diminishes as a
consequence of the larger momentum ratio. This
behaviour is consistent with the influence caused by the
M ratio in the curves of minimum dilution.

3 Conclusions
The dilution analysis held and the comparison with
the curves of minimum dilution, show good agreement
with the experimental values obtained by & Wilson
[1988] and Saathoff et al [1998], for small gaseous
emissions at the level of the top of the buildings. From
the analysis of concentration fluctuations profiles, the
general behaviour is similar to that obtained by other
authors [Li & Meroney, 1983, Fackrell & Robins, 1982],
noting similar tendencies, even if the results obtained
by Li & Meroney [1983] are smaller at the plume
boundaries.
On the other hand, it is important to note that there is
a major asymmetry between the upper and lower
regions, consequence of the buoyancy generated by
the emission of a light gas as the helium. In both
theoretical models, for the analysis of the standard
deviation of the fluctuations, and in the experimental
studies, this asymmetry is not considered.
The plume emission and buoyancy conditions do
influence the process, but this influence is relevant only
in the configurations corresponding to the single
building near the emission source.
The distributions of the mean concentrations and of
the concentration intensities are, in general, similar to
gaussian configurations. The values measured in this
work, are in good agreement with the results of
theoretical models as well as experimental values
obtained in other studies. However, the possible
asymmetry of the concentration profiles is seldom
studied in the literature. This behaviour requires a
probabilistic
analysis
of
the
instantaneous
concentrations which is under study, including
considerations regarding the inttermitence in the
dispersion process.

104

Dmin (M = 0.245)
Dmin (M = 0.414)

1
0,1

1,0

10,0

x/H

Figure 10. Experimental values of the dilution and

the curves of minimum dilution.

Acknowledgments
The authors are gratefull to CAPES and CNPq for
then financial support conceded to two first authors.

References
Blessmann, J. The Boundary Layer Wind Tunnel of the
UFRGS. Journal of Wind Engineering and
Industrial Aerodynamics, Amsterdam, vol.10, 1982.
pp. 231-248.
Chui, E., Wilson, D. [1988], Effect of varing wind
direction on exhaust gas dilution, Journal of Wind
Engineering and Industrial Aerodynamics, 31, 87104.
Fackrell, J. E, Robins, A. G. [1982], Concentration
fluctuations and fluxes in plumes from point
sources in a turbulent boundary layer, J. Fluid
Mech., vol. 17, pp. 1-26.
Hanna, S. R. [1984], Concentration fluctuations in a
smoke plume, Atmospheric Environment, vol. 18,
No. 6, pp. 1091-1106.
Li, W., Meroney, R. N. [1983], "Gas dispersion near a
cubical model Building. Part II: Concentration
fluctuation measurements", Journal of Wind
Engineering and Industrial Aerodynamics, 12, 3547.
Robins, A., Castro, I., Hayden, P., Steggel, N., Contini,
D., Heist, D. [2001], A wind tunnel study of dense
gas dispersion in a neutral boundary layer over a
rough surface, Atmospheric environment 35,
2243-2252.
Saathoff, P., Stathopoulos, T., Wu, H., [1998], The
influence of the turbulence en near field dilution of
exhaust from building vents, Journal of Wind
Engineering and Industrial Aerodynamics, 77&78,
741-752.
Zannetti, P. [1990], "Air pollution modeling: Theories,
computational methods and available software",
Comp. Mech. Publications, Van Nostrand Reinhold,
New York.

Improved Building Dimension Inputs for AERMOD Modeling of the

Mirant Potomac River Generating Station
Ronald L. Petersen, John J. Carter
CPP, Inc.
Fort Collins, Colorado USA
rpetersen@cppwind.com
jcarter@cppwind.com

Abstract - Awind tunnel study was conducted for

the Mirant Potomac River Generating Station
because a USEPA recommended computer
dispersion model, AERMOD, predicted high impacts
of plant emissions on a nearby high-rise residential
tower. The heights of the stacks at power station
were restricted due to the proximity of the power
plant to Reagan National Airport. The purpose of the
wind tunnel study was to obtain a better
understanding of the concentration spatial
distribution on and around residential tower and to
provide site-specific building dimension inputs
(Equivalent Building Dimensions or EBD) for
AERMOD. This paper focuses on the methods used
to obtain the EBD and provides a comparison
between the EBD and the AERMOD building preprocessor (BPIP) determined input values.
Key words Dispersion Modeling, Equivalent Building
Dimensions, EBD, AERMOD.

1 Introduction
This paper describes a wind tunnel study conducted
for the Mirant Potomac River Generating Station
(MPRGS) located as shown in Figure 1. The study was
commissioned by Mirant because a USEPA
recommended computer dispersion model, AERMOD
[1], predicted high impacts of plant emissions on a
nearby high-rise tower (Marina Towers) as shown in
Figure 2. The tower was built near the plant in the
1970s, without the benefit of site-specific modeling or a
wind tunnel study, long after the MPRGS was built in
1949. The heights of the stacks at MPRGS were
restricted due to the proximity of the power plant to
Reagan National Airport.

105

Figure 1. Location of area modeled (near Washington,

DC, USA
Since AERMOD was predicting high concentration
levels on the Marina Towers and at various ground level
locations surrounding MPRGS, a wind tunnel study was
undertaken to obtain a better understanding of the
concentration spatial distribution on and around the
Marina Towers and to provide site-specific building
dimension inputs for AERMOD.[2]
Since AERMOD has advanced building downwash
and plume rise modeling capabilities, it was anticipated
that if the correct building dimensions are input into the
model it will produce accurate concentrations estimates.
The Equivalent Building Dimension (EBD) values are
the building height, width, length and position that
should be input into AERMOD to allow the model to
produce an accurate representation of concentration
spatial distributions. When a single solid rectangular
building is adjacent to a stack and the wind flow is
perpendicular to a building face, the actual building
dimensions are the appropriate inputs. An estimate of
these dimensions for each wind direction is normally
determined using the Building Profile Input Program
(BPIP). For more complicated situations, such as for this
application, the use of EBD values for model input will
result in more accurate concentration estimates, and

hence, an optimal determination of any required

mitigation.
The only practical manner for determining EBD is
through the use of wind tunnel modeling using a detailed
physical model of MPRGS and nearby structures
including the Marina Tower. [3,4,5,6,7] The wind tunnel
testing was divided into two main phases for both the
current and future design of the MPRGS: EBD
determination for predicting ground-level concentrations
where the downwash effects of MPRGS and the Marina
Towers are present; and EBD determination for
determining concentrations on the Marina Towers with
the downwash effects of MPRGS present.
This paper focuses on the methods used to obtain
the EBD and provides a comparison between the EBD
and BPIP determined input values. The general use of
the EBD values is discussed in a paper by Petersen. [8]
The use of the EBD values developed in this study is
discussed in Shea. [9]

2 Similarity Requirements
To model plume trajectories for EBD determination,
the velocity ratio, R (Ve/Uh), and density ratio, (s /a)
were matched in model and full scale, where Uh = wind
velocity at stack top (m/s), Ve = stack gas exit velocity
3
(m/s), s = stack gas density (kg/m ), and a = ambient
3
air density (kg/m ). In addition, the stack gas flow in the
model was fully turbulent upon exit as it is in the full
scale.
To simulate the airflow and dispersion around the
buildings, the following criteria were met as
recommended by EPA [10]: 1) all structures within a
518-m radius of the stacks were modeled at a 1:300
scale reduction; 2) appropriate mean and turbulent
approach boundary layer was established; 3) building
Reynolds number independence was verified through
testing; 4) a neutral atmospheric boundary layer was
established simulating an approach surface roughness
of 0.79 m for wind directions of 175-360 (urbanized
sector) and 0.15 m for all other wind directions (water
and low roughness sector).
The above scaling parameters were used to
determine the model operating conditions. It should be
noted that the use of these scaling parameters is the
recommended method for determining GEP stack
heights by EPA [11] and have been used on past EBD
studies. The use of these scaling parameters does not
include an exact simulation of full buoyancy, and as a
result, full-scale plume rise is underestimated (i.e., a
conservative scaling approach). If one wants to compare
the wind tunnel results with AERMOD, the full scale
source parameters have to be back calculated from the
conditions set in the wind tunnel by using the
appropriate buoyancy and momentum scaling method.
[10] These full scale conditions are provided in Table 1.

Table 1. Model inpus

3 Model Construction and Setup

A 1:300 scale model of the MPRGS and surrounding
structures and terrain was constructed. The model
included all significant structures within a 518-m radius
of the center of the MPRGS. Figure 2 is a close-up plan
view drawing of a portion of the area modeled. The
model was placed on a turntable so that different wind
directions could be easily evaluated. Photographs of the
model are provided in Figures 3-5. Stacks were
constructed of resin using a rapid-prototyping technique
and were supplied with a heliumhydrocarbon (or
nitrogen-hydrocarbon) mixture of the appropriate
density. Measures were taken to ensure that the flow
was fully turbulent upon exit. Precision gas flow meters
were used to monitor and regulate the discharge
velocity.

Figure 2.Close-up plan view of area modeled

106

located in each of 7 rows perpendicular to the wind

direction. Background samples collected upwind of the
stacks. The receptor arrays were designed to define the
maximum concentration in the lateral and longitudinal
directions.
For the site structures portion of the Marina Towers
testing, the Marina Towers were instrumented with 46
sampling taps that closely matched the locations
modeled with AERMOD (see Figure 4). Roof-top, sidewall and ground-level locations were evaluated. For the
EBD portion of the Marina Towers testing, a flagpole
receptor grid was constructed such that sampling
locations were the same as when the Marina Towers
were in place (see Figure 5). The EBD structures were
placed directly upwind of the stack on the model
turntable with the MPRGS removed.
Figure 3. Photograph of model in the wind tunnel

4 Determination of EBD Values

4.1 Ground Level

Figure 4. Close-up of MPRGS and Marinta Tower

Figure 5. Close-up of MPRGS and flagpole receptor grid

A set of solid rectangular structures was fabricated
for placement directly upwind of each stack for EBD
testing. The structures had height-to-width-to-length
ratios of: 1:2:1; 1:3:1; and 1:4:1. For the ground-level
EBD tests, the stacks in Table 1 and the idealized
buildings were tested with the turntable model removed
from the wind tunnel and a uniform roughness installed
in its place. The uniform roughness was constructed
such that it provided the same surface roughness as the
site surroundings (i.e., 0.79 m for the urbanized
approach, and 0.15 m for the water and lower
roughness
approach).
For
the
ground-level
concentratino testing, sampling taps were installed on
the surface of the model so that at least 46 locations
were sampled simultaneously for each simulation. A
typical sampling grid consisted of 5 to 7 receptors

107

The basic modeling approach for determining EBD

values is to first document the ground level
concentration profiles as a function of wind direction with
all significant nearby structure wake effects included in
the wind tunnel model. Next, a ground level
concentration profile is collected with an equivalent
building positioned directly upwind of the stack in place
of all nearby structures. This testing is conducted for
various equivalent buildings until an equivalent building
is found that provides a profile of maximum ground-level
concentration versus downwind distance that is similar
(within the constraints defined below) to that with all site
structures in place. This effort is repeated for each wind
direction of interest.
The criteria for defining whether or not two
concentration profiles are similar is to determine the
smallest building which: 1) produces an overall
maximum concentration exceeding 90 percent of the
overall maximum concentration observed with all site
structures in place; and 2) at all other longitudinal
distances, produces ground-level concentrations which
exceed the ground-level concentration observed with all
site structures in place less 20 percent of the overall
maximum ground-level concentration with all site
structures in place.
To demonstrate the method for specifying the EBD
values, consider Figure 6 which shows a typical result
from this study. The figure shows the maximum groundlevel concentration versus downwind distance for five
different equivalent buildings and the maximum
concentration measured with site structures in place.
Within this figure, the concentration profile for EBD 11.5
meets the first criterion in that the maximum measured
concentration is at least 90 percent of the maximum
concentration measured with the site structures in place.
(Note, the 11.5 is building height in model centimeters.
Multiply by 3 to obtain the full-scale height in meters).
However, the EBD 11.5 profile fails the second criterion
at the third actual site data point (at approximately 180
m downwind) where the lower bound of the error bar is
greater than the interpolated concentration value for
EBD 11.5. Therefore, the equivalent building for the test
case shown in Figure 6 is EBD 12, since EBD 12 is the
smallest equivalent building which meets both criteria.

Figure 7. Typical flagpole EBD results for BS4 with BS5

also operating
Figure 6. Typical ground level EBD results for BS4

5 Results

4.2 Marina Towers

In addition to traditional EBD values used to predict

ground-level concentrations, EBD values were also
determined to serve as an AERMOD input to predict
concentrations on the Marina Towers.
In this case, the approach for determining EBD
values was to first document concentration levels on the
Marina Towers as a function of wind direction with all
nearby structures included in the model. Next, the
MPRGS and the Marina Towers were removed but the
rest of the site remained the same (i.e., surrounding
buildings, terrain, etc.). The Marina Towers were
replaced with a flagpole receptor grid with sampling
points at the same locations as those obtained when the
Marina Towers were in place to replicate the way the
receptors are represented in AERMOD. The stack under
evaluation was placed in the appropriate location and
various EBD structures were placed directly upwind of
the stack.
The determination of EBD values for concentrations
on the Marina Towers is similar to that described in the
previous section with the exception of the type of data
profile evaluated. Concentration data for the flagpole
receptors were ranked from largest to smallest for both
the site structures and EBD tests. The site structure
profiles were then compared with the EBD profiles. The
criteria for defining whether two concentration profiles
was the same as described above for the ground level
tests.
Figure 7 shows typical results from this study. The
profiles for the EBD 8 1:4:1, EBD 9 1:2:1, EBD 9 1:4:1,
and EBD 10 1:4:1 meet the first criterion in that the
maximum measured concentration is at least 90 percent
of the maximum concentration measured with all site
structures in place. (Note: the three numbers following
the building height specify the building height to width to
length ratios). However, EBD 8 1:4:1 and EBD 9 1:2:1
both fail the second criterion of 20 percent of the overall
maximum ground-level concentration with all site
structures in place.
Of the two profiles that meet both criteria, the lowest
value can be chosen as the EBD structure. Therefore,
the EBD for this case is EBD 9 1:4:1 denoted by a white
diamond in Figure 7.

108

5.1 Ground Level

For the ground-level EBD portion of the study, wind

tunnel tests were first conducted for the existing and
merged (future) exhaust stacks for the wind directions of
interest with all site structures in place as shown in
Figure 4. The full-scale exhaust information for the
various exhausts is listed in Table 1. The MPRGS
consists of five boiler exhaust stacks with stack
identification labels of BS1, BS2, BS3, BS4, and BS5
representing stacks one through five as shown in Figure
2. Stacks BS1 and BS2 were simulated with the same
source parameters (i.e., exit diameter, volume flow,
temperature, etc.) while BS3, BS4 and BS5 were
simulated with the same source parameters.
Due to the proximity of the airport, MPRGS cannot
significantly raise the stack heights to increase the
plume height and escape building wake effects.
However, it is possible to merge the exhaust streams to
accomplish the desired objective of redesigning the
plant to reduce concentrations on the Marina Towers.
Therefore merged exhausts, MS1 and MS2, were
evaluated at the stack positions normally occupied by
stacks BS1 and BS4, as shown in Figure 2. MS1
represents combining the BS1 and BS2 exhausts
through one merged-flue stack, while MS2 represents
combining the BS3, BS4, and BS5 exhausts through
another merged-flue stack. The stacks (BS1, BS4, MS1
and MS2) were evaluated for wind directions of 10
through 360 degrees at ten degree increments.
The EBD structures initially tested had height-towidth-to-length ratios similar to those used by Huber and
Snyder [12,13] for development of the ISC2 downwash
algorithm (H:W:L = 1:2:1). For cases where the
traditional EBD did not provide an adequate
concentration profile, alternate EBD configurations were
assessed. For example, wider EBD structures with the
ratios of 1:3:1,1:4:1, etc. were very effective. For
certain cases, the best EBD configuration that resulted
in the proper profile was with the EBD turned at a 45degree angle to the approach flow resulting in a corner
vortex affecting the exhaust plume. Unfortunately,
AERMOD does not allow this type of configuration for
input. Petersen[2] provides a listing of building
dimensions that were evaluated and the EBD values
chosen for each exhaust stack and wind direction
scenario.
To illustrate the difference between the BPIP and
EBD determined building dimension inputs, only the
results for BS4 will be discussed in detail. The variation

in building dimensions and building location versus wind

direction for the two methods are shown in Figures 8-12.

BS4

50.0

BS4

0.0

50.0
45.0

10

20

30

40

50

60

70

80

90

100

110

120

130

140

150

160

170

180

190

200

210

220 230 240 250 260 270 280 290 300

310

320 330 340 350 360

-50.0

XBADJ (M)

40.0

BUILDHGT (m)

35.0
30.0

-100.0

-150.0

25.0

-200.0
20.0

-250.0

15.0
10.0

-300.0
Wind Direction (degrees)

5.0
EBD

0.0

10

20

30

40

50

60

70

80

90

100

110

120

130

140

150

160

170

180

190

200

210

220

230 240

250 260 270

280 290

300

310

320

330 340

350 360

EBD

BPIP

Figure 8. Building height, BUILDHGT, determined using

BPIP and EBD methods

BS4

BS4

60.0

40.0

180.0
160.0

20.0

YBADJ (M)

140.0
120.0
BUILDWID (M)

BPIP

Figure 11. Building position in X direction, XBADJ,

determined using BPIP and EBD methods

Wind Direction (degrees)

0.0

10

20

30

40

50

60

70

80

90

100

110

120

130

140

150

160

170

180

190

200

210

220

230 240 250 260

270 280 290 300

310

320 330 340

350 360

-20.0

100.0
-40.0

80.0
60.0

-60.0

40.0
-80.0
Wind Direction (degrees)

20.0
EBD

0.0

10

20

30

40

50

60

70

80

90

100

110

120

130

140

150

160

170

180

190

200

210

220 230 240 250

260 270 280 290 300

310

320

330 340 350 360

Wind Direction (degrees)

EBD

BPIP

Figure 9. Building width, BUILDWID, determined using BPIP

and EBD methods

BS4

180.0
160.0
140.0

BUILDLEN (M)

120.0
100.0
80.0
60.0
40.0
20.0
0.0

10

20

30

40

50

60

70

80

90

100

110

120

130

140

150

160

170

180

190

200

210

220 230 240 250

260 270 280 290 300

310

320

330 340 350 360

Wind Direction (degrees)

EBD

BPIP

Figure 10. Building length, BUILDLEN, determined using

BPIP and EBD methods

109

BPIP

Figure 12. Building position in Y direction, YBADJ,

determined using BPIP and EBD methods
For the building height comparison, shown in Figure
8, BPIP designates the same dimension for all wind
directions with the exception of 330 to 360 degrees
which show an increased dimension that corresponds
with the height of the Marina Towers structure. The EBD
values change by wind direction which reflects the true
variation of downwash based on the wakes created by
all site structures and structure positions relative to the
stack.
Figure 9 shows the building width comparisons for
BPIP and EBD. It is apparent that BPIP is using the long
dimension of the power plant for winds from the east
and west and the short dimension for winds from the
north and south. From this plot, inclusion of the Marina
Towers structure upwind of the power plant is not
apparent. The BPIP and EBD values show similar
trends, with the exception of the 140 and 160 degree
wind directions. For these cases, wider EBD dimensions
were necessary to replicate the downwash attributed to
corner vortex shedding on the power plant structure.
Figure 10 shows significantly different values for the
building length component produced by BPIP and EBD.
In this plot, BPIP appears to be choosing the long
dimension of the power plant when winds are from the
north and south and the short dimension when winds
are from the east and west. Again, from this plot,
inclusion of the Marina Towers structure upwind of the
power plant is not apparent. The EBD values do not
demonstrate a drastic change in the dominant length
component.
Figure 11 is an excellent indicator that BPIP is
choosing the Marina Towers structure to the north of the

power plant as the dominant structure when winds are

from 330 to 360 degrees. From 20 to 100 degrees, the
BPIP and EBD values are similar. For all other wind
directions, the BPIP values are slightly greater or less
than the EBD values depending upon wind direction.
Figure 12 shows that all EBD structures were
centered on the exhaust stack for all wind directions.
The BPIP values, on the other hand shift depending on
wind direction. At the 330 through 360 directions, it is
likely the values are attributed to the Marina Towers
structure upwind of the power plant.

5.2 Marina Towers

Not all wind directions were relevant for determining

EDB values to predict concentrations on the Marin
Towers. Only 150, 160, 170, and 180 degrees were
simulated with the site structures in place as these
encompass the wind directions where the MPRGS
affects the Marina Towers. Measurements were
obtained for various exhaust stack and wind direction
combinations as specified in Petersen.[2]
For these cases, traditional EBD values with the
1:2:1 relationships were initially evaluated and if
necessary, other types of EBD configurations were
evaluated. The values chosen for each exhaust stack
and wind direction scenario are listed in Petersen.[2]
Shea [9] discusses the use of these values in AERMOD
for predicting the concentratinos on the Marina Towers
and also compares the results of the estimates to field
observations.

6 Conclusions
This analysis has demonstrated that EBD and BPIP
determined building dimension inputs are significantly
different. The EBD building dimension inputs are based
on a characterization of the wake effects created by all
site structures. The BPIP determined building inputs are
based on logic algorithms that consider building tiers,
building spacings and building angles to the wind.
Unfortunatly, the BPIP inputs may or may not be
appropriate to characterize building wake effects for the
site under evaluation and may result in over- or
underpredicted concentrations.
For this particular study, the effect of the EBD values
on concentration predictions is discussed elsewhere.[9]

References
1. Cimorelli, A.J., S.G. Perry, A. Venkatram, J.C. Weil,
R.J. Paine, R.B. Wilson, R.F. Lee, W.D. Peters, and
R.W. Brode. AERMOD: A Dispersion model for
Industrial Source Applications. Part I: General
Model Formulation and Boundary Layer
Characterization. JAM, 44, 682-693. American
Meteorological Society, Boston, MA. (2005).
2. Petersen, R. and J. Reifschneider. Wind Tunnel
Modeling Evaluation for the Mirant Potomac Power
Generating Station, CPP, Inc. Wind Engineering
and Air Quality Consultants Report 05-3527, Fort
Collins, CO., August 2006.
3. Tikvart, J.A. , Chief, Source Receptor Analysis
Branch, United States Environmental Protection
Agency, Letter to Brenda Johnson, Regional
Modeling Contact, Region IV and Douglas Neeley,
Chief Air Programs Branch, Region IV, July 25,
1994.
4. Petersen, R. L., D. N. Blewitt, and J. A. Panek,
Lattice Type Structure Building Height

110

Determination for ISC Model Input, 85th Annual

AWMA Conference, Kansas City, Missouri, June
2126, 1992.
5. Petersen, R. L., and B. C. Cochran, Equivalent
Building Height Determinations for Cape Industries
of Wilmington, North Carolina, CPP Report No.
930955, CPP, Inc., Ft. Collins, Colorado, 1993.
6. Petersen, R. L., and B. C. Cochran, Equivalent
Building Dimension Determination and Excessive
Concentration Demonstration for Hoechst Celanese
Corporation Celco Plant at Narrows, Virginia, CPP
Report No. 931026, CPP, Inc., Ft. Collins,
Colorado, 1995a.
7. Petersen, R. L., and B. C. Cochran, Equivalent
Building Dimension Determinations for District
Energy St. Paul, Inc. Hans O. Nyman Energy
Center, CPP Report No. 93-0979, Ft. Collins,
Colorado, 1995b.
8. Petersen R. L., and J. Carter, Evaluation of
AERMOD/PRIME for Two Sites with Unusual
Structures, 99th Annual AWMA Conference, New
Orleans, Louisiana, June 20-23, 2006.
9. Shea, D., O. Kostrova, A. MacNutt, R. Paine, D.
Cramer, L. Labrie, A Model Evaluation Study of
AERMOD Using Wind Tunnel and Ambient
Measurements at Elevated Locations, 100th
Annual AWMA Conference, Pittsburgh, PA, June
2007.
10. Snyder, W. H., Guideline for Fluid Modeling of
Atmospheric Diffusion, USEPA, Environmental
Sciences Research Laboratory, Office of Research
and Development, Research Triangle Park, North
Carolina, Report No. EPA600/881009, 1981.
11. EPA, Guideline for Use of Fluid Modeling to
Determine Good Engineering Practice Stack
Height, USEPA Office of Air Quality, Planning and
Standards, Research Triangle Park, North Carolina,
EPA450/481003, July 1981.
12. Huber, A. H, and W. H. Snyder, Building Wake
Effects on Short Stack Effluents, Preprint Volume
for the Third Symposium on Atmospheric Diffusion
and Air Quality, American Meteorological Society,
Massachusetts, 1976.
13. Huber, A. H., and W. H. Snyder, Wind Tunnel
Investigation of the Effects of a RectangularShaped Building on Dispersion of Effluents from
Short Adjacent Stacks, Atmospheric Environment,
Vol. 16, No. 12, pp. 28372848, 1982.

Physical modeling of the downwash effect of rooftop structures

A. Gupta
Centre for Building Studies
Concordia University
amit_concordia@yahoo.com

P. Saathoff
Centre for Building Studies
Concordia University
saathoff@bcee.concordia.ca

T. Stathopoulos
Centre for Building Studies
Concordia University
statho@bcee.concordia.ca

Abstract
A wind tunnel study was carried out to investigate the influence of a rooftop structure (RTS) on the dispersion
of exhaust from a downwind rooftop stack. Previous studies have shown that downwash produced by the RTS
can significantly increase roof level concentrations compared to those obtained with a flat-roofed building
[Saathoff et al. (2002, 2003), Gupta et al. (2005)]. These studies indicated that the severity of RTS downwash
depends on a number of factors such as building height, wind direction and crosswind width of the RTS. Of
primary importance is the location of the RTS with respect to the separated flow region on the roof of the
building. For downwash to occur, the RTS must be outside the separated flow region.
In the present study, the downwash effect of an RTS was investigated for different stack locations. The
experiments were performed on a square plan building with a height of 15 m and a width of 50 m. The RTS
height (h) and length (l) were 4 m and 8 m, respectively, and the width (w) varied from 10 m to 50 m. Figure 1
shows a plan view of the test building.
Wind tunnel experiments were carried out for five stack locations (S1-S5), as indicated in Figure 1. The distance
from the stack to the RTS (xc) varied from 0.5h to 2.5h. Results were obtained for stack heights of 0.25h to
1.75h and for wind directions () of 0 and 45 degrees. A key parameter is the exhaust momentum ratio, M,
defined as the ratio of exhaust speed (we) to wind speed at building height (UH). Values of M varied from 1 to 5,
which is a typical range for building emission sources. Experiments were carried out with and without the RTS
using SF6 as the tracer gas. Mean concentration measurements were obtained on the roof surface along the
plume center-line.
The study aims at proposing simple rules of thumb in order to determine the required stack height (hreq) to avoid
the downwash effect of an RTS. Figure 2 shows the variation of normalized hreq with exhaust momentum M for
stacks S1 to S5 for an RTS with w/h = 7.5. As expected, the value of hreq decreases significantly as M and xs
increase. Figure 2 shows that for a typical design M-value of 2.0, the required stack height is reduced from 2.5h
(10m) to approximately 1.5h (6m).
= 0

UH

= 45
UH

w/h = 7.5

x RTS
y
S1
S3

L = 50

S1

S2

S2

h s /h

xS

S4

S5

S3
S5

h
H = 15m

Stack with height hs

Stack
S1
S2
S3
S4
S5

S4

Figure 1 Plan view of the test setup

0
0

xs/h
0.5
1.0
1.5
2.0
2.5

Exhaust momentum
(M)
M

Figure 2 Stack height required to avoid downwash due to an RTS

References
Saathoff, P., Gupta, A., Stathopoulos, T., Lazure L., 2003. Effect of roof top structures on the plume from a
nearby stack. Proceedings of 11th International Conference on Wind Engineering, Lubbock, Texas, USA, June 2-5.
Saathoff P., Stathopoulos T., Lazure L, Peperkamp H. (2002) The influence of rooftop structure on the dispersion of
exhaust from a rooftop stack ASHRAE Transactions, v.108, Pt.2

Gupta, A., Saathoff, P., Stathopoulos, T., (2005) Effect of building orientating on downwash effect of rooftop
structures, Proceedings of Physmod 2005, London, Ontario, Canada, August 24-26

111

Dispersion from an area source in urban-like roughness

Frauke Pascheke12 , Janet.F. Barlow1 and Alan Robins2

Department of Meteorology, University of Reading, Reading, RG6 6BB, UK

2
School of Engineering, University of Surrey, Guildford, GU2 7XH, UK
E-mail: f.pascheke@surrey.ac.uk

Abstract - The ventilation of scalars from urban-like

geometries at neighbourhood scale was investigated
using two geometries with an equal plan and frontal
density of p = f = 25% but different height distributions: a uniform height roughness (C10S) and a nonuniform height roughness (RM10S). In both configurations the area source was represented by a sub-unit
of the idealised geometries coated with a thin layer
of naphthalene. The naphthalene sublimation method
(Barlow and Belcher, 2002) was used to measure directly total area-averaged transport of scalars out of
the complex geometries. At the same time, naphthalene vapour concentrations controlled by the turbulent fluxes were detected using a fast Flame Ionisation
Detection (FID) technique.
Particular emphasis was given to testing the
Reynolds number independence of the measured
naphthalene concentrations since for low windspeeds, transfer from the naphthalene surface is determined by a combination of forced and natural convection (Goldstein and Cho, 1995). The comparison
with a propane point source release, showed that for
the naphthalene area source a 25% higher free stream
velocity was needed to yield Reynolds number independent concentration fields.
The ventilation transfer coefficients wT /U derived
from the naphthalene sublimation method showed
that, whilst there was enhanced vertical momentum
exchange due to obstacle height variability, advection
was reduced and dispersion from the source area was
not enhanced. Thus, the height variability of a canopy
is an important parameter when generalising urban
dispersion.
Fine resolution concentration mappings in the
canopy showed the effect of height variability on dispersion at street scale. Individual high-rise buildings
play an important role on local ventilation processes.
At street level, the tallest obstacle (heights are indicated by numbers) affected the flow around nearby
roughness elements. Rapid vertical transport in the
wake of individual high-rise obstacles was found to
generate elevated point-like sources.
A more comprehensive description and discussion
of this study including further particulars is given in
Pascheke et al. (2007) and should be referred to when
cited.
Key words - Area source, area-averaged turbulent

fluxes, concentration measurements, naphthalene

sublimation, urban heat and pollutant dispersion.

Introduction
Urban pollutant concentrations are controlled by two
dispersion processes, which are acting on similar
time scales: horizontal advection through the urban canopy and turbulent exchange with cleaner air
above (ventilation). Within the urban canopy layer

local flow characteristics are determined by the surrounding buildings. Wake interactions result in highly
efficient mixing and diffusion processes for momentum and scalar properties. The scales of the larger
eddies involved in that process are determined by
the building dimensions. Therefore, a greater surface roughness caused by geometrical diversity and
variable building heights induce high turbulence levels which in turn intensify dispersion processes and
canopy ventilation. On the other hand, increased
momentum exchange from the flow to the surface
due to form drag causes a reduction of the advection
velocity through the canopy. Whether urban canopy
ventilation is increased or decreased as roughness
is increased thus depends on the relative magnitudes of these opposing effects (Britter and Hanna,
2003).
Flow parameterizations are often based on morphological descriptors such as plan area density
p and frontal area density f . Oke (1987) for instance described the flow in and above the canopy
as isolated, wake interference or skimming flow, with
thresholds based on values of the frontal index f .
However, based on systematic windtunnel experiments over a wide range of urban-like roughness arrays, Hall et al. (1996) pointed out the importance
of height variability in inhibiting the skimming flow
regime. Cheng and Castro (2002) performed windtunnel experiments over idealised urban-like rough
surfaces and found that that an array with nonuniform height roughness elements with the same
plan area density p and frontal area density f as
an uniform height array is more efficient in generating surface stress. Both studies indicate that certain
dynamical mechanisms acting in more realistic geometries may not be captured in flow parameterizations based on p and f .
Another important factor in a morphologically
based parametrization of urban canopy turbulence
may be the skewness of the building height distribution. This concept is supported by results from
a recent work of Heist et al. (2005). They studied
the effect of a single high-rise building on residence
times of pollutants in a regular array of cubical buildings and found that directly upwind of the tall building
characteristic time scales of the pollutant venting decrease significantly as the building height increases.
This finding suggests that ventilation is particularly
enhanced by rapid vertical dispersion in the wake of
tall buildings.
Urban pollution is caused by numerous (mobile)
point sources. However, for studies at neighbourhood scale it is helpful to define an area source.
The naphthalene sublimation technique provides a
means of directly simulating an area source and

113

also measuring the total area-averaged transport of the plan area density and frontal area density were
scalars out of any given geometry. The ventilation equal, i.e. p = f = 25%.
efficiency of the surface can be quantified by calculating a transfer coefficient (equivalent to a Stanton number, in engineering terms). A full description
of the technique and its use for the study of scalar
transfer from complex surfaces is reviewed by Goldstein and Cho (1995). Recently, Barlow and Belcher
(2002) and Barlow et al. (2004) applied the method
to study the ventilation of idealised 2-dimensional
street canyons.
A key objective of the present study was to investigate the effect of varying roof height on scalar
1: Sketch of 3D and plan view of one unit for array with uniform
dispersion within and above the urban canopy. The Figure
heights (C10S). All dimensions in mm, element heights are indicated on
rough surfaces chosen for the study were two of the plan view. Mean flow along x-axis.
those studied by Cheng and Castro (2002), having equal f and p but with respectively uniform
and non-uniform height buildings. The use of Flame
Ionisation Detectors (FIDs) for the measurement of
naphthalene concentrations was required. The development of this technique, which is entirely novel,
is described in Section 1. A careful study of the effect of Reynolds number on the measurements is
presented in Section 2. Finally, the transfer coefficients and concentration fields for the two surfaces
are presented and discussed in Sections 3 and 4.

Figure 2: Sketch of 3D and plan view of one unit for array with non-uniform
heights (RM10S). All dimensions in mm, element heights are indicated on
the plan view. Mean flow along x-axis.

Experimental Set-up

1.1 Boundary layer windtunnel

The experiments were performed in the A boundary
layer wind tunnel of the Environmental Flow and Research Centre (EnFlo). The test section of the opencircuit wind tunnel is 0.9 0.6 4, 5 m (W H L).
The wind tunnel set-up is identical to the flow experiments by Cheng and Castro (2002) using urbanlike roughnesses labelled C10S and RM10S. In each
case, the entire tunnel floor was covered with repeated units of the specified roughness and no additional devices commonly used to establish boundary
layer conditions, such as spires or turbulence grids,
were used. The test section blockage and the
pressure gradient along the test section were negligible:
=

Amodel
0.02,
Awindtunnel

p 2
< 0.006.
x U2

1.3

Naphthalene sublimation technique

To simulate a scalar area source, the naphthalene sublimation technique was used (Barlow and
Belcher, 2002; Goldstein and Cho, 1995). Naphthalene sublimes readily at room temperature, and
if applied to a surface, a thin layer of saturated naphthalene vapour is rapidly formed and maintained in
the immediate vicinity of the surface. Given that
the surface temperature is uniform and constant for
the time of an experimental run, the naphthalene
vapour pressure above the surface will also be uniform and constant, representing a constant concentration boundary condition. The area- and timeaveraged flux from the surface F is given by:
F = w T (S ) ,

(1)

where over-bars denote temporal averages and angled brackets denote spatial averages. w T is the
transfer velocity and (S ) is the difference between the mean naphthalene vapour density at the
active surface and the mean naphthalene vapour
density in the free stream. The naphthalene vapour
density in the free stream was negligibly small
( 5 108 kg m3 , whith typical surface vapour
1.2 Roughness surfaces
densities of S 5 104 kg m3 ). The flux density
The two urban-like roughnesses with different ge- F adjacent to the source is derived directly from the
ometries were composed of many repeated units, mass loss m of naphthalene from the coated area
each 80 80 mm in size and made of moulded A taking place during an experimental run of time t:
plastic material (Acrylonitrile Butadiene Styrene). A
single unit consisted of 16 uniform height cubes
m
F =
.
(2)
(C10S), or non-uniform height cuboids respectively
At
(RM10S), which were arranged in staggered arrays
as shown schematically in Figures 1 and 2. The
The ideal gas law gives the mean naphthalene
cube side length was 10 mm. For both geometries, vapour density at the surface for a given surface tem(X, Y, Z) are the stream-wise, lateral and vertical
coordinates respectively. The plane X = 0 is at the
beginning of the roughness, Y = 0 denotes the test
section centre line and the plane Z = 0 is at ground
level.

114

perature
S =

eS
RN TS

(3)

where RN = 64.95 J mol1 K1 is the gas constant

for naphthalene and TS is the surface temperature
in Kelvin. The naphthalene saturation vapour pressure at the surface can be derived from the following
empirical expression (C.R.C., 1993)
eS = 10

13.57 3729
TS

(4)

Combining Equations 1 and 2, the transfer velocity

is given by
m
.
A t S

wT =

(5)

The relationship between transfer velocity and reference windspeed U is an independent function for
each surface geometry and source location:
wT = CT U

(6)

where CT is the transfer coefficient. Barlow and

Belcher (2002) demonstrated that the relationship is
close to linear for sufficiently high wind speeds, thus
allowing easy comparison of the transfer coefficient
for different urban surface geometries.
In terms of dispersion, the flux density F is equivalent to the source area emission rate QA . Implicit in
Equation 6 is the fact that the emission rate for naphthalene vapour is dependent on windspeed, which
contrasts with more traditional trace gases which are
released at a fixed emission rate. Hence, for sufficiently high windspeeds, the measured naphthalene
vapour density is invariant with windspeed (see section 1.5.4 for discussion of this) and only requires
normalization by the surface vapour density to give

N = .
S

(7)

For a fixed emission rate source, the vapour density would be normalized by a given reference windspeed Uref and area emission rate QA to give
F =

Uref
.
QA

(8)

Recalling that F QA and 0, by substitution of Equations 1 and 6 into Equation 7, it can be

shown that
N = CT F ;

(9)

that is, the normalized naphthalene vapour density

field need only be divided by the transfer coefficient
CT to be fully equivalent to the normalized vapour
density field due to a fixed emission rate source.

X = 2960 mm and placed symmetrically about the

wind tunnel centreline (Y = 0 mm). For each surface, an area source unit with geometrical features
as shown in Figures 1 and 2 was manufactured from
aluminium. Since aluminium has a high thermal conductivity, temperature gradients across the unit surface were negligible. Each unit was briefly dipped
into a beaker of molten naphthalene to coat the
street surface only (simulating a ground level area
source). A reproducibility of the naphthalene layer
thickness of naph = 0.7 0.1 mm was achieved. Any
spilling and misplaced naphthalene residuals were
accurately removed to ensure only the street was
coated. Under the chosen experimental conditions,
a recoating of the surface was required after approximately three hours to ensure full surface coverage
at all times.

1.5
1.5.1

Measurement techniques
Temperature

A computer controlled procedure was developed

specifically to keep the laboratory temperatures constant to within 0.2 C. The surface temperature
of the coated unit, as well as the ambient air
temperature inside the wind tunnel, were continuously recorded using negative temperature coefficient 10 k thermistors. Both thermistors were
calibrated against a precision platinum resistance
thermometer with an absolute accuracy better than
0.01 C. The surface thermistor was plugged into a
small recess in the bottom of the coated surface of
the unit and itself coated with heat sink compound
to ensure good thermal contact with the aluminium
unit.

1.5.2

Mass

To measure the mass loss from the coated surface

a mass balance with an accuracy of 0.1 mg was
used. The coated surface unit was weighed prior
to a run, and then removed from the wind tunnel
and re-weighed as quickly as possible after a run
to minimise losses and hence errors due to natural
convection. On average, the total duration for the
mounting, dismounting and weighing of the test unit
took no longer than 90 seconds. Nevertheless, the
mass loss rate due to natural convection was determined from additional tests, which included all steps
in the procedure except the running of the wind tunnel. For typical test runs with a free-stream velocity
Uref between 5 and 20 ms1 , the error in transfer velocities due to natural convection was determined to
be < 0.8% . All subsequent transfer velocities were
corrected for the mass loss due to natural convection.

1.5.3

Velocity

The mean boundary layer flow windspeed profiles were measured using a miniature hot wire
anemometer probe (55P11, Dantec). During all exThe objective was to measure fluxes and vapour periments, the reference free stream velocity was
densities due to an area source of one unit size continuously recorded with a Pitot-static tube which
within the roughness array, with leading edge at was mounted at (Xref , Yref , Zref ) = (2.7, 0, 0.45)m and

1.4 Area sources

115

its value was used to close the tunnel speed control obstacle height H as a characteristic lengthscale of
loop to maintain velocity within 0.05ms1 .
the obstacles giving

1.5.4

ReH =

Concentrations

The naphthalene vapour concentrations were measured using a fast flame ionisation detector (HFR400
fast FID, Cambustion). As the measurement of the
naphthalene vapour (C10 H6 ) using a FID system is
entirely novel, the following procedures were developed to ensure accurate concentration measurements.
Due to the lack of available naphthalene calibration gas, the FID was calibrated against propane
calibration gas at concentrations of 151 ppmv and
749 ppmv. An automated recalibration every 30 minutes was included in the experimental procedure;
background concentrations were also measured every five minutes. Any calibration drifts, and the rising background concentration, were removed from
the data using linear de-trending. For a given surface temperature, the concentration of naphthalene
vapour at the surface, CS , can be expressed in units
of ppm by volume according to
CS = 106

eS
p

(10)

where p is the air pressure. A normalization of measured concentrations with the corresponding saturated source concentration CN = C/CS provides temperature independent results 1 .
Since the concentration sampling system was calibrated against propane but the source concentrations were calculated using Equation 3 and 10, a
methodology was required to convert the measured
propane-equivalent concentrations into naphthalene
concentrations. For this, the inner walls of a small
plastic cylinder were coated with molten naphthalene (cylinder volume: V = 5 cm3 , coated surface:
A 23.5 cm2 ). The cylinder was placed in a temperature controlled water bath. Assuming that the
vapour pressure inside the cylinder was saturated after a reasonably short duration, concentrations and
temperatures were measured at the cylinder centre
for a temperature range of 13 32 C. A clear linear
relationship between the measured and the theoretical concentrations in the studied temperature range
Cpropane = 2.27 Cnaphthalene 11.31

(11)

with R2 = 0.9997 was found.

Particular emphasis was given to testing whether

flow and concentration measurements were independent of Reynolds number. For urban-like geometries, the Reynolds number is usually based on UH ,
the mean velocity at obstacle height, and the mean
1 Note that C/C is equivalent to / in Equation 7 when surS
S
face temperature TS and air temperature T are equal. Given that
all experiments were performed under neutral conditions, this requirement was met.

(12)

where is the kinematic viscosity of air. Critical

values vary between 2100 and 15000 Uehara et al.
(2003), and the range is presumably caused by different geometrical complexities and upstream flow
conditions of the individual studies.
In the EnFlo A tunnel a maximum free stream velocity of 25 ms1 could be realised. Thus, additional
tests to check for Reynolds number independence
were performed for both set-ups. Table 1 shows the
reference velocity range which was covered in the
following experiments along with corresponding velocities at building height and Reynolds numbers calculated using Equation 12. The velocities at building
height were derived from a spatially averaged velocity profile for each surface.
Table 1: Values of free stream velocity Uref , corresponding velocity at building height UH and the Reynolds number based on UH and building height
H for the tested surfaces.

Uref
[ms1 ]

C10S
UH
[ms1 ]

ReH

RM10S
UH
[ms1 ]

Re H

24
16
8

5.3
3.5
1.8

3520
2333
1200

4.3
2.9
1.4

2880
1933
933

2.1

Flow profiles

Within both roughness arrays, mean flow profiles

were measured at several positions using a single
hot wire. The near surface flow was considered to
be highly turbulent with turbulence intensities much
higher than 20%. Thus, hot wire measurements inside the canopy were only expected to give qualitative results. At some measurement positions within
the canopy, evidence of changing wind profile was
found for the lowest reference windspeed (8 ms1 ).
Above the canopy at a height of Z 2H all profiles collapse for the tested windspeed range and
no windspeed dependencies were found for all studied locations (for more details see Pascheke et al.
(2007)).

2.2

Reynolds number independence

UH H

Concentration

In contrast to momentum transfer, a critical Reynolds

number cannot be defined for scalar transfer as it
is ultimately limited by molecular diffusion. However, practically speaking, experiments should be
performed at windspeeds such that the Reynolds
number dependence is small.
Prior to using
the naphthalene area source, a simpler test was
conducted for both surfaces using a continuous
propane release from a point source at the location (X, Y) = (2965, 5) mm. Excessive source discharge rates may produce vertical jets and thus influence the flow pattern in the given geometry. The
average horizontal velocity close to the ground was

116

estimated to be approximately 0.05Uref and it was assumed that a similar outflow velocity, WS , would be
appropriate to ensure minimal effect on the flow field.
For each windspeed tested, the source strength was
adjusted to keep a constant ratio of WS /Uref = 0.04.
Figure 3a shows concentration measurements at
a number of locations within the canopy for both surfaces. Concentrations are given in dimensionless
form
C, ps =

C UH H2
Q

(13)

to allow for slight windspeed and source strength

variations during the experiments. Q is the emission
volume flow rate. Overall, it can be seen that systematic changes of concentration values occur for
windspeeds below 16 ms1 , but the effect is weaker
for the RM10S surface. This might be attributable
to the more turbulent flow which was observed over
the RM10S surface by Cheng and Castro (2002).
These results also show a different dependence on
Reynolds number for different measurement locations. Figure 3b shows the percentage deviations
from the average concentration in the windspeed
range 16 24 ms1 . The grey area indicates the experimentally derived reproducibility of the concentration measurements ( 1.5%).

Figure 3: a) Effect of windspeed on normalized concentrations from a point

source at different locations below canopy height for C10S and RM10S
surfaces. b) Percentage deviation from average concentration in the windspeed range 16 24 ms1 (grey area indicates 1.5% reproducibility
range).

The naphthalene sublimation technique can be

compromised at low windspeeds because transfer
from the surface is then determined by a combination of forced and natural convection (Goldstein and
Cho, 1995). For the same reason, the low velocity
limitation may add to the Reynolds number dependence present for scalar concentrations measured
above both surfaces for windspeeds below 16 ms1 .

1.4 for both surfaces. Figure 4b shows the percentage deviation from the average concentration
in the windspeed range 20 24 ms1 . Generally,
there is a more systematic dependence of concentration on windspeed for the naphthalene area
source compared to the point source (Figure 3), and
only for U > 20 ms1 do variations appear to approach a constant limit, within the experimentally derived reproducibility of the measurements, marked
in grey (3%). In comparing measurements made
at Uref = 10 ms1 , Figure 4b shows percentage deviations from the high speed results of up to 40%
for C10S and 30% for RM10S in comparison to 15%
and 10% respectively for the point source. Again,
the magnitude of the deviation depends on measurement location with respect to recirculation zones
around the obstacles.
These tests indicated that concentration fields
measured within the canopy using the naphthalene
area source showed undetectable changes when the
reference windspeed was increased above 20 ms1 ,
and hence Uref = 20 ms1 was used throughout all
measurements of concentration fields presented in
Sections 4.1 and 4.2.

Transfer velocities

Transfer velocities wT were measured in order to

quantify the effect of surface roughness type on
canopy ventilation from a ground level area source.
The naphthalene sublimation method as described
in Section 1.3 was used for three different surface
roughness configurations. In addition to the C10S
and RM10S surfaces, a local and mild roughness
change was simulated by replacing the active source
unit within the C10S roughness by a single RM10S
unit, which will be denoted as CRM10S.

Figure 5: Transfer velocities as a function of wind speed UH derived from

the naphthalene sublimation method. Comparison of three different surface geometries: C10S, uniform cubes; RM10S, non-uniform cuboids;
CRM10S, single non-uniform cuboid unit within uniform cubes. a) all windspeeds, b) high windspeeds only.

Figure 5 shows transfer velocity wT as a function of windspeed UH for each surface. Taking the
complete windspeed range into account, a linear relationship between transfer velocity and windspeed
was observed for all tested roughness geometries
with R2 0.99 (Figure 5a, and Table 2). Following
the discussion in Section 2 on the low windspeed
Figure 4: a) Effect of windspeed on concentrations from naphthalene area limitation of the naphthalene sublimation method, it
source at different locations below canopy height for C10S and RM10S
surfaces. b) Percentage deviation from average concentration in the was concluded that the non-zero offset was a result
windspeed range 20 24 ms1 (grey area indicates 3% reproducibility of mixed convection processes at low windspeeds.
range).
However, when data points for lower windspeeds
were excluded, the measurements were sufficiently
Figure 4a shows the results from concentration well described by zero offset linear relationships,
measurements over a range of windspeeds using with R2 0.92 (Figure 5b and Table 2). A repeatathe naphthalene area source described in Section bility of 1.5% was determined for the fit parameters

117

Table 2: Observed linear relationships between transfer velocity wT and wind speed Uref or UH respectively, for the complete windspeed range, and for
high windspeeds only.Dynamical properties of each surface included for comparison.

Dynamical
propertiesa

C10S
RM10S
CRM10S
a Taken

u
Uref

UH
Uref

0.058
0.063
-

0.22
0.18
-

Complete windspeed range

wT = a Uref + c ;
wT = b UH + c ;
All coefficients 103
a
b
c
R2

High windspeeds
wT = d Uref ;
wT = e UH ;
All coefficients 103
d
e
R2

1.43
1.27
1.54

1.78
1.68
1.89

6.52
7.07
8.56

6.9
8.9
8.9

8.10
9.32
10.5

0.92
0.97
0.95

from Cheng and Castro (2002), based on u determined in inertial sublayer.

from repeated measurements for the RM10S roughness.

Focussing on the high windspeed results in Table
2, coefficients d and e are each equivalent to the dimensionless transfer coefficient wT /U (Equation 6),
which represents the ventilation efficiency of the surface. Using the free stream velocity Uref to derive the
transfer coefficient (coefficient d), the C10S roughness has a 6% higher ventilation efficiency than the
RM10S roughness. This might be explained by comparison with UH /Uref for the two surfaces (column 3):
the windspeed at mean canopy height is 4% higher
for the C10S surface. Coefficient d for the CRM10S
surface is 12.5% larger than for the RM10S surface,
however the dynamical properties were not known
for this surface. Given that the configuration represents a smooth-to-rough roughness change, a deceleration in stream-wise flow would be expected at
roof height, which might be compensated by lateral
and vertical deflection of the mean streamlines (by
continuity), which in combination lead to better ventilation.
Using UH to derive the transfer coefficient (coefficient e) is equivalent to comparing ventilation efficiencies for the same windspeed at rooftop. In this
case, e is 15% higher for the RM10S surface. This
might be explained by the increased roughness of
the surface, as u /Uref is 10% higher for the RM10S
surface. The dynamical properties of the CRM10S
surface were not known, but an increase in surface
stress would be expected: this might explain the
value of e, which is 30% higher than for the C10S
surface. The higher values of both d and e for the
CRM10S surface suggest that a localised increase
in the standard deviation of building height is particularly effective in enhancing ventilation from street
level.

0.99
0.99
0.99

Concentration Mappings

4.1 Above the source area

High resolution concentration measurements were
carried out in two horizontal planes, at Z = 0.3H
and Z = 1.2H, for Uref = 20 ms1 . Figures 6 and
7 show the normalized concentration fields, which
were derived from point measurements using a triangulation method. Note that all coordinates are
normalized using the mean element height H with
(X/H, Y/H) = (0, 0) being the centre of the source
area.

Figure 6: Concentration field 3 mm above the active source area. a) C10S

surface, b) RM10S surface.

Figure 7: Concentration field 12 mm above the active source area. a) C10S

surface, b) RM10S surface.

Figure 6a shows the mean concentration distribution at Z = 0.3H for the C10S surface. The concentration distribution shows a very regular pattern with
increasing magnitude in the direction of the flow (left
to right). This is due to stream-wise advection of
naphthalene vapour. A regular pattern of concentrations is observed around individual roughness elements: lower concentrations ahead of the roughness
element, and higher concentrations behind. This is
consistent with LES simulations of flow for the same
experimental layout (Z. Xie, pers. comm.), which
show higher velocities in the impact zone in front
of an roughness element, and lower velocities and
strong updraughts in the wake of each roughness element. Hence, the vertical transport is controlled by
the repeated flow structures generated by individual
roughness elements.
Figure 6b shows that roughness element height
variation has a clear influence on the transport processes near the ground. The concentration pattern
is no longer regular and the stream-wise increase

118

Figure 8: Concentration plume for Z = 6 mm = 0.6H. a) C10S surface, b)

RM10S surface.

Figure 10: Concentration plume for Z = 18 mm = 1.8H. a) C10S surface,

b) RM10S surface.

Figure 9: Concentration plume for Z = 12 mm = 1.2H. a) C10S surface,

b) RM10S surface.

Figure 11: Concentration plume for Z = 24 mm = 2.4H. a) C10S surface,

b) RM10S surface.

in concentration due to advection is less continuous.

In general, lower concentrations are found in front of
roughness elements and higher concentrations behind, but the interaction of the roughness element
wakes influences the transport processes. This is
especially evident around the highest roughness element (17.2 mm). The low concentrations in front of
the roughness element and to the sides suggest the
formation of the so-called horseshoe vortex, which
not only creates a strong downwash at the roughness element front, but also affects concentrations in
the wakes of the two upstream roughness elements.
Figure 7 shows the concentration patterns for
Z = 1.2H. The measurement plane is just above the
roughness elements for the C10S surface. The striking difference for the RM10S surface (see Figure 7b)
is that roughness elements intersect the measurement plane and large concentrations are observed
in their wakes.

tions in the wakes of roughness elements intersecting the measurement plane act as point-like sources.
The plume shape is thus strongly influenced by
the locations of these elevated sources. For the
measurement planes at Z = 1.8H and Z = 2.4H, the
plume shape becomes more consistently Gaussian
further downstream, indicating effective mixing.

4.2 Concentration fields downstream

of the source area
Additionally, concentration measurements were
made in horizontal planes downstream of the source
area to determine far field dispersion characteristics. Figures 8, 9, 10 and 11 show the measurement planes at Z = 0.6H, 1.2H, 1.8H and 2.4H respectively. Some general features can be seen by
looking at Figures 8 to 11 in sequence. For Z = 0.6H
the plume concentration pattern is elongated downstream for the C10S surface, which may be explained by increased windspeeds within the roughness elements, leading to greater advection. The
lateral spread of the plume is greater for the RM10S
surface, which is particularly clear at Z = 1.8H and
Z = 2.4H. The influence of rapid vertical transport in
the wake of individual high roughness elements can
clearly be seen in Figure 9, which shows the concentrations at Z = 1.2H for both surfaces. Concentra-

Conclusions

Naphthalene coated surfaces were used to represent area sources at street level and thereby to
study ventilation from two different urban-like geometries at neighbourhood scale: one with uniform height roughness elements, the other with nonuniform heights. Area-averaged scalar fluxes were
measured and the results represented in terms of a
dimensionless transfer coefficient wT /Uref . Naphthalene concentrations were measured with high spatial
resolution in two horizontal planes above the active
source area to capture the scalar transport at street
scale. Coarse mesh concentration measurements
at four different heights in the wake of the active surface were also conducted to study differences in the
plume spread arising from the different geometrical
surface characteristics. A novel Fast Flame Ionisation Detection technique was developed to measure
naphthalene vapour concentrations.
For both of the geometries studied, the plan
area density as well as the frontal area density
was p = f = 25%. Two different height distributions were studied, one uniform and the other randomly (Gaussian) distributed but with the same
mean height. The difference in height distribution
had a significant effect on the ventilation efficiency of
the geometries considered. The transfer coefficient
was calculated using two different reference velocities. Using UH , which represents the flow at the top
of the canopy, resulted in an increased ventilation

119

coefficient for the non-uniform height array. Using

the free stream velocity, Uref , which is unaffected by
the surface roughness, yielded a larger transfer coefficient for the uniform height roughness. The latter definition of the transfer coefficient is probably
a more useful representation of ventilation from the
canopy as the incident flow at higher elevation drives
the exchange and is largely unaffected by changes
in the surface morphology. The results demonstrate
that whilst there is enhanced vertical momentum exchange due to height variability, the dispersion from
a limited area source within the canopy is not enhanced. This is because the advection is reduced.
The height variability of a canopy must therefore be
considered as an important parameter when generalising urban dispersion characteristics based on
morphometric methods - its influence on both advection and dispersion should be acknowledged.
Fine resolution concentration measurements
close to the source elucidated the effect of height
variability on dispersion at street scale. Both array geometries generated similar area-averaged
concentration values close to the active surface
at Z = 0.3H but the area averaged concentration
at Z = 1.2H was 18% higher for the non-uniform
surface (see Pascheke et al. (2007)). The important
role of individual high-rise roughness elements on
local ventilation processes was seen in the concentration measurements at both heights. At street
level, the tallest roughness elements affected the
flow around nearby roughness elements, showing
that the lateral extent of their influence is large.
Surface released scalars were also transported
upwards in the wake of individual high roughness
elements, thus generating point-like sources at
elevated heights.
Coarse resolution concentration measurements
downstream of the source showed significant differences in the location and shape of the plume. The
non-uniform height array generated additional turbulence within the canopy, resulting in more efficient
mixing processes. As a consequence, the lateral
and vertical plume spread was increased and plume
dilution was intensified in and above the RM10S surface. Plume shape was consistent with height for
the non-uniform height surface, whereas the plume
narrowed with height for the uniform height surface
indicating an adjustment from the dispersion characteristics within the canopy to those in the boundary
layer above.
The results presented show the potential of
using a naphthalene area source in wind tunnel
studies of urban dispersion problems. Bulk fluxes
of scalars can be derived from the naphthalene
sublimation technique to quantify area-averaged
ventilation. The novel measurement of naphthalene
vapour concentrations using a Fast Flame Ionisation
Detector technique proved to be sensitive enough
to capture variability of concentration fields within
the urban arrays. Thus information on the exchange
processes between neighbourhood and street
scales could be gained. The combined use of the
sublimation method of flux determination and FID
plume mapping has been found to be a powerful
research tool. Its utility will be further enhanced
in future studies by concurrent measurements
of the flow field, for example using pulsed wire,

laser Doppler anemometry (LDA) or particle image

velocimetry (PIV).

Acknowledgments
The authors are very grateful for the outstanding
technical support from Paul Hayden, Tom Lawton and Allan Wells at the EnFlo laboratory. The
project was funded by the Engineering and Physical Sciences Research Council, grant number
GR/S71798/01

References
Barlow, J. F. and S. Belcher, 2002: A wind tunnel
model for quantifying fluxes in the urban boundary
layer. Boundary-Layer Meteorol., 104, 131150.
Barlow, J. F., I. N. Haman, and S. Belcher, 2004:
Scalar fluxes from urban street canyons. part i:
Laboratory simulations. Boundary-Layer Meteorol., 113, 369385.
Britter, R. E. and S. R. Hanna, 2003: Flow and dispersion in urban areas. Annu. Rev. Fliud Mech.,
35, 469496.
Cheng, H. and I. P. Castro, 2002: Near wall flow over
urban-like roughness. Boundary-Layer Meteorol.,
104, 229259.
C.R.C., 1993: Handbook of Chemistry and Physics.
CRC Press Inc. 74th edn.
Goldstein, R. J. and H. H. Cho, 1995: A review of
mass transfer measurements using naphthalene
sublimation. Exp. Thermal and Fluids Sci., 10,
461434.
Hall, D. J., R. Macdonald, S. Walker, and A. M. Spanton, 1996: Measurements of dispersion within
simulated urban arrays a small scale wind tunnel study. BRE Client Report 178/96, Build. Res.
Establ., Garston, Watford, UK.
Heist, D. K., L. A. Brixley, S. G. Perry, and G. E.
Bowker, 2005: Residence time measurements in
an array of buildings. In Proceedings of PHYSMOD 2005, Int. Workshop on Physical Modelling
of Flow and Dispersion Phenomena.
Oke, T. R., 1987: Boundary Layer Climates. Routledge, Taylor & Francis Group, 435 S, 2 edition.
Pascheke, F., J. F. Barlow, and A. Robins, 2007:
Wind tunnel modelling of dispersion from a scalar
area source in urban-like roughness. BoundaryLayer Meteorol., submitted.
Uehara, K., S. Wakamatsu, and R. Ooka, 2003:
Studies on critical reynolds number indices for
wind-tunnel experiments on flow within urban areas. Boundary-Layer Meteorol., 107, 353370.

120

Dispersion of Traffic Exhausts in

Urban Street Canyons with Tree Plantings
- Experimental and Numerical Investigations C. Gromke1, J. Denev2, B. Ruck1

For further variations of characteristic tree

planting parameters, e.g. tree spacing, crown
shape and size or leaf free stem height, see
Gromke and Ruck, 2006; Gromke and Ruck,
2007a; b; c; d. Furthermore, for the purpose of
comparison, numerical investigations with a commercial CFD-code using a LVEL k-H turbulence
closure scheme have been performed and will be
discussed.

Laboratory of Building- and Environmental

Aerodynamics, Institute for Hydromechanics
2
Institute for Chemical Technology
University of Karlsruhe
Karlsruhe, Germany
gromke@ifh.uka.de
Abstract - Wind tunnel experiments and numerical computations have been performed in
order to investigate the influence of avenuelike tree plantings on the dispersion of traffic
exhaust in an urban street canyon. Reduced
natural ventilation and enhanced pollutant
concentrations have been found in the presence of trees. A comparison of experimental
and numerical results shows qualitative good,
but quantitative rather poor agreement. In the
numerical computations, flow velocities are
underestimated and pollutant concentrations
are overestimated.

3 Experimental Model Setup,

Measurement Instrumentation
and Numerical Model
The wind tunnel model (scale 1:150) consists of
two parallel aligned rows of houses forming an
isolated urban street canyon with aspect ratios
H/W = 1 and L/H = 10 (Figure 1). Model tree arrangements with different crown porosities have
been placed in one row along the street center
axis. This arrangement was subjected to a boundary layer flow with profile exponent D= 0.30 according to the power law (Figure 9), approaching
perpendicular to the street axis (Gromke and
Ruck, 2005). The Reynolds number Re, calculated with the building height H and the velocity
u(H) of the undisturbed flow at building height H,
amounts to Re = 37000 and ensures a ReynoldsNumber independent flow field.

Key words - trees, pollutant dispersion, urban

street canyon, urban air quality, traffic exhausts

1 Introduction
The emission of traffic exhausts is one of the major sources for airborne pollutants in urban areas.
In particular, narrow urban street canyons with
large traffic volume are subject to high pollutant
concentrations. Thus, it is important for the healthiness of inhabitants that sufficient natural ventilation is ensured which dilutes and removes the
traffic emissions. A large number of studies have
been devoted to pollutant dispersion processes in
urban street canyons in the past, however, all of
them were restricted to empty, obstacle-free street
canyons (Chang and Meroney, 2003; Gerdes and
Olivari, 1999; Kastner-Klein et al., 2001; Pavageau and Schatzmann, 1999). By contrast, the
central question of the present investigation is, to
what extent avenue-like tree plantings in urban
street canyons influence the natural ventilation.

y
line source
model tree
x

L = 180 m

u
roughness
elements
concentration
measurement
taps
u(z)
D = 0,30

2 Approach

z
A

W = 18 m

H = 18 m

Figure 1. Model setup of street canyon

(scale 1:150)

In order to clarify the influence of tree plantings on

the dispersion of traffic exhausts, wind tunnel experiments at a small-scale urban street canyon
model have been performed. Velocity and concentration measurements for an approaching boundary layer flow perpendicular to the canyon center
axis have been carried out for different tree planting configurations. In the present article, emphasis
is placed on variations of the crown permeability.

In-between the houses, a tracer gas emitting

line source is embedded at street level for simulating the release of traffic exhausts (Meroney et al.,
1996). Sulfur hexafluoride (SF6) was used as
tracer gas to model the traffic emissions. Measurement taps were applied along the inward canyon walls to sample the near-wall canyon air.
These samples were analyzed by an Electron

121

aspect ratio L/H > 7, a distinct region with the canyon vortex being the solely dominating vortex
structure can be observed.
In Figure 3, the normalized pollutant concentrations at the inward canyon walls are shown. These
concentrations have been made dimensionless
according to the formula

Capture Detector (ECD) yielding mean concentrations. Laser-Doppler-Velocimetry (LDV) was used
to measure flow velocities in the street canyon
and to determine the rates of vertical air exchange
between street canyon and atmospheric flow at
roof top level.
The numerical computations have been performed by using the commercial CFD-code
FLOVENT. This software package offers a LVEL
k-H turbulence model to close the ReynoldsAveraged-Navier-Stokes (RANS) equations. In
near-wall regions, the turbulent viscosity Qt is determined by a blending of the turbulent viscosity
Qt,k-H, calculated by the classical k-H approach, and
the turbulent viscosity Qt,LVEL, calculated by an
algebraic LVEL approach using a characteristic
length and velocity scale. Furthermore, wall functions for the treatment of wall-adjacent cells are
included. The governing equations are numerically
solved on a structured, staggered grid, using the
finite volume method. Underlying discretization
schemes are first order upwind and second order
central differences for convection and diffusion
terms, respectively. The turbulent Schmidt number
Sct in the advection-diffusion equation is 1 and
can not be varied (FLOVENT 6.1, 2005).

(1)

z/H

Wall A

1
0.5
-5

-4

-3

-2

-1

z/H

y/H
Wall B

1
0.5
-5

-4

-3

-2

-1

0
y/H

4.1 Reference Case - Tree-free Street Canyon

Before presenting and discussing the measurement results obtained at street canyons with avenue-like tree plantings, the flow- and concentration
field at the tree-free (so called reference) canyon,
will be illustrated. In Figure 2, a schematic sketch
of the flow field at the reference canyon is shown.

u(z)

[ ]

with cmeas measured concentration, Lref reference

length characterizing a building dimension (Lref =
H), uref reference velocity characterizing the atmospheric flow (uref = uH) and QT/l tracer gas
source strength per unit length of the line source.
Furthermore, the lengths have been normalized
by the reference length Lref = H.

4 Measurement Results

Wall A

c meas Lref uref

QT l

c

Figure 3. Normalized pollutant concentration c+ [-]

at canyon walls for reference case
Two characteristics of the concentration distribution at the inward canyon walls can be seen at
once. First, the decline in concentration towards
the street ends at both walls A and B, and second
the much higher pollutant concentrations at wall A.
The first characteristic is due to the superposition
of the two different vortex structures at the street
ends, providing enhanced ventilation. The second
observation can be explained by following the
canyon vortex for one rotation. At the roof top, an
exchange between street canyon air and the
skimming atmospheric flow takes place. Unpolluted air of the atmospheric flow is entrained into
the rotating canyon vortex, which moves downwards into the street canyon in front of wall B. On
the reverse flow from wall B towards wall A, released traffic exhausts near the bottom are getting
accumulated in the canyon vortex. The upward
motion of the vortex carrying the pollutants leads
to high concentrations in front of wall A.
This is confirmed by the results of LDV measurements in a vertical cross section at y/H = 0.5.
Figure 4 shows the normalized mean vertical velocity w+ = w/uref of the rotating canyon vortex
measured in the reference canyon without trees.
Ascending air flow in front of wall A and descending air flow in front of wall B are clearly visible.
The maximum vertical velocity in the canyon vortex amounts to 25 % of the flow velocity at building height H.

Wall B
Canyon Vortex
Corner Eddy

Figure 2. Flow field at empty street canyon

(reference case)
Two essential vortex structures can be identified. The canyon vortex in the middle section of
the street canyon and the corner eddies at the
outer parts of the canyon. Driven by the atmospheric flow skimming over the roof top level, the
canyon vortex is serving for vertical air exchange
and thus providing pollutant dilution and removal
in the middle section of the street canyon. Due to
the corner eddies at the street ends, ambient air is
entering the street canyon horizontally and serving
for ventilation, too. At the street canyon ends, a
superposition of both vortex structures takes
place. According to Hunter et al., 1990/91, only in
the middle section of long street canyons with

122

At the leeward wall A, high increases in concentration are evident. Three pronounced areas of
increases are evident. One is in the center part of
the street canyon at y/H = 0 and the others at the
canyon outer parts at 2.5 < y/H < 4. The high
relative increases at the canyon outer parts are
easy to explain. Due to the tree crowns, the corner
eddies are effectively hindered in entering the
street canyon laterally. Thus, one vortex structure
which was serving for ventilation in the outer regions of the tree-free street canyon is missing or
at least significantly reduced in strength. Comparing the flow fields in front of wall A at y/H = 0.5 in
Figure 4 and Figure 6 reveals that the maximum
velocity w+ is only slightly smaller in the presence
of the tree planting. However, because of the limited passage width, the volume flow rate of the
remaining canyon vortex is reduced significantly.
Consequently, the upward streaming flow in front
of wall A contains higher concentrations of pollutants.

1.2

z/H

0.8

0.6

0.4

0.2

-0.4

-0.2

0.2

0.4

x/H
1.2

Figure 4. Normalized mean vertical velocity w+ [-]

between buildings at y/H = 0.5 for reference case

4.2 Continuous Tree Planting with impermeable Crown of rectangular Cross Section
The relative change in pollutant concentration at
the canyon walls in the presence of a continuous
tree planting with block shaped cross section of 9
m x 12 m when compared to the reference case
(Figure 3) is presented in Figure 5. This configuration represents an avenue-like tree planting with a
leaf-free stem height of 6 m and crowns interfering
with each other, which is common for the urban
environment. In this example, the crown permeability was neglected and Styrofoam was used to
model the trees.

0.8
z/H

impermeable crown
cross section:
9 m x 12 m

0.6

0.4

0.2

-0.4

-0.2

0.2

0.4

x/H

z/H

Figure 6. Normalized mean vertical velocity w+ [-]

between buildings at y/H = 0.5

-5

At the windward wall B, decreases in concentration are visible (Figure 5). Considering the velocity plots of Figure 4 and Figure 6 shows modified flow fields at the roof top level. The upward
moving stream in front of wall A extends farther
into the skimming atmospheric flow in the case of
the street canyon with tree planting. This means
that the polluted air is discharged into higher layers of the above roof wind and is thus better diluted. In addition, analyses of the fluctuating part
of the vertical velocity (not shown here) show enhanced turbulence intensities in the presence of
the tree planting. The air which is entrained into
the canyon in front of wall B is less polluted, leading to the smaller concentrations observed.

Wall A

1
0.5
-4

-3

-2

-1

z/H

y/H
Wall B

1
0.5
-5

-4

-3

-2

-1

0
y/H

Figure 5. Relative change in pollutant concentration [%] for street canyon with impermeable, continuous crown of cross section 9 m x 12 m when
compared to reference case (Figure 3)

123

4.3 Continuous Tree Planting with permeable

Crown of rectangular Cross Section

only negligible deviations in the mean vertical

velocity component w+.

z/H

Now, the relative change in pollutant concentration for the same street canyon arrangement as
addressed before, but with a permeable instead
an impermeable crown is discussed. In order to
take the crown porosity and permeability into account, an open-pored foam material (foam ppi 10)
with a volume porosity of 97 % and a pressure
loss coefficient O = 250 Pa (Pa m)-1 was used to
model the trees. Figure 7 shows the relative
change in concentration when compared to the
reference case (Figure 3), (top) and when compared to the impermeable crown arrangement,
(bottom).

0.8
z/H

permeable crown
cross section:
9 m x 12 m

0.6

Wall A

1
0.5
-5

1.2

0.4

-4

-3

-2

-1

z/H

y/H

-5

0.2

Wall B

1
0.5
-4

-3

-2

-1

-0.4

z/H

y/H

-5

-4

-3

-2

-1

z/H

-5

-3

-2

-1

0.4

5 Numerical Results
The first step in numerical modeling was to generate the approaching flow and boundary conditions
of the wind tunnel experiments (Gromke and
Ruck, 2005). Figure 9 shows the computed and
measured vertical profiles of mean flow and turbulence intensity profiles u(z) and Ixyz(z) at the origin
of the co-ordinate system (Figure 1) for the empty
domain. The agreement between experimental
and numerical data is quite satisfactory.

Wall B

-4

0.2

Figure 8. Normalized mean vertical velocity w+ [-]

between buildings at y/H = 0.

y/H
1
0.5

0
x/H

Wall A

1
0.5

-0.2

y/H

Figure 7: Relative change in pollutant concentration [%] for street canyon with permeable, continuous crown of cross section 9 m x 12 m when
compared to reference case (Figure 3), (top)
and relative change in pollutant concentration [%]
when compared to impermeable crown, (bottom).

mean flow velocity u(z)

1.0
0.8
z [m]

The typical characteristics of the concentration

distribution known from the impermeable crown
arrangement with increased concentrations at the
leeward wall A and decreased concentrations at
the windward wall B are found (Figure 7 top). A
direct comparison of the two setups reveals only
minor relative differences at wall A and moderate
relative differences at wall B (Figure 7 bottom).
The effect of crown permeability is of secondary
importance for the pollutant concentrations at the
canyon walls. At wall A, the relative average
change in concentration amounts to - 3 % and at
wall B to - 31 %. The absolute changes in concentrations at wall B are small, too. However, due to
intrinsic lower concentrations at this wall (Figure
3), the relative changes appear to be enhanced
when compared to wall A. Considering the flow
fields for the impermeable and permeable crown
arrangements in Figure 6 and Figure 8, reveals

u (CFD)

0.6

u (wind tunnel)

0.4
0.2
0.0
0.0

2.0

4.0
u [m/s]

6.0

8.0

mean turbulence intensity Ixyz(z)

1.0

z [m]

0.8
I_xyz (CFD)

0.6

I_xyz (wind tunnel)

0.4
0.2
0.0
0

10

20

30
Ixyz [%]

40

50

60

Figure 9. Vertical profiles of mean flow velocity

u(z) [m/s] and mean turbulence intensity Ixyz(z) [%]

124

5.1 Reference Case - Tree-free Street Canyon

and windward edge of building A. In a first attempt, numerical computations were performed
without any special gird refinement (coarse grid)
near the building walls. The resulting flow fields at
the roof level above the windward edge of building
A and the street canyon are shown in Figure 11.

The next step was to model the reference case,

i.e. the empty, tree-free street canyon. One of the
well-know deficiencies of k-H turbulence models is
the stagnation point anomaly. This anomaly leads
to overstated pressure values around the stagnation point at the buildings windward face. Furthermore, an excessive production of turbulent
kinetic energy k is predicted at the buildings leading roof edge. The overestimation of k causes an
enhanced transport of momentum from higher
layers downward to the roof. This net downward
momentum flux reduces the separation bubble or
even suppresses the separation of the flow at the
leading roof edge. However, since separation
phenomena are important for the flow field and the
pollutant dispersion in the wake of the building, i.e.
in the street canyon, some effort was made to
generate the flow separation at the leading edge.
LDV measurements of the mean flow around
the windward building A and above the street canyon at roof top level at y/H = 0.5 are shown in
Figure 10. The model buildings are made of Perspex and have a very smooth surface. As expected, the characteristic eddy at the bottom in
front of the building and the separation bubble at
roof level can be seen clearly (Figure 10 top). In
the vector plot showing the flow field at roof level
above the street canyon (Figure 10 bottom), the
upper part of the canyon vortex can be identified.
As can be seen, the canyon vortex protrudes into
the skimming atmospheric flow above the buildings.
1.6

windward edge of building A

Figure 11. Vector plots of mean flow at roof level

at y/H = 0.5 for coarse grid case
Neither flow separation at the windward edge
nor a canyon vortex protruding into the skimming
above roof flow can be seen. The canyon vortex
remains completely inside the street canyon, suggesting a weak air exchange in the canyons middle part.
In order to obtain flow separation at the leading
edge, a refined grid was generated at the roof top
level of building A. The corresponding flow fields
above the windward building A and the street canyon are shown in Figure 12.

flow field in front of building A

1.4
1.2

windward edge of building A

z/H

1
10

0.8
8

0.6

Reference Vectors
0
2
4
6
8
10

0.4

0.2

0
-2.4

-2.2

-2

-1.8

-1.6

-1.4

-1.2

-1

-0.8

-0.6

Figure 12. Vector plots of mean flow at roof level

at y/H = 0.5 for refined grid case

x/H

z/H

flow field at roof top level above street canyon

This time, flow separation and a separation

bubble above building A can be observed (Figure
12 top). Furthermore, regarding the upper street
canyon flow field (Figure 12 bottom) reveals a
canyon vortex protruding into the above-roof flow.
Typical flow field characteristics as measured at
the wind tunnel setup are evident.
The normalized pollutant concentrations at both
canyon walls A and B for the coarse and refined
grid are shown in Figure 13. A comparison to the
measured pollutant concentrations (Figure 3) reveals considerably higher values in the case of the
numerical computations, one reason for this is the

1.1
1
-0.5

-0.25

0.25

0.5

x/H

Figure 10. Vector plots of mean flow around

windward building A and at roof top level above
street canyon at y/H = 0.5 (LDV measurements)
As stated before, special attention was paid to
generate flow separation at the windward edge of
building A. Additionally to embedded refined grids
around the street canyon arrangement, a grid
refinement study was performed at the roof top

125

z/H

relatively high value of Sct = 1.0. In average, the

numerically predicted concentrations at wall A are
1.9 times higher for the coarse grid case and 1.3
times higher for the refined grid case. The corresponding factors for wall B are 2.9 and 2.8, respectively. Concerning the general shape of the
pollutant concentration distributions at the canyon
walls, a better agreement with the measurements
is found for the refined gird case with vertical concentration gradients in the middle part of wall A
and horizontal gradients at wall B. Although the
numerically predicted separation bubble and recirculation zone (Figure 12 top) are much smaller
than experimentally measured (Figure 10 top), the
concentration field inside the street canyon is
highly sensitive to separation phenomena at the
windward edge.

z/H

0.8

0.6

0.4

Wall A

1
0.5
-5

1.2

-4

-3

-2

-1

0.2
1

z/H

y/H
Wall B

1
0.5
-5

-4

-3

-2

-1

z/H

-5

-3

-2

-1

-4

-3

-2

-1

z/H

0.4

The tree planting already described and studied

experimentally in chapter 4.2 has been investigated numerically, too. Since the wall average
concentrations as well as the shape of the concentration distribution at the canyon walls show
more similarity to the measurement results of Figure 3 for the refined grid case (Figure 13 bottom),
the numerical computation was performed with the
refined grid at the roof level of leeward building A.
The relative change in concentrations when compared to the reference case (Figure 13 bottom) is
shown in Figure 15.

y/H

-5

0.2

5.2 Continuous Tree Planting with impermeable Crown of rectangular Cross Section

Wall B

1
0.5

Figure 14. Normalized mean vertical velocity w+ [-]

at y/H = 0.5 for reference case with refined grid

Wall A

-4

-0.2

x/H

y/H

1
0.5

-0.4

y/H

Figure 13. Normalized pollutant concentration c+ [-]

for reference case with coarse grid (top)
and refined grid (bottom)

z/H

The contour plot of the normalized vertical velocity component w+ inside the street canyon with
refined grid at the roof level of windward building
A is shown in Figure 14. In comparison to the LDV
measurements of the wind tunnel experiment
(Figure 4), the numerical computation results in
significantly lower flow velocities. A reason for
underestimating the velocity is the combination of
first order upwind convection scheme with the
present LVEL k-H model. This means that the canyon vortex strength is weaker and less air is exchanged between canyon flow and atmospheric
above-roof flow. Consequently, the pollutant concentrations, as predicted by the numerical computation, are higher. A comparison between the numerical results of the coarse and refined grid reveals similar vertical velocity fields, but slightly
smaller flow velocities in the case of the coarse
grid (not shown here). This is because of reduced
momentum exchange between the skimming flow
and the air masses below the roof top level, which
is driving the canyon vortex (Figure 11 and Figure
12).

Wall A

1
0.5
-5

-4

-3

-2

-1

z/H

y/H
Wall B

1
0.5
-5

-4

-3

-2

-1

0
y/H

Figure 15. Relative change in pollutant concentration [%] for street canyon with impermeable, continuous crown of cross section 9 m x 12 m when
compared to reference case (Figure 13 bottom)
Like in the experimental results (Figure 5), pronounced increases in concentration are visible at
wall A. Whereas the increases in concentrations
at the canyons outer part at 3.5 < y/H < 4.5
are of same size, both in numerical and experimental results, the relative change in concentration at the canyons middle part is significantly

126

higher in the numerical simulation. Furthermore,

the decrease in concentration at y/H = 5 is
stronger in the numerical results. At wall B decreases in concentration can be found in wide
parts. However, at the middle part around y/H = 0,
weaker concentration decreases are predicted,
differing from the experimental results (Figure 5).
Figure 16 shows the contour plot of the normalized vertical velocity component w+ inside the
street canyon at y/H = 0.5. The magnitude of the
updraft in front of the leeward wall A is comparable to the numerical reference case (Figure 14).
However, in combination with the limited passage
width due to the tree arrangement, the volume
flow rate of the rotating canyon vortex is considerably reduced to 51 % of the volume flow rate of
the numerical reference case. In front of wall B, a
modified flow field is present. Contrary to the experimental results (Figure 6), a recirculation vortex
develops in the gap between the tree crown and
the windward wall B (Figure 17).

6 Summary
Flow and pollutant dispersion processes inside a
typical urban street canyon with and without trees
have been investigated. Avenue-like tree plantings
consisting of crowns interfering with each other
and forming a rectangular cross-sectioned continuous block have been considered. Wind tunnel
experiments as well as numerical computations
have been performed for single-row tree arrangements placed along the street center axis. Tree
plantings with different crown permeabilities, covering 33 % of the street canyon volume have been
investigated.
For all tree planting arrangements investigated,
an increase in the overall pollutant concentration
inside the street canyon was found when compared to the empty street canyon without trees
(reference case). A more detailed analysis revealed increased pollutant concentrations at the
leeward canyon wall of the upwind building and
decreased pollutant concentrations at the windward canyon wall of the downwind building. High
relative increases were found at the street ends
towards the intersections. The entrainment conditions at the roof top level as well as at the sidewise end cross sections were considerable modified due to tree plantings.
Qualitatively, the CFD results are in agreement
with the wind tunnel experiments. However, the
quantitative agreement is limited. The numerical
computations result in higher pollutant concentration levels and lower flow velocities inside the
street canyon. Furthermore, the importance of
flow separation at the windward edge of building A
for the concentration field inside the street canyon
was demonstrated by the CFD-simulations. As a
result of the grid refinement study, the LVEL k-H
turbulence model together with the first-order upwind convection scheme predicts flow separation
only in the case of a very fine grid near the windward edge of building A.
In Table 1 the normalized wall average concentrations of all configurations investigated are
summarized. The values in the brackets denote
the relative change in concentration when compared to the reference case with refined grid.

1.2

0.8
z/H

impermeable crown
cross section:
9 m x 12 m

0.6

0.4

0.2

-0.4

-0.2

0.2

0.4

x/H

Figure 16. Normalized mean vertical velocity w+ [-]

between buildings at y/H = 0.5

configuration
ref. case
(coarse grid)
ref. case
(refined grid)

exp. result
wall A wall B

19.6

impermeable
crown

(+71 %)

33.5

permeable
crown

(+66 %)

32.6

num. result
wall A wall B
37.3

15.9

26.1

15.0

5.4

3.6

(-33 %)

2.4

(-55 %)

67.1

9.7

(157%)

(-35 %)

Table 1. Normalized wall average concentrations

c+ [-], values in brackets denote relative change in
concentration when compared to refined grid case

Figure 17. Vector plot of mean flow in street canyon with impermeable crown at y/H = 0.5

127

Figure 18 shows the profiles of the normalized

mean vertical velocity component w+ at height z/H
= 0.7 in a vertical cross section at y/H = 0.5. The
corresponding normalized integrated vertical volume flow rates V+ per unit spanwise length are
summarized in Table 2. When multiplying these
values with the reference velocity uref and the reference length Lref = H, the vertical volume flow rate
per unit spanwise length in m2 s-1 is obtained.

References
Chang, C. and Meroney, R.N., (2003). Concentration and flow distributions in urban street
canyons: wind tunnel and computational data,
Journal of Wind Engineering and Industrial
Aerodynamics, vol. 91, pp. 1141 - 1154.
FLOVENT 6.1 (2005). Users Manuel, Flomerics
Limited, September 2005.
Gerdes, F. and Olivari, D., (1999). Analysis of
pollutant dispersion in an urban street canyon,
Journal of Wind Engineering and Industrial
Aerodynamics, vol. 82, pp 105 - 124.
Gromke, C. and Ruck, B., (2005). Die Simulation
atmosphrischer Grenzschichten in Windkanlen, Proc. 13. GALA Fachtagung "Lasermethoden in der Strmungsmesstechnik", Cottbus,
September 2005, pp. 51-1 - 51-8.
Gromke, C. and Ruck, B., (2006). Der Einfluss
von Bumen auf das Strmungs- und Konzentrationsfeld in Straenschluchten, Proc. 14.
GALA Fachtagung "Lasermethoden in der
Strmungsmesstechnik", Braunschweig, September 2006, pp. 59-1 - 59-10.
Gromke, C. and Ruck, B., (2007a). Influence of
trees on the dispersion of pollutants in an urban
street canyon - experimental investigation of
the flow and concentration field, Atmospheric
Environment, vol. 41, pp. 3387 - 3302.

LDV-Measurements
+

w [-]
0.3
reference case

0.2

impermeable crown

x/H [-]

0.1

-0.50

permeable crown

0.0
0.00
-0.1

-0.25

0.25

0.50

-0.2
-0.3
CFD-Computations
+

w [-]
0.3
0.2

x/H [-]

0.1

-0.50

-0.25

ref. case (coarse grid)

ref. case (refined grid)
impermeable crown

0.0
0.00
-0.1

0.25

0.50

http://dx.doi.org/10.1016/j.atmosenv.2006.12.043

Gromke, C. and Ruck, B., (2007b). Trees in urban street canyons and their impact on the dispersion of automobile exhausts, Proc. 6th International Conference on Urban Air Quality,
Cyprus, March 2007.
Gromke, C. and Ruck, B., (2007c). Flow and
dispersion phenomena in urban street canyons
in the presence of trees, Proc. 12th International Conference on Wind Engineering, Australia, July 2007.
Gromke, C. and Ruck, B., (2007d). Effects of
trees on the dilution of vehicle exhaust emissions in urban street canyons, Special Issue
on Urban Air Pollution in International Journal
of Environment and Waste Management
(IJEWM), paper accepted for publication, paper
submitted on invitation.
Hunter, L.J., Watson, I.D. and Johnson, G.T.,
(1990/91). Modelling air flow regimes in urban
canyons, Energy and Buildings, vol. 15, pp.
315 - 324.
Kastner-Klein, P., Fedorovich, E. and Rotach,
M.W., (2001). A wind tunnel study of organised
and turbulent air motions in urban street canyons, Journal of Wind Engineering and Industrial Aerodynamics, vol. 89, pp. 849 - 861.
Meroney, R.N., Pavageau, M., Rafailidis, S. and
Schatzmann, M., (1996). Study of line source
characteristics for 2-D physical modelling of
pollutant dispersion in street canyons, Journal
of Wind Engineering and Industrial Aerodynamics, vol. 62, pp. 37 - 56.
Pavageau, M. and Schatzmann, M., (1999). Wind
tunnel measurements of concentration fluctuations in an urban street canyon, Atmospheric
Environment, vol. 33, pp. 3961 - 3971.

-0.2
-0.3

Figure 18. Normalized mean velocity w+ [-] profiles

at y/H = 0.5 and height of z/H = 0.7,
experimental (top) and numerical (bottom) results
configuration
ref. case
(coarse grid)
ref. case
(refined grid)

exp. result
wall A wall B

0.082

-0.015

0.034

-0.020

(-57 %)

permeable
crown

(-59 %)

0.040

-0.023

0.049

-0.029

0.025

-0.006

-0.057

0.035

impermeable
crown

num. result
wall A wall B

(-74 %)

(-65 %)

(-49 %)

(-79 %)

Table 2. Normalized vertical volume flow rates

V+ [-] per unit spanwise length inside the street
canyon at y/H = 0.5 and height z/H = 0.7,
values in brackets denote relative change in volume flow rate when compared to refined grid case

Acknowledgement
The authors gratefully acknowledge the financial
support of the Deutsche Forschungsgemeinschaft
DFG, grant-no. Ru 345/28.

128

Flow and dispersion study in the simplified urban area

H. ednkov, Z. Jaour

Institute of Thermomechanics, v.v.i.

Academy of Sciences of the Czech Republic
Prague, Czech Republic
hased@seznam.cz
Abstract The aim of this experiment was a flow
measurement inside urban canopy layer and also
the study of pollutant diffusion from the point
source placed inside the urban area. The work took
place at the environmental wind tunnel of The
Institute of Thermomechanics, Academy of
Sciences of the Czech Republic. The Laser Doppler
Anemometry (LDA) was used as a flow
measurement technique. For the concentration
measurement the Flame Ionisation Detector (FID)
was used.
Key words urban canopy layer, gas dispersion, point
source

Introduction
Nowadays the pollutant diffusion is a big problem in
the urban build-up area. The character of the flow in the
rural boundary layer is well understood but in contrast
the flow inside the urban boundary layer needs more
exploration. The same situation hold for the diffusion of
the air pollution.
The task of this experiment is to study the flow in the
urban canopy layer - the layer of air in the urban canopy
beneath the mean height of the buildings and trees. Also
the pollutant diffusion from the point source placed
inside the urban area is investigated.

Wind tunnel is constructed as a open-circuit facility

that is operated as a fan driven one with constant profile
1.5 x 1.5 m2. Entrance is placed in suction tower (6 x 6 x
12) m3. There are dust filters inside its walls. Knee with
regulated vanes and semi-porous nets follow entrance.
Whole this section provides for stabilization of velocity
profile and inhibition of turbulent fluctuations. Wind
tunnel goes ahead for 20.5 m (zone with spires and
roughness elements where flow and boundary layer is
established) and comes to test section (2 m long). Test
section has glass side walls and traverse system which
provides remote control of measuring probes.
The Laser Doppler Anemometry (LDA) was used as
a flow measurement technique. 2 components of the
velocity vector were measured simultaneously.
For the concentration measurement the slow Flame
Ionisation Detector (FID) was used.

2 Experimental set-up
As the model of the urban build-up area the array of
cubes was used. Two model variants were studied.
Figure 2 shows the scheme of the arrangement of both
models.
4

4
3

2
2

y/H

y/H

1 Experimental Equipment

0
-1

The study took place in the environmental wind

tunnel of The Institute of Thermomechanics, Academy
of Sciences of the Czech Republic which is in Nov
Knn. The sketch of the tunnel is on the Figure 1.

-2

-2

-4

-3
-1

x/H

x/H

10

Figure 2 arrangement of the models ground plan

In each model, after the field of roughness elements,
26 rows of regularly placed cubes were used. The width
of street canyons was the same like the height of
buildings H = 50 mm. The scale of the model is
approximately 1:200.

3 Results
3.1 Flow field

All values in following graphs are time average over

100 s in model, which means around 5.5 hours in reality.
Figure 3 shows typical measuring point between the
buildings where blue points are instantaneous values of
one component of velocity vector and red points are
averages from zero time. It is evident, that measuring
time is long enough.

Figure 1 environmental tunnel of the Institute of

Thermomechanics AS CR, Nov Knn.

129

b)
3

measured values
average

10

U [m/s]

1
0
-1

nS ww(n)/u2* [-]

10-1

-2
-3
0

20

40

60

80

100

time [s]

10-2

Figure 3 time averaging

measured z/H = 4
Kaimal

The vertical wind profiles above both models are on

the Figure 4. The roughness parameter z0 and the
power law exponent D have the appropriate value for
very rough terrain.

10-3

10-2

10-1

n = fz/U [-]

100

101

Figure 5 smooth normalized spectral distribution of

the turbulent kinetic energy

measured data model 1

data fit
measured data model 2
data fit

8
7

There are horizontal cross-sections on the Figure 6.

For both models z/H = 0.5. Color of the vectors means
their magnitude. One can see that field is strongly
affected by the building blocks both in direction and in
magnitude.

D1 = 0.27
z01= 0.87
D2 = 0.3
z02= 0.94

z/H

4
1.5

3
2

magnitude

0
1.5

2.5

u [m/s]

y/H

1
3.5

0.5

Figure 4 vertical wind profile

Another important characteristic of the boundary
layer is the spectral distribution of the turbulent kinetic
energy. The Figure 5 shows the situation above the
model 1 at the height z/H = 4, a) is for the u-component
of the velocity vector and b) is for w-component.

-0.5

-1
-1

-0.5

0.5

x/H

a)

1.5

1.2
1.1
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1

z/H = 0.5
2.5
2

10

magnitude
1.2
1.1
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1

1.5

y/H

nS uu(n)/u2* [-]

10-1

10

0.5

-0.5
-1
-2

-1.5

measured z/H = 4
Kaimal
Karman
Simiu&Scalan
10

-3

10

-2

10

-1

n = fz/U [-]

10

-1

x/H

Figure 6 horizontal cross-section of the flow field in

the height z/H = 0.5
Following plot (Figure 7) shows the same situation as
the previous plots, but for z/H = 1.5.

10

130

buildings. In the model 1 is much bigger than in the

model 2.
1.5

3.2 Concentration field

In the model 1 the dispersion study from a point

source was done. The point source was placed inside
the build-up area in different positions in the face of
buildings. Outlet of the source was always on the ground
level. Ethane was used as a tracer gas and FID was
used for measuring the concentration.
Dimensionless concentration c*, which is used in
following plot is defined as

y/H

magnitude

0.5

2.00
1.94
1.88
1.81
1.75

-0.5

c*
-0.5

0.5

x/H

1.5

Figure 7 - horizontal cross-section of the flow field in

the height z/H = 1.5
As one can see, the direction of the velocity vector is
not affected by the obstacles. The same behavior was
measured also above the model 2.
Horizontal profiles of the velocity vector can be seen
on the Figure 8, which is on the separate page. On the
first plot (model 1) there is a situation in the middle of
the street canyon which is parallel with the direction of
the main flow. Streamlines are nearly parallel to each
other and there is no eddy. The second plot shows flow
field in the middle of the street canyon which is
perpendicular to the direction of the main flow. The
regular leeward wake formed between the buildings.
Third plot displays shorter street canyon in the model 2.
There is a foretaste of some eddy, but it is not
completely formed.
Also Reynolds stress was measured. Figure 9
illustrates the same street canyons as previous figure
but with Reynolds stress. Black dots are measuring
points, the values of this variable are interpolated from
measuring points only for illustration.
a)
mean (u'w')
0.06
0.04
0.02
0
-0.02
-0.04
-0.06
-0.08
-0.1
-0.12
-0.14
-0.16

z/H

0.5

0
-0.5

0.5

x/H

3
2.5
2

x/H = 2.5
x/H = 4.5

1.5

x/H = 6.5
x/H = 10.5

x/H = 14.5

0.5
0
-10

-5

10

y/H

b)

1.4

1.2

1.5

x/H = 2.5

0.8

x/H = 4.5
x/H = 6.5

c*

where cmeas represents measured concentration of

the tracer gas, v is referent wind speed, H is height of
the buildings and Q is source strength.
Measuring points were at the height z/H = 0.5. In the
Figure 10 are measured concentration in the case a)
from the point source placed in the middle of crossroads and in the case b) the source was placed in the
middle of the street canyon, which is perpendicular to
the direction of the main flow.
a)

c*

-1
-1

c meas [ ppm] v[m.s 1 ] ( H [m]) 2 10 6

Q[m 3 .s 1 ]

b)

0.6

x/H = 10.5

0.4

z/H

1.5

mean (u'w'): -0.13 -0.12 -0.10 -0.09 -0.08 -0.07 -0.05 -0.04 -0.03 -0.01 -0.00 0.01 0.02 0.04 0.05

x/H = 14.5

0.2

-8

-6

-4

-2

10

y/H
0.5

0
-1.5

-1

-0.5

0.5

x/H

1.5

2.5

Figure 9 Reynolds stress (vertical cross-section),

a) model 1, b) model 2.
In both cases typical area of negative Reynolds
stress appears around the roof level. Also area of
positive Reynolds stress is developed in lower parts of
the street canyons on the windward side of the

131

Figure 10 dimensionless concentration from point

source

2
1.5
1

y/H

0.5
0

magnitude: 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 1.6

-0.5
-1

1.4

-1.5

-1

x/H

1.2
1

z/H

0.8
0.6
0.4
0.2
0
-1.5

-1

-0.5

0.5

x/H

1.5

2
y/H

magnitude: 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5

z/H

1.5

-1

-1

x/H

0.5

0
-1.5

-1

-0.5

0.5

x/H

1.5

2.5

1.8

2
1.5
1

1.6

y/H

0.5
0

1.4

-0.5

magnitude: 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 1.6

-1
-1.5

z/H

1.2

-1

x/H

1
0.8
0.6
0.4
0.2
0
-1.5

-1

-0.5

x/H

0.5

1.5

Figure 8 vertical cross-section of the flow field, model 1 - first and second plot, model 2 third plot

132

4 Conclusion
The flow field inside two different arrangement of the
model of the urban canopy layer was done. For both
model was found out that the obstacles dont affect the
horizontal direction of the flow at the height z/H = 1.5
above the ground. For model 1 was also discovered that
the center of the leeward eddy in the vertical profile is
moving towards up at the position behind the cube.
Measurement of the pollutant diffusion from the point
source was done for the model 1. The dependency of
the concentration field on the source placing (in the
middle of the cross-roads, in the street canyon) was
found.

Acknowledgments
This work was supported by Institutional Research
Plan AVOZ20760514 and by COST 732.

References
Bedn J., Zikmunda O. (1985): Fyzika mezn vrstvy
atmosfry, Academia, Praha
Stull R. B. (1988): An Introduction to Boundary Layer
Meteorology,
Kluwer
Academic
Publishers,
Dordrecht
Jaour Z. (2001): Modelovn mezn vrstvy atmosfry,
Karolinum, Praha

133

Measurements on the Dispersion of Pollution in Urban Environment of

Cubic Building Arrays with Different Wind Directions
Bao-Shi Shiau1,2,

Yen-Shen Lin2

Institute of Physics, Academia Sinica, Taipei 115, Taiwan

E-mail: bsshiau@gate.sinica.edu.tw
2
Department of Harbor and River Engineering, National Taiwan Ocean University, Keelung 202, Taiwan
E-mail: b0085@mail.ntou.edu.tw
Abstract - In this study, experiments on the
dispersion of pollution in urban environment of
cubic building array in-line configuration with
different wind directions were conducted in wind
tunnel. Effects of wind attack angle, and
building array gap, G/H (G: building array gap, H:
cubic building height) on the short range
dispersion were investigated. Measurements of
flow field show that the building array gap
increases, the mean velocity increases, and the
building array gap becomes smaller, the
turbulence intensity is higher. Results of the flow
field and concentration measurements indicate
that the decrease of the array gap will be
favorable for pollutants dispersion in the urban
environment. The lateral concentration profiles
at downstream stations of source are shown to
be approximately Gaussian distribution when
the building
array gap G/H=3 with wind attack
angle =00. As the building array gap G/H=2 and
1 with wind attack angle >00, the lateral
concentration profiles at downstream stations of
source are exhibited to be non-Gaussian form.
The wind attack angle of 45 degree shows that
the lateral dispersion length scales are larger
than that of other wind attack angles. The
alternation of building array gap has shown no
significant effect on the change of the vertical
length scales.
Key words: dispersion scales, wind attack angle,
Gaussian distribution, building array gap

1 Introduction
Dispersion of emitted airborne pollutants in urban
environment is mainly affected by the buildings density
and wind attack angles on the buildings. Due to the
complexity of the buildings arrangements in the urban
region, it is difficult to predict precisely the dispersion of
pollutant by the numerical model. Field study can achieve
the goal in a more precision status. But works of the field
investigation [1] cost much. Wind tunnel experimental
simulation is therefore a feasible alternative. Hanna et al.
[2] had conducted the model simulations and observations
of mean flow and turbulence within simple obstacle arrays.
Mavroidis and Griffiths [3] had studied experimentally the
local characteristics of atmospheric dispersion within
building arrays. Mfula et al. [4] performed physical model
study in wind tunnel for urban building exposure to outdoor
pollution.
In this study, experiments on the dispersion of
pollution in urban environment of cubic building array
in-line configuration with different wind attack angles were

conducted in wind tunnel. Effects of wind attack angle, and

cubic building array gap on the short range dispersion
were investigated.

2 Experimental Works
Experiments were conducted in the environmental
wind tunnel. The test section of the wind tunnel had the
cross section of 2m by 1.4m and 12.6m long. The tunnel
was an open suction type and it contracted to the test
section with an area ratio of 4:1. The turbulence intensity
of empty tunnel is less than 0.5 % at the mean velocity of 5
m/s.
Four spires were placed at the entrance of the test
section and roughness elements succeeded to be
arranged 9 m long. The fully developed turbulent boundary
layer was generated as the approaching flow.
An X-type hot-wire incorporating with the TSI IFA-300
constant temperature anemometer was applied to
measure the turbulent flow signals. Output of the analog
signals for turbulent flow was digitized at a rate of 4 K Hz
each channel through the 12 bit Analog-toDigital
converter. Since none of the analog signals containing
significant energy or noise above 1 K Hz, with the Nyquist
criteria, a digitizing rate of 2 K Hz was sufficient. The low
pass frequency for the analog signals is set as 1 K Hz in all
runs of the experiments.
Methane was used as tracer gas and it mixed with the
standard gas. The mixed gas emitted from the stacks as
the sources in the experiments. The rake of sampling
tubes was placed at the sampled position. The rake is
composed of ten tubes. The cam system is adopted as the
pump function to suck the samples to the airbags through
tubes simultaneously. Each sample was taken for 5
minutes. The sampled tracer gas in airbag was analyzed
with FID (Flame Ionization Detector) to count the methane
concentration.
Three kinds of the cubic buildings arrangement are
made, and Fig.1 is the schematic diagram of one kind of
arrangement. In the figure, G is the gap between cubic
buildings, and H is the width and height of cubic building.
The source with a height of 0.5H is placed at 2H distance
in front of the first building of the building group. Wind
attack angles of 0o, 15o, 30o, and 45o are executed in the
experiments.

3 Analysis of Concentration
Dispersion Parameters
3.1 Scaling of the concentration

135

and

For analysis of the measured concentration variation,

dimensionless concentration parameter is adopted. The
dimensionless concentration, K is obtained from scaling
the measured concentration C by the building height H,
emission discharge rate Q, and mean wind speed of the
approaching flow at the building height UH .
K

CU H H 2
Q

WIND

U (Z )
U ref

Z n
)
Z ref

(6)

where U(Z) is the mean velocity at height of Z , Uref is the

free stream velocity, and Zref is the boundary layer
thickness. In the present study, an urban terrain type of
neutral atmospheric boundary layer was simulated with a
model scale of 1/1500. The measured mean velocity
profile is shown in Fig.2. Results indicate that the power
exponent n is 0.21. This value lies in the range of 0.17 to
0.23 as proposed by Counihan [5] for the rural terrain type
of neural atmospheric boundary layer flow.
The simulated longitudinal turbulence intensity profile
is shown in Fig.3. It is seen that the simulated longitudinal
turbulence intensity close to the wall is about 23%.
Counihan [5] had found that the longitudinal turbulence
intensity close to the ground in the rural terrain areas fell in
the range of 20% to 35 %.

(1)

G=H

Source

2H
H
H

Fig.1 Schematic diagram of the arrangement of the cubic

buildings and location of the source; wind attack angle
0
=0

Zref = 1m
n=0.21

0.8

3.2 Lateral and vertical dispersion parameters

( y  y ) C ( x, y, z)dy
C ( x, y, z)dy

0.6

Z / Zref

For the non-Gaussian distributions of airborne

pollutants dispersion in the urban environment, the
parameters V y and V z can be viewed as the standard
deviation of the concentration distributions in lateral and
vertical directions, respectively. They are used as the
indicators of the pollution dispersion length scales. The
parameters are defined by,

0.4

V 2y

V z2

0.2

(2)

( z  z ) 2 C ( x, y , z )dz

C ( x, y, z)dz

0.5

zC ( x, y, z)dy
C ( x, y, z)dy

0.8

0.9

Fig.2 Mean velocity profile of approaching flow

1

(4)

n = 0.21
Zref = 1m

0.8

0.6

(5)

Z/Zref

0.7

U(z) / Uref

where the centroids of the lateral and vertical

concentration distributions y and z are calculated as
follows.

yC ( x, y, z)dy
C ( x, y, z)dy

0.6

(3)

0.4

4 Results

0.2

4.1 Approaching flow

The neutral atmospheric turbulent boundary layer flow
was simulated as the approaching flow. The urban terrain
type of neutral turbulent boundary layer was generated as
the approaching flow. Mean velocity profile of the
simulated turbulent boundary layer flow is approximated
by the power law.

0
8

12

16

Iu (%)

20

Fig.3 Longitudinal turbulence intensity profile of

approaching flow

136

24

In Fig.4, the simulated longitudinal turbulent velocity

spectrum at Z/Zref=0.2 is compared with the von Karman
spectrum equation [6].
2u '2 Lux

(7)

2cfLux 2 6
U ( z )[1  (
) ]
U ( z)

The spectrum density, Su(f) and frequency, f are

normalized and they are denoted by USu ( f ) / u '2 Lux and
fLux / U , respectively. Here u' 2 is the mean square of
longitudinal velocity fluctuation; c is coefficient; Lux is the
integral length scale of longitudinal velocity in x direction;
U is the longitudinal mean velocity at the height of z. The
integral length scale is obtained by multiplying the integral
time scale, TE with the longitudinal mean velocity, U. And
the integral time scale, TE is computed by integrating the
longitudinal velocity autocorrelation coefficient function,
Ru (W ) .

TE

R (W )dW
0

1.4

1.2
velocity (Z/H=0.5)
open terrain

(8)

Fig.4 shows the comparison of the present the

simulated spectrum with the von Karmans spectrum
equation. It is found that a satisfactory agreement is
achieved between the simulation result and von Karman
theoretical prediction.

U(x) / U(hS)

Su ( f )

As the wind direction is normal to the building arrays,

the mean velocity variations at different heights along the
downstream stations of the source for various building
array gaps are shown in Fig.7 (a)~(d). Results exhibit that
the change of mean velocity is stronger as we narrow the
building array gap. This is due to the existence of the
building array in the flow filed. MacDonad and Griffith [7]
had reported the similar phenomenon as they made field
measurement of flow filed in the regular arrays of cubic
structures. The mean velocity varies significantly at the
lower height. The mean velocity At the height far from the
building, i.e. X/H>3, the mean velocity does not show large
variations along the downstream stations of the source for
three cases of G/H=1, 2, and 3.

G/H=1
G/H=2

G/H=3
0.8

0.6

10

0.4

10

15

20

25

USu(f)/u'2Lx u

X/H
0.1

Fig.5 Mean velocity variations along the downstream

station of source for different building gaps; Z/H=0.5, =0o

Z / Zref = 0.2
observational value
von karman
0.01

TIu
open terrain

50

G/H=1

0.001
0.001

0.01

0.1

fLxu/U

10

100

Fig.4 Longitudinal turbulence velocity spectrum;

Z/Zref=0.2

4.2 Flow field

The flow field, such as the mean velocity and
turbulence intensity are measured for the approaching flow
which is normal to the building array, i.e. the wind attack
angle =0o. The mean velocity variations along the
downstream stations of source for different building gaps
at X/H=0.5 are shown in Fig.5. It is seen that the mean
velocity decreases sharply at some downstream station,
X/H=3 for three building array gap cases. As the building
array gap increases, the mean velocity increases.
Fig.6 is the longitudinal turbulence intensity variations
along the downstream station of source for different
building gaps at Z/H=0.5. Results reveal that the building
array gap becomes smaller, the turbulence intensity is
higher.

longitudinal turbulence intensity TIU (%)

G/H=2
G/H=3
40

30

20

10
0

10

15

20

25

X/H

Fig.6 Longitudinal turbulence intensity variations along the

downstream station of source for different building gaps;
o
Z/H=0.5, =0

137

shows the concentration distributions at different

downstream stations of source are Gaussian distributions
for the case of open terrain. Results of Fig. 9 (b), (c), and
(d) indicate that for the case of G/H=3, the concentration
distributions at various downstream stations are also in
Gaussian distribution forms. For the case of G/H=2, the
concentration distributions at various downstream stations
deviate from the Gaussian form. When the building array
gap becomes G/H=1, the concentration distributions at
various downstream stations are almost in non-Gaussian
forms. This can be seen form the flow filed variations (see
Fig.5 ~Fig.7) that the local flow is strongly disturbed for the
high density arrangement of the building (i.e. building array
gap is small).

U(z) / U(H)
0.80.9 1 1.1 0.80.9 1 1.1 0.80.9 1 1.1

0.80.9 1 1.1

0.80.9 1 1.1

Z /H

4
3
2
1
0
0

10

12

14

16

18

20

22

24

26

X/H

(a)
U(z) / U (H)
0.60.70.80.9

(K)

4
3
2
1
0

9E-008
5
7E-008

10

12

14

16

18

20

22

24

Y/H

Z /H

0.60.70.80.9 0.60.70.80.9 0.60.70.80.9 0.60.70.80.9

26

X/H

(b)

5E-008

1
-1

4E-008

-3

3E-008

-5

U(z) / U(H)
0.50.60.70.80.50.60.70.80.50.60.70.8

0.50.60.70.8

0.50.60.70.8

10

12

14

16

18

20

22

24

X/H

2.5E-008

2.2E-008

(a)

4
3
2
1
0

Z /H

10

(K)
8
7.2E-008

10

12

14

16

18

20

22

24

26

Y/H

U(z) / U(H)
0.70.80.9 1

5.2E-008

(c)
0.6 0.8 1

6.2E-008

X/H

0.6 0.8 1

4.2E-008
0
3.2E-008
-2

0.6 0.8

3E-008

-4

Z /H

4
3
2

2.6E-008

-8

2.2E-008

-10

2.8E-008

-6

10

12

14

16

18

20

22

24

X/H

10

12

14

16

18

20

22

24

(b)

26

X/H

(K)
8

(d)

2.5E-007

4.3 Concentration

8E-008
4
7.2E-008
2

Y/H

Fig.7 Mean velocity variations at different heights along the

downstream station of the source for various building array
gaps, =0o;(a)open terrain, (b)G/H=3, (c)G/H=2, (d)G/H=1

6.2E-008

5.2E-008
4.2E-008

-2

The concentration contours for different building array

gap cases at Z/H=0.5, =0o are presented in Fig.8 (a)~(d).
As comparing the concentration contours, it is seen that
the case of smaller building array gap presents larger
dispersion extent. In other word, small building array gap is
favorable to the dispersion of pollutants in urban
environment.
Fig.9 (a)~(d) are the lateral concentration distributions
variations along the downstream stations of source for
different buildings array gaps at Z/H=0.5, =0o. Fig. 9(a)

3.2E-008
-4
3E-008
-6

2.5E-008

-8

2.2E-008
2

10

12

14

X/H

(c)

138

16

18

20

22

24

(K)

8E-008
6
5.2E-008

Y/H

G/H=2
X/H=1
street 1(X/H=4)

4.32E-008

street 2 (X/H=7)

-4

X/H=25

3E-008

-6

-8

X/H=13

6E-008

3.2E-008

-2

2.3E-008
2

10

12

14

16

18

20

22

X/H

4E-008

(d)
Fig. 8 Concentration contours for different building array
gap at Z/H=0.5, =0o; (a)open terrain, (b)G/H=3, (c)G/H=2,
(d)G/H=1
2E-008
-8

-4

Y/H

1.2E-007

(c)

Open terrain
X/H=1

8E-008

street 1 (X/H=4)
1E-007

X/H=13

G/H=1
X/H=1

X/H=25

street 1 (X/H=3.5)

street 2 (X/H=7)

street 2 (X/H=5.5)

8E-008

X/H=13.5

6E-008

X/H=23.5

6E-008

4E-008

4E-008

2E-008
-8

-4

Y/H

2E-008

(a)

-8

-4

Y/H

8E-008

(d)

G/H=3
X/H=1

Fig.9 Lateral concentration distributions variations along

the downstream station of source for different buildings
array gaps at Z/H=0.5, =0o; (a)open terrain,
(b)G/H=3,(c)G/H=2,(d)G/H=1

street1 (x=22.5)
street2 (x=42.5)
X/H=12.5
6E-008

X/H=24.5

The variations of maximum concentration axis at

Z/H=0.5 and G/H=1 for different wind attack angles is
shown as Fig.10.

4.4 Dispersion parameters

4E-008

2E-008
-8

-4

Y/H

(b)

Dispersion parameters are generally used as

indicators of pollutants diffusion extent. Fig.11 shows the
vertical dispersion parameters, V z variations for different
building array gaps at various downstream stations with
wind attack angle 0o. The vertical dispersion length scales
in open terrain are smaller than other cases with building
array. The vertical dispersion length scales increases as
increasing the downstream distance. And at some far
downstream distance, the vertical dispersion length scales
approach to a constant value for all cases of gaps. The

139

alternation of building array gap has shown no significant

effect on the change of the vertical length scales.
5

sigma y / H

-5

-10

Y/H

sigma (G/H=3)
wind direction: 0deg
wind direction: 15deg

-15

concentreation line
G/H=1
wind direction:0 deg

wind dirction: 45deg

wind direction:15 deg

-20

wind direction:30 deg

wind direction:45 deg

-25
0

10

X/H

15

20

25

Fig.10 Variations of maximum concentration axis at

Z/H=0.5 and G/H=1 for different wind attack angles

1.1

sigma z / H

1
Sigma z
open terrain
S/H=1
S/H=2
S/H=3
0.8

0.7

0.6
0

10

X/H

15

20

X/H

12

16

Fig.12 Lateral dispersion parameters variations for

different building array gaps at various downstream station
with wind attack angle 0o

5 Conclusion

1.2

0.9

wind dirction: 30deg

25

Fig.11 Vertical dispersion parameters variations for

different building array gaps at various downstream station
o
with wind attack angle 0

Wind tunnel experiments on the dispersion of

pollution in urban environment of cubic building array
in-line configuration with different wind directions were
conducted. Effects of wind attack angle, and building
array gap, G/H (G: building array gap, H: cubic building
height) on the short range dispersion were investigated.
Some conclusions are drawn from the experimental results
as follows:
(1) Measurements of flow field show that the building
array gap increases, the mean velocity increases, and
the building array gap becomes smaller, the
turbulence intensity is higher. Results of the flow field
and concentration measurements indicate that the
decrease of the array gap will be favorable for
pollutants dispersion in the urban environment.
(2) The lateral concentration profiles at downstream
stations of source are shown to be approximately
Gaussian distribution when the building array gap
G/H=3 with wind attack angle =00. As the building
0
array gap G/H=2 and 1 with wind attack angle >0 ,
the lateral concentration profiles at downstream
stations of source are exhibited to be non-Gaussian
form.
(3) The wind attack angle of 45 degree shows that the
lateral dispersion length scales are larger than that of
other wind attack angles. The alternation of building
array gap has shown no significant effect on the
change of the vertical length scales.

References

The Lateral dispersion parameter, V y variations for

different wind attack angles at various downstream
stations with building array gap G/H=3 are presented in
Fig.12. The lateral dispersion length scales increase as
the downstream distances increase. The wind attack angle
of 45 degree shows that the lateral dispersion length
scales are larger than that of other wind attack angles.

[1] Davison, M.J., Mylne, k.R., Jones, C.D., and Phillips,

J.C., (1995), Plume dispersion through large groups of
obstacles - A field investigation, Atmospheric
Environment, Vol.29, pp.3245-3256
[2] Hanna, S.R., Tehranian, S., Carissimo, B., MacDonald,
R.W., and Lohner, R., (2002), Comparison of model

140

simulations with observations of mean flow and

turbulence within simple obstacle arrays, Atmospheric
Environment, Vol.36, pp.5067-5079
[3] Mavroidis, I., and Griffiths, R.F., (2001) Local
characteristics of atmospheric dispersion within
building arrays, Atmospheric Environment, Vol.35,
pp.2941-2954
[4] Mfula, A.M., Kukadia, V., Griffiths, R.F., and Hall, D.J.,
(2005) Wind tunnel modeling of urban building
exposure
to
outdoor
pollution,
Atmospheric
Environment, Vol.39, pp.2737-2745
[5] Counihan, J. (1975), Adiabatic Atmospheric Boundary

Layers: A Review and Analysis of the Data from the

Period 1880-1972, Atmospheric Environment, Vol. 9,
pp. 871-905
[6] Kaimal, J.C.,, Wyngaard, J.C.,, Izumi, Y., and Cote,
O.R., (1972) Spectral characteristics of surface-layer
turbulence, Quart. J. R. Met. Soc. 98, 563-589
[7] MacDonald R. W. and Griffiths R. F., (1997) Field
Experiments of Dispersion through Regular Arrays of
Cubic Structures Atmospheric Environment, Vol. 31,
pp. 783-795

141

Specifying Exhaust Systems that Avoid Fume Reentry and Adverse

Health Effects
Ronald L Petersen and
John J. Carter
CPP, Inc.
Fort Collins, Colorado USA
rpetersen@cppwind.com
jcarter@cppwind.com

1. INTRODUCTION

Figure 1: ASHRAE 110 test setup for evaluating the

performance of a fume hood to contain chemical vapors.

Figure 2: Photographs of wind tunnel simulations showing

fumes exiting exhaust stacks. From the photograph one
should ask, are the air intakes safer than a worker at the
fume hood? Only a detailed dispersion modeling analysis
will provide the answer.

Studies have shown that there is a direct relationship

between indoor air quality and the health and productivity of
building occupants. [1, 2, 3 ] Historically, the focus of indoor
air quality has addressed emission sources emanating from
within the building. For example, as manufactured and as
installed containment specifications are required for fume
hoods to ensure that the worker is not exposed to toxic
chemicals as illustrated in Figure 1. An often-overlooked
aspect of indoor air quality within a laboratory is external
emission sources that may be re-ingested into the building
through closed-circuiting between the buildings exhaust
stacks and air intakes.
If the exhaust sources and air intakes are not properly
designed, higher concentrations of emitted chemicals may
be present at the air intakes than at the front of the fume
hood, where the chemical was initially released.
Furthermore, in the event of a toxic spill within the fume
hood, the individual working at the fume hood can take
corrective action by closing the sash and/or leaving the
immediate area, reducing their exposure to the released
chemical vapors. Conversely, the presence of the toxic
fumes at the air intake, which can distribute the chemical
vapors throughout the building, typically cannot be easily
mitigated. The only option may be to evacuate the entire
building, resulting in immediate loss of productivity and longterm reduction in the occupants satisfaction with the
working conditions.
To ensure a good exhaust and intake design,
dispersion modeling is required to determine the amount of
fume reentry, or the concentration levels expected at
building air intakes. Intakes include any potential pathway for
external air to enter into the building, including mechanically
driven air intakes, naturally ventilated intakes such as
operable windows and entrances, and leakage through
porous walls.
Petersen et al.,[4] provides a more technical description
of various aspects of exhaust and intake design. Some of
the challenges to specifying a good design mentioned in that
article include the existing building environment, aesthetics,
building design issues, chemical utilization, source types,
local meteorology and topography. For example, if a new
laboratory building is planned to be shorter than surrounding
buildings, it will be difficult to design a stack such that the
exhaust will not impact neighboring buildings. The effect of a
taller downwind or upwind building is illustrated in Figure 2.
The figure shows how the plume hits the face of the taller
building when it is downwind and how, when it is upwind, the
wake cavity region of the taller building traps the exhaust
from the shorter building. In either case the plume impacts
the face of the taller building. Typically, laboratory stack
design is a balance between various constraints and
obtaining adequate air quality at surrounding sensitive
locations (air intakes, plazas, operable windows, etc.). The

143

lowest possible stack height is often desired for aesthetics,

while exit momentum (exit velocity and volume flow rate) is
limited by capital and energy costs, noise, and vibration.
This paper summarizes a best practices guide that
was prepared for Laboratories for the 21st Century (Labs
21), a joint program of the U.S Environmental Protection
Agency and the U.S. Department of Energy. Geared toward
architects, engineers, and facility managers, this is one of a
series of guides that provide information about technologies
and practices to use in designing, constructing and
operating safe, sustainable, high performance laboratories.

10. Consider the effect of architectural screens. An

ASHRAE funded research project [8] found that
screens can significantly increase concentrations
on the roof.
11. Avoid a direct line-of-sight between exhaust stacks
and air intakes. An ASHRAE research project [9]
demonstrated that there is a distinct reduction in air
intake concentrations from roof top exhaust stacks
when air intake louvers are hidden on sidewalls
rather than placed on the roof.

3. Exhaust Design Criteria

2. Qualitative Information on Acceptable

Exhaust Designs

Fume hood emissions are often considered in a laboratorys

design; however, other pollutant sources may also be
associated with the project. These could include emissions
from emergency generators, kitchens, vivariums, loading
docks, traffic, cooling towers, boilers, and others. Each
source needs its own air quality design criteria. An air quality
acceptability question can be written:

Several organizations have published standards or

recommendations regarding laboratory exhaust stack design
as summarized below.
General Design Guidelines or Standards
1.
2.

3.
4.

5.

6.

7.

8.

9.

C max < C health / odor ?

Maintain a minimum stack height of 10 ft (3.0 m) to

protect rooftop workers. [5]
Locate intakes away from sources of outdoor
contamination such as mobile traffic, kitchen
exhaust, streets, cooling towers, emergency
generators and plumbing vents. [6]
Do not locate air intakes within the same architectural
screen enclosure as contaminated exhaust outlets. [6]
Avoid locating intakes near vehicle loading zones.
Canopies over loading docks do not prevent hot
vehicle exhaust from rising up to intakes above the
canopy. [ 6]
Combine several exhaust streams internally to
dilute intermittent bursts of contamination from a
single source, as well as producing an exhaust with
greater plume rise. Additional air volume may also
be added to the exhaust at the fan to achieve the
same end. [6]
Group separate stacks together (where separate
exhaust systems are mandated) in a tight cluster
to take advantage of the increased plume rise
from the resulting combined vertical momentum.
[6] Note that all the exhausts must operate
continuously to take full advantage of the
combined momentum. However, if not all of the
exhausts are operating at the same time, such
as in an n+1 redundant system, the tight
placement of stacks may be detrimental to their
performance.
Maintain an adequate exit velocity to avoid stacktip-downwash. The American National Standards
Institute/American Industrial Hygiene Association
Standard for Laboratory Ventilation, Z9.5-2003[7]
suggests that the minimum exit velocity from an
exhaust stack should be at least 15 m/s (3000
fpm), while ASHRAE [6] recommends a minimum
exit velocity of 10 to 15 m/s (2000 to 3000 fpm).
Apply emission controls where viable. This may
include installing restrictive flow orifices on
compressed gas cylinders; scrubber systems for
chemical specific releases; low NOx units for
boilers and emergency generators; and oxidizing
filters or catalytic converters for emergency
generators.
Avoid rain caps or other devices that limit plume
rise. Alternate design options are provided in
Chapter 44 of the ASHRAE Handbook-HVAC
Applications. [6]

(1)

where Cmax is the maximum concentration expected at a

sensitive location (air intakes, operable windows, pedestrian
areas), Chealth is the health limit concentration and Codor is
the odor threshold concentration of any emitted chemical.
When a large number of potential chemicals are emitted
from a pollutant source, a variety of mass emission rates,
health limits, and odor thresholds need to be examined. It
then becomes operationally simpler to recast the
acceptability question by normalizing (dividing) Equation 1
by the mass emission rate, m:

C
C
?
<
m
max m health / odor

(2)

The left side of the equation, (C/m)max, is only dependent on

external factors such as stack design, receptor location, and
atmospheric conditions. The right side of the equation is
related to the emissions and is defined as the ratio of the
health limit, or odor threshold, to the emission rate.
Therefore, a highly toxic chemical with a low emission rate
may be of less concern than a less toxic chemical emitted at
a very high rate. Three types of information are needed to
develop normalized health limits and odor thresholds:
1) a list of the toxic or odorous substances that
may be emitted,
2) the health limits and odor thresholds for each
emitted substance, and
3) the maximum potential emission rate for each
substance.
Health limits, Chealth, are typically based on the ANSI/AIHA
Standard Z9.5 on Laboratory Ventilation [7 ], which specifies
that air intake concentrations should be no greater than 20%
of the acceptable indoor concentrations for routine emission
and 100% of acceptable indoor concentrations for accidental
releases. Acceptable indoor concentrations are frequently
taken to be the short-term exposure limits (STEL) and can
be obtained from the American Conference of Governmental
Industrial Hygienists (ACGIH), the Occupational Safety and
Health Administration (OSHA), and the National Institute of
Occupational Safety and Health (NIOSH), as listed in
ACGIH.[10,11] ACGIH also provides odor thresholds,
Codor.[12]
For laboratories, the emission rates are typically based
on small-scale accidental releases, either liquid spills or
emptying of a lecture bottle of compressed gas. For other
sources, such as emergency generators, boilers, and
vehicles, chemical emissions rates are often available from
the manufacturer. Typical design criteria for various sources
are provided in Table 1.

144

Table 1: Typical Design Criteria

4. Dispersion Modeling Methods

Concentration predictions (C/m) at sensitive locations can
be obtained with varying degrees of accuracy using three
different methods: 1) a full-scale field program; 2) a reduced
scale wind tunnel study; or 3) a mathematical modeling
study.
A full-scale field program may provide the most
accurate predictions of exhaust behavior but may be very
expensive and time consuming. If the nature of the study is
to estimate maximum concentrations for several stacks at
several locations, it may require many years worth of data
collection before the maximum concentrations associated
with the worst case meteorological conditions are measured.
In addition, it is not possible to obtain data for future building
configurations.
Wind Tunnel modeling is often the preferred method for
predicting maximum concentrations. Wind tunnel modeling
is recommended because it provides the most accurate
estimates of concentration levels in complex building
environments. [6] A wind tunnel modeling study is much like
conducting a full-scale field study of a project before it is
built. Typically, a scale model of the building under
evaluation, along with the surrounding buildings and terrain
within a 1000-1500 ft radius (300-450 m), is placed in an
atmospheric boundary layer wind tunnel. A tracer gas is
released from the exhaust sources of interest. Concentration
levels of the tracer gas are then measured at receptor
locations of interest and converted to full-scale concentration
values. These values are then compared against the
appropriate design criteria to evaluate the acceptability of
the exhaust design. ASHRAE [6] and the EPA [13] provide
more information on scale-model simulation and testing
methods. Due to the highly technical nature of such studies,
care should be taken when selecting a wind tunnel modeling
consultant. Such factors as past experience and staff
technical qualifications are extremely important.
Mathematical models can be divided into three
categories: geometric, analytical, and Computational Fluid
Dynamic (CFD) models. The geometric method [6] defines
an appropriate stack height based on the string distance

between the exhaust stack and a nearby receptor location.

This method is entirely inadequate for exhaust streams that
contain toxic or odorous material since it does not provide
estimated concentration values at air intakes or other
sensitive locations. Hence, no information is provided to
avoid concentrations in excess of health or odor limits.
Analytical models assume a simplified building configuration
and provide concentration estimates based on assumed
concentration distributions (i.e., Gaussian). These models
do not consider site-specific geometries that may
substantially alter plume behavior and, thus, concentration
predictions. ASHRAEs HVAC Applications Handbook [6]
discusses exhaust stack design in some detail. When
properly applied, the analytical equations will tend to provide
conservative results for an isolated building or one that is the
same height or taller than the surrounding buildings and with
air intakes on the roof. As such, the analytical model can be
useful for screening out sources that are not likely to be a
problem, thus reducing the scope of more sophisticated
modeling. Neither the geometric nor the analytical models
are appropriate for complex building shapes or when taller
buildings are nearby.
Designers use Computational Fluid Dynamics (CFD) to
resolve fluid transport problems by solving a subset of the
Navier-Stokes equations at finite grid locations. CFD models
are currently being used successfully to model internal flow
paths within such areas as vivariums and atriums and in
external aerodynamics for the aerospace industry. However,
the aerospace CFD turbulence models are ill suited to model
the atmospheric turbulence in complex full-scale building
environments due to the difference in the geometric scales.
This is exemplified in the conclusions of Castros [14] recent
evaluation of applying CFD to the built environment:
Despite considerable effort over the last two decades, there
is no agreed modeling approach which will automatically
yield accurate results for the surface pressure field on and/or
the flow field around buildings in the wind Only LES
techniques genuinely have the potential to yield adequate
mean and fluctuating data, but these have yet to be fully
developed for complex bluff body flows Based on the
current state-of-the-art, CFD models should be used with
extreme caution when modeling exhaust plumes. Current

145

building roof. Plume rise increases with increased exit

momentum and decreases with increased wind speed. [13]
Reducing the diameter to increase exit velocity will increase
the exit momentum and thus, the plume rise. However, there
are limitations to how much the exit velocity can be
increased before developing noise and vibration problems.
Therefore, it is often preferable to increase the plume rise by
augmenting the volume flow rate, possibly by bringing in
additional air via a by-pass damper at the fan inlet. In
addition, plume rise is adversely affected by atmospheric
turbulence since the vertical momentum of the exhaust jet is
more quickly diminished. Therefore, in areas of high
turbulence the only method for obtaining an adequate plume
centerline may be to increase the physical height of the
stack.
If the ratio of exit velocity to approach wind speed is to
low, the plume can be pulled downwards into the wake of
the stack structure creating negative plume rise, a condition
referred to as stack-tip-downwash. This downwash defeats
some of the effect of a taller stack and can lead to high
concentrations. A rule of thumb to avoid stack-tip-downwash
is to have the exit velocity be at least 1.5 times the wind
speed at the top of the stack. [6] This stack top wind speed
is commonly taken to be the 1% wind speed, which can be
obtained from ASHRAE for various worldwide metropolitan
areas. [15] Note that the ASHRAE-provided wind speed
must be adjusted from the anemometer location to the stack
top. [16]
Variable volume exhaust systems should be designed
to maintain adequate exit velocity during turndown
operation. The exit velocity should be sufficient to avoid
stack-tip-downwash at all times. A high exit velocity can be
maintained by either having adjustable make-up air at the
exhaust stack via a by-pass damper or by employing several
stacks that can be brought on/off line in stages as flow
requirements change. Products are also available that can
change the geometry of the stack exit in an attempt to
maintain a high exit velocity with variable volume flow rates.
However, many of these devices do not properly condition
the flow as it exits the stack, which reduces the vertical
momentum and ultimately the plume rise out of the stack.
Alternatively, smart control systems can be used to set
minimum exit velocity requirements based on the current
wind conditions measured at a nearby anemometer.

research indicates that CFD models can both over and

under predict concentration levels by orders of magnitude,
leading to potentially unsafe designs. If a CFD study is
conducted for such an application, supporting full-scale or
wind tunnel validations studies should be provided.

5. Effective Stack Height and InducedAir Fans

Induced-air fan manufacturers often quote an effective
stack height for their exhaust fan systems. Many designers
interpret this value to be a physical stack height and
compare it to the height requirement defined from a
dispersion modeling study. This is an incorrect interpretation
of the effective stack height. The manufacturers specified
effective stack height is actually a prediction of the final
height of the exhaust plume centerline based on a
mathematical plume rise equation [6]. This final height
typically occurs far downwind of the exhaust stack (on the
order of 100-200 ft [30-60 m]). A more general mathematical
equation is available [13] that predicts the height of the
plume centerline as a function of downwind distance. What
the manufacturers should state is the effective increase in
the plume height versus downwind distance that may be
obtained over a conventional exhaust fan system. The
stated improvement may not be as great as one might
expect, as the following analysis points out.
Figure 3 shows the predicted plume centerline height
versus downwind distance for an induced-air exhaust stack
and a conventional exhaust fan system at a 20 mph (9 m/s)
stack height wind speed. The curves indicate that the
difference in the plume height between the two exhaust
systems is only 1 to 2 ft at 20 ft downwind (0.2 to 0.6 m at 6
m downwind) with a maximum difference of 6 ft (2 m) after
both plumes have reach their final rise. Therefore, the use of
an induced-air fan may reduce the necessary stack height
only slightly, depending upon the location of the nearby air
intakes. The effective stack height specification is
misleading.

7. Energy Issues
Several factors affect exhaust system energy consumption,
including: 1) the design and operation of the laboratory,
specifically the relative location of exhaust sources and air
intakes, the presence of nearby building elements such as
screen walls and penthouses, the exhaust volume flow rates
and exit velocities, and the chemical utilization within the
fume hoods; 2) the environment surrounding the laboratory,
involving the presence of nearby structures, air intakes, and
other critical receptor locations; and 3) the local
meteorology, specifically the distribution of local wind
speeds and wind directions.
Chemical utilization is the basic criterion used to judge
whether a specific exhaust/intake design is acceptable. An
overly conservative judgment regarding potential toxicity of
an exhaust stream may result in a high-energy-use exhaust
system as volume flow or exit velocity is increased
unnecessarily. A more accurate assessment of the intended
chemical use, with some consideration of future program,
will result in an exhaust system yielding acceptable air
quality while consuming a minimum amount of energy.
Local wind speeds may be used to set exit velocity
targets, as discussed previously. However, exhaust
momentum is the true parameter governing exhaust plume

Figure 3: Plume centerline height for conventional and

induced-air exhaust systems.

6. Plume Rise and Exit Velocity

Adequate plume rise is important to ensure that the exhaust
escapes the high turbulence and recirculation zones on the

146

rise and dispersion. In cases of high-volume flow-rate

exhausts (i.e., 30,000 cfm [14 m3/s] or greater), it has been
shown that exit velocities as low as 1000 fpm (5 m/s) can
produce acceptable air quality. Specific designs should be
evaluated on a case-by-case basis, regardless of exhaust
design parameters, to ensure that adequate air quality is
maintained at all sensitive locations.
Figure 4 was developed using the laboratory fume hood
criteria and the analytical dispersion model discussed
above. The figure shows that as volume flow rate increases,
shorter exhaust stacks can be used to meet the design
criteria. However, the shorter stacks are obtained at the cost
of increased exhaust fan power. The figure also
demonstrates the advantage of manifolding exhaust
systems. For example, a single stack operating at 5000 cfm
(2.4 m3/s) should be approximately 22 ft (6 m) tall to achieve
the design criterion at a receptor 160 ft (50 m) downwind.
Conversely, five stacks operating at 1000 cfm (0.5 m3/s)
would need to be nearly 38 ft (11 m) tall to provide the same
air quality at the same receptor location.

Figure 4: Stack height above top of intake required to meet a

specified design criterion for various exhaust volume flow
rates at a range of downwind distances.
Figure 5 shows how fan power may increase with
exhaust flow rate for various system designs. The figure
illustrates the relationships between the design volume flow
rate, Q, and the fan power requirements for two typical
induced-air systems and for a conventional system at three
different exit velocities. For the conventional exhaust
systems, the figure shows the benefit of decreasing the exit
velocity for a given design flow rate, always assuming that
the specified system meets the design goal, discussed
previously.

Figure 5: Required fan power versus design exhaust volume

flow rate, Q.

inches water column (W.C.) (1 kPa) to adequately ventilate

the building. An assessment of the exhaust plume shows
that a 10 ft tall, 30,000 cfm exhaust fan with 2500 fpm exit
velocity (2.1 m, 14 m3/s, 13 m/s) would meet the design
criterion established for the exhaust stack. Figure 5 shows
that a conventional exhaust system meeting these
parameters requires fan power of approximately 27 bhp (20
bkW). An equivalent induced-air system requires between
32 and 42 bhp (24 to 31 bkW) to exhaust the same flow from
the building, a 19% to 55% increase.
The above discussion illustrates the importance of using
dispersion modeling to evaluate exhaust performance to
ensure that acceptable air quality is achieved while
considering fan energy costs.

8. Summary And Conclusions

An accurate assessment of exhaust dispersion can be
used to produce exhaust/intake designs optimized for
energy consumption. No matter what type of exhaust system
is used, the important design parameters are: physical stack
height, volume flow rate, exit velocity, expected pollutant
emission rates, and concentration levels at sensitive
locations. Whether conventional or induced-air exhaust
systems are used, the overall performance should be
evaluated using the appropriate criterion, i.e., ensuring
acceptable concentrations at sensitive locations.
1

Fisk, William, J., Review of Health and Productivity

Gains from Better IEQ, Proceedings of Healthy
Buildings, Vol 4,pp. 23 - 34, 2000.

Yates, Alan, Quantifying the Business Benefits of

Sustainable Buildings (Draft), Building Research
Establishment, Ltd., Project Report 203995, 2001.

Kats, Greg, The Costs and Financial Benefits of Green

Buildings, Californias Sustainable Building Task
Force, Capital E, 2003.

Petersen, R.L., B.C. Cochran, and J.J. Carter,

Specifying Exhaust and Intake Systems, ASHRAE
Journal, August, 2002.

NFPA, Fire Protection for Laboratories Using Chemicals.

ANSI/NFPA Standard 45-96. National Fire
Protection Association, Quincy, MA, 1996.

ASHRAE, ASHRAE Handbook-HVAC Applications,

Chapter 44, American Society of Heating,
Refrigerating and Air-Conditioning Engineers, Inc.,
1791 Tullie Circle, N.E., Atlanta, GA, 30329, 2003.

ANSI/AIHA, American National Standard for Laboratory

Ventilation; Standard Z9.5-2003. American National
Standards Institute/American Industrial Hygiene
Association, 2003.

Petersen, R.L., J.J. Carter, and M.A. Ratcliff, Influence

of Architectural Screens on Roof-top Concentrations
Due to Effluent from the Short Stacks, ASHRAE
Transactions, Vol. 105, Pt. 1, 1999.

Petersen, R.L., J.J. Carter, and J.W. LeCompte,

Exhaust Contamination of Hidden vs. Visible Air
Intakes, ASHRAE Transactions, Vol. 110, part 1,
2004.

10

To better understand the data presented in Figure 5,

consider the following example. A building exhaust system
requires 30,000 cfm (14 m3/s) at a static pressure of 4

147

ACGIH, Guide to Occupational Exposure Values 2003, American Conference on Governmental

Industrial Hygienists, 2003.

11

ACGIH, 2002 Threshold Limit Values for Chemical

Substances and Physical Agents, American
Conference of Governmental Industrial Hygienists,
2003.

12

ACGIH, Odor Thresholds for Chemicals with

Established Occupational Health Standards,
American Conference of Governmental Industrial
Hygienists, 1989.

13

EPA, Guideline for Use of Fluid Modeling of

Atmospheric Dispersion, U.S. Environmental
Protection Agency, Office of Air Quality, Planning
and Standards, Research Triangle Park, North
Carolina, EPA-600/8-81-009, April 1981.

14

Castro, I.P., CFD for External Aerodynamics in the

Built Environment, The QNET-CFD Network
Newsletter, Volume 2, No. 2, July 2003.

15

ASHRAE, ASHRAE Handbook-Fundamentals, Chapter

27, American Society of Heating, Refrigeration, and
Air-Conditioning Engineers, 1791 Tullie Circle N.E.,
Atlanta, GA, 30329, 2001.

16

ASHRAE, ASHRAE Handbook-Fundamentals, Chapter

16, American Society of Heating, Refrigeration, and
Air-Conditioning Engineers, 1791 Tullie Circle N.E.,
Atlanta, GA, 30329, 2001.

148

Comparison of near-field behaviour of jets and plumes in a crosswind

E Savory, M Restorick, Z Duan

Department of Mechanical and Materials Engineering

University of Western Ontario
London, Ontario, Canada, N6A 5B9
e-mail: esavory@eng.uwo.ca
Abstract - Single and multiple momentum jets
in a crossflow (JICF) and buoyant plumes
exhibit both similarities and differences in their
near-field
behaviour.
Some
of
these
differences arise as a result of jets issuing
through ground plane openings (nozzles of
diameter D), whilst plumes often issue from
stacks within a thick turbulent atmospheric
boundary layer. In the literature many different
characteristics of JIFC have been examined,
including
velocity
and
turbulence
distributions, vorticity and circulation and the
jet wakes, whereas plume studies tend to
focus on spreading rates and species
concentrations. Studies on single and twin
jets, issuing from circular nozzles flushmounted in the ground plane, show that jet
penetration (z/D) is enhanced when the jets are
arranged in-line with the crossflow and
reduced when they are side-by-side (or normal
to the crossflow), when compared to a single
jet of the same jet velocity/crossflow velocity
ratio. The buoyant plumes issuing from two or
three stacks in-line or side-by-side exhibit
greater penetration for all the in-line plume
cases, whereas the side-by-side plumes only
had enhanced penetration for the three-stack
case.
Key words multiple jets, plumes in a crossflow,
LIF.

Introduction
The need to understand the behaviour of emissions
from multiple stacks is a matter of considerable interest
in air pollution control studies. Building exhaust systems
often incorporate two or more stacks placed only on a
few diameters apart, for which the effect of factors such
as stack separation distance, stack array pattern and
effluent exit velocity ratio should be considered.
However, in standard design texts such as the ASHRAE
Handbooks, only single exhaust sources are considered
in any detail. ASHRAE [1,2] suggest that a cluster of
stacks may be considered to act as a single stack,
where a cluster is defined as a group of sources all of
which lie within a radius of two stack diameters from the
centre of the grouping. However, there is little physical
basis for this statement nor has is been confirmed by
existing data and analysis of underlying mechanisms.
The well-studied jet in a crossflow (JICF)
phenomenon from aerodynamics is also very relevant to
buoyant plumes where the density distribution, due to
the initial jet-to-crossflow differences and/or crossflow
stratification, plays a crucial role [3-6]. The dominant

149

vortical mean-flow feature of buoyant plumes is the

secondary-flow contrarotating vortex pair (CVP).
It should be mentioned that hundreds of papers have
studied the single JICF problem in detail, while the
literature on the twin JICF or multiple JICF problem is
relatively scarce [7-13]. These papers give useful
information predominantly on global characteristics such
as jet path or vortex trajectory, on multiple or confined
multiple jets in crossflow by varying the number of jets in
a row, or by considering specific geometrical
configurations of nozzles (including their overlapping).
The present paper considers the case of buoyant
plumes issuing from two or three stacks arranged either
in-line (parallel to the wind direction, termed the
crossflow) or side-by-side (normal to the crossflow)
with the overall length of the stack line, S, measured
between the centres of the stacks at each end, being
equal to 4D, where D is the stack internal diameter. In
the present work, for momentum jets issuing from
ground plane nozzles, one geometry has been studied,
namely a spacing of 5D with in-line, side-by-side and
staggered (45o) orientations to the crossflow.

1 Experimental details
1.1 Plume measurements

The experiments were carried out in a pollutant

dispersion modeling water flume at the University of
Waterloo (Ontario, Canada), shown in figure 1. The
12.8m long flume has a 1.2m x 1.2m x 2.4m long test
section bounded by tempered glass panes for flow
visualization. A barrier, screens and ground roughness
elements were used to simulate the surface layer of the
atmospheric boundary layer (ABL) at reduced scale of
1:200, for a nominally suburban terrain.

Figure 1. Hydraulic flume working section looking

upstream (left), from side with laser light sheet (right)
A
three-component
micro-Acoustic
Doppler
Velocimeter (ADV), accurate to r1% of the mean value,
was used to determine the fluid velocity field. The
boundary layer parameters were: log law roughness
length z0 = 1.8 mm, power law exponent = 0.29, integral
length scale Lu = 40 cm at z = H, where H is the stack
height. The results showed that the flow in the flume
was very uniform (within r5% of the average value
across the flume width at any given height) and that the

repeated with the dye first issuing from all the stacks
and then from each stack in turn. By examining the
different data sets it was possible to observe the spread
of each plume individually and assess its contribution to
the overall plume at each downstream location. The LIF
technique requires a careful calibration to convert image
intensity to concentration. Hence, for each crosssection, a calibration measurement was first carried out,
as described above. The results showed that a linear
relationship had been achieved for all the experiments.
The plume cross-section concentration fields were
obtained from the video images by applying the
calibration charts of pixel intensity against concentration.

1.2 Jet measurements

The twin momentum jet measurements were carried

out in a wind tunnel, using the standard crossed hot-wire
anemometry technique. The tunnel working crosssection was 0.62 m width u 0.75 m height and the twin
jet nozzles of diameter D = 13.5 mm were placed flush
with the ground plane. All experimental data were
obtained for a nozzle centre-to-centre separation of
approximately 5D. The jet outlet velocity was UJ = 25.0
4
m/s, resulting in ReJ | 2.24x10 . Only one jetvelocity/crossflow-velocity ratio, R = UJ/UC, was studied,
R = 8. Hence, the crossflow velocity was set to be UC =
3.125 m/s. The velocity fields, that is the projections of
the mean velocity vectors in the planes x = Const., were
determined in five (or four) rectangular cross sections
located at x/D = 10, 12.5 (not measured for an oblique
nozzle arrangement), 15, 17.5 and 20.

2. Results and discussion

2.1 Plume trajectories and growth rates

It has been found that a concentration cross-section

through a plume may be approximated by a Gaussian
distribution [15]. The Gaussian peaks were assumed to
be at the plume centrelines and the standard deviation
Vy (lateral) and Vz (vertical) were found by a leastsquares fit to the concentration data using equations of
the form (e.g. for Vy):
(1)
where C is the time-averaged concentration and Cmin is
the minimum concentration at each downstream
location. In order to check the accuracy of the present
experimental simulations the results for the plume rise
and spreading parameters from the single stack case
were compared with existing results from the literature,
as shown in figures 2 and 3.

Non-dimensionalized
plume rise (z/D)

turbulence profiles scaled well with a typical neutral

ABL.
A platinum resistance temperature detector (RTD)
system was applied to measure and control the stack
exit flow temperatures and the upstream ambient flume
flow temperature. The crossflow velocity was maintained
at UC = 69 mm/s (measured at stack height) for all the
experiments, whilst the plume velocity at the stack exit
was UJ =138 mm/s, corresponding to a velocity ratio of
R = 2. The H=100 mm high model stacks with inside
diameter D=11.5 mm were located on the flume floor.
The geometrical centre of any stack arrangment was the
midpoint near the test section entrance. The water for
the stack supply was heated to give turbulent buoyant
plumes having a Froude number of Fr2 = 4.1, this being
a typical practical value. A number of stack
arrangements were studied, namely a single stack
(denoted by 1S in the legend on the graphs presented
here), two tandem (2SIL) and then two side-by-side
(2SBS) stacks (centre-spacing of 4D) and three tandem
(3SIL) and then three side-by-side (3SBS) stacks
(centre-spacing of 2D). In every experiment all the
stacks were operational, with the plumes issuing at the
same velocity and Froude number.
The planar laser-induced fluorescence (LIF) method
was used to measure the plume cross- section
concentration field. A 5W Ar-Ion laser provided
excitation of a fluorescent dye tracer (Rhodamine WT)
at a wavelength of 514nm, with the dye then re-emitting
light at a longer wavelength (580nm). Under appropriate
experimental conditions, the intensity of the fluoresced
light is proportional to the local dye concentration and
laser light intensity [14]. The LIF system was composed
of a laser, a scanning mirror, and a calibrated CCD
camera with a 555nm high pass cut-off filter to remove
the primary laser light, and a video image acquisition
and processing system. The laser light was guided
through a fibre optic cable and onto the scanning mirror
that was placed above the working section. The output
signal of the CCD camera was recorded and then
analysed by digital image processing software.
The plume cross-section concentration fields were
obtained from the video images by applying calibration
charts of pixel intensity against concentration. The
concentrations fields are presented here in the form (C
Cmin) / Cs, where C is the time-averaged concentration,
Cmin is the minimum concentration at each downstream
location and Cs is the concentration at the stack exit.
The accuracy in the measurement of C was estimated to
be r5%.
Each plume experiment comprised of three stages.
First, a 20-second sequence of images of a laserilluminated calibration box was recorded. The calibration
box had four rectangular plexiglass chambers, each
containing dye of a different known concentration. These
concentrations (0.5, 0.25, 0.1, and 0.05 ppm) were
chosen to be within the linear region of the intensityconcentration relationship. For the same conditions
(video camera located at the same distance from the
area of interest, with the same aperture and zoom) the
plume cross-section images were recorded for two
minutes in order to obtain the concentration field.
Simultaneously, the temperature of the source, Ts, and
of the ambient fluid, Ta, were measured by the RTD
system.
This complete procedure was repeated at four
different distances from the stack: at x = 100, 200, 400,
and 800 mm (where x is defined as distance measured
from the centre of the stack array), corresponding to
8.7D, 17.4D, 34.8D, and 69.6D. The experiments were

3
2.5
2
1.5
1
0.5
0

Int e g ral mo d e l
Exp e rime nt
2 / 3 p o we r law

0
20
40
60
80
Non-dimensionalized downstream
distance (x/D)

Figure 2. Comparison of single plume rise from

experiment with prediction model data

150

Normalized standard deviation

(s/D)

6
5
4
3
2

Sigma y/D
Sigma z/D

1
0
0

20

40

60

80

Normalized downstream distance (x/D)

Figure 3. Downstream variation of lateral (Vy)

and vertical (Vz) plume spread parameters
The measured trajectory in figure 2 is compared with
that calculated from the 2/3rd power law [16], and by
using an integral plume rise model [15]. The data are in
good agreement, which clearly illustrates the value of
the LIF technique for mapping plume development in a
crossflow. The data in figure 3 illustrate an
approximately linear growth rate for the spread
parameters, as expected for a plume in a neutral
atmosphere, with a ratio of dVz/dVy |0.8 that is within the
range of values reported previously [16]. These growth
rates are also in agreement with those conventionally
approximated using Vy = Iy x and Vz = Iz x, where Iy and
Iz are the average turbulence intensities in the ABL
along the plume.

The structure of a single plume is illustrated by the

flow visualisation images in figures 4 (time-averaged)
and 5 (instantaneous), the latter also highlighting the
planes in which mean concentrations were assessed.
The large-scale motion, associated with interconnected
vortex rings, is clearly well developed at x/D=17.4 but is
obscured, due to turbulent diffusion, by x/D=69.6. It is
these structures, deformed by the shearing action of the
crossflow, that give rise to the contrarotating vortex pair
(CVP) observed in the time-averaged flow associated
with both jets [17] and plumes in a crossflow. The ability
of the CVP to entrain fluid from the boundary layer is
much reduced, when compared to jets issuing from
ground plane nozzles, due to the elevation of the plume
by the stack.
Figures 6 and 7 compare the centre-line (plane y =
0) trajectories of the plumes between the different
cases, expressed as a plume rise (the height of the
geometrical centre of the plume above H), with the
geometrical locations normalized using the conventional
buoyancy length scale, lb. The data plots include results
from the 2/3rd power law model and also the Combined
Flux Model [16,18]. This incorporates an enhancement
factor (En, for n stacks) that modifies the 2/3rd power law
for multiple stack cases, the simplest version [18] using
En =
as the maximum possible enhancement.
It may be clearly seen that the CFM cannot
accurately predict the trajectories for side-by-side cases
(Figure 6 (a) and (b)) but it does predict the in-line stack
plume rise very well (Figure 6 (c) and (d)). The CFM
over predicts both the 2SBS and 3SBS plume rise by a
factor of 1.26. For the 2SIL case the plume rise
enhancement is slightly higher than the maximum of

31 3

Figure 4. Pseudo-colour images of time-averaged

concentrations in near-exit cross-sections of the single
plume at x/D=8.7 (top) and x/D= 17.4 (bottom)

Figure 5. Instantaneous LIF image of single plume

trajectory and structure

151

from the CFM model, this result being similar to

that found in previous work for two in-line stacks with a
stack spacing of S=2D [3].
Figure 7 shows that the trajectories are higher for the
in-line cases than for either the single stack (1S, some
points are hidden behind others on the plot), and the
side-by-side
cases.
Whilst
direct
quantitative
comparison with jet in crossflow studies (where the jets
issue from nozzles in the ground plane) is not possible
because of the different boundary conditions, these
results demonstrate a more rapid combining of the inline plumes (as in the in-line, or tandem, jet case [17]),
leading to increased penetration into the crossflow.
As an illustration, the profiles of concentration across
two side-by-side plumes at three consecutive
downstream planes, shown in Figure 8, clearly show
that the plumes have still not mixed together very well as
far as x/D=17.4 downstream. Because entrainment of
fluid only happens where the two plumes adjoin and the
CVP vortices have opposite rotations when the plumes
are side-by-side, mixing and re-organization are not
accelerated but, rather, delayed. By the time the two
plumes have re-organized into a single plume (around
x/D | 34.8), the plume has passed the initial phase and
has become bent over. In addition, an accumulation of
plume material at the underside of the combined plume
lowers the centre of mass and causes a downwash
effect [4], so that little enhancement occurs.

1000

2/3 Law

1000

CFM model

1S Experim ents
S
z/lb

z/lb

3SIL Experim ents

1S Experiments

100

CFM m odel

Uc

2SBS Experiments

Uc

2/3 Law

100

10
10

100

1000

10

x/lb

10

100

1000

x/lb

(a) Two side-by-side stacks (2SBS, S=4D)

1000

(d) Three stacks in-line (3SIL, S=2D)

2/3 Law

Figure 6. Comparison of multiple stack plume rise with

rd
2/3 law and Combined Flux Model (CFM)

CFM model
1S Experiments
Uc

3SBS Experiments

z/lb

S
S

100

1000

2/3 Law
1S Experiments
2SIL Experiments
3SIL Experiments
2SBS Experiments
3SBS Experiments

10

100

1000

x/lb

z/lb

10

100

(b) Three side-by-side stacks (3SBS, S=2D)

1000

2 /3 L a w
CF M m o de l

Uc

10

2 SIL E x p e rim e n ts
S
z/lb

10

1 S E x p e rim en ts

100

100

x/lb

1000

Figure 7. Comparison of single and multiple plume

centre-line trajectories

10
10

100

1 00 0

x /lb

(c) Two stacks in-line (2SIL, S=4D)

152

SINGLE
24

2SBS8.7L

21

21

18

18

SIDE-BY-SIDE
24

0 .0

0.14

2SBS8.7R

0.03

21

18
0.14
0. 25

0.03

0. 3

5
0 .1

0 .2

0. 25
14

0.

21

18

18

15

15

0.0 3

14
-0 .

-0 . 03

0.03

03

-0. 03

0.0 3

36

0.

-6

0.14

-0.03

-9

03

0. 47

0.

-3

Y/D

2SBS34.8A

0.

0 .2

0.03

0. 14

0 .0

-0. 03

0.03

12

5
-0 .26
-0 .3

-0 .1

.3

0. 25

-0

.2 5

Z/D

-0

9
6

Normalized lateral distance (y/D)

-0 .0

21

-0.01

OBLIQUE for Y>0

24

-0 .0

Z/D

Y/D

24

-0 .0

Y/D

12

0
1

0 . 03

-0 .

(C-Cmin)/Cs

0.01

03

Y/D

2SBS34.8L

0.02

(C-Cmin)/Cs

OBLIQUE for Y<0

2SBS17.4L

0. 2

Y/D

2SBS17.4R

-1

6
3

0. 0

0.02

-2

2SBS17.4A

-3

0 .0

0.04
0.03

0. 14

0.36
0.03

0.03

Z/D

0. 03

Normalized lateral distance (y/D)

0.1 4

12

407.3

0 .0

0.

0.03

36

Z/D

-0.01

-1

0.03

0.4

-2

12

0.36

-3

0. 03

0.

03. 1
0

15

0.

12

0.01

0 .1

0.02

15

0. 14

15

0 .2

0.

(C-Cmin)/Cs

24

0.03

0.04

TANDEM

2SBS8.7A

03

0.05

2SBS34.8R

0.01
0
-3

-2

-1

-0.01
Normalized lateral distance (y/D)

Figure 9. Comparison of vorticity contours at x/D = 15

for the different jet nozzle arrangements.

Figure 8. Concentration laterally across the plume

geometrical centre at x/D = 8.7, 17.4 and 34.8 for two
side-by-side stacks (legend: A = all stacks traced by
dye, L = left hand stack traced by dye,
R = right hand stack traced by dye)

16
14

2.2 Jet vortex trajectories and circulation

153

12
10

Z/D

The streamwise vorticity distributions, Zx (normalized

by Uc and D), for the different jet nozzle arrangements
are shown in Figure 9, for x/D = 15. These distributions
indicate that the gross vorticity features of the CVP (with
the exception of a slight asymmetry for the oblique
arrangement) are similar to those of a single JICF [19].
For the oblique case the peak vorticity values are on
average slightly lower than those of the tandem
arrangement and much higher than those of the side-byside arrangement. The z-locations of the vorticity peaks
are a little higher than those of the side-by-side
arrangement and clearly lower than those of the tandem
arrangement having a systematic upward z-shift with
downstream position (of about 25% of downstream
change of position). For comparison purposes Figure 10
shows the z-location of vorticity peaks for both in-line
(tandem), oblique (averaged due to asymmetry) and
side-by-side TJICF arrangements and a single JICF
measured for the same range of downstream positions.

8
6

SINGLE
TANDEM
OBLIQUE (AV.)
SIDE-BY-SIDE

4
2
0

10

12

14

16

18

20

X/D
Figure 10. The z-location of vorticity peaks for the
different jet nozzle arrangements.

applicable to plumes. The effects of the presence of the

stack, with respect to ground plane nozzles, in terms of
the near-field behaviour, remains a topic for further
research, although recent work on the detailed structure
of the near wakes of finite cylinders and stacks [20,21]
should help in that regard.

15

Circulation

12

Acknowledgments

9
6

SINGLE
TANDEM
OBLIQUE (AV.)
SIDE-BY-SIDE

3
0

The valuable contributions of our good friend and

colleague, the late Prof Robert Macdonald (1961-2004),
and Mr Brian Kim of the Department of Mechanical
Engineering at the University of Waterloo, Ontario,
Canada, are gratefully acknowledged. The enormous
contribution of Dr Vaclav Kol of the Institute of
Hydrodynamics, Prague, Czech Republic, to the twin jet
studies is also acknowledged.

10

12

14

16

18

20

X/D

Figure 11. Variation of integrated vortex circulation with

downstream distance for the different jet configurations
The variation with downstream distance of the total
circulation associated with one of the CVP vortices is
shown in Figure 11 (the data have been averaged from
the two vortices in the case of the oblique arrangement).
It may be seen that the in-line (tandem) case provides
the strongest circulation closely followed by the oblique
arrangement, illustrating constructive mixing between
the two jets. The side-by-side and single jet circulations
are significantly weaker.
The centreline vorticity exchange between the single
vortices of the CVP is a basic vorticity transport feature
of the CVP [17,19]. Turbulent vorticity transport can be
characterized and visualized in a straightforward manner
by turbulent vorticity fluxes. Indeed the centreline
vorticity exchange is the main cause of the cancellation
of the vortex strength of the CVP.

3 Conclusions
The results from the single stack buoyant plume
experiments are in good agreement with previous work.
In comparison with single stack experiments, there was
no plume rise enhancement but, rather, some
diminishment for the two stack side-by-side case.
However, some plume rise enhancement occurred for
three stack side-by-side case, indicating that both the
number of stacks and the stack separation distance
affect the plume rise. As expected, plume rise was
greatly enhanced for the in-line stack cases, because
the individual plumes can quickly mix and re-organize
into a single plume with more buoyancy flux. In all the
experiments, the maximum rise enhancements have
approached the theoretical limit of
13

for 2 stacks and

3 for 3 stacks. Further work is required to assess a

wider number of stack arrangements, including circular
arrays, in addition to the stack lines studied here.
Although stack plumes have not previously been
studied in any depth from the point of view of their
vorticity dynamics, the similarities with momentum jet
behaviour, in terms of enhanced plume rise (and, hence,
higher CVP trajectory) for in-line (constructively mixing)
plumes and reduced rise for side-by-side (destructively
mixing) plumes, suggest that the vorticity transport
processes in the jet and, hence, the resulting mean
vorticity distributions and circulation strengths, are also

References
1

ASHRAE. (1997). Airflow around buildings,

ASHRAE Handbook-Fundamentals, Chapter 15.10.
2
ASHRAE. (2003). Building air intake and exhaust
design, ASHRAE Handbook of HVAC Applications,
Chapter 44.1.
3
Contini, D. and Robins, A., (2001). Water tank
measurements of buoyant plume rise and structure
in neutral crossflows, Atmospheric Environment,
vol. 35, pp. 6105-6115.
4
Contini, D. and Robins, A., (2004). Experiments on
the rise and mixing in neutral crossflow of plumes
from two identical sources for different wind
directions, Atmospheric Environment, vol. 38, pp.
3573-3583, 2004.
5
Jirka, G.H., (2004). Integral model for turbulent
buoyant jets in unbounded stratified flows. Part I:
Single round jet, Environmental Fluid Mechanics,
vol. 4, pp. 1-56.
6
Macdonald, R.W., Strom, R.K., Slawson, P.R.,
(2002). Water flume study of the enhancement of
buoyant rise in pairs of merging plumes,
Atmospheric Environment, vol. 36, pp. 4603-4615.
7
Barata, J.M.M., Duro, D.F.G., Heitor, M.V. &
McGuirk, J.J., (1991). Impingement of single and
twin turbulent jets through a crossflow, AIAA
Journal , vol. 29, pp. 595-602.
8
Gregoric, M., Davis, L.R., Bushnell, D.J., (1982).
An experimental investigation of merging buoyant
jets in a crossflow, ASME Journal of Heat Transfer,
vol. 104, pp. 236-240.
9
Holdemann, J.D. and Walker, R.E., (1977). Mixing
of a row of jets with a combined crossflow, AIAA
Journal, vol. 15, pp. 243-249.
10 Isaac, K.M. and Jakubowski, A.K., (1985).
Experimental study of the interaction of multiple
jets and a crossflow, AIAA Journal, vol. 23, pp.
1679-1683.
11 Karagozian, A.R., Nguyen, T.T., Kim, C.N., (1986).
Vortex modeling of single and multiple dilution jet
mixing in a cross flow, ASME Journal of Propulsion
and Power, vol. 2, pp. 354-360.
12 Makihata, T. and Miyai, Y., (1979). Trajectories of
single and double jets injected into a crossflow of
arbitrary velocity distribution, ASME Journal of
Fluids Engineering, vol. 101, pp. 217-223.

154

13
14

15
16

17

18
19
20

21

Savory, E. and Toy, N., (1991. Real-time video

analysis of twin jets in a crossflow, ASME Journal
of Fluids Engineering, vol. 113, pp. 68-72.
Crimaldi, J.P. and Koseff, J.R., (2001). Highresolution measurements of the spatial and
temporal scalar structure of a turbulent plume,
Experiments in Fluids, vol. 31, pp. 90-102.
Schatzmann, M., (1979). An integral model of
plume rise, Atmospheric Environment, vol. 13, pp.
721-731.
Briggs, G.A., (1984). Plume rise and buoyancy
effects, Atmospheric Science and Power
Production, Office of scientific and Technical Info.
Center,
U.S.
D.O.E.
DOE/TIC-27601
(DE84005177), 327-366.
Kol, V., Takao, H., Todoroki, T., Savory, E.,
Okamoto, S. and Toy, N., (2003). Vorticity
transport within twin jets in crossflow, Experimental
Thermal and Fluid Science, vol. 27, pp. 563-571.
Overcamp, T.J., (1982). Plume rise from two or
more adjacent stacks, Report PPRP-67, Maryland
Power Plant Siting Program, September.
Kol, V., Savory, E. and Toy, N., (2000).
Centerline vorticity transport within a jet in
crossflow, AIAA Journal, vol. 38, pp.1763-1765.
Adaramola, M.S., Akinlade, O.G., Sumner, D.,
Bergstrom, D.J., Schenstead, A.J., (2006).
Turbulent wake of a finite circular cylinder of small
aspect ratio, Journal of Fluids and Structures, vol.
22, pp. 919928.
Adaramola, M.S., Sumner, D., Bergstrom, D.J.,
(2007). Turbulent wake and vortex shedding for a
stack partially immersed in a turbulent boundary
layer, Journal of Fluids and Structures (accepted).

155

Comparison of experimental and modelled plume rise in stable

environments
D. Contini, A. Donateo, D. Cesari

A. G. Robins

CNR-ISAC, Istituto di Scienze dellAtmosfera e del Clima,

Str. Prv. Lecce-Monteroni km 1.2,
73100 Lecce, Italy
d.contini@isac.cnr.it
Abstract In this work the trajectories of plumes
developing in a linearly stratified stable crossflow
have been studied. Several plumes trajectories, in
different physical conditions, have been simulated
and measured in a water tank. The characteristics of
the plumes have been discussed and compared
with the ones of plumes in neutral crossflow.
Results have also been compared with available and
widely used numerical and analytical models. A new
analytical model able to follow the oscillations of
the plumes around the equilibrium height is
presented and tested against experimental results.
Key words water tank, plume rise, stable crossflow.

Introduction
The accurate knowledge of the rise and spread of
plumes in a crossflow stably stratified is important in the
analysis of pollution dispersion in atmosphere. The
interest in the details of plumes development arises
because the understanding of the plume-crossflow
interaction is still incomplete despite the large
experimental and theoretical efforts of researchers.
What is more the study of plume-crossflow interaction
can be useful in several other applications like, for
example, VTOL aircraft, disposal of sewage into rivers
and estuaries and cooling of blades in turbine. The
frequent occurrence of plumes developing into a
crossflow has lead to numerous studies, in the last
several years, that face the problem from a theoretical,
numerical or experimental point of view (Bornoff &
Mokhtarzadeh-Dehghan 2001; Briggs 1984; Huq 1997;
McDonald et al 2002; Overcamp, 2007). In this paper
nineteen plumes in different physical conditions have
been simulated in the water towing tank at EnFlo
(University of Surrey) by using a stable cross-flow with a
linear stratification. The main characteristics of plume
development including trajectories and vertical spread
are discussed and compared with the predictions of
different models. The analysis represents an extension
of the study reported in Contini et al (2005) that was
limited to plumes developing in a neutral environment. In
particular it has been used the Briggs (1984) formula, an
integral model already described and used in Contini &
Robins (2005) and a new analytical model. The new
analytical model has been developed extending the one
reported in Slawson & Csanady (1971) in order to be
used in the general case of plumes in which both
momentum and buoyant fluxes are important. Results
indicate that the new model is able to predict the
oscillations of the plumes around their equilibrium height
but the frequency is underestimated as well as the
damping of the oscillations.
A fitting procedure has been used in order to
evaluate the entrainment coefficients that best fit

157

EnFlo, School of Engineering,

University of Surrey,
GU2 7XH Guildford, Surrey, UK
a.robins@surrey.ac.uk
experimental results. An important results is that the
entrainment coefficients obtained from the analytical
models are depending on the strength of the stability.

1 Experimental method and data

elaboration
Experiments have been carried out in the water towing
tank at EnFlo (University of Surrey). A linear density
gradient was obtained by filling the tank with a mixture of
fresh and salted water released at the bottom of the tank
with a PC-controlled filling procedure. Measurements
have been performed in an up-side-down configuration
even if results are reported in this paper in the usual
coordinates with the z-axis directed upwards. Buoyant
emissions were simulated by releasing, at the top of the
tank, a salt-water solution from stacks immersed for
about five diameters (D). A small amount of a blue
vegetable dye was added to the emissions and it acts as
a tracer for the plume visualisation system. The crossflow was obtained by towing the stacks through the still
water at a constant speed Ua referred to as the flow
speed.
Quantitative plume visualisations were obtained by
digital recording a side view of the back-lit plumes,
typically for 10-20 seconds (at 24 frames/s), showing the
overall path and structure of the dye concentration field.
Processed images were obtained by averaging over the
whole record and subtracting an associated background
average image in order to obtain the mean pixel
intensity I(x,z), at any position (x,z) in the image. The
averaging period was long enough to obtain statistically
stable results for I(x,z). The final images were then
analysed to determine the centre of mass trajectory,
<z(x)>, as function of the downwind distance x, of the
plume by considering I(x,z) proportional to the
concentration field C(x,y,z) integrated over the ydirection:

C(x, y, z)dy

I( x , z ) v

(1)

over the plume

The trajectory of the centre of mass, <z(x)>, downwind

of the sources is obtained from:

 Z( x ) !

f
0 zIz, x )dz
f

0 Iz, x )dz

(2)

The spatial resolution of the analysis is given by the

pixel-to-pixel geometrical distance that was typically 1.3
mm (vertical) and 1.5 mm (horizontal).

Other details about the experimental set-up as well

as the post-processing of data can be found in Contini &
Robins (2004) where it is also shown the validity of the
post-processing method using both an analysis of mass
balance and a comparison between the pixel intensity
profiles and the results of local concentration
measurements with a colorimeter system. The
repeatability of plume trajectories measured in nominal
identical conditions is estimated around 3-4% in the
range of distances from the source studied (i.e.
0<=X<=100D).
The vertical spread of the plume is studied using the
standard deviation z of the vertical pixel intensity
profile:
1

Vz

f z   Z ! 2 Iz, x ) dz 2
0

Iz, x ) dz

(3)

and also using the vertical radius of the plume rz(x)

defined as the distance between the top and bottom
outline of the plume in the final average image at any
distance x.

2 Plume rise models used in the

comparison
The plume rise models (analytical and numerical)
considered in this work for the comparison with
measured trajectories are widely used models in
dispersion codes. All the models are based on some
approximations to solve the general equation set that
governs plume development in a crossflow. The
equations used in this paper are derived with the
following assumptions: (I) the plume has no effect on
the properties of the environment, (II) the plume is
continuous, slender and with circular cross-section, (III)
properties are uniform within the plume (this is known as
top-hat profile). With these assumptions it is possible to
write the equation of motion for the plume, that have to
be integrated with the continuity equation for stationary
&
flows: div UU 0 , referring to Figure 1 for notations
(Contini & Robins, 2001):

d
ds
d
ds
d
ds
d
ds

Ur 2Us  2rUa Ve
Ur 2Us Us cosJ   Ua  Ur 2Us sinJ  dUdza
Ur 2Us2 sinJ  gr 2 Ua  U
gr2Us U  Ua  Ua Us sinJ r 2Z2

(4)
(5)
(6)
(7)

Equation (4) represents the entrainment assumption

through an entrainment velocity Ve. Equations (5) and
(6) represent the momentum equations in the x and z
direction respectively and g is the gravitational
acceleration. The symbol Ua indicates the flow speed
and Us is the plume tangential speed. Equation (7) is a
buoyancy conservation equation and Z is the BruntVisl frequency that can be written, for incompressible
fluids, as: Z2  g dUa .
Ua dz

Wind direct ion

U Ua
a
z

ds
U

Us
J

2r

Plumecent reline
Plume element
x

Figure 1. Scheme and notation used to describe the

trajectory and the properties of the plume.
Some simplifications and assumptions applied to the
general equations (4) to (7) allows to obtain a new set of
equations:

dr
U dx Ve

dM F
dx U

M
dF
Z 2

U
dx
2
M Ur w

2 Ua  U
F Ur g U

(8)

The simplifying assumptions are: the plume is bent-over

and it is basically moving in the horizontal direction at
the speed of the crossflow: Ua; the density difference
between the plume and the environment is important
only in the buoyancy equation. The set of equation is
generally closed with an entrainment hypothesis in
which the entrainment velocity Ve is given as function of
other known parameters usually involving the relative
motion between the plume and the crossflow. In the
case of constant Ua and with a linear density profile (i.e.
Z constant) and If the entrainment velocity is described
by Ve=EW being W the vertical plume velocity and E a
constant entrainment coefficient the set of equation
produces the Briggs equation, also reported in Weil
(1988) and in Davidson (1989):
1/ 3

z( x )

3Fo
Zx
Zx
3Mo


sin
2 2 2 1  cos

2
Z E U a
U a ZE U a U a

where:

Fo
Mo

grs2 Ws

U a  Us 
Ua

Us 2 2
Ws rs
Ua

(9)

(10)

(11)

being Ws the exit vertical velocity of the release, rs the

radius of the stack and U s
2U a  U m is the density
of the release of an equivalent plume released upwards

158

as in the usual coordinate system and Um is the real

emission density.
Equation (9) describes the plume rise, of a point
source, in a linearly stratified stable environment up to
the maximum plume height zmax. Afterwards the model is
not able to describe plume trajectories giving also
negative values of z. This happens because during
plume oscillations W becomes negative giving a
negative entrainment velocity that is not physically
acceptable. A possibility to overcome this problem is to
use an entrainment velocity given by Ve E W as was
firstly proposed in Slawson & Csanady (1971) for pure
jets. With this change in the entrainment the equation
set can be solved analytically separating the cases in
which W>0 (plume moving upwards) and the cases in
which W<0 (plume moving downwards). This cases are
separated by the relative maxima and minima in the
plume trajectories that happens at time ti given by:

ti

1
ZMo (i)S
arctg 

Z
Fo Z

i 1, 2, 3, ...

(12)

The general set of equation can be solved for each

temporal step generalising the results reported in
Slawson & Csanady (1971) to the cases in which both
the plume initial buoyancy flux and momentum fluxes
are important. The solution is an infinite series each one
representing a step between a relative maxima and a
relative minima of the trajectory. The first step is the only
one starting from the stack position.

z( t )

1/ 3
3Fo

 1i 1 cos(Zt )  (1)i cos(Zt i ) 

2
2
Z U a E

3
i
z( t i )  (1)  3Mo (1) i sin(Zt )  (1) i 1 sin(Zt )  R i
i
E
ZU a E 2

i
 E

R i 1

R i  E Z( t i  1 )  Z( t i )

for t i d t d t i 1 with i

0,1,2.....

(13)

The previous equation uses the point source

approximation and it has to be initialised with R0=0 and
t0=0 and z(t0)=0. Equation (13) is given in the time
domain, however it can be easily obtained in the space
domain by using the bent-over plume approximation in
which t=x/Ua. The first step of the new model describes
the plume trajectory up to the trajectory maximum that
happens at the time:

t1

1
ZMo S
arctg 

Z
Fo Z

(14)

and it is equivalent to the Briggs equation (9). It is

possible to demonstrate that the limit Z o 0 for
neutral crossflow of the new model is giving exactly the
Briggs equation reported, for example, in Contini &
Robins (2001). Therefore the new equation (13)
proposed is completely coherent with known limiting
cases.
A third model compared with the measurements is
an integral model, similar to the one described in Ooms

159

& Mahieu (1981) that iteratively solves the general

equation set using an entrainment velocity given by:

Ve

D 'U t  E 'U n

(15)

where D and E are the entrainment constant coefficients.

The Ve is proportional to the relative tangential velocity
Ut and the relative vertical velocity Un of the plume.
This integral model was already been applied to the
analysis of plume trajectories in neutral crossflows
(Contini & Robins 2001, 2005). It has the advantage,
with respect to the mentioned analytical models, that it is
not based on further simplifying assumptions and it can
be applied, for example, in presence of wind shear or in
non linear stratification. The integral model can easily be
implemented on a PC and it can run in a few seconds.

3 Characteristics of plumes in stable

crossflow
During experiments 19 plumes have been reproduced in
the water tank in the different conditions reported in
Table 1.
Run

Ua
cm/s
4.6

wU / wz
4
kg/m
-300

D
mm
7
7
10

Q
l/min
1
1
2

Ua
kg/m3
1062

Us
kg/m3
1098

6a
8a
4.6
-284.1
1087
1200
9a
4.85
3
10.36
-75.7
7
2
1032
1127
5
19.75
7
9
4.85
-75.7
7
1
1032
1127
11
10.36
-78.2
7
1
1032
1129
13
19.75
15
1
20
-81.9
7
1031
1129
17
2
19
7.56
-81.9
7
2
1031
1129
21
6.23
-81.9
7
2.7
1031
1129
19.75
23
10.36
-35.1
7
2
1032
1125
25
4.85
27
29
4.85
-114.8
7
2
1036
1141
31
10.36
-117.4
7
2
1036
1141
33
19.75
Table 1. Summary of the different plumes simulated
during experiments.
The density gradient wU / wz used in the measurements
goes from slightly stable environment to a strong
stratification. There are three main differences between
the plumes developing in stable crossflow with respect
to the ones developing in neutral crossflow (Contini &
Robins, 2001 and 2005). The first difference is that
plumes rising in stable crossflow can reach an
equilibrium height ze and, in cases of strong density
gradient and important momentum flux at the source, ze
is reached through a series of damped oscillations in
which the plume reaches a maximum height zmax. In our
experiments evident oscillations are present, near the
source, in 9 cases out of 19; an example is reported in
Fig. 1
The second difference with respect to plumes
developing in neutral crossflow is that the vertical
spread rz of the plume is relevant up to zmax; afterwards
rz is more or less constant (or at least its growth is
strongly reduced) and the plume tends to spread in the

carried out using E=0.38 (i.e. equal to the average ratio

rz/<z>). Results indicate that the comparison is
reasonably good.
Calculated position of xmax (mm)

horizontal direction. Our side views of the plumes do not

allow to determine ry; however this behaviour has also
been observed in other experiments in which a ratio ry/rz
up to 6 has been observed in plumes developing in
stable crossflows (Huq, 1997). The measured values of
rz are well correlated with <z> as shown in Fig. 2. The
average ratio rz/<z> is 0.38 and this has to be compared
with the value obtained for plumes developing in neutral
plumes 0.46 (Contini et al, 2005).

600
y = 0.8409x

500

R = 0.991

400
300
200
100
0
0

100
200
300
400
500
Measured position maximum xm ax (mm)

600

Calculated frequency (Hz)

Figure 3. Comparison of calculated and measured

values of xmax.

Z
Figure 1. Example of a plume final average image
showing the oscillations around ze. Reference system is
added to the figure.

R = 0.9712

10

11

12

13

Measured frequency (Hz)

Rz = 0.3168 <Z> + 8.1744

2
R = 0.9129

350

40
20
0
0

50

100

150
200
<z> (mm)

250

300

350

Figure 2. Measured vertical spread rz against average

plume heigth <z>.
The values of rz and Vz are well correlated and the
average ratio rz/Vz is 2.3 and it is basically the same
value that has also been obtained for plumes developing
in neutral crossflows (Contini et al, 2005).
The calculated position of the maximum height xmax
are compared with the measured values in Fig. 3. The
linear fit (made forcing the fitted line to pass through
zero) shows that analytical models underestimate xmax.
A similar underestimation is also observed for the
calculated frequencies of the oscillations around the
equilibrium height as shown in Fig. 4.
The values of zmax calculated by the analytical
models of Eq. (9) and eq. (13) are depending on the
entrainment coefficient used:
1/ 3

3F
3M o

3Fo2
o

 sin(Zt1)

2
2
2
3
2
ZU E
Z U a M oE
Z U aE
a

(16)

A comparison with measured values of zmax is

reported in Fig. 5 in which calculations have been

Calculated Zmax (mm)

60

300

y = 0.9912x

250

R = 0.8933

200
150
100
50
0
0

50

100

150
200
250
Measured Zmax (mm)

300

350

Figure 5. Comparison of calculated and measured

values of zmax obtained using E=0.38.
300
Calculated Ze (mm)

rz (mm)

80

Zmax

y = 0.8843x

Figure 4. Comparison of calculated and measured

frequencies of the oscillations around ze.

120
100

12
11
10
9
8
7
6
5
4
3
2

y = 0.9408x

250

R = 0.9652

200
150
100
50
0
0

50

100
150
200
Measured Ze (mm)

250

300

Figure 6. Comparison of calculated and measured

values of ze obtained using E=0.38.
In Figure 6 is reported a comparison of the calculated
and measured values of the equilibrium height ze. The
ratio K between zmax and ze can be calculated from our
measurements in 9 cases out of 19 and its average
value is: 1.19 (+/- 0.05 as one standard deviation). The

160

In order to compare the model predictions with the

measured trajectories it is necessary to use specific
values of the entrainment coefficients. From an analysis
of the scientific literature (Davidson, 1989; Huq, 1997)
the best entrainment coefficient E for Eq. (9) is around
0.4 and it is reasonable to assume the same value also
for the new model in Eq. (13) because the two models
are based on the same set of equations and uses the
same approximations. However there are some
differences in the entrainment coefficient determined
from the trajectories (fitting a model to the data) and the
one determined from the growth of the plume through
the definition of the entrainment velocity. As matter of
fact the analytical models used are based on the
following growth of the plume r=E<z>. Where r is the
radius of the plume that, with the assumption of
cylindrical plume, is the same in the z and y direction so
that r=ry=rz. If the entrainment coeffcient is determined
through the growth of the plume usually is found 0.4 in
stable crossflows (Davidon 1989; Huq 1996; Huq 1997;
Weil 1988). In our experiments the analysis of the
growth of the plume gives 0.38 as a coefficient.

Calculated E from fit

0.7

Integral model

0.65

Briggs model

0.6

90

RUN 6a

80
70

Plume rise Z (mm)

4 Comparison of measured and

modelled trajectories

However it is reasonable to use a fitting procedure in

order to find the best coefficient that fits experiments.
The model in Eq. (9) is fitted only up to the maximum of
the trajectories (if a maximum is reached) instead the
new analytical model as well as the integral one are
fitted also in the descending parts of the trajectories
when oscillations of the plume are present. The results
of the fit are reported in Fig. 7 as function of a stability
parameter Z/Ua and in Table 2 in terms of average and
standard deviation.

60
50

Measurements
Briggs model
Integral model
New model

40
30
20
10
0
0

50

100

150

200

250

Distance X (mm)

300

350

400

Figure 8. Comparison of measured trajectory for run 6a

with the predictions of the different models.
120

RUN 7
100

Plume rise Z (mm)

calculated average value from Eq. (9) (that is the same

of the one of Eq. 13) is 1.21 (+/- 0.02 as one standard
deviation). This calculated value is independent on the
entrainment coefficient because if E changes the values
of zmax and ze changes in the same way so that the ratio
is constant. The average ratio calculated with the
integral model is 1.21 (+/- 0.04 as one standard
deviation).

Measurements
Briggs model
Integral model
New model

80
60
40

New model

0.55

20

0.5
0.45

0
0

0.4
0.35
0.3
0

10

20

30

Z/Ua

40

E New model
E Integral
D Integral

Z/Ua>6

Z/Ua<6

0.45
(0.11)
0.44
(0.11)
0.49
(0.05)
0.01
(0.03)

0.39
(0.06)
0.37
(0.05)
0.48
(0.04)

0.58
(0.05)
0.56
(0.06)
0.5
(0.05)
0.026
(0.04)

~0

300

400

Distance X (mm)

500

600

700

350
300

Plume rise Z (mm)

E Briggs

All data

200

Figure 9. Comparison of measured trajectory for run 7

with the predictions of the different models.

Figure 7. Calculated values of the entrainment

coefficient E against the stability parameter Z/Ua for the
different models used.

Parameter/mode
l

100

250
200

RUN 27

Measurements
Briggs model
Integral model
New model

150
100
50
0
0

Table 2. Summary of average values of the entrainment

coefficients obtained from the fitting procedure on single
plumes. In parenthesis the standard deviation.

161

100

200

300

400

Distance X (mm)

500

600

700

Figure 10. Comparison of measured trajectory for run 27

with the predictions of the different models.
It has to be put in evidence that the entrainment
coefficient calculated from the Briggs formula and also
from the new analytical model are depending on the

strength of the stability. In near neutral crossflows

(Zo0) the calculated values of E are around 0.58 that is
a value similar to the one (0.6) obtained in neutral
crossflows (Contini et al 2005) instead in relatively
strong stable conditions the value are around 0.38 that
is a value similar to other results reported in the
scientific literature (Davidson 1989; Huq 1997). The
entrainment coefficient E obtained from the integral
model is almost constant at the different stability,
however the values of D are increasing in near neutral
conditions. This could be interpreted thinking that in
neutral crossflow becomes important the entrainment
due to the relative tangential velocity between the plume
and the crossflow.
In Figures 8, 9 and 10 some examples of measured
and simulated trajectories are reported for a range of
different release and crossflow conditions.
To obtain information about the entrainment
parameters that can be considered representative of all
the measurements carried out it has been developed a
procedure that minimize a global error function H in order
to find the entrainment coefficients that best fit
simultaneously all the data available. The global error
function is calculated starting from the error function Hi of
each plume included in the fit:

Hi

1 N
z meas x k   z mod el x k 2

Nk 1
1 M 2
H
Mi 1 i

(17)

(18)

where N is the number of point considered in the fit of

single plume and M is the number of plumes considered
in the global fit. Results of the global fits are reported in
Table 3. Comparison of results in Table 3 and Table 2
shows that there are some difference between the
average of fit of single plumes and the results of global
fit even if the general pattern is similar. In particular the
main differences are in the results of the integral model.
These differences arise because in the average of
single fits every plume has the same weight of the other
instead in the global fit the weight of each plume
depends on the error function chosen.

Parameter/mode
l
E Briggs
E New model
E Integral
D Integral

All data

Z/Ua>6

Z/Ua<6

0.39
0.38
0.47
0.005

0.36
0.35
0.42
0.014

0.53
0.53
0.51
~0

Table 3. Summary of average values of the entrainment

coefficients obtained from the fitting procedure multiple
plumes.

5 Conclusions
The behaviour of buoyant plumes developing in a
linearly stratified stable crossflow has been studied with
experiments in a water towing tank. Measured
trajectories have been compared with the analytical
Briggs model, with an integral model and with a new
analytical models that is presented in this work. The new

model is an extension of the one proposed by Slawson

& Csanady (1971). The trajectories of plumes in stable
environments often show damped oscillations around an
equilibrium height that is reached after a maximum
height. During oscillations the vertical spread of the
plumes is limited because of the collapse of the vertical
mixing and the plumes develop in a thin layer spreading
horizontally.
All the models analysed are based on a closure of
the general set of equation based on entrainment
parameters in which the entrainment velocity is
expressed as function of the relative motion between the
plume and the crossflow.
The Briggs formula is not able to describe the
mentioned oscillations; instead both the new analytical
model proposed and the integral one can reproduce this
behaviour even if the frequency of the oscillations is, on
average, underestimated of about 12% and the damping
is also underestimated. The predicted position of the
maximum is smaller than the observed one (on average
of about 16%) instead the ratio zmax/ze is correctly
predicted by both the new analytical model and the
integral model. The entrainment coefficients obtained by
the Briggs model and by the new model through a fitting
procedure depends on stability and they approach 0.6
for near neutral conditions and 0.38 for stable
conditions. The entrainment coefficient E obtained fitting
the results of the integral model to the measured plumes
is almost constant (around 0.5) with respect to stability,
however the parameter D is increasing in near neutral
conditions going from zero up to 0.026.
The integral model is more versatile than the
analytical ones because it can be employed in different
conditions like, for example, variable density gradient
and wind shear that are conditions not easily solved
analytically. However, in the conditions analyzed, there
is not any evident advantage in using the integral model
with respect to the new analytical model proposed here.
This conclusion was also reached in Contini et al (2005)
studying the behaviour of buoyant plumes in neutral
crossflows.
A further development of the new analytical model is
to include a factor k that describes the added mass to
the plume. This approach is described in Briggs (1984),
Davidson (1989) and in Weil (1988) and it can modifies
the frequency of the oscillations and the position of the
trajectory maximum even keeping constant the
entrainment coefficients. This will be matter of future
analysis on this data set, however preliminary studies
show that a factor k around 0.3 could improve the
prediction of both analytical models.

Acknowledgments
This research was funded by Ministero della Universit
e della Ricerca Scientifica e Tecnologica in
collaboration with the Conferenza dei Rettori delle
Universit Italiane (CRUI), by British Council and by the
European Community through the Large Scale Facilities
section of the Training and Mobility of Researcher
program, contract ERBFMGECT980117. The authors
wish to thank Dr Paul Hayden and Mr. Tom Lawton at
EnFlo, University of Surrey, for their help in setting up
the experiments and in developing software and
equipment.

162

References
Bornoff R. B., Mokhtarzadeh-Dehghan, M.R. (2001)
Numerical study of interacting buoyant coolingtower plumes, Atmospheric Environment, vol. 35,
pp. 589-598.
Briggs G. A. (1975) Plume Rise Prediction in Lectures
on Air Pollution and Environmental Impact
Analyses, Workshop Proceedings, pp. 59-111,
American Meteorological Society.
Briggs G. A. (1984) Plume rise and buoyancy effects,
in Atmospheric science and power production,
Randerson D. Eds., pp. 327-366, DOE/TIC
27601, USA Dept. of Energy.
Contini D. and Robins A. (2001) Water tank
measurements of buoyant plume rise and structure
in cross flow, Atmospheric Environment, vol. 35,
pp. 6105-6115.
Contini D., Robins A. (2004) Experiments on the rise
and mixing in neutral crossflow of plumes from two
identical sources for different wind directions,
Atmospheric Environment, vol. 38, pp. 3573-3583.
Contini D., Donateo A., Robins A. (2005) Experimental
and modelled plume rise in neutral environments
proceedings of the International Workshop
PHYSMOD 2005, pp. 2-3.
Davidson G. A., (1989). Simultaneous trajectory and
dilution predictions from a simple integral plume
model, Atmospheric Environment, vol. 23, pp. 341349.
Huq P, Stewart E.J. (1996) A laboratory study of
buoyant plumes in laminar and turbulent
crossflows, Atmospheric Environment, vol. 30, pp.
1125-1135.
Huq P. (1997) Observations of jets in density stratified
crossflows, Atmospheric Environment, vol. 31, pp.
2011-2022.
MacDonald R. W., Strom R. K., Slawson P. R. (2002)
Water flume study of the enhancement of buoyant
rise in pairs of merging plumes, Atmospheric
Environment, vol. 36, pp. 4603-4615.
Ooms G. and Mahieu A. P. (1981) A comparison
between a plume path model and a virtual point
source model for a stack plume, Applied Scientific
Research, vol. 36, pp. 339-356.
Overcamp T.J. (2007) Stable plume rise in a shear
layer, Journal Of The Air & Waste Management
Association, vol. 57, pp. 328-331.
Slawson P. R., Csanady G. T. (1971) The effect of
atmospheric conditions on plume rise, J. Fluid
Mech., vol. 47, pp. 33-49.
Weil J. C. (1988) Plume Rise in Lectures on air
pollution modelling, A. Venkatram, J. C. Wyngaard
Eds., 121-166, American Meteorological society.

163

Validation of LES on plume dispersion in the convective

boundary layer capped by a temperature inversion
Tetsuro Tamura, Hiromasa Nakayama and Kenichi Ohta
Department of Environmental Science and Technology
Tokyo Institute of Technology
4259 Nagatsuta, Yokohama, Kanagawa, Japan
Email : tamura@depe.titech.ac.jp

Abstract - In order to investigate unsteady

properties of the concentration field under unstable
thermal condition, Henn and Sykes [1] and Mason
[2] carried out LES of concentration fluctuations in
the convective boundary layer. Transport process in
thermally stratified turbulent boundary layers, which
can be generally seen in the actual atmospheric
boundary layer, is much influenced by turbulence
structures. In this research, we carry out LES
analysis of dispersion in spatially-developing
convective turbulent boundary layers capped by a
temperature inversion. We employ the dynamic
Smagorinsky model as subgrid-scale modeling of
LES for flow, temperature and concentration fields.
The objective of this research is mainly to validate
LES model by comparison with the experimental
data [3] and to investigate the dispersion
characteristics of plumes emitted from a source
point in spatially-developing thermally stratified
turbulent boundary layer, including concentration
fluctuations and their peak values. Figure 1 shows
a numerical model for the simulation of thermally
stratified boundary layers. In order to generate the
inflow turbulence, in the driver region we used
Lunds method [4] that the turbulence parameters
for velocity at recycle station are rescaled and
introduced to the inlet boundary. Temperature is
treated as a passive scalar in the driver region. The
generated inflow data are imposed at entrance of
the main computational region, and the buoyancy
effect is introduced into the flow field of the main
region. This process allows a small-domain
computation of spatially developing convective
boundary layer with a capping inversion layer. We
analyzed concentration field in the way to release
passive scalar from an elevated source point at the
main region. Figure 2 is typical snapshots of the
plume in the vertical section under unstable, neutral
and stable conditions. Two iso-surfaces indicate 3%
and 0.3% of emit intensity. Dispersion in stable
boundary layer is more suppressed than in neutral
boundary layer. A large plume meandering is
frequently seen in unstable boundary layer.

165

Fedorovichs experimental data [3] for the unstable

case are mainly simulated for LES validation.
Quantitative estimation for accuracy of the present
LES will be performed.

Figure 1

Fig. 2 : Instantaneous concentration field in vertical

section (Unstable(top), Neutral (middle), Stable 1
(bottom))

References
D. S. Henn and R. I. Sykes, 1992, Atmos. Environ. 26A,
3145-3159.
P. J. Mason et al., 1990, Boundary-Layer Meteorol. 53,
117-162.
E. Fedorovich and J. Thater, 2002, Atoms. Environ. 36,
2245-3355.

Simulation of long time averaged concentration

under actual meteorological conditions

Haruyasu Nagai and Takashi Hayashi*

Tomohiro Hara, Ryohji Ohba, Kazuki

Okabayashi and Jiro Yoneda

Nagasaki R&D Center, Mitsubishi Heavy Industries,

Ltd., Fukahori machi 5-717-1, Nagasaki, Japan
Ryohji_ohba@mhi.co.jp

Nuclear Science and Engineering Directorate,

Japan Atomic Energy Agency
Shirakata 2-4, Toukaimura, Ibaragiken, Japan
*:Former Affiliation

Abstract We simulated a meandering effect

of wind direction fluctuation on horizontal gas
diffusion over Mt. Tsukuba near Tokyo, using a
rotating turntable in the wind tunnel experiment.
Experimental results of wind velocity and gas
concentration were validated with field data
observed by Japan Atomic Energy Research
Institute (JAERI), 1989 and 1990. This technique can
be applied to environmental assessment based on
the air quality standard usually defined by 1 hour or
30 min. averaged concentration.
Recently,
mesoscale
meteorological
models become able to simulate local scale
phenomena in the mesh size up to about 10m, by
improving turbulence closure model. We simulated
actual unsteady phenomena of airflow and gas
diffusion over Mt. Tsukuba, and compared the
calculated results with field data. Both data agreed
well under neutral, stable and unstable atmospheric
stabilities.
Key words Meandering, atomospheric stability, wind
tunnel, meteorological model, diffusion.

1 Introduction
Variation of gas concentration in the field depends
on unsteady meteorological conditions of wind velocity,
wind direction and atmospheric stability. However,
conventional
wind
tunnel
experiments
and
Computational Fluid Dynamic (CFD) models cannot
simulate these unsteady phenomena, because they
assume steady state meteorological conditions.
Environmental
assessments need a long time
averaged concentration for 30 min., 1 hour or 1 year in
actual site. Some empirical formula for a meandering
factor have been used for these needs. However, the
meandering factor depends on a terrain and a stability
condition at each site.
We simulated a meandering effect of wind direction
fluctuation on horizontal gas diffusion over Mt. Tsukuba
near Tokyo, by our original technique using a rotating
turntable. In the wind tunnel experiment, and by the
mesoscale meterological model (RAMS/HYPACT).

2 Field experiments

Field experiments were carried out at Mount

Tsukuba region ( Fig. 1) in Japan.

167

Fig.1 Map of Mt. Tsukuba region. Locations of

measuring stations are denoted by closed circles.
The observed data refer to two atmospheric diffusion
experiments carried out in 1989 from 13 to 20
November, and in 1990 from 10 to 18 November, in the
Tsukuba area as part of the Experiment to Demonstrate
the Propriety of Atmospheric Dispersion Evaluation
Method for Safety Analysis (Hayashi et al., 1999a,
1999b), conducted by Japan Atomic Energy Research
Institute in cooperation with the Japan Weather
Association. In Fig. 1, the campaign domain with the
locations of the measurement points are shown also.
The available meteorological observations for the
station at the top of Mt. Tsukuba (named TOP) were
collected from the AMeDAS (Automated Meteorological
Data Acquisition System of the Japan Meteorology
Agency) dataset. The AMeDAS system measures wind
speed, wind direction, temperature, and precipitation
automatically, recording 10-minute-averaged data every
hour.
At the AA station, a Doppler sodar was used,
providing wind speed and direction measures, and
standard deviations of velocity fluctuation, every hour as
10-minute-averaged data. Turbulence data were
available for the 1990 campaign, from the Doppler sodar
at the AA station and from sonic anemometers at the U1
and U2 stations.
Fig. 2 shows diagrams of the occurrence frequency
for wind direction for AA at 100 m height AGL and for
W5 at 5.5 m AGL.

Overlapping
method
under
the
corresponding
meteorological condition to the field experiment at Mt.
Tsukuba under 3 kinds of wind fluctuation variannce;
=4.5, 9.0 and 12.0 deg. (Hayashi et al.
2001) .Correlation of these results in Table 1 indicate
that wind fluctuation variance of 9.0 deg seems to be
good.
Table 1 Correlation coefficient of concentration data
Wind fluctuation
Correaltion
Regression
4.5 deg
0.82
0.59
9.0 deg
0.90
0.89
12.0 deg
0.89
0.98

Fig 2. Occurrence frequency (%) of observed wind

directions at 100m AGL on AA (left) and 5.5m AGL on
W5 (right), for the periods 11/13 - 11/19 1989 (top)
and 11/11 - 11/16 1990 (bottom)
Tracer gas of SF6 was released at A and B point
from pressured vessel lifted by a balloon at 100m height
for 90 minutes, several times under neutral, stable and
unstable conditions. Gas concentration was sampled by
sampling bags at 63points during the last 30 minutes in
180 minutes of gas release time.

3 Wind tunnel experiments

Ide et al. of MHI (1994) developed a new method to
simulate a long time averaged concentration by rotating
a circular terrain model on a turn-table shown in Fig. 3
and 4, and named it Overlapping method. Their
rotating speed is in inverse proportional to the
occurrence probability of wind direction, and ground
level concentration is measured at more than 400
sampling points, simultaneoursly and continuously.
We conducted a validation experiment of the

Fig. 4 Schematic illustration of overlappking eqipment in

wind tunnel (top-view above and side-view below)

Fig. 3 Overlapping system to simulate a meandering effect on gas diffusion in wind tunnel

168

a) Conventional method (constant wind direction:=0.0)

b) Overlapping method (turn table method; =9.0)

Fig. 5 Ground level concentration distribution around Mt. Tsukuba without and with Overlapping
method, where broken curves represent field data (RUN 89-3) and solid ones wind tunnel
The results of conventional wind tunnel experiment
minutes sampling time, they are lower than they
and the Overlapping one were compared with field
value of =9.0 in correspondance with 30
data, as shown in Fig. 5. The Overlapping method
minutes sampling time.
simulates a wide lateral spread caused by
Fig. 7 Lateral plume spreads of wind tunnel
10000.00
`
a
b
c
d
e

1000.00

meandering effect for 30 minutes.

Axial ground-level concentrations were compared
with field data in Fig.6. It was found from Fig. 6 that
the
conventional
method
overestimated
the
concentration due to the underestimation of lateral
plume spread, while the Overlapping method
reproduced well the field data. Standard deviation of
lateral wind fluctuation was 2 degree and 9 degree in
the wind tunnel with and without overlapping method,
respectively. It was confirmed from Fig. 6 and 7 that
axial concentration is in inverce propotion to lateral
wind spread y.

=
0deg

100.00

Field
Conventional W/T
Overlapping W/T

=
9.0deg

10.00
1000

10000

100000

@
w
i

Next, we examined terrain effect on axial ground

level concentration and lateral plume spread,as
shown in Figs. 8 and 9. There are some differences of
terrain effect between two conditions of wind direction
fluctuation for axial concnetration and lateral plume
spread.

Fig. 6 Axial ground-level concentrations

Because P-G curves of A to F in Fig. 7 were

obtained from field observed data for few

169

without wind direction fluctuation

100

Relative concentrationUC/Q (m -2)

withoutterrain
with terrain

3 2

60

1
`
5

However,.it may underestimate the axial

concnetration over complicated terrain to use the
same value of meadering factor () as PG formula ,
because PG formula was obtained from experimental
data over flat terrain.

10

10000

1
1
X(km )

10

Lateralplum e spread

0.1

with wind direction fluctuation

100

Relative concentration UC/Q ( 10-6)

withoutterrain
with terrain

1000

W ithoutterrain

100
W ithoutterrain

Overlapping case
10

Conventionalcase

10
100

1000

10000

Downwind disatance

Fig. 9 Lateral plume spread with and without

terrain model

1
0.1

1
X(km)

10

4 Meteorological model simulation

Fig. 8 Axil ground level concentration with and

without terrain model

RAMS version 5.05 was adopted to simulate

meteorological fields around Mt. Tsukuba, located in
the North of the Kanto region of Japan. To provide
high spatial resolution in the area including Mt.
Tsukuba in RAMS simulations, four nested grids were
used, with resolution of 16 km, 4 km, 1 km and 250 m
and domains of approximately 975 x 750, 245 x 245,
40 x 50 and 20 x 25 km, respectively, as shown in Fig.
10. A stretched vertical grid was used, starting from a
first layer about 50 m deep and with a stretch ratio of
1.2, with a maximum z of 1000m, up to about 15 km.
Klemp-Wilhelmson lateral boundary conditions were
chosen for the velocity component perpendicular to
the boundaries and a zero-gradient inflow and outflow
for the other variables. Landuse, vegetation and
topographical data from RAMS libraries were used for
the coarser grids 1 and 2, while on grids 3 and 4 a 50
m resolution topography dataset by the Geographical
Survey Institute of Japan was input. ECMWF
reanalysis data with horizontal resolution of 0.5
degree were used as initial input and for the nudging
procedure, applied at the lateral boundaries of the
coarse domain during the run.

It was found from Fig. 8 and 9 that the terrain

effect on concentration depends on a meandering
effect; without meandering effect by conventional
method, there is significant change of concentration
by terrain effect, while with meandering effect by
Overlapping method, there is no significant chenge.
Next, we compared the terrain effect on lateral
plume spread without and with meandering effect, in
Fig,. 10. It was also found that the terrain effect on
lateral plume spread depends on the meandering
factor, as well as concentration. Sometime,
concentration of wind tunnel for short sampling time
have been transformed into one for long sampling
time by multipling the meandering factor (), in the
same way as P-G formula for environmental
assessment.

PG (1hr )

D u PG (3 min)

170

1990/11/10 15:00-11/16 3:00 (Top)

360

Observation

W ind direction (deg.)

315

Sim ulation (RAM S)

270
225
180
135
90
45
0

12km

15

4km mesh

21

15

21

15

21

9 15
Time

21

1990/11/10 15:00-11/16 3:00 (W 3)

15

21

15

21

Observation

360

Sim ulation (RAM S)

W ind direction (deg.)

315
270
225
180
135
90
45
0

15

250m mesh

1km mesh

21

15

21

15

21

9 15
Time

21

15

21

15

21

Observation

1990/11/10 15:00-11/16 3:00 (W 5)

Sim ulation (RAM S)

360
W ind direction (deg.)

315

Fig. 10 Multi grid configulations of calculation

domain
HYPACT is a dispersion model aimed at simulating
the motion of atmospheric tracers, driven by the
atmospheric flow which, in this case, is simulated by
RAMS. The advantage of using HYPACT lays in its
hybrid Lagrangian-Eulerian approach. The tracer is
represented by Lagrangian particles near the source
region, where concentration gradients are large and
the atmospheric dispersion is not yet dictating the
broadening of the plume. At appropriate large
distances downwind, where the plume is well mixed
and broadly spread, an Eulerian treatment is adopted
to estimate the concentrations. Here, the gas
dispersion was simulated with the version 1.2 of
HYPACT with Lagrangian mode. At each time step,
corresponding to 30 seconds, 20 particles were
emitted from about a height of 100m, at AA or BB
station according to which experiment was simulated.
Concentration was computed by counting the number
of the particles within each grid cell at each output
time.
Fig.11 and Fig.12 show time series of wind
direction and wind speed at the top of Mt. Tsukuba,
W3, and W5, respectively. It is found that the
simulated results are in good agreement with the
observed ones.
Fig.13 shows time series of turbulent kinetic
energy at AA, U1, and U2. The simulated results were
improved by using a new closure model implemented
by Castelli(2006), and reproduced the observed ones
very well.

270
225
180
135
90
45
0
15

21

15

21

15

21

9 15
Time

21

15

21

15

21

Fig.11 Time series of wind direction at Top of

Mt.Tsukuba(top), W3(middle), and W5(bottom).
1990/11/10 15:00-11/16 3:00 (Top)

20
W ind speed (m /s)

Observation
Sim ulation (RAM S)

15
10
5
0
15

21

15

21

15

21

15

21

15

21

15

21

Tim e

1990/11/10 15:00-11/16 3:00 (W 3)

W ind speed (m /s)

20
Observation

15

Sim ulation (RAM S)

10

0
15

21

15

21

15

21

15

21

15

21

15

21

15

21

Tim e

1990/11/10 15:00-11/16 3:00 (W 5)

W ind speed (m /s)

20
Observation

15

Sim ulation (RAM S)

10

0
15

21

15

21

15

21

9
15
Tim e

21

15

21

Fig.12 Time series of wind speed at Top of

Mt.Tsukuba(top), W3(middle), and W5(bottom).

171

The results of the plume model show a different

behaviour than RAMS-HYPACT modelling system.
The maximum is found at about 1500 m from the
emission point and after the maximum, the values
keep being always higher than RAMS-HYPACT
simulated results and than observed data. This may
be due to not considering the meandering effect
affecting the wind direction during a sampling time of
30 minutes. Since the sampling time of the Plume
model using Pasquill chart of plume spreads is a few
minutes, the meandering effect is not taken into
account and the axial ground level concentrations are
overestimated.
Run 90-5, covering the period from 19:30 to 21:00
of 12th November in 1990, is characterized by a
stable stratification of Pasquill class F, the mean wind
velocity is nearly calm and it takes values less than 2
ms-1 at the foothill of Mt. Tsukuba. Wind direction is
not steady and shows a consistent fluctuation.

turbulentkinetic energy (m 2/s2)

1990/11/10 15:00-11/16 3:00 (AA)

14
12

Observation

10

Sim ulation (RAM S)

8
6
4
2
0
15

21

15

21

15

21

11/10

9 15
Tim e

21

15

21

15

21

11/16

turbulentkinetic energy (m 2/s2)

1990/11/10 15:00-11/16 3:00 (U1)

14
12

Observation

10

Sim ulation (RAM S)

8
6
4
2
0
15 21

15 21

15 21

11/10

9 15 21
Tim e

15 21

15 21

11/16

turbulentkinetic energy (m 2/s2)

1990/11/10 15:00-11/16 3:00 (U2)

14
12

Observation

10

Sim ulation (RAM S)

a) Field observation

6
4
2
0
15
11/10

21

15

21

15

21

9 15
Tim e

21

15

21

15

21

11/16

Fig.13 Time series of turbulent kinetic energy at

AA(top), U1(middle), and U2(bottom).
Gas concentrations simulated by HYPACT code
are compared with field data by contour maps, as
shown in Fig. 14 and Fig. 15. The release height of
the tracer gas was about 100m AGL and the release
period was 90 minutes. The release occurred at the
same location and for an equal period for both years,
1989 and 1990. Simulated results of gas
concentration were also averaged during the last 30
minutes of release period, in the same way as the
field observations.
The simulated results of axial ground level
concentration are compared with the observed data,
as shown in Fig. 16 and Fig. 17. The available data for
the measured concentrations are sparse, so that they
do not strictly indicate the axial ground level
concentration. Only their distance from the source
identifies the available observations, no coordinates
are specified and their horizontal topographical
location is provided only graphically. To avoid a nonprecise selection of the data along the real plume
axis, we decided to plot all the measured data. The
predicted data are instead defined on a regular grid,
and it is possible to identify the central axis of the
simulated plume.
To provide a comparison with the performance of
the methodology usually adopted in environmental
assessment, the results obtained from a plume model
are also shown in the same figures.
In the following, a discussion of the results for the
cases Run 89-2 and Run 90-5 is proposed.
Run 89-2, covering the period from 10:00 to 11:30
of 15th November in 1989, is characterized by nearly
neutral conditions, described by Pasquill stability class
C-D; the mean wind speed ranges between 2 and 4
ms-1 at the foothill of Mt. Tsukuba and the wind
direction is almost steady from NE.
Simulated results of axial ground level
concentration
show that
the maximum of
concentration is found at a distance of about 330 m
from the emission point. On the other hand, the first
observed data are available only at a distance of 1800
m.The simulated results give in general a good
agreement with measured data.

b) Simulation

Fig.14 Contours of concentration on the ground

(CASE: Run89-2)

172

model(Level 2.5) is the most popular turbulence

model in meteorological model, like the k-epsilon
turbulence
model
in
Computaitonal
Fluid
Dynamic(CFD) model. MY model usually assumes a
2D boundary layer approximation and neglects a
horizontal diffusion terms.

a) Field observation

a) Mellor-Yamada turbulent model(Level2.5)

1000
750

b) Simulation

500
25000

250
0
0

20000
5000

15000
10000

10000

15000

5000
20000 0

b) Castelli 3D turbulent model

1000
750

500
25000

250
0
0

Fig.15 Contours of concentration on the ground

(CASE: Run90-5)

20000
5000

15000
10000

10000

15000

RUN89-2

C/Q(*10-9)s/m 3

100000

Fig. 18 3D view of particle dispersion around Mt.

Tsukuba calculated by RAMS/HYPACT codes
(CASE: Run90-5)

1000
100
10
0

2
3
4
Downwind distance (km )

Fig.16 Relative axial ground level concentration

(CASE: Run89-2: Neutral stability)
RUN90-5

100000
C/Q(*10-9)s/m 3

5000
20000 0

Observation
Sim ulation(RAM S/HYPACT)
Plum e equation

10000

1000
100
10
0

2
3
4
Downwind distancek

Fig.17 Relative axial ground level concentration

(CASE: Run90-5: Stable stability)
It is found in Fig.17 that the simulated axial ground
level concentrations correspond with the observed
data. The concentration field calculated by the plume
model is shifted farther from the source, at about 4000
m and it is heavily inconsistent with the observed
data. This may indicate a limit of plume model under
low wind velocity and unsteady wind direction.
Next, we compared 3D view of particle dispersion
around Mt. Tsukuba for two kinds of turbulent models,
as shown in Fig. 18. Mellor-Yamada(MY) turbulence

173

dE
dt

w
wE
KE
P
wz
wz

dE
dt

w
wE
KE
P
wx
wx

b) Castelli 3D turbulence model

Observation
Sim ulation (RAM S/HYPACT)
Plum e equation

10000

We developed a new 3D turbulence model with

Castelli(2006) for total kinetic energy(E), as follows.
a) Mellor-Yamada 2D turbulence model (Level2.5)

The upper result of MY model shows relatively

narrower diffusivity than Castelli model, as shown in
Fig. 18. This seems to be due to a difference of
horizontal and vertical diffusion by each model. And
the plume axis of MY model flows behind Mt. Tsukuba,
while Catelli model flows straightly. This difference is
due to the horizontal wind patterns shown in Fig.19;
there exists low wind region around Mt. Tsukuba in
Fig. 19a) of MY model, which means strong stable
layer appears over the surface of the ground. This
may be due to the difference of vertical diffusivity by
both models

a) MY model
We are now developing a new assessment
system to estimate the pollutant concentration for one
year with RAMS and HYPACT codes, in stead of
conventional assessment scheme using the plume
model, as shown in Fig. 20.

Conventional
method

New method

Meteorological
observation for one
year

Meteorological
simulation for one
year

Gas diffusion
calculation by plume
model, discretely

Gas diffusion
simulation for one
year, continuously

b)Castelli model
Fig. 20 Flowchart of a conventional and a new
environmental assessments

Acknowledgements
Wind tunnel experiments were conducted under
the sponcership by Japan Atomic Energy Research
Institute (Japan Atomic Energy Agency at present).

References

Hayashi, T. et al., (1999a), Data of atmospheric

diffusion experiments (Tsukuba, 1989), JAERIData/Code 99-036 (in Japanese)
Hayashi, T. et al., (1999b), Data of atmospheric
diffusion experiments (Tsukuba, 1990), JAERIData/Code 99-037 (in Japanese)
Hayashi, T. et al., (2001), Effective stack heights
obtained from wind tunnel and atmospheric
diffusion experiments, JAERI-Tech 2001-034 (in
Japanese)
Ide, Y. et al. (1994), Development of overlapping
modeling for atmospheric diffusion, Atmospheric
Environment, Vol.28, NO.11
Kakuta, T. and Hayashi, T. (1986), Results of
atmospheric
diffusion
experiments
Vol.2;
TOKAI82/TOKAI83,
JAERI-M
86-097
(in
Japanese)
Trini Castelli S., T. Hara, R.Ohba and C.J.Y
Tremback, 2006. Validation studies of turbulence
closure schemes for high resolutions in
mesoscale meteorological models, Atmospheric
Environment, 40, pp.2510 -2523

Fig. 19 Horizontal wind vectors over ground surface

5 Conclusions

It was found from this study that

(1) Overlapping method can simulate the
meandering effect on gas diffusion around
complicated terrain.
(2) RAMS and HYPACT codes can simulate well
the actual diffusion experiment in Mt. Tsukuba with a
new 3D turbulent model.

174

A proposal for a new atmospheric boundary layer wind tunnel

in the Netherlands
Peter Builtjes

Han van Dop

TNO, Dep. of Air Quality and Climate

Apeldoorn, The Netherlands

Inst. For Marine and Atmospheric Sciences (IMAU),

Univ. Utrecht
Utrecht, The Netherlands

Bert Holtslag

Harm Jonker

Wageningen Univ., Meteorology and Air Quality

Wageningen, The Netherlands

Dep. of Multi-Scale Physics, Delft Univ.

Delft, The Netherlands

Abstract-In this paper a short description is given of

the plans to establish a new atmospheric boundary
layer wind tunnel in the Netherlands. The main
feature is that next to the normal neutral boundary
layer, also stable and convective boundary layers
will be simulated. The concept is to develop a multipurpose/modular wind tunnel, also suited to
investigate for example air-sea exchange processes.
Key words Wind tunnel development, stable
boundary layers, thermal stratification

Introduction
TNO in Apeldoorn has been active in the field of
wind tunnel research for more than three decades.
Research activities have been carried out in the field of
wind nuisance, atmospheric dispersion in urban areas,
wind loading on constructions, heavy gas dispersion,
chemical reactive plumes etc.
The most recent wind tunnel in Apeldoorn is,
however ,already more than 20 years old. Furthermore,
due to a reorganisation focussed on centralizing most
environmental research in the Netherlands in and
around Utrecht, the TNO Environment Institute will be
transferred from Apeldoorn to Utrecht. This has led to
the decision to investigate the possibility of building a
new wind tunnel. The concept is to develop a wind
tunnel which is suited for applied as well as for more
fundamental research. TNO, in cooperation and
discussions with universities and other research
institutes, has made a first out-line for such a state-ofthe-art wind tunnel. The main new aspect is the
possibility to generate also thermally stratified flows over
a heated or cooled surface.
The wind tunnel should cover the following areas of
research:
x Air Quality and Safety
x Climate Change
x Built Environment
x Miscellaneous
An analysis of the requirements for these different
areas has led to a first technical lay-out of the wind
tunnel.

175

1 Air quality and safety

At the moment,the major air quality concerns are the
exceedances throughout Europe of the limit values of
PM10/PM2.5 and NO2. Critical situations occur around
traffic lines, especiallly in urban areas. Although CFDapproaches have grown in strength and reliability over
the last decade, the use of the wind tunnel to investigate
dispersion in complex areas is indispensble. In general
these investigations are performed in neutral boundary
layers, but it is well-known that the highest
concentrations occur in stable conditions. Little is known
about the interaction between stable boundary layers
and the flow in urban areas.
Safety aspects of the dispersion of LNG/LPG after a
failure in for example pipelines or off-loading by ships
have been investigated in wind tunnels especially
between 1980-1990. Currently there is an increasing
demand for new research in this area, due to a stronger
drive for more independent fuel-delivery, and also in
relation to possible terroristic acts involving biologic,
toxic or explosive matter. This type of research requires
besides stable boundary layers, the possibility to
operate the wind tunnel at low windspeeds, yet in a
controlled/stable manner.

2 Climate Change
In particular, the description of stable boundary
layers in current numerical weather prediction models,
air quality models and climate models ,is one of the
weakest elements in these models. Although field and
laboratory experiments have been carried out, in fact
more in convective than in stable cases, there is a
general lack in detailed experimental data under
controlled conditions . In case such experimental data
would become available, this would enable the
validation of the current model descriptions more
accurately, including the validation of LES-models,
again especially under stable conditions, see Holtslag,
2006
Furthermore, the current descriptions are valid over
homogeneous surfaces, related to the grid-averaging in
the models. Sub-grid scale effects caused by
heterogeneously heated surfaces, and frictional
convergence driven by a change in roughness length,
are all aspects which could be studied in the new
windtunnel.
The impact of climate change on the processes
close to the surface is a general topic of research. This

includes vertical exchange between air and water/sea

(possible even including some wave-interaction as well
as biological active processes caused by plankton), and
earth-air and vegetation-air exchange processes
including dry deposition. Also the katabatic (dense) flow
over ice could be a topic of research, since it plays an
important role in the glacier energy budget, which is
closely related to the widespread retreat of land-ice.
This type of research requires the possibility of
stable
and
convective
boundary
layers
by
heating/cooling the floor and the installation of an airflow heating unit. Furthermore, the test section of the
wind tunnel floor should be made modular to enable the
insertion of floors with a water or vegetation surface.

x -The test section of the wind tunnel should be

modular,the wind tunnel floor should be
exchangeble by water/ earth/vegetation surfaces
Although the main focus is on an atmospheric
boundary layer wind tunnel, using the wind tunnel at full
scale, as a ventilator, with applications in for example
the sport-research, should be possible.
The wind tunnel should be fully contained in a
building. It might be worthwhile to make room in the
same building to carry out smaller experiments also (for
example a rotating water table, or a small facility for
experimental cloud research). In this manner, a
laboratory for Geophysical Environmental Flows could
develop.

3 Built Environment

6 Conclusions and discussion

The main aspects here are wind nuisance caused by

high-rise buildings and wind forces on constructions.
Also wind safety around airports, the layout of
exhaust pipes on ships and luxurious vessels, and the
location of helicopter desks on oil platforms fall into this
category of research.
This type of research in general does not require
stable and convective boundary layers, but can also be
investigated in neutral flows at reasonably high
windspeeds

4 Miscalleneous
The following items can be found in this category,
based on the projects carried out over the years:
x The layout of wind turbine farms, often off-shore,
and the investigations of wind turbine wakes
x Testing of instruments like sonic anemometers
x Force reduction for skaters and bikers, posture,
helmet, garments
In general, this type of research requires high wind
speeds in the tunnel.
Another research area might be the study of cloud
microphysical aspects. A pre-humidified air flow could
be led into a turbulent flow , further downwind seeded
with droplets of a particular size (say 1-10 Pm). The
basis research question is the interaction between
droplets and turbulence and the possible enhancement
of droplet collision coalescence.
Because the wind tunnel would be run in cooperation
with three universities, it will also be applied for
educational purposes.

A first set-up of a new atmospheric boundary layer

wind tunnel is given, in which also convective, and
especially stable boundary layers would be simulated.
A major decision still to be made is whether the wind
tunnel would be a closed wind tunnel (preference for the
time being), or an open one. It is interesting to note that
the Japanese thermally stratified wind tunnel is an open
wind tunnel (Ohya et al, 1996), Further investigations
and discussions are needed to reach a final decision.
In anticipation of a further management decision, a
more detailed technical set-up will be made.
Assuming that the final decision to build the wind tunnel
will be made by the end of 2007, the wind tunnel could
be operational by the beginning of 2009.

References
Holtslag, A.A.M. (2006) GEWEX Atmospheric
Boundary-layer Study (GABLS) on stable boundary
layers. BoundaryLayer Meteorology 118, 243-246
Ohya, Y et al (1996) A thermally stratified wind tunnel
for environmental flow studies.Atm. Env. 30, 16,
2881-2887

5 First main technical aspects

The above described research areas lead to the
following requirements:
x The test section should be 2 x 3 m2, with a length of
15-20 m
x The minimal operational velocity should be 0.3-0.5
m/s, the maximum velocity about 20 m/s
x The floor temperature for (part of) the test section
should be variable between about 10 C and 50 C
x Vertically a temparure gradient should be possible
of about dT/dz of 500-800 C/m over a small
vertical distance. The maximum temperature of the
air flow should be about 80-100 C, at a wind speed
of 0.5 m/s

176

Active driving of a multi-fan wind tunnel

Koji SASSA* and Hiromori MIYAGI**
*
**

Dept. of Applied Science, Kochi University

Dept. of Applied Physics, Miyazaki University

We need well-controlled high Reynolds number turbulence fields to investigate characteristics of atmospheric turbulence fields in which turbulence Reynolds numbers, R, may be
more than 1000. Some large wind tunnels can realize such high Reynolds number flow fields.
However, there are few ones can easily control turbulence characteristics.
The multi-fan wind tunnel in Miyazaki University as shown in Fig. 1 has 99 fans
controlled independently with each other by a computer. We can form various velocity
profiles and add various fluctuations in the test section. In the present work, we tried to realize
quasi-isotropic and homogeneous turbulence fields with large R.
The fans were driven by a single basic wave starting from the different phase with each
other. We checked two basic waves, one was transformed from the Karman type spectrum
and the other was a low-pass filtered random noise. Though each fan induce only streamwise
velocity fluctuation less than 4 Hz, the local shear between neighboring fans can generate
lateral velocity fluctuations. We measured two components of fluctuating velocity by using an
X-probe and hotwire anemometers. Measured velocities were sampled at 4 kHz by a 16bit
digital recording unit and stored in a computer.
The resultant turbulence fields were found to have large turbulence intensity more than
0.1 of mean velocity even near the end of test section. Their homogeneity and isotropy
became better as they flowed downstream, and reached the levels of conventional grid
turbulence. The energy in the lower wavenumber region was excited and the energy gap was
observed in the energy spectra in X < 3 m. But, the spectra in the downstream region of the
test section have clear inertial subrange more than 2 order of bandwidth as shown in Fig. 2.
The turbulence Reynolds number was R=880. It is still less than the value of atmospheric
turbulence, we expect to achieve R>1000 by improving the driving mode.

Fig. 1 Multi-fan wind tunnel in Miyazaki

University

Fig. 2 Energy spectra at X=10 m

177

Determination of Spatial Structure of Internal Gravity Wave

by Multi-channel Thermo-Anemometer Measurement
H. MAKITA

K. OHBA

Department of Mechanical Engineering,

Toyohashi University of Technology,
Toyohashi, Japan
makita@mech.tut.ac.jp

Department of Digital Engineering,

Numazu College of Technology,
Numazu, Japan
ooba@numazu-ct.ac.jp

Abstract - The present study aims to clarify the

spatial structure of an internal gravity wave
developing in a strongly stably-stratified mixing
layer and its effects on the mechanism of heat and
momentum transfer in it. Multi-point measurement
was conducted on temperature and velocity
fluctuations, using a ladder probe composed of 7
cold- and hot-wires. The resultant instantaneous
isothermal lines illustrated the process of
streamwise evolution of the internal gravity wave.
The vertical scale of the internal gravity wave
evolved downstream. At the same time, spatial
correlation
was
strengthened
between
the
temperature fluctuations at different points. In the
downstream region, the internal gravity wave began
to collapse and the smooth profile of the stablystratified mixing layer was locally destructed around
its wave crests. Then, random components were
produced through the nonlinear wave-wave
interaction.
Key words Stratified Flow, Internal Gravity Wave,
Spatial Measurement, Negative Turbulence Production.

Introduction
Generally, the occurrence of internal gravity wave
keenly affects heat and momentum transfer and
turbulence production in natural strongly stably-stratified
flow fields [1],[2]. Buoyancy force, originating in the large
negative density gradient of the stably-stratified flow with
local Richardson number Ri0.25, induces internal
gravity waves in it. Its transition makes quite important
roles in the thermo-fluid dynamics of the transfer
problems. In our previous studies, we analyzed the
transition process of the spontaneously generated
internal gravity waves and their effects on heat and
momentum transfer in the same experimental situation
by thermo-anemometer single-point measurements [3][5]. The instantaneous information of local temperature
gradient is also important to understand the occurrence
of turbulence production and counter-gradient heat flux
in the downstream collapse process of the internal
gravity wave.
In the present experiment, a strongly stably-stratified
mixing layer was realized by giving a large positive
temperature gradient in a low turbulence wind tunnel
equipped with a thermal stratification generator. Multipoint simultaneous measurement on temperature and
velocity fluctuations was made using a ladder of 7 coldand hot-wires.

1 EXPERIMENTAL SET-UP AND

MEASUREMENT SYSTEM
The wind tunnel has a test section of 0.420.42m2 in
cross section and 8.0m in length [3]-[5]. In order to
minimize background turbulence, a thermal stratification
generator composed of a ladder of sixty coil heaters was

179

installed upstream of a contraction nozzle of the lowturbulence wind-tunnel. It can realize arbitrary
temperature profiles in the test section by independently
regulating the electrical power of each coil heater. A
temperature-controllable ceiling and adiabatic sidewalls
made of vacuum glass were employed for the test
section and thermal contamination disturbing inside
temperature profiles from the outside was almost
completely rejected.
Figure 1 illustrates the ladder probe used for the
spatial measurement. Each three set of I-I type probe is
arranged in the upper and the lower sides of the central
I-X type probe every 6mm. Each sensor is numbered
from Ch.1 to 7 from upper one as denoted in fig.1. They
gave the measuremental span of 36mm. Simultaneous
measurement must be made on the temperature
fluctuation component, , and the velocity fluctuation
components, u and w, in order to get higher order
correlation terms indispensable to minutely understand
the turbulent thermal transfer problems. We employed a
high precision thermo-anemometer system with S/N
ratio of 60dB and a frequency range of DC~5kHz. It is
equipped by a digital delay circuit which can equivalently
compensate the error due to the gap between the coldwire and the hot-wire, X=1mm, based on the frozen
pattern hypothesis, X=UT. Detailed analysis on the
characteristics of streamwise change in spatial structure
of the internal gravity wave and its effect on the heat
transfer and turbulent production are of the authors
special interest.

2 EXPERIMENTAL CONDITIONS
The initial flow conditions of the present experiment
are shown in table 1. A stable stepwise mixing layer with
20.0mm thickness was produced at the entrance of the
test section, X/D=0, where D=0.42m is the span of the
test section. Mean velocity was U0=3.0m/sec in the nonheated region and the maximum temperature difference
was max=26.8K. The maximum local temperature and
velocity gradients reached about 1200K/m and 9.7/sec
at the center of the mixing layer of X/D=0 as shown in fig.
2. The local Richardson number exceeded Ric=0.25 [6]
across the mixing layer where only the internal gravity
wave can be spontaneously generated in it.

3 EXPERIMENTAL RESULTS AND

CONSIDERATION
3.1 STREAMWISE CHANGE IN MEAN FLOW
Figure 2 shows the streamwise change in mean
temperature and velocity differences measured by the IX probe. The arrows indicate the height of the maximum
local temperature gradient and the hatched regions
show the range of spatial measurement conducted by
the ladder probe. The mixing layer with smooth and
quasi-stepwise temperature distribution has a maximum

Table 1 Mean flow characteristics at X/D=0.

'l=1mm

Mixing layer thickness d0 [mm]

20.0

Bulk Richardson number Ri

0.81

Ch.7

local temperature gradient of about (d/dZ)max=

1200K/m at X/D=0. It decreased as the thickness of the
mixing layer became gradually enlarged downstream by
the usual down-gradient thermal diffusion until X/D=5.
The profiles of the mixing layer are suddenly destructed
to have complex profiles at X/D=9 and the temperature
gradient decreased remarkably in the enclosed regions,
where the heat transfer is no more dominated by the
simple gradient-type diffusion.

d
NB
g
,
        (1)
dZ
where g is the gravitational acceleration and is the
coefficient of cubic expansion. Energy density level is
quite low and no peaks are clearly observed at X/D=0. It
remarkably increases downstream below the B-V
frequency. The temperature spectrum at X/D=5 clearly
shows three frequency components of 1.2, 1.8 and
3.0Hz satisfying the three-wave resonance condition,
f1+f2+f3=0 [7]. These facts show that the buoyancy
actively forces on the motion of fluid elements inside the
layer and the internal gravity wave was enhanced from
the smallest perturbations in the upstream region. At
X/D=9, the energy density level is observed to increase
drastically even in the frequency range above NB. It
suggests that the grown-up internal gravity wave begins
to collapse and the turbulence components were
produced through the nonlinear wave-wave interaction.

3.3 SPATIAL STRUCTURE OF INTERNAL

GRAVITY WAVE

Figures 4(a)-(c) illustrate the timewise change in

isothermal lines every 1K derived by interpolating the
seven temperature signals at X/D=0, 5 and 9. The
longitudinal axis indicates the vertical distance from the
position of Ch.4. At X/D=0, the isothermal lines become
dense around Ch.4, because the ladder probe was
located at the position of the maximum temperature
gradient. There, each isothermal line was slowly and
randomly fluctuating with so small amplitudes that we
cannot recognize any relevance between the isothermal
lines. This means that any coherent motion having a
vertical scale exceeding the sensor distance of 6.0mm
cannot be observed at X/D=0.
However, some periodic low-frequency fluctuations
synchronizing across the several isothermal lines
became observed intermittently at X/D=5. Their periods
seem about T=0.3-0.7sec, which agree with those of
the constitution components satisfying Thorpes threewave resonance condition [7]. The instantaneous
isothermal lines visually illustrate the aspects of
development of the internal gravity wave in the present

X/D=0

11

1200K/m

d0
Z/d0

-6

0
0
/max , U/Umax

Fig. 2 Streamwise evolution of mean temperature and

velocity distributions at U0=3.0m/s, max=26.8K.
:/max, :U/Umax
   : height of maximum temperature gradient
  : measurement region by ladder probe
10

3.6Hz
4.8Hz
5.4Hz
6.0Hz

10
X/D=11
Z/d0=1.72

10

-4

Figure 3 gives the streamwise evolution of

temperature spectra at the maximum temperature
gradient at each X/D. The Brunt-Visl frequency gives
the maximum frequency of internal gravity wave as
follows,

Prong (X type)
I=0.3mm

Fig. 1 Schematic figure of multi cold- and hot-wire probe.

E(f)/max [1/Hz] (2.010 /div)

3.2 STREAMWISE EVOLUTION OF

TEMPERATURE SPECTRA

Hot-wire
I=2.5mm
l=1.0mm

Cold-wire
I=2.5mm
l=1.0mm

0.21
1200

Maximum velocity difference Umax [m/s]

Maximum temperature gradient d/dZ [K/m]

Flow
Ch.4

26.8

Maximum temperature difference max [K]

Prong (I type)
I=0.2mm

6mm

Ch.1

3.0

Mean velocity (unheated region) U0 [m/s]

1.8Hz

1.2Hz

X/D=9
Z/d0=1.51

10

3.0Hz

X/D=5
Z/d0=1.10

X/D=0
Z/d0=1.00

NB=6.3

10

20

Frequency [Hz]

Fig. 3 Streamwise evolution of temperature spectra.

     : Brunt-Visl frequency, NB.
thermal mixing layer. The existence of fluctuations inphased among several isothermal lines surely indicates
that the wave motion with strong vertical correlation
passed through the probe position and the almost flat
waveforms show the duration of no wave motion being
there.
Furthermore,
such
quasi-synchronized
fluctuations were clearly detected in the various areas
measured by the ladder probe. With these results, the
vertical scale of the internal gravity waves was
approximately from 20 to 30mm. Here, the internal
gravity wave is known to be generated unsteadily
somewhere in the mixing layer.

180

(a) X/D=0, Z/d0=1.00

18

X/D=5, Z/d0=1.10

23

12

15

0
-6

-12
-18

Time [sec] (0.5/div)

(b) X/D=5, Z/d0=1.10

18

1
0

Ch.4-5

1
0

Ch.4-6

15

[K]

6
Z[mm]

Ch.4-3

12

-6

Time [sec] (0.5/div)

(c) X/D=9, Z/d0=1.51

18

Ch.4-7

1
0

NB=5.2Hz

Ch.4-3

1
0

Ch.4-5

1
0

Ch.4-6

1
0

Ch.4-7

1
0

Ch.4-1

Ch.4-2

NB=3.2Hz

20

40

Frequency [Hz]

Frequency [Hz]

14

Now, we discuss the spatial structure of the internal

gravity wave from the view point of coherence between
each channel and the center probe. The coherence is
defined as follows.
1/2

[K]

6
Z[mm]

40

3.4 STREAMWISE CHANGE IN COHERENCE

12

10

0
-6

-12
-18

20

X/D=9, Z/d0=1.51

Fig. 5 Coherence of temperature fluctuations between

channel 4 and others at X/D=5 and 9.
     : Brunt-Visl frequency, NB.

-12
-18

Coherence [-]

Ch.4-2

[K]

Z[mm]

Ch.4-1

5
Time [sec] (0.5/div)

Fig. 4 Comparison of instantaneous isothermal lines

at X/D=0, 5 and 9.
At X/D=9, the isothermal lines were apparently
curved with increasing their amplitudes. Sometimes, the
wave front of the internal gravity wave became much
steeper. Intermittent fluctuations with the period of about
T=0.2-0.3sec became observed which correspond to
the higher harmonics of the internal gravity wave
generated by the nonlinear wave-wave interaction, as
observed in the temperature spectra in fig. 3. Random
high frequency components with smaller amplitudes
were also generated. It suggests that the internal gravity
wave begins to collapse through the nonlinear
interaction among the resonant components and their
higher harmonics. The amplitudes of the isothermal
lines change greatly, which suggests that
the
occurrence of the internal gravity wave was not a
steady-state event. The facts suggest that the spatial
structure of the internal gravity wave also has
unsteadiness in the collapse process.

181

Coh(f)

S (f) 2
4i

44 (f)Sii (f)

(2)

here, S44(f) and Sii(f) express the spectral

components of 4(t) and i(t). S4i(f) is a cross spectrum
of 4(t)i(t) in which the subscriptsdenote the channel
number. Figures 5 illustrate the coherences of Ch.4-1
and Ch.4-2 from the top. These results were obtained at
the same positions as denoted in fig. 4. The dashed
lines in the figures show NB at each cross section.
In X/D=5 where the internal gravity wave is
developing, the coherence of all channels approached 1
in the low frequency region below NB. The increase in
coherence means that the spatial structure of the
internal gravity wave developing in the mixing layer was
apparently detected by the ladder probe. This fact
shows that the internal gravity wave develops as the
large-scale wave motion accompanied by the strong
correlation in the vertical direction. At X/D=9, several
peaks appear for the frequencies higher than NB in the
temperature spectrum shown in fig. 3. There, the
internal gravity wave began to collapse through the
complicated nonlinear wave-wave interaction. Contrary,
the coherences of Ch.4-3 and Ch.4-5 almost
approached 1 for the frequency region more than 10Hz.
However, the coherences of Ch.4-1 and Ch.4-7 at each
end of the ladder probe fell to the lower levels from the
higher frequency region. Possibly, the sensor at the both
ends of the ladder probe captured the crests of the

(a) vertical heat flux

(b) turbulent intensities

6
5

Z/d0

X/D=0

Ch.1

Ch.1

Ch.7

Ch.7

X/D=0

9
Ch.1

Ch.1

Ch.1

Ch.7

Ch.7

Ch.7

9
Ch.1
Ch.7

Counter- Downgradient gradient

-6

-w /U0 max, d /dZ

0
: w/U0 (0.001/div)
: / max (0.004/div)

Fig. 6 Streamwise change in vertical distributions of (a) heat flux and temperature gradient
and (b) turbulent intensities of w and .
: vertical heat flux (110-5/div),
: temperature gradient (200[K/m]/div)
internal gravity waves in the process of turbulence
production during the collapse of the wave front.

3.5 VERTICAL HEAT FLUX AND TURBULENT

PRODUCTION

Figure 6(a) compares the distributions of the timeaveraged vertical heat flux, -w/U0max, and the local
temperature gradient, d/dZ. The intensities of the
vertical component of velocity fluctuation, w, and the
temperature fluctuation, , are shown in fig. 6 (b), in
which the arrows indicate the vertical position of each
sensor shown in the figs. 4 and 5. The positive and the
negative values of the heat flux correspond to the downand the counter-gradient heat transfer, respectively. The
initial turbulent intensity is so low that the heat transfer is
quite small at X/D=0 as shown in fig. 6 (a). The local
temperature gradient denoted by a solid line and the
vertical heat flux have similar one-peak distributions at
X/D=0. That is, the heat transfer is dominated by the
usual down-gradient mechanism proportional to the local
temperature gradient there.
At X/D=5, the heat flux and the local temperature
gradient became not to keep such similar profiles to
each other and the heat flux was apparently suppressed
in the region of Z/d0=0.7~1.6. In such a region, the
internal gravity wave is assured to grow up downstream,
as estimated from figs. 4 and 5. The heat flux across the
wave front was suppressed as the internal gravity wave
develops with strong correlation suggests that the
internal gravity wave exerts decisive influence on the
aspects of heat transfer in the stably stratified mixing
layer.
The counter-gradient heat flux is clearly observed in
the hatched region at Z/d0=0.6 and 2.4 of X/D=9, though
the local temperature gradient is still kept positive.
These regions correspond to the upper and the lower
sides of the ladder probe. There, the counter-gradient
heat flux apparently grew larger, as mentioned before.
These heights catched the wave crest of the internal
gravity wave when estimated by the isothermal lines and
the coherences. It must be noticed that the wcomponent of the velocity fluctuation was produced
where the counter-gradient heat flux occurred as shown
in fig. 6(b). With these results, the internal gravity wave
began to collapse from around the wave crests and the
counter-gradient heat flux induced the negative turbulent

production by converting large potential energy stored in

the internal gravity waves to the turbulent kinetic energy.
Conclusively, the spatial structure of the internal
gravity wave developing in a strong stably-stratified mixing
layer was experimentally analyzed. Buoyancy force keenly
affects the characteristic features of transport phenomena
occurring in the stable mixing layer, and generates a
strong counter-gradient heat flux. Negative turbulence
production is conducted through the nonlinear interaction
between the thermal and velocity fields.

Acknowledgments
This work was partially supported by the Japanese
Ministry
of
Education
through
grants-in-aid
(No.15560138).

References
[1] Turner J.S., (1973), Buoyancy Effects in Fluids,
Cambridge University Press.
[2] Lighthill J., (1978), Waves in Fluids, Cambridge
University Press.
[3] Makita H, Mori S, Yahagi A., (1994), Spontaneous
generation of internal gravity wave in a wind tunnel,
Stably Stratified Flows:Flow and Dispersion over
Topography, pp. 81-91.
[4] Makita H, Ohba K, Sekishita N., (2002),
Occurrence of internal gravity waves and countergradient heat flux in a strongly stably-stratified
mixing layer, Advances in Turbulence IX, pp. 605608.
[5] Makita H, Ohba K, Sekishita N., (2005),
Experimental Analysis on the Transition Process
of Internal Gravity Waves in a Strong StablyStratified Mixing Layer, Proceedings of iTi
Conference on Turbulence 2005, p. 89.
[6] Hazel P., (1972), Numerical studies of the stability
of inviscid stratified shear flows, Journal of Fluid
Mechanics, vol. 51, pp. 39-61.
[7] Stewart R.W., (1969). Turbulence and waves in a
stratified atmosphere, Radio Science, vol. 4, pp.
1269-1278.
[8] Thorpe S.A., (1966). On wave interactions in a
stratified fluid, Journal of Fluid Mechanics, vol. 24,
pp. 737-751.

182

POSTERS

Use of detection of coherent flow structures for better understanding of

3D flow fields in urban environment
Tams Rgert, Ph.D., Istvn Goricsn, Ph.D., Mrton Balcz, Kroly Czder, Prof.
Tams Lajos
Department of Fluid Mechanics
Budapest University of Technology and Economics
regert@ara.bme.hu

Abstract - The rapid development of flow

measurement (particularly PIV) and numerical
simulation results in a huge amount of information:
2D or 3D velocity, pressure and turbulent kinetic
energy distribution of high resolution. Particularly in
case of 3D distribution it is not at all easy to
understand the flow field. Without clear
understanding of the flow field characteristics it is
difficult to develop proposals addressing e.g.
improvement of a given air pollution situation. This
paper discusses the use of advanced flow field
structure
extraction
methods
enabling
a
comprehensive understanding of the flow field
structure presented on cases of parts of urban
environment like streets, square, intersection of
streets, as well as on simplified models of built
environment. The turbulent flow is measured or
modeled via RANS approach solved by means of
Finite Volume Method using the commercial
software FLUENT 6 or MISKAM 5. Post-processing
of the results is based on simultaneous observation
and analyses of streamlines, wall streak lines,
vortex cores, iso-surfaces of the second invariant of
the velocity gradient tensor (usually denoted by Q)
and iso-surfaces of the total pressure (in case of
CFD). Applying these methods, one can not only
show the structure of separated flow regions but
can determine which of these structures are
predominantly influential in terms of development of
aerodynamic
characteristics
and
transport
processes of the flow in urban areas. The gathered
additional information may extend the support of
city and traffic planning in making conceptual or
optimization decisions.
Key words
structures in flow

Post-processing,

CFD,

coherent-

Introduction
The dispersion and transport of pollutants in urban
environment is a task of high significance in fluid
dynamics research. The flow past buildings can be
characterized by very high Reynolds number. The
critical point of the investigation of this type of flow is the
lack of accurate reference data on the real phenomenon
due to the reliability problems of field tests (Schafer et
al. 2005). The first phenomenon that is usually, however
not always (see f.e.g. Uehara et al. 2000) neglected is
the thermal stratification. The constant density
investigations restrict only to cases when atmospheric
wind is dominant. Due to the very complex geometry of
a city, mostly wind tunnels are the most appropriate
tools for obtaining answers on the required pollution

185

related questions (Plate, E.J. (1982), VDI Guideline

3783 Part 12 (2000)).
The other, popular and very effective investigation tool is
Computational Fluid Dynamics (CFD) that gains more
and more significance on this field. Due to the high
Reynolds number the flow is usually modeled by the
concept of Reynolds-Averaged Navier-Stokes equations
that nessecitates the application of turbulence models to
compute the values of the Reynolds stresses (Franke et
al. 2004). Robust turbulence models, like the widely
used k-H model were developed for high Reynolds
number flows that fit well for various applications.
However, the regions of the flow field which are in the
vicinity of walls or in separation bubbles and vortices
can be characterized by conditions that are not properly
modeled by these turbulence models. Even more
problematic is the modeling of fine details like trees,
windows, balconies that influence the generation or
dissipation of turbulence and deflect the flow. These
problems might introduce discrepancies in the
characteristics of pollutant propagation, extension and
diameter of vortices, even regarding their existence at
all (Franke et al. 2004, Gromke and Ruck, 2007).
The following main area of the investigation is postprocessing of the results obtained either via
experiments, or via CFD. In most cases tracer gas is
introduced into the flow to play the role of pollution and
then the air is sampled at certain locations in the area of
the city and magnitude of local immission will be
determined. Although the resultant pollution load in the
locations of interest is known, the reason why the given
immission values were obtained, i.e. the flow structures
resulting the given transport remain unknown.
Several works are nowadays focusing on the
determination of the flow field both by means of
experiments (e.g. Dezs et al. 2003) and CFD (Li et al.
2006). In this way not only the resultant loads but the
whole flow mechanism will be available that can
establish the right method for controlling the transport
characteristics. This paper discusses post-processing
methods that are well established at other fields of fluid
dynamics but may be useful for the interpretation of very
complex three dimensional flow fields in urban canopy.
These methods can be applied for both steady and
unsteady computations, but in this paper only steady
computational results are shown.

1 Investigation of coherent structures

on the streets
It is well known from several works on the field of urban
flow modeling that vortices develop in street canyons,
influencing significantly both the dispersion of pollutants
and wind comfort conditions. The concept of coherent
structures has been proposed by Hussain 1986 to
describe the phenomenon of turbulence on a

deterministic but chaotical way besides the widely

applied stochastic, statistics-based approach. Nowadays
coherent structure approach is the most powerful tool for
turbulence research (Hussain 1986, Adrian 2007,
Mathur et al. 2007). It is known that this approach was
originally applied for unsteady flows as the definition of
coherent structures is expressed in terms of coherence
in time (Hussain 1986), not in space. The classical
coherent structures are vortices [Jeong and Hussain
1995, Hussain 1986) that play crucial role in the
mechanism of turbulence, but alternative proposals are
also present, like the most recently introduced concept
of repellor and attractor curves by Mathur et al. 2007.
In the present case authors adopt coherent structure
detection techniques from the turbulence research field
and apply them on large scale steady computational
results, but it can be also applied for results from
measurements. Here authors would like to note that the
flow structures extracted from a steady flow field by
means of coherent structure extraction techniques are
not coherent structures strictily in the original meaning.
This way regions characterized by higher vorticity as
opposed to deformation can be extracted. Similar
approach was applied by Kolar 2006 for jets in a cross
flow and by Lohsz et al. 2006 for the results of Large
Eddy Simulation over a bluff body.
In this paper authors apply techniques for extraction of
coherent structures on steady flow fields over a portion
of Budapest city centre and over a simplified matrix of
buildings. The flow field was obtained by numerical
simulation using software Miskam and Fluent 6. Beside
the well known but mostly confusing streamline patterns
the authors present here more interpretable skeletons of
complicated flow fields. The results of computations
were post-processed by software Tecplot 360.

1.1 Application of the second invariant of the

velocity gradient tensor (Q)

The skeleton builds up from the iso-surfaces of the

second invariant of the velocity gradient tensor which
property is usually denoted by Q (Jeong and Hussain
1995). The Q property is defined in Eq. (1) and
interpreted as follows: Q expresses the domination of
swirling flow as opposed to the deformation of the fluid
particles.

1
tr ( T )  tr (S S T )
2

1
 w iu j w j ui
2

(1)

where is the anti-symmetric part and S is the

symmetric part of the velocity gradient tensor, ui and uj
are the mean velocity components (i, j = 1, 2, 3; where 1
= x, 2 = y, 3 = z in traditional frame of reference
notation). As vorticity induces potential-like vortex flow in
the environment of the vortex filament due to the BiotSavart law, the Q quantity thus detects structures in the
flow field that are of high dynamical significance on their
environment. This can be expressed as follows: if these
structures are modified then large scale modification in
the surrounding area can be achieved.

1.2 Application of the total-pressure based

method

'p tot

r2

U u u rot u   Qrotrot (u ) dr

(2)

r1

where U[kg/m3] is the density of the air, Q[m2/s] is the

kinematic viscosity of the air. It can be shown that the
second term under the integral in Eq. (2) is negative in a
cross section of a practically occurring vortex, the value
of the difference in total pressure is increasing when the
value of the rotation increases, i.e. the total pressure is
decreasing at approaching the axis of the vortex. The
drawback of the total-pressure based coherent structure
extraction method is that its value decreases strongly
not only in vortices but also in shear layers and
boundary layers where the value of rotation is high but
also the deformation of fluid particles is significant.

1.3 Analysis based on wall streak-lines

Wall streak-lines show lines the tangent of which are the

shear force vectors. Separation and reattachment
regions can be detected as well, as regions
characterized by large shear stress that influences the
wind comfort conditions.

1.4 Vortex core analysis

Vortex cores can be extracted based on the theory of

critical points (Perry et al. 1990, Haimes et al. 1999).
This type of analysis gives directly the vortex filaments
in the given frame of reference but there is no
information on the extension and strength of vortices.
However, the results provide a kind or skeleton of the
flow field.

2 The investigated cases

The flow was modeled by Miskam and Fluent past two
geometries. The model for Miskam was a part of the
downtown of Budapest. This software simplify the real
geometry of the buildings significantly, i.e. shapes
different from rectangular block cannot be modeled.
Roofs and domes are thus missing from the buildings.
The models used for the Miskam and Fluent
investigations can be seen in Figure 1a and b
respectively.
The model where only Fluent was used for flow
simulation was a simplified test area consisting of simple
blocks arranged in a matrix with slightly randomly
varying distances between the blocks (Yee and Biltoft
(2004)). The reason why also Fluent simulations were
used here, is the possibility of extracting information on
the surfaces of the buildings, which is not possible in
Miskam, yet. The information on the wall surfaces are
the wall-shear vectors that make wall-streakline
visualization possible.
In case of the Miskam model the wind direction was
285 in the global orientation system when using the
map of Budapest. In this case the hills and mountains in
Buda are upstream from the investigated area, thus their
effects are included. In the Fluent model the matrix of
buildings are blown perpendicular to their longer side
faces.

In case of flow with rotating fluid particles the total

pressure is dependent on the integral value of the
rotation integrated along the radius of the vortex
(mathematically represented by the rotation or curl
operation) (Eq.(2)).

186

Convective and diffusion terms of the governing

equations were discretized by second order upwind
scheme for Fluent simulations and SIMPLE method was
used for pressure-velocity coupling. In case of Miskam
computations, however, the discretization was of first
order accurate for all terms of the equations.
The computational grid consisted of fully hexahedral
cells (non-equidistant Cartesian grid in case of MISKAM
simulations) for both geometries. In case of Miskam
simulations the buildings were simply blocked out from
the grid thus it was not body fitted type, while the mesh
for Fluent simulations was block structured and body
fitted. The number of cells for the Budapest city centre
(Miskam) model and for the simplified matrix of buildings
(Fluent) were 5million and 1.5million, respectively.
Reference velocity was 1m/s (FLUENT) and 10m/s
(MISKAM).

4 Interpretation of the flow field

4.1 Flow field represented by streamlines

b
Figure 1. Investigated models by Miskam (a) and Fluent
(b) software

3 Description of numerical modeling

The horizontal extension of the computational domains
can be seen in Figure 1. The vertical extension
corresponds to 300m height which is the assumed
thickness of the atmospheric boundary layer, the models
were constructed in the scale of 1:1. The brick shaped
domains are characterized by a slip boundary condition
on the top surface simulating the free stream off from
the boundary layer, a no-slip boundary condition on the
wall surfaces, buildings and streets. The sides of the
brick domain were of variable boundary conditions to
ensure adjusting the direction of wind. Thus the domain
had one or two neighboring surfaces of velocity inlet
boundary conditions where atmospheric boundary layer
flow was prescribed according to the VDI standards.
The opposite faces of the computational domain were
outlet surfaces where constant static pressure was
prescribed.
For all the cases steady state simulations were carried
out based on the Reynolds-Averaged Navier-Stokes
Equations using turbulence models to take the unsteady
effect of turbulence into account. For both software the
k-Hturbulence model was used which was found to be
appropriate for such large scale flow problems (Franke
et al. 2004, Li et al. 2006). Near-wall treatment was
solved by means of wall-function approach (with
standard logarithmic wall functions).
Throughout all computations constant density was
assumed, so no thermal stratification and thus no
buoyancy effects were modeled.

187

In the city model only the inner part is detailed and

resolved sufficiently thus that region is shown. Some
streamlines are shown only to provide ability to follow
them through the streets of the city. An overall view and
a close-up to the building of the Hungarian Academy of
Sciences are represented in Figure 2a and b. As it can
be immediately concluded, this way of representation
can be effectively used only if small portions of the flow
field are visualized. If streamlines were placed over the
whole region, no flow structures would be visible. The
other conclusion is the well known fact that all the
buildings are in interaction with each other and they
determine the flow field together, none of them can be
handled individually. This is illustrated by the present
example where a large vortex with vertical axis is formed
beside the entrance of the Hungarian Academy of
Sciences (Figure 2b) that involves the sample of the
local air quality. The air is transported upwards along the
vortex core and is lifted up into the faster upper wind.
Streamlines show that this sample of air moves into the
courts and gardens of the neighboring group of buildings
and more downstream it mixes into the longitudinal
street-canyon-like vortex along the street on the right
lower part of Figure 2a.
Those places where vortices are present might be
important from pollution transport aspects as vortices
preserve the same air sample for long time along its
whole extension. Such a previously mentioned vortex is
the street-canyon vortex. In Figure 3 the concentration
of pollutants released from the square and the main
road passing almost horizontally on the lower half of
Figure 2a can be seen. It can be observed that the isosurface of the concentration of pollutants is surrounded
by the streamlines indicating that the pollutants tend to
stay inside the vortex extending along the street. On the
upstream side of the square in the middle of Figure 3 a
backward facing step type separation bubble is formed
that contains the polluted air. It is known, however, that
the vortices forming in the streets are the elements of
major influence onto the pollutant propagation in urban
environments, differently from the classical pointsource models over free fields.

recirculation zones are forming. However, in case of a

city, an extremely broad size range of vortices are
present out of which the largest ones are the streetcanyon vortices, the diameter of which is extending the
width of the street. These vortices are usually coherent
in time and space and can be very well predicted for a
certain city and wind direction. The next known size
scale is present along the leading upper and side edges
of buildings due to the separation of local boundary
layer on the individual buildings. The diameters of these
vortices are in the order of 1/6th of the building height.
These small vortices can be well visualized by the isosurfaces of the second invariant of the velocity gradient
tensor, Q, see Figure 4.

Figure 2. An overall view (a) and the zoom of the region

of the Hungarian Academy of Sciences (b) showing
streamlines based on Miskam modeling

b
Figure 4. Iso-Q surfaces over the city model (a) and the
simplified matrix of buildings (b). Flow from the upper
left corner to the lower right for both figures.

Figure 3 Iso-surfaces of the concentration of pollutants

released from the main road and Erzsbet square

4.2 Dynamically significant structures, the isoQ surfaces

The region of a city is characterized obviously by bluff

bodies from fluid mechanics point of view. As the local
boundary layer separates from bluff bodies, vortices and

It can be seen in Figure 4 that the iso-Q surfaces that

enclose a certain region are located mainly along the
upstream upper leading edges of the buildings for both
test geometries. The vortices enclosed by iso-Q
surfaces are characterized by high rotation rate and are
mainly unstable (Jeong and Hussain 1995, Kolar 2006)
thus they are usually detaching from the leading edges
and shed into the flow. Thus their main role in the
mechanism of the flow is turbulent mixing, these
vortices are the start of the well known turbulent
cascade. As these vortices are responsible for the
diffusion-like mixing that governs the dispersion of
pollutants besides pure convection, the character of
their position and overall pattern might provide useful

188

information on the expectable spreading processes, and

as so, an effective tool for influencing it.
Nevertheless, it has to be noted that the iso-Q structures
presented here are deduced from a rather coarse
computational grid thus their location, size and shape
are strictly qualitative. For more accurate prediction of
these structures, much finer grids are needed than the
present ones.
In Figure 4b on the left hand side an iso-Q surface is
surrounded by streamlines that show the formation of a
dynamically significant vortex. In this vortex the
velocities are expectably high as rotation of fluid
particles is higher than their deformation velocity. Similar
high-Q vortices produce the lift for delta-wing airplanes.
Also vortex cores, represented by spheres can be seen
in Figure 4b. These vortex cores are shifted from the
centers of vortices mainly due to the coarse grid but
practically should be placed exactly in the axis of them.

high values are remain invisible. By applying the

methods shown above simultaneously, more detailed
information on the flow field can be gathered and the
main flow mechanisms can be understood.
Understanding the flow field might help in the
determination of influencing it effectively.

4.3 Wakes, iso-surfaces of total pressure

The separation bubbles that are well captured by

streamline visualization are mainly characterized by very
low velocity comparing to the free stream values. In
these dead regions vorticity is almost negligible thus
the Q property is not applicable to detect them. Mainly
due to losses, total pressure decrease inside them, thus
iso-surfaces of it can enclose recirculation regions. Total
pressure was obtained for the simplified matrix of
buildings and is represented in Figure 5.

Figure 6. Surface flow pattern in a portion of the

simplified building matrix

5 Conclusion
In this paper different flow visualization methods were
shown that form a useful tool for understanding the flow
field around complicated geometries, like urban
environments. For the investigation of the flow field two
city models were applied and the flow was determined
via numerical modeling. The model that represented the
inner city of Budapest was used for numerical
simulations by software Miskam. On this model the wall
surfaces did not contain any information thus
characteristics regarding to the volume of the flow field
were extracted only. To show additional information
originating from the wall surfaces flow patterns an
additional simplified city model was applied over which
the flow was simulated by software ANSYS-Fluent.
Surface flow patterns are in strong relation with the
discomfort parameters of the pedestrians.
The flow field was visualized by different properties
deduced from the variables.

Figure 5. Iso-total pressure surfaces around the

buildings for the simplified matrix arrangement
The iso-surfaces of total pressure enclose the whole
region behind the building models where recirculation
was present, as well, as regions which were also
detected by the Q property.

4.4 Wall streaklines

Traces of vortices and main flow directions can be

determined from the surface flow pattern which is
based on the shear forces exerted on the surfaces by
the flow. Such a surface flow pattern can be seen in
Figure 6.
In Figure 6 also vortex cores are represented that help
in understanding the flow pattern. The surface flow is in
relation with the local discomfort parameters as this
region of the flow field is in direct connection with the
pedestrians. From, for example, sand-erosion
experiments one can map the regions characterized by
high wall shear but the flow structures producing these

189

Acknowledgments
Authors of this paper wish to express their thanks for
NKFP 3A/088/2004 for providing the financial basis to
make these investigations possible.

References
Adrian, R. J. (2007) Hairpin vortex organization in wall
turbulence, Physics of Fluids, Vol. 19, Issue 4, pp.
041301-041301-16
Dezs-Weidinger, G., Stitou, A.; van Beeck, J.,
Riethmuller, M. L. (2003) Measurement of the
turbulent mass flux with PTV in a street canyon
Journal of Wind Engineering and Industrial
Aerodynamics, vol. 91, pp. 1117-1131
Franke, J., Hirsch, C., Jensen, A. G., Krs, H. W.,
Schatzmann, M., Westbury, P.S., Miles, S.D.,
Wisse,
J.A.,
Wright,
N.
G.
(2004)

Recommendations on the use of CFD in wind

engineering, Proceedings of the International
Conference on Urban Wind Engineering and
Building Aerodynamics, VKI, Belgium, May 5-7, pp.
C.1.1-C.1.11
Gromke, C., Ruck, B. (2007) Influence of trees on the
dispersion of pollutants in an urban street canyon-Experimental investigation of the flow and
concentration field Atmospheric Environment, 2007
, 41 , 3287-3302
Haimes, R., Kenwright, D. (1999) On the velocity
gradient tensor and fluid feature extraction, AIAA
paper No. 99-3288
Hussain, A.K.M.F. (1986) Coherent structures and
turbulence, Journal of Fluid Mechanics. Vol. 173,
pp. 303-356
Jeong, J., Hussain, F. (1995) On the identification of a
vortex, Journal of Fluid Mechanics, vol. 285, p. 6994
Kolar, V. (2006) On universal vortical features of jets in
crossflow, Proceedings of Physmod 2006
conference, pp. 8-9
Li, X., Liu, C., Leung, D. Y., Lam, K. (2006) Recent
progress in CFD modelling of wind field and
pollutant transport in street canyons, Atmospheric
Environment, vol. 40, pp. 5640-5658
Lohsz, M. M. Rambaud, P. Benocci, C. (2006) Flow
Features in a fully developed ribbed Duct Flow as a
Result of MILES,
Flow, Turbulence and
Combustion, 77:59-76
Mathur, M., Haller, G., Peacock, T., Ruppert-Felsot, J.
E., Swinney, H. L. (2007) Uncovering the
Lagrangian skeleton of turbulence, Phys. Rev. Lett.
98, 144502
Perry, A.E., Chong, M.E., Cantwell, B.J. (1990) A
general classification of three-dimensional flow
fields, Physics of Fluids A 2 (5), May, 765-777
Schafer, K.; Emeis, S.; Hoffmann, H.; Jahn, C., Muller,
W. J., Heits, B., Haase, D., Drunkenmolle, W. D.,
Bachlin, W., Schlunzen, K. H., Leitl, B., Paschek, F.
Schatzmann, M. (2005) Field measurements within
a quarter of a city including a street canyon to
produce a validation data set, International Journal
Of Environment And Pollution, vol. 25, pp. 201-216
Uehara, K., Murakami, S., Oikawa, S. Wakamatsu,
S.(2000) Wind tunnel experiments on how thermal
stratification affects flow in and above urban street
canyons Atmospheric Environment, vol. 34, pp.
1553-1562
Plate, E.J. (1982) Engineering Meteorology, ed. E.J.
Plate, Elsevier Scientific Publishing Company,
Amsterdam-Oxford-New York
VDI Guideline 3783 Part 12 (2000) Environmental
meteorology, Physical modelling of flow and
dispersion processes in the atmospheric boundary
layer, Application of wind tunnels, VDI/DINHandbuch Reinhaltung der Luft, Band 1b
Yee, E., Biltoft, C. A. (2004) Concentration fluctuation
measurements in a plume dispersing through a
regular array of obstacles, Boundary-Layer
Meteorology, Kluwer Academic Publ, vol. 111, pp.
363-415

190

Air-quality and spatial planning

Dr F.L.H. Vanweert, Ing. J.I.J.H. van Rooij

Cauberg-Huygen Raadgevende Ingenieurs BV

P.O. Box 480, 6200 AL MAASTRICHT, The Netherlands
E-mail: f.vanweert@chri.nl; j.vanrooij@chri.nl

Abstract - Building in city centres is often made

difficult by the high concentrations of NO2 and fine
dust (PM10). Projects resulting in extra traffic cannot
be implemented in areas where the limit values of
the 2005 Dutch National Air Quality Decree [Besluit
luchtkwaliteit 2005] are exceeded. Nevertheless, it
can be demonstrated for many such projects that
constructing new buildings may have a positive
effect on air quality. Using specialised calculation
techniques, it is possible to determine that new
structures may influence the spread of air-polluting
substances in such a way that the air quality is
improved by the construction of the building. In the
case of an optimal urban substantiation of the plan,
the improvement of the air quality exceeds the air
quality deterioration resulting from increased traffic.
Using practical examples, this form of balance is
explained. In addition, the urban plans having a
positive effect on the spread of air-polluting
substances are presented.

1 Spatial planning in the Netherlands

In the Netherlands, the European air quality
standards have been incorporated into the Air Quality
Decree 2005 (Besluit Luchtkwaliteit 2005) and a number
of related laws and regulations. Plans for developing
housing, offices, industrial estates and roads can only
proceed if the concentrations of air pollutants are below
the limit values set out in the Air Quality Decree 2005.
These limit values are tested in 3 stages:
if the limit values are respected on completion
of the plan, the plan can proceed even if air
quality deteriorates as a result of the plan.
if the limit values are exceeded on completion
of the plan, the plan must not result in a
deterioration of the air quality.
if the limit values are exceeded and the
concentrations of air pollutants increase as a
result of the plan, the plan may only proceed if
the air quality improves elsewhere. This
improvement must compensate for the harmful
consequences of the plan. This is called a
balance approach.
This approach to the issue of air quality has a major
impact on spatial planning and economic development
in the Netherlands. In many construction projects and
industrial developments, extensive air-quality surveys
must be carried out to show that the limit values are not
exceeded, that the air quality is not deteriorating or that
measures are being taken (elsewhere) to improve air
quality in accordance with the balance approach. The
demand for detailed air-quality surveys has produced
computational methods including CFD calculations
(Computational Fluid Dynamics).

191

2 Air-quality survey with CFD

techniques
Almost all spatial planning and industrial
developments result in more traffic and increased
emissions of air pollutants. Under the Dutch system, no
further development is permitted on sites where the
concentrations of air pollutants are already high or the
limit values are being exceeded. However, as well as
attracting additional traffic, erecting a building in itself
also has an effect on the dispersion of air pollutants.
New buildings affect the wind profiles and can if
efficient urban development procedures are followed
bring about an improvement in the dispersion of air
pollutants. The siting of sound barriers alongside roads
also affects the dispersion of air pollutants from
vehicles. At Cauberg-Huygen Raadgevende Ingenieurs
BV a CFD system has been adjusted for the
Netherlands, in which both the negative effect of
attracting additional traffic and the (positive) effect of
new buildings and barriers on the concentrations of air
pollutants can be determined. This article explains the
CFD system in detail and provides a number of
examples to illustrate the practical applicability of the
technology.

3 Dispersion calculations using

WinMISKAM
MISKAM
[1]
(Mikroskaliges
Klimaund
Ausbreitungsmodell [Microscale Climate and Dispersion
Model]) uses CFD techniques for the microscale
prediction of the wind climate and associated emission
concentrations of air pollutants in built-up areas.
Numerical comparisons are made by the model to
determine the wind profiles and concentrations. These
comparisons are based on the laws of preservation of
mass, energy and momentum. WinMISKAM [2] is a
user-friendly user interface for the three-dimensional
flow and dispersion model, MISKAM. There follows
below a qualitative examination of the way in which a
number of basic physical principles have been
incorporated into the model.

3.1 Concentration calculations per wind direction

Just as with flow obstacles, WinMISKAM presents

emission sources as block structures: volume sources. If
the source to be studied has a certain spatial
extendedness, e.g. as is the case with a traffic route, the
emissions from the road as a whole are calculated first.
A spatial distribution of these emissions is then arranged
over several block structures or grid cells. Just as in the
case of flow obstacles, the grid resolution chosen by the
user for sources with a spatial extendedness determines
the level of detail that is used to model the sources.

By analogy with the wind climate, WinMISKAM is

used to calculate a balanced situation for each inflow
direction for the concentrations of air pollutants in the
model area: concentrations inside a grid cell change by
less than 0.1% compared with the concentration in the
same grid cell in a previous time step. This calculation is
based on the previously calculated wind climate (wind
velocity vectors) and information on the location and
strength of emission sources.

3.2 Limiting conditions

A number of limiting conditions must be taken into

account when determining the building and grid
configuration in order to guarantee the independence of
the model results from the height of the upper boundary
of the model and from the horizontal area size. Even the
"sub-scale" processes (processes which are completed
faster than the smallest integration step of the model or
which have a spatial extendedness which is smaller than
the finest grid resolution) produce a number of limiting
conditions. Below is a summary of the main limiting
conditions which must be taken into account by the user
in this context:
- the lowest height at which concentrations above
ground level can be reliably calculated is at least four
times the local roughness length
- a distance of two grid cells is recommended as the
minimum distance from the ground plane and
surrounding buildings to a point at which the wind
climate or the concentrations can be reliably viewed;
- survey points must not be situated in or adjacent to
grid cells which produce emissions;
- the highest relevant building in the model area must
not be higher than one third of the total model height;
- the total model height must be selected so that the
total frontal surface of flow obstacles is less than 10% of
the total inflow surface for each inflow direction
More general limiting conditions and computational
rules for producing reliable calculations have been set
out in VDI Guideline 3783, Part 9 "Environmental
meteorology - Prognostic microscale wind field models,
Evaluation for flow around buildings". MISKAM has been
designed in accordance with this guideline.

4 Computational parameters
The prevailing concentrations of the pollutants are
determined by calculating:
- the background concentrations: the contribution
from an unspecified origin;
- the contribution from local traffic.
In the Netherlands, the concentrations of NO2 and
PM10 are most likely to exceed the limit values. The
chemical conversion of NO to NO2 is taken into account
on the basis of an empirical formula. The number of
days on which the 24-hour mean limit for PM10 is
exceeded is determined on the basis of the annual
mean
concentration.
The
above
mentioned
computational parameters are discussed in detail below.

The OPS model is also used to forecast future

background concentrations[3].
At survey sites situated close to major sources, care
must be taken that the contribution from this source is
not double-counted: firstly as a contribution to the
background concentration and secondly as a separately
calculated contribution from the local source. A doublecounting adjustment must be used to prevent double
counting. For busy traffic routes, the double-counting
adjustment for NO2 is [4]:
'C a , jm >NO2 @ C 25 m >NO2 @

20  0,53C

and:
'C a , jm >NO2 @ C 25m >NO2 @

a , jm*

>NO2 @  0,82C25 m >NO2 @ 5000  x

100

20  0,53C

a , jm*

4500

>NO2 @  0,82C25m >NO2 @

100

,  x ! 500 m 

.x d 500m 

The adjustment for the other parameters is:

'C a , jm
and:

'C a , jm

p.C 25 m

5000  x
, ( x ! 500m)
4500

p.C 25m .( x d 500m)

where:
Ca, jm
: the adjustment for the background
concentration
C25m : the contribution from the local traffic route at
a distance of 25 m
p 0.08
Ca, jm*
:
the
unadjusted
background
concentration
x
: the distance from the local traffic route

4.2 The contribution from the local traffic route

The contribution from the local traffic route is

determined on the basis of emission indices established
for five speed classes per category of vehicle. The
following speed classes are used:
motorway: average driving speed is 100 km/hour;
country road: road with a maximum speed limit of 70
km/hour (average 44 km/hour);
city through traffic: through traffic inside the built-up
area, city street (average 26 km/hour);
normal city traffic: average speed 19 km/hour;
obstructed traffic: the flow of traffic is obstructed,
average 13 km/hour.
The emission indices for 2007 are shown in Table
3.1. Emission indices for 2010, 2015 in 2020 are also
available in the same way [5]:

4.1 The background concentrations

In the Netherlands, the background concentrations

are determined using a national model, the OPS model,
which contains all the relevant national and international
sources of air pollutants. The OPS model is used to
determine concentrations of the pollutants on a 5 x 5 km
grid. The results of these calculations are compared to
measurement data from a national measuring network
which consists of approximately 50 measuring stations.

192

Speed type
NOx

PM10

Private cars

A
0,448

0,367

0,527

0,589

E
0,648

Medium-size goods vehicles

5,916

7,447

9,129

9,783

12,952

Goods vehicles

8,143

10,986

13,192

14,050

18,600

Private cars

0,041

0,037

0,054

0,061

0,070

Medium-size goods vehicles

0,173

0,249

0,332

0,364

0,466

Goods vehicles

0,226

0,321

0,401

0,432

0,546

4.3 NO-NO2 conversion

Some NO is converted to NO2 due to the effects of
ozone. The NO2 concentration is established by
first determining the NOx concentration and then
deriving the NO2 concentration from it. The annual
mean concentration contribution of traffic to NO2 is
calculated using the following formula [4]:
Cb ,i >NO2 @

f NO 2 .Cb ,i >NOx @ 

B.C a ,i >O3 @.Cb,i >NOx @.(1  f NO2 )

Cb ,i >NOx @.(1  f NO2 )  K

where:
Cb,i [NO 2] :

annual mean concentration

contribution of traffic to NO2
concentration
from wind sector
i;
Cb,i [NO x] :
annual mean concentration
contribution of traffic to NOx
concentration
from wind sector
i;
Ca,i [O3]:
annual mean concentration of
ozone (O3) in the background
from wind sector i;
fNO2:
weighted fraction of directly
emitted NO2 : 0.05[-];
B, K:
empirically determined parameters
for converting NO to NO2: B= 1.0[-]
and K=100 [g/m3].
4.4 Number of days on which

PM10 is exceeded

In the Netherlands, extensive measurements have

established a correlation between the annual mean
PM10 concentration and the number of days per year
that the 24-hour mean limit values of 50 g/m3 are
exceeded. The number of days on which the 24-hour
mean concentration of suspended particles (PM10) is
higher than the limit value of 50 g/m3 is calculated on
the basis of the total annual mean concentration of
suspended particles (PM10). The formula used depends
on the level of the annual mean concentration of
suspended particles (PM10) [4]:
If Cjm [PM10] > 31,2 g/m3:
ODPM10 = 5.367.Cjm[PM10 132.4]
If 16 g/m3 Cjm [PM10] 31.2 g/m3:

193

2
ODPM10 = 0.10498.(Cjm[PM10 ]- 31.2) +3.1092.Cjm[PM10
]-31.2)+35
If Cjm [PM10] > 16 g/m3:
ODPM10 = 12

where:
Cjm[PM10]: annual mean concentration of suspended
particles (PM10), calculated using formula 1.1.
ODPM10: the number of days on which the 24-hour mean
concentration
of
PM10
exceeds
50 g/m3.

5 Practical examples
In recent years, the CFD method has been used in a
variety of air-quality surveys. CFD provides added value
compared with the traditional transfer calculations in
scenarios where:

the built-up areas alongside roads are not

continuous and vary in height;
buildings exhibit a screening effect which prevents
the dispersion of air pollutants;
turbulence around high-rise buildings exceeding 30
m in height has a major impact on the wind profiles in
the dispersion of air pollutants.
A practical example of this type of scenario is
discussed below.

5.1 Varying shapes of buildings

5.1.1
The plan

The plan involves renovating an inner-city area to

improve parking, traffic flow and the living environment.
The current scenario consists of approximately 600
above-ground parking spaces, about 20 housing units (2
x 2, built side-by-side) and a small apartment building.
The plan is adjacent to a road carrying a lot of through
traffic which is obstructed by a number of road junctions.
The limit values laid down in the Quality Decree 2005
are already exceeded in the current scenario.

The new plan provides for six apartment buildings

containing approximately 300 housing units. A car park
for over 600 cars will be provided underneath the
apartments. The car park will be mechanically
ventilated. All traffic to and from the car park will have to
travel along the already busy through road. A number of
junctions will be removed to improve traffic flow. The
terrain behind the apartments fulfils the function of
public parkland. The plan can only be put into effect if it
does not result in a deterioration of the air quality. The
impact of the additional traffic attracted, the changed
parking facilities and building volumes and the improved
traffic flow on the concentrations of air pollutants is
modelled on the basis of a CFD calculation.

5.1.2
The CFD model
A three-dimensional CFD model is created of both
the existing and the future scenario in the plan
area and its surroundings. Developments which
could be expected if the proposed plan was not
put into effect are extrapolated from the existing
scenario. We call this "autonomous development".
The CFD models of the autonomous development
and the scenario on completion of the plan are
shown in Figs. 4.1 and 4.2.

Scenario after completion of plan

Fig. 4.3: annual mean PM10 concentrations (g/m3)

The results of the survey show that the

concentrations of air pollutants do not increase despite
the traffic-attracting effect of the plan. The positive
impact of the improved traffic flow, the higher buildings
with openings that allow greater wind speeds on the
street side and the screening effect of the new building
is greater than the negative impact of the additional
emissions of air pollutants caused by the increase in
traffic.
This detailed approach adopted in the air-quality
survey has made it possible to show that the plan does
not have a negative impact on air quality and that the
plan can go ahead.

5.2 High-rise buildings

5.2.1
The plan

Fig 4.1 CFD model of the autonomous development

Fig 4.2 CFD model of the scenario after completion

of the plan

5.1.3

The results

The annual mean PM10 concentrations for

autonomous development and in the scenario after the
proposed plan has been put into effect have been
calculated for the year 2010. Fig. 4.3 shows the results
of the calculations.

In the centre of a major Dutch city a bank is planning

to renovate and expand its existing office. As part of the
plan, an office building 105 m in height is to be built,
providing 56,000 m2 of office space. The existing
parking capacity is also to be extended with an
underground car park accommodating 600 cars. The
additional traffic to and from the office complex will lead
to increased emissions of air pollutants in the local area.
The plan can only be put into effect if the limit values
laid down for concentrations of air pollutants are not
exceeded or if the plan does not result in a deterioration
of the air quality. A CFD model has been created
because the high-rise building also affects the
dispersion of air pollutants.

5.2.2
Autonomous development

The model

The CFD calculations take into account of the impact

of additional traffic and the impact of the high-rise
building. The concentrations of air pollutants are
determined for the autonomous development of the
surrounding area and in the scenario after the plan has
been put into effect. The CFD models of the

194

autonomous development and the scenario

completion of the plan are shown in Fig. 4.4.

on

Autonomous development

Autonomous development

Scenario after completion of plan

Fig 4.4: CFD model of the autonomous development
and after completion of the plan

The results

The annual mean NO2 concentrations are

determined for the year 2012 to ascertain whether limit
values will be exceeded and whether the plan will result
in a deterioration of the air quality. These concentrations
are determined for the autonomous development of the
surrounding area and in the scenario after the plan has
been put into effect. The results of the calculations are
shown in Fig. 4.5.

Scenario after completion of plan

Fig 4.5: annual mean NO2 concentrations for
autonomous development and after completion of the
plan
The CFD calculation clearly shows that the
construction of the high-rise building will lead to a
reduction in the concentration of air pollutants in the
local area despite the additional emissions of air
pollutants due to the additional traffic to and from the
new office building. The reduction in concentrations is
due to the increase in the wind speeds at living level as
a result of the high-rise building, which improves the
dispersion of air pollutants. Fig. 4.6 shows the change in
the wind profile and the increase in wind velocities
caused by the high-rise building.

195

6 References
[1] MISKAM, Handbuch zu Version 4, Institut fr Physik
der Atmosphre, Dr Joachim Eichhorn, Johannes
Gutenberg University Mainz
[2] WinMISKAMMiskam for Windows Manual,
Ingenieurbro Lohmeyer GmbH & Co. KG
[3]
Concentratiekaarten
voor
grootschalige
luchtverontreinigingen in Nederland rapportage
2006, MNP Bilthoven Nederland.
[4]
Meeten
rekenvoorschrift
bevoegdheden
luchtkwaliteit VROM Nederland
[5] Handleiding CAR II, versie 5.1, TNO ApeldoornNederland

Autonomous development

Scenario after completion of plan

Fig. 4.6: wind profiles in autonomous development
and after completion of the plan in a south-westerly wind

196

Characteristics of the low-speed wind tunnel of the LMF, France

L Perret

Laboratoire de Mcanique des Fluides, UMR CNRS 6598

Ecole Centrale de Nantes,
Nantes, France
laurent.perret@ec-nantes.fr
Abstract The aim of this poster contribution is to
present the new wind tunnel of the Laboratoire de
Mcanique des Fluides (Laboratory of Fluid
Mechanics) in Nantes, France, which is dedicated to
study interactions between the atmospheric
boundary layer and urban areas. This wind tunnel is
specially designed to allow the use of the Particle
Image Velocimetry which has now proved to be a
mature and reliable method to measure velocity
fields in turbulent flows. The use of the PIV is
expected to provide new insights on the structure
and dynamics of the flow within and above urban
canopies and valuable support to the development
and validation of numerical models, as it can
provide access to instantaneous bidimensional
maps, with two components, of the velocity field.
Key words PIV, urban areas, wind tunnel.

Introduction
Laboratory investigations of flows over urban- or
rural-type roughness above and within the canopy
region are still mainly limited to the evaluation of one- or
two-point statistics that are accessible via hot-wire or
Laser Doppler Velocimetry measurements. Even if ideal
to determine turbulent statistical properties, these
techniques can only provide limited information about
the detailed time-dependent structural characteristics of
the flow and its coherent structures whereas the
important role of these structures on the dynamics of the
flow, on the energy transfers and on the transport
mechanisms is well-established (Finnigan, 2000,
Jimenez, 2004, Zhu, 2007). As pointed out by Castro et
al. in their conclusion (Castro, 2006), the most adequate
experimental technique to investigate the spatiotemporal organization of the flow within and above urban
or vegetation canopies appears now to be the Particle
Image Velocimetry (PIV), which can give access to
instantaneous bidimensional velocity fields with up to
three velocity components. The extensive use of this
technique in the study of the structure of the smoothwall boundary layer (among a wide variety of turbulent
flows) has proved its reliability and efficiency (see for
instance the work of Stanislas and coworkers in France
or Adrian and coworkers in the USA). With its ability to
provide spatially dense measurements, the PIV has
enabled to draw a rather precise picture of the spatiotemporal structure of boundary layer based on the
model of clusters of hairpin vortices (Adrian, 2007).
Recently, PIV has been applied by Zhu et al. (Zhu,
2007) to the study of the flow structure and turbulence in
a wind-tunnel model canopy.
Furthermore,
because
of
its
principle
of
measurement (which will be detailed below) based on
the cross-correlation of small areas of two consecutive
images of the studied flow, velocity fields obtained by
PIV are very close in their nature to those computed by

197

Figure 1. Top: photograph of the new wind tunnel of

the LMF, Nantes, France; bottom: sketch of the wind
tunnel setting in the building
Large Eddy Simulation (LES). As a consequence, by
carefully choosing the spatial resolution or with the
adequate post-processing (spatial averaging for
instance), PIV measurements can be of great help in the
validation process of LES subgrid models. In particular,
in flows over inhomogeneous roughness representing a
specific urban area, measurements can be directly
compared to LES results, the mesh of which generally
including several buildings.
In order to conduct studies of the interaction between
the atmospheric boundary layer and urban areas, and
take advantage of all the ability of the PIV technique, the
team Dynamics of the inhabited atmosphere
(Dynamique de l'Atmosphre Habite) of the Laboratory
of Fluid Mechanics (Laboratoire de Mcanique des
Fluides, LMF) in Nantes, France, has chosen to build a
new wind tunnel specially designed to allow the use of
the Particle Image Velocimetry. In this poster
contribution, the characteristics of this wind tunnel are
first presented. Then, a brief presentation of the basic

principles of the PIV technique is given and the retained

PIV setup as well as its expected performances are
detailed.

Wind tunnel setup

Figure 1 shows a photograph and the setting of this

new wind tunnel. It has a total length of 37.5 m. The
retained configuration is the open-loop one. Before
reaching the test section, the air enters a contraction to
pass through a honeycomb and several wire meshes
designed to reduce the free stream turbulence (figure 2).
To allow the natural development of the boundary layer,
the test section is 27 m long. This channel presents a
square cross-section of 2x2 m. The vertical walls are
entirely made of glass. In the measurement section,
located in the most downstream part of the channel, the
roof of the wind tunnel is made of Plexiglas to provide
optimal optical access, which is necessary to perform
PIV measurements. This part of the roof is adjustable to
prevent the flow from blockage effects that can be
induced by the use of large models. Furthermore, the
measurement section (in red in figure 1, bottom) is
equipped with a moving floor which allows the rapid
change of urban area models. The end of the test
section is connected by a short contraction to a
centrifugal blower, which is driven by a 45 kW electric
motor. With this setup, mean flow velocities up to 10 m/s
can be reached. To simulate the atmospheric boundary
layer, the entire floor of the channel will be mounted with
roughness elements consisting of regular or random
arrays of cubes. Thanks to the three transparent walls,
numerous configurations of PIV measurements can be
achieved by simply varying the location of the camera
and the orientation of the laser sheet. Moreover, the
optical access from both sides of the wind tunnel will
allow in the future the use of a stereoscopic PIV system
which requires two cameras located on both sides of the
laser sheet.
Preliminary measurements of the longitudinal mean
velocity were performed with a Pitot-static tube 20,5 m
after the end of the contraction in order to check the
spanwise homogeneity of the flow before mounting any
roughness on the wind tunnel floor, and the possibility of
reaching the expected maximum velocity of 10 m/s.
Obtained distributions of the longitudinal mean
velocity at various heights and spanwise locations in the
test section of the wind tunnel, for four different rotation
speeds of the fan, are presented in figure 3.

Figure 2. Honeycomb at the entrance of the contraction.

Figure 3. Distribution of the longitudinal mean

velocity across the test section of the empty wind tunnel

198

0.75 m

110

85

1.30

1.00 m

150

115

1.80

1.50 m

220

170

2.60

Table 1. Typical size and resolution of vector fields obtained

with the present PIV setup

Figure 4. Experimental arrangement for particle image

velocimetry after Raffel et al. (2000)
The flow exhibits good spanwise homogeneity in the
entire cross-section, except near both vertical walls
where the effect of the boundary layers developing on
these walls can be seen. In effect, in the range of
investigated velocities, with a development length of
approximately 20 m, the thickness of boundary layers
over smooth walls are expected to be of the order of 30
cm, which is confirmed in the present case.
Furthermore, these measurements confirm that the
chosen setup enables to reach velocities up to 10 m/s.
Preliminary hot-wire measurements conducted at the
same longitudinal location, at 70 cm above the floor
(e.g. above the boundary layer) showed that the
turbulence level is comprised between 0.43 % and
0.51% for mean flow velocities ranging from 3 to 10 m/s.
Further measurements are required to fully
characterize the flow in the wind tunnel. In particular,
mean flow as well as Reynolds stresses will be
measured and spectral content of the velocity
components will be investigated in order to detect any
perturbation that could be induced by the fan for
instance.
This
characterization
also
includes
measurement of the longitudinal pressure gradient and
its variation in function of the position of the adjustable
roof. In a second step, once the flow over the smooth
wall fully investigated, roughness elements will be
mounted on the floor to model a neutrally stable
atmospheric boundary layer over an urban environment.

PIV principles and setup

In this section, the basic principles of the PIV

technique are presented and characteristics of the
retained PIV setup as well as its expected performances
are detailed.

2.1

PIV principle

Particle image velocimetry is an optical technique

that relies on the measurement of displacement during a
known short time interval of particles seeding the flow.
To this end, particles are generally illuminated by a laser
sheet at least twice and images of the scattered light are
recorded by a CCD camera. A typical PIV setup is
presented in figure 4. Even if some particle tracking
algorithms exist (which aim at following individually each
particle from one image to the other), displacement
information is generally extracted from the image pairs
by statistical approaches based on the cross-correlation
of a small area of the first exposure with a small area of
the second exposure.
Distance CCDLight sheet

Lx (mm)

Ly (mm)

Resolution
(mm)

199

Consequently, the obtained displacement can be seen

as the average displacement of the particles included in
the chosen interrogation area. Thus, one displacement
vector is obtained for each interrogation area. The
cross-correlation calculation is usually performed by
using Fourier transforms to ensure a minimal
computational cost. As recorded images are sampled on
a limited number of pixels, displacement is known with
an accuracy of 0.5 pixel. Sub-pixel displacement
evaluation schemes are then employed to reduce the
uncertainty. Once the displacement correctly evaluated,
its value is divided by the value of the time interval
separating the two exposures of a same pair in order to
provide an estimation of the corresponding velocity. As
a consequence, if the spatial resolution of the
measurement, which is directly given by the
magnification factor of the employed optical setup, is
greater than the smaller scales present in the flow, the
PIV algorithms act as a low-pass filter and then provide
velocity fields that are very close to those obtained by
large eddy simulation. It should be noted here that
stereoscopic PIV systems are now commercially
available. These setups, by employing a combination of
two CCD cameras and a laser, enable the
measurements of the three components of the velocity
field. Moreover, new lasers and cameras that permit
measurements at high frequencies (up to several
kilohertz) have been recently developed and give
directly access to the spatio-temporal evolution of high
Reynolds number flows. For further details on the
different methods and a more precise description of the
existing algorithms developed to improve the quality and
the accuracy of PIV measurements, the reader is
referred to the review of PIV techniques by Raffel et al.
(Raffel, 2000).

2.2

PIV setup

In the present work, the retained PIV setup

employed to perform acquisition of two-component
velocity fields is based on a Nd-YAG laser Quantel
(2x35 mJ), the wavelength of which is 532 nm, and a
Dantec HiSense Mk2 CCD camera of resolution
1344x1024 pixels providing images coded with 4096
gray levels. These two components are synchronized in
time and PC-controlled via the Dantec Flow-manager
software. The same software is used to perform the PIV
analysis of the recorded images. Acquisitions are
performed in double frame mode to allow crosscorrelation PIV processing in order to obtain accurate
velocity fields. The flow is seeded either by DEHS
particles generated via a Laskin nozzle fed with
pressurized air or particles generated by a commercially
available smoke generator. In both case, the diameter of
the generated particles presents a distribution with a
mean value close of 1 m. The spray generator is
located at the entrance of the contraction of the wind
tunnel. The camera is currently equipped with a 60mm
objective lens. With this setup, typical PIV processing
based on cross-correlation of 50% overlapping
interrogation areas of 32x32 pixels provides velocity
maps containing 84x64 vectors. The physical size and

resolution, which depend on the distance between the

camera and the light sheet, of the obtained velocity
maps are summarized in table 1.

Conclusions

The main characteristics of the new wind tunnel of

the LMF, Nantes, France, have been presented as well
as the very preliminary measurements performed to
assess the flow quality. Thanks to its design which
offers an optimal optical access, this wind tunnel
constitutes an ideal platform to perform studies based
on optical techniques of interactions between the
atmospheric boundary layer and urban areas. In
particular, the use of the PIV is expected to provide new
insights on the structure and dynamics of the flow within
and above urban canopies and valuable support to the
development and validation of numerical models.
Near future work will first deal with the full
investigation of the boundary layer flow developing over
the smooth floor. The next step will consist in designing
and mounting roughness elements on the wind tunnel
floor in order to create a neutrally stable atmospheric
boundary layer over an urban environment.

References
Adrian, R. J. (2007). "Hairpin vortex organization in wall
turbulence", Physics of Fluids, vol. 19, no 4.
Castro, I. P., Cheng, H. and Reynolds, R. (2006).
"Turbulence over urban-type roughness: deductions
from wind-tunnel measurements", Boundary-Layer
Meteorology, vol 115, pp. 109-131
Finnigan, J. (2000). "Turbulence in plant canopies",
Annual Review of Fluid Mechanics, vol. 32, pp. 519
- 571.
Jimenez, J. (2004) "Turbulent flows over rough walls",
Annual Review of Fluid Mechanics, vol. 36, pp. 173
- 196.
Raffel, M., Willert, C. E. and Kompenhans, J. (2000).
"Particle Image Velocimetry", Springer.
Zhu, W., van Hout R. and Katz J. (2007) "On the flow
structure and turbulence during sweep and ejection
events in a wind tunnel model canopy", Boundary
Layer Meteorology, published online.

200

Evaluation of Pollution Dispersion Prediction Using RANS with

Turbulence Models Available in Fluent 6.3
R. Izarra-Garcia, J. Franke, W. Frank

Department of Fluid- and Thermodynamics, FB11,

University of Siegen, Siegen, Germany
rafael.izarra@uni-siegen.de
Abstract - The numerical prediction of pollution
dispersion within a street canyon by means of
solution of the statistically steady Reynolds
Averaged Navier Stokes (RANS) equations is known
to be strongly dependent on the turbulence models.
In the case of pollution dispersion turbulence
models do not only have to be used for the
Reynolds stresses but also for the turbulent scalar
fluxes. While the influence of several turbulence
models for the Reynolds stresses on the dispersion
in 2D and 3D street canyons has been examined
already several times, the turbulent scalar fluxes
were exclusively modelled by the simple gradient
diffusion assumption. In the present work therefore
the influence of more advanced, ansiostropic
algebraic models for the turbulent scalar fluxes on
the dispersion inside a 2D street canyon is
analysed. The performance of the models is
evaluated with the measurements of Pavageau and
Schatzmann [1].

immediate result of this comparison was that both

k- models are unable to predict the flow field in
the street canyon and therefore lead to erroneous
concentration results. Concerning the performance
of the turbulence models for the scalar fluxes
preliminary results indicate that there is hardly any
improvement over the simple gradient diffusion
model.
Key words Street canyon, CFD, Turbulence models,
Pollutant dispersion, Scalar flux models.

References
[1] Pavageau M., Schatzmann M. (1999). Wind tunnel
measurements of concentration Fluctuations in an
urban street canyon, Atmospheric Environment,
33, 3961-3971.
[2] Fluent 6.3 Users Guide, (2006).
[3] Daly B. and Harlow F., (1970) Transport Equations
in Turbulence. The Physics of Fluids. Vol. 13,
Number 11.
[4] Kim J., Moin P., (1989) Transport of passive scalar
in a turbulent channel flows. Turbulent Shear
Flows, Vol.6 Springer, Berlin 85-96.
[5] Abe K., Suga K., (2001) Towards the development
of Reynolds-averaged algebraic turbulent scalarflux model. International Journal of Heat and Fluid
Flow, 22 19-29.

To that end three anisotropic algebraic flux models

were implemented in the commercial software
FLUENT 6.3 [2], comprising the generalized gradient
diffusion hypothesis (GGDH) model of Daly and
Harlow [3], the model of Kim and Moin [4] and the
model of Abe and Suga [5]. These three models and
the already available gradient diffusion model were
used together with seven different turbulence
models for the Reynolds stresses, comprising three
k-H models, two k- models, one second moment
closure model and the Spalart-Allmaras model. One

0.06

0.06

0.02
0
-0.06

Leeward
side
-0.04

0.04

Windward
side
-0.02

0.02

0.04

0.06

0.02

(b)

Canyon Symmetry Line

1
Leeward
side

0.8

0
-0.06

-0.04

Windward
side
-0.02

0.02

0.04

y/H

0.04

(a)

1.2

0.1
0.08

0.1
0.08

0.6
0.4

0.06

0.2
0
0

(c)
Exp

S-A

50

SKE

100
C*

KERE

RNG

150

KW

200

SST

LRR-IP

Streamlines details for (a) Realizable k-H and (b) standard k-. (c) Non-dimensional concentration C*=CHL/Qs in the
centre line of the canyon.

201

FLOW AND DISPERSION AROUND TALL BUILDINGS

Tom Lawton and Alan Robins
School of Engineering
University of Surrey
Guildford, UK
t.lawton@surrey.ac.uk, a.robins@surrey.ac.uk
assumes well-mixed near-wake conditions, which is
clearly not ideal for tall, thin buildings and will lead to a
decrease in the likely accuracy of model predictions in
such cases. Applications highlighted a further and more
important deficiency in the original algorithms that arose
because streamline deflections for a given building
orientation to the wind were scaled solely on HB, which is
clearly not an appropriate assumption when L/HB 1 and
W/HB 1.
Streamline deflections over an obstacle and its
recirculating near-wake are a function of wind direction,
being greatest for winds aligned with the roof diagonal
(due to the formation of roof vortices) and least for normal
orientations. Experimental work with cubes suggests that
in the 'strong' downwash case (orientation, = 45) the
effective height of a roof level emission is reduced almost
to zero, whereas there is little or no loss of height in the
'weak' case ( = 0).
Emissions or plumes above a certain height are not
affected by a building; this height is the upper boundary of
the flow region perturbed by the presence of the building.
Mean streamline paths, zSL, are defined in ADMS BUILD
in terms of the orientation, the shape of the recirculation
region boundary, zR(x), and the height of the building
effects region, zB, as:

Abstract - A detailed wind tunnel study of flow and

dispersion in the wake of tall, narrow buildings in a
simulated atmospheric boundary layer is described.
The objectives are: a) to develop better
understanding of the physical processes involved; b)
to support development of the ADMS buildings effect
model and place it on a firmer foundation when
applied in such circumstances (the results are
however equally relevant to other advanced
dispersion models, such as AERMOD-PRIME); c) to
provide high quality test cases for evaluating the
performance of advanced computational methods,
such as LES simulation.
Key words building effects, tall buildings, wind tunnel
studies

Introduction
That tall buildings are capable of dispersing roof level
emissions to the ground and ground level emissions to
roof height has been long known; Figure 1 illustrates such
behaviour for a building of rectangular plan and height 3.6
the characteristic cross-section side.
Clearly, the
assumption of well-mixed near-wake conditions isnt
satisfied here and a more complex balance prevails
between vertical dispersion within the near-wake and
local loss from it into the main wake, leading to a nonuniform vertical concentration distribution.
The ADMS building effects model (Robins et al, 2000;
2001) was developed from extensive wind tunnel
investigations of flow and dispersion around cubes and
cuboids, generally with either L/HB > 1 or W/HB > 1, where
L is the building length, W the width and HB the height.
(The same is broadly true of AERMOD-PRIME). ADMS

dzSL
w
dz z z (x) | |
=
= R B SL
dx
UH
dx zB zR (x) 4

(1)

with w the mean vertical velocity and the horizontal

velocity taken to be UH = U(HB), the roof level wind speed.
For passive releases, which are assumed to follow the

1.00
Height,
z/HB
0.75
Roof level
emissions
Roof source, Zs/Hb = 1; 0

0.50

Zs/Hb = 1; 34
Zs/Hb = 1; 90
Ground source, Zs = 0; 0

0.25

Zs = 0; 34

Ground level
emissions
0.00
0

10
Concentration,
CUHHB2/Q

15

20

Figure 1. Vertical profiles of non-dimensional concentrations on the centre-line, 0.1HB downstream of the rear face of a
tall building for a ground and roof level source; building dimensions: HB/3, HB/4.4, HB.

203

mean streamlines, the plume trajectory, zP(x), can be

calculated analytically as:

zB zP (x) zB zR (x) (4| |/ )

zB zP (x S ) zB zR (x S )

(2)

(3)

2 Results
The effect of a tall building in increasing ground level
concentrations downstream is demonstrated by Figure 2
for the HB/L = 3 building at orientation 45. The figure
shows data for four source heights from zS/HB = 1.02 to
1.5 compared with equivalent results in the undisturbed
flow. The recirculation region of this building reaches
about x = 2L (x/HB = 0.67) downstream and high
concentrations can be seen there for the two lower source
heights, zS/HB = 1.02 and 1.07; indeed the maximum
ground level concentration lies in the recirculation region
in the former case. The maximum ground level
concentration for the source at height 1.02HB is
approximately the same as would be produced by a
source at height 0.5HB in the undisturbed flow.
The figure once again shows that material released
above roof level can be brought rapidly to the surface in
the near-wake, but to a degree that falls rapidly with
increasing source height. The processes at work here are
downward plume deflection in the roof vortex system and
entrainment into the recirculating flow. Downward
deflection continues downstream because the wake
structure enhances downward dispersion, and wake
decay and the persistence of the roof vortices both create
down-flow. The net result is that ground level
concentrations are significantly raised above the values
that would exist in the absence of the building for all the
source heights considered.
The high spatial resolution of the laser anemometry
data has allowed mean streamlines to be calculated, as
illustrated in Figure 3, where the recirculation region and
above-roof streamline deflections are visualised.
The roof vortex system is clearly observed in the LDA
data, as Figure 4a illustrates. This shows secondary flow
vectors in the y-z plane at x/L = 1 from the centre of the
HB/L = 3 building at orientation 45. A well organised
secondary flow is clear, with downward flow above the
roof centre and upward flow in the near-wake
recirculation. Further downstream, at x/L = 2 (Figure 4b),
the secondary flow is basically inflow, with a substantial

(4)

that was used both to modify the rate of plume

downwash, (1) and the plume height range over which
downwash decreased to zero. Equation (1) was replaced
by:

dzSL
w
dz z' z (x) | |
=
= R B SL
dx
UH
dx z'B zR (x) 4

(5a)

when zSL z'B , otherwise:

dzSL
w
=
=0
dx
UH

(5b)

where zB is the height of the building effects region and

z'B the modified building effects height scale:

z'B = H B + (zB H B )

D = 1m
z0 = 1.5mm
u*/Uref = 0.055
Uref = 2.5ms-1

Three building models have been studied, one round

and two square in cross section, with diameter/side
dimensions of 100mm and heights of 300 and 400mm; i.e.
HB/L = HB/W = 3 and 4. Detailed velocity measurements
have been made around the 300mm buildings using a two
component laser Doppler anemometer. Horizontal,
passive emissions at and above roof level were used in
the dispersion studies. Emissions contained a small
amount of hydrocarbon trace gas that was detected by
flame ionisation based systems, one capable of
measuring mean concentrations at sixteen points
simultaneously and the other of measuring fluctuating
concentrations.
The velocity field results have been used to determine
the wake structure, in particular the extent of the
recirculation region, the nature of secondary flow fields
and mean streamline patterns above the roof and
downstream. Concentration measurements described the
variation in ground level and near-wake concentrations
with source height and building orientation and the full
concentration field (i.e. mean concentration, fluctuation
and intermittency) at selected downwind locations.

Relating the maximum mean streamline deflections

over the near-wake to the building height is clearly
inappropriate for tall, narrow buildings, where a zero
deflection limit is to be expected for all wind directions as
the building side to height ratio tends to zero. A
straightforward algorithm revision was introduced to
provide this sensitivity, whilst still providing the observed
level of downwash over cuboids. There was very little
experimental evidence on which to base the initial
revision, so although the desired sensitivity was attained
further refinement may still be warranted as sufficient data
become available.
The revision introduced a downwash scaling factor, :

H (L + W ) /2
B
= min1,

HB

boundary layer depth,

roughness length,
friction velocity,
reference speed,

where xS denotes the source location. The plume

trajectory is horizontal for normal alignment, = 0, but
follows the boundary of the recirculation region, to an
extent depending on source height and the magnitude of
, for oblique wind incidence. For non-passive releases
the following plume trajectory equation is solved:
PlumeRise
dzP dzSL dzP
=
+

dx
dx dx

atmospheric boundary layer at the model location, 14m

into the working section, are:

(6)

Reduced plume downwash implies reduced

entrainment into the near-wake as the approach of the
plume centre-line to the near-wake boundary is
decreased.

1 Methodology
Experiments were and are being carried out in the
20x3.5x1.5m EnFlo wind tunnel at the University of
Surrey. The characteristics of the simulated neutral

204

area of downward motion from HB/2 and above. The

vortex system is no longer apparent at this location,
masked in the downflow. Figure 5 shows the overall effect
on the path and dispersion of an emission from just above
roof level. A streamline drawn through this field from a
point just above roof level falls to a height of about 0.3HB
at about x = 4HB. (c.f. ~0.5HB for the round building, and
~0.6HB for the square section building at 0). Deflections
diminish rapidly with increasing release height.
1.25
zs/H = 1.02
no building
zs/H = 1.17
no building
zs/H = 1.33
no building
zs/H = 1.5
no building

Concentration
1.00
CUH2/Q
0.75
0.50
0.25
0.00
0.0

5.0

10.0x/H
Fetch

15.0

20.0

Figure 3. Streamlines released at x=100 in the wake of a

building H = 3L = 300mm at orientation 45 visualised with
OpenDX

400

400

300

300
Height, mm

Height, mm

Figure 2. Ground level centre line concentrations

downwind of a building of height H = 3L = 300mm at 45
orientation compared with results for the undisturbed flow.

200

100

100

0
-200

200

-100
0
100
Lateral position, mm

0
-200

200

-100
0
100
Lateral position, mm

200

Figure 4. Secondary flow velocity vectors in the wake of a building of height H = 3L = 300mm at orientation 45. A vector
length of 100mm is equivalent to 1ms-1. 4(a) left, at x = 100mm; 4(b) right, at x = 200mm from centre of building

500

Height, mm

400

1000
100
10

300
200
100
0
-100

100

300

500
700
Distance downstream, mm

900

1100

1300

Figure 5. Velocity vectors and non-dimensional concentration contours downstream of a building of height H = 3L =
300mm at orientation 45. A vector length of 100mm is equivalent to 1ms-1. The source height is 320mm.

205

These experiments will include PIV mappings and flow

visualisation studies, in addition to further high-resolution
LDA and concentration measurements.

3 Discussion
The results illustrate how combined velocity and
concentration field measurements reveal the complexity
of wake and dispersion processes and the role of specific
effects such as streamline deflection. They show that the
approach adopted in ADMS for treating dispersion near
tall, narrow buildings is not unreasonable, at least for
mean streamline deflections. Further experimental work
will address a wider range of geometries so that the
details of the model revisions can be made more assured.
Attention will also be focussed on behaviour in the nearwake.
It is quite straightforward to write a mass balance
model for the near-wake in which vertical transport within
the wake is included; e.g. the horizontally averaged
concentration, CA(z), at any level z is given by:

wA A

dCA
dwc
+A
+ sun c = 0
dz
dz

Robins, A G, McHugh, C A & Carruthers, D J (1997)

Testing and evaluating the ADMS building effects
module. Int. J. Environment and Pollution, 8, 3-6,
708-717
Robins, AG and Apsley, DD (2000) Modelling of building
effects in ADMS. ADMS Technical Specification
Paper ADMS3 P16/01N/00, CERC, Cambridge
Robins, AG and McHugh, C (2001) Development and
Evaluation of the ADMS Building Effects Module. Int.
J. Environment and Pollution, 16, 1-6, 161-174

(7)

where A is the cross-sectional area and s the perimeter

of the recirculation region, wA the mean vertical velocity
and wc the vertical turbulent flux averaged over A, and
unc the flux across the perimeter. The flux wc may be
modelled by a gradient transfer process as:

wc =

References

dCA
dx

(8)

with a diffusivity, and (7) solved with boundary

conditions CA = CR at z = HR (with HR the boundary of the
well mixed region), dCA/dz = 0 at z = 0.
As written, (7) includes vertical transport in the
recirculation region by both an area-averaged mean flow
and a turbulent flux. In practice, there are zones of
ascending mean flow and zones of descending mean flow
within the recirculation region, the distribution of
concentration in the horizontal is non-uniform, and the
structure is height dependent. These features make
estimation of the effective vertical mean velocity, wA,
almost impossible. With wA=0, equation (7) and (8)
reduce to:

d 2CA
= sun c
dx 2

(9)

The area and perimeter may be approximated by:

A = LRW B ; s = 2LR + W B

(10)

and the flux and diffusivity written in parametric form

as:

un c = U H CA ; = U H W B

(11)

which is then readily solved.

The near-wake
experiments will be used to examine the usefulness of
this approach.
Two or three characteristic dispersion problems for tall,
narrow buildings will then be chosen and subject to
detailed experimental study. The main aims of the further
phase of the research will be to develop deeper
understanding of the flow and dispersion processes
involved, and to provide high quality data that can be
used for evaluating the performance of advanced
computational methods, most likely LES simulations.

206

Quantifying the temporal representativeness of flow and dispersion

measurement in a complex urban area
Meike Rix, M. Schatzmann, B. Leitl
Meteorological Institute,
ZMAW, University of Hamburg,
Hamburg, Germany
meike.rix@zmaw.de, bernd.leitl@zmaw.de

Abstract The paper illustrates the temporal

representativeness of flow and dispersion
measurements within a complex urban
environment. Based on systematic wind
tunnel measurements within the central
business district of Oklahoma City, the
temporal uncertainty of locally measured wind
and concentration data was quantified.
Statistically representative long-term time
series
of
wind
velocity
and
tracer
concentration measurements were analyzed
for different averaging intervals and the
uncertainty of the ensemble averages was
calculated
for
typical
measurement
locations within the model area.

high turbulence intensity. Measurement locations within

street canyons are compared directly with results from
above roof stations, even though they must have a
different temporal representativeness due to the
difference in mean wind speed present at these
locations.
To
achieve
the
same
statistical
representativeness
of
flow
and
dispersion
measurements at different locations within an urban
area, significantly different averaging times would be
needed.
As varying conditions are often present in urban
areas, this paper illustrates the scatter in the temporal
representativeness
of
flow
and
dispersion
measurements within a complex urban geometry,
especially the quantification of errors due to differences
in the representativeness and also the determination of
appropriate averaging intervals for different locations
within a city.

Key words temporal representativeness, urban area,

flow and concentration measurements, wind tunnel

2 Experimental setup
The experiment to acquire the data used for the
analyses was conducted at the Environmental
Windtunnel Laboratory (EWTL) of the Meteorological
Institute at the University of Hamburg in the large
boundary layer wind tunnel WOTAN. The
measurements were carried out using a detailed model
of the Central Business District of Oklahoma City, USA.
The model had a scale of 1:300 and the modeled area
was about 1.6 km x 1.6 km.
During the experiment several hundred systematic
flow and dispersion experiments were conducted, for all
measurements statistically representative long-term
time series were recorded and later analyzed for
different length of averaging intervals. During the whole
experiment the wind direction was kept constant (
180). For the dispersion measurements the tracer g as
used was Ethan ( C 2 H 6 ).

1 Introduction
The understanding and the prediction of the
characteristic flow and dispersion processes within
complex urban areas is a critical issue, as around the
world more and more people tend to live within urban
areas. Especially in the case of an emergency, for
example the release of toxic gases, a high number of
people is influenced and without a well functioning
warning and forecasting system the casualties are
expected to be high. Therefore models to simulate and
forecast such scenarios have been developed. To test
the liability and accuracy of such models they are
validated using data from numerous field campaigns
that are often very time-consuming and expensive. The
field data is expected to represent the reality and the
results from field measurements are commonly used as
a reference for numerical modeling or physical
modeling in boundary layer wind tunnels.
In the context of using field data for model validation
and further analysis little attention is paid to the
temporal representativeness of the measurements. In
general the results are averaged over time intervals of
15 or 30 minutes. It is questionable however whether
such short time intervals are sufficiently long to assure
that the average is representative.
The fact, that the flow and turbulence structure plays
an important role in determining the temporal
representativeness, is also often neglected. At
measurement locations with a uniform flow structure
averaging over a shorter interval may result in the same
statistical significance as the same parameter averaged
over a longer interval at a measurement location with

The measurement locations were placed at

characteristic locations within the area, within street
canyons, on street crossings or on rooftops. For these
measurement locations a set of open questions was
investigated. To determine how the representativeness
in time changes locally measurements were carried out
in a grid of nine measurement locations. The distance
between probe positions was 1.5 m and 4.5 m,
respectively. Also the change in representativeness
with height was analyzed. The four heights at which
measurements were carried out were 3m or 6m, as the
lowest measuring position, half the building height, roof
level and twice the building height. The dispersion
measurements were especially analyzed with focus on
the distance of the measurement location to the source
of the tracer gas.

207

Some of the locations for the flow measurement

were placed according to measurement locations during
the Joint Urban 2003 field campaign and the results of
the field data analyses were compared with those of the
wind tunnel data analyses.

The statistical analysis was performed for the uand v-component of the wind, the turbulent fluxes u 'v'
and for the dimensional concentration c*.

4 Results

3 Data analysis

4.1 Flow measurements

Two different aspects were evaluated in terms of the

flow measurements, first the change in temporal
representativeness with changing averaging time and
second the differences in temporal representativeness
for different measurement locations when the same
averaging time was used.
In all cases the extension of the averaging interval
leads to a reduction in scatter of the short interval
values. A typical example of the reduction in scatter of
the interval values with extension of averaging time can
be seen in Figure 1. The reduction in scatter of the
average wind speed was evaluated for an extension of
the averaging interval from 7.5 to 30 minutes as the
ratio of the variation range for the 7.5-minute interval
and the variation range of the 30-minute interval.
For 80 % of the analyzed measurements the ratio
Vr30 min adopted values between 0.3 and 0.5 for the
Vr7.5 min
average wind speed. For the measurement shown in
Figure 1 this ratio is 0.3, which indicates that this is one
of the cases where the scatter is strongly reduced.
These values showed no dependence on height or
measurement location and were similar for both
analyzed wind components. The reduction in scatter of
the other analyzed statistical moments showed a
significantly higher discrepancy in the ratio of the 7.5and 30-minute measurements and no general
statement about the reduction of scatter with increased
averaging time could be made.

During the measurement campaign long-term time

series of the wind components u and v were taken,
were u is the longitudinal and v the lateral wind
component. In case of the dispersion measurements
time series of the concentration were recorded. The
wind tunnel data was scaled with a height of 80 m full
scale and a wind speed of 5 m/s, which was a typical
wind speed during the field campaign. The
concentrations were converted into dimensionless
concentrations c*.

c* =

c u ref L2ref
Q

c := measured concentration [ppm]

u ref

:= reference wind velocity [m/s]

Lref

:= reference length scale [m]

Q := source strength [l/min]

For the statistical analysis of the experimental data,
first the time series was analyzed over the whole length
of about 24 hours full scale which appeared to be the
averaging interval necessary to obtain repeatable
values at all measurement positions. The 24h-averages
were later used as reference values for comparisons
with the results of the shorter averaging intervals. The
time series was then broken down into shorter
averaging intervals of 7.5, 15, 30 and 60 minutes.
The statistical parameters used for the analysis are
the probability distribution and the first four statistical
moments, average

1
a = ai
n

i = 1,,n

1
ai a
n 1

skewness

1 n
ai a
n i =1
Sk =
s3

1
ai a
n
Fl = i =1 4
s
n

-0.2

and flatness

7.5 min
15 min
30 min
60 min
Long-term average

0.2

v [m/s]

standard deviation

0.4

-0.4
0

1000

2000
dt [s]

3000

4000

Figure 1: Typical reduction of scatter of the short-term

averages of the mean wind speed with extension of
averaging time

The analysis of the change of the probability

distributions with averaging time indicated that for all
measured points a minimum averaging interval of 30
minutes was required to receive a reliable result. For
many measurement locations longer averaging intervals
were required to receive acceptable results. For
measurement
locations
where
the
probability
distribution of the average wind speed was close to a
normal distribution an averaging time of 30 minutes
proved to be long enough, an example is given in
Figure 2 that shows the reduction of scatter for an
extension of averaging time for an almost normal
distribution. In Figure 3 the probability distributions of
the short-term averaging intervals are plotted against

They were calculated over the whole time series

and for the shorter averaging intervals. From the results
of the short-term intervals for each measurement
location and time interval the maximum and minimum
values were identified and the variation range was
calculated.

Vr = a max a min

The variation range allows an estimate of the

maximum error occurring at that location for the
analyzed time interval.

208

time for the whole length of the time series to illustrate

shifts in the maximum probability or the distribution
shape with time. In both plots it can be seen that the
results of the averaging interval of 30 minutes produce
a reliable picture of the long-term distribution, even
though Figure 3 indicates that there are still shifts in the
maximum probability.
1

0.4

0.6

0.4

0
-3

-2

-1

v [m/s]

0.4

0.2

0.6

0.4

v [m/s]

-2

-1

v [m/s]

-2

-1

800

t [min]

1000

1200

-3

1400

200

400

600

800

t [min]

1000

v [m/s]

v [m/s]

0.95
0.9
0.85
0.8
0.75
0.7
0.65
0.6
0.55
0.5
0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05

-1

-2

-1

-2

200

400

600

800

t [min]

1000

1200

-1

1400

-3

v [m/s]

200

400

600

800

t [min]

1000

1200

0.4

0
-2

-1

v [m/s]

15 min
1

200

400

600

800

1000

t [min]

1200

-2

1400

200

400

600

800

1000

t [min]

1400

frequency [-]

60 min

0.95
0.9
0.85
0.8
0.75
0.7
0.65
0.6
0.55
0.5
0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05

-1

Figure 3: Short-term probability distribution for 30minute averages plotted against time for a normal
distribution (same as Figure 2)

1200

30 min

-1

-2

1400

-1

frequency [-]

60 min

0.6

-2

1400

30 min

-3

1200

v [m/s]

600

0.2

-1

-2

400

0.4

v [m/s]

v [m/s]

v [m/s]

-1

3600s

7.5 min

200

0.6

15 min

v [m/s]

7.5 min

-1

Figure 4: Typical reduction of scatter of the probability

distribution with an extension of averaging time for a
bimodal distribution

0
-2

0.8

0
-2

Figure 2: Typical reduction of scatter of the probability

distributions with an extension of averaging time for a
normal distribution

-3

v [m/s]

0.2

0
-3

0.4

1800s

0.2

-1

0.8

frequency [-]

frequency [-]

0.6

-2

3600s
0.8

0
-3

-1

1800s
0.8

0.6

0.2

0
-2

frequency [-]

v [m/s]

v [m/s]

0.4

v [m/s]

-1

0.6

200

400

600

800

1000

t [min]

1200

-2

1400

200

400

600

800

1000

t [min]

1200

1400

Figure 5: Short-term probability distribution for 30minute averages plotted against time for a bimodal
distribution (same as Figure 4)

However in complex urban environments most of

the
measurement
locations
show
probability
distributions that differ significantly from a normal
distribution. For probability distributions with very high
skewness or flatness and those with a bimodal
distribution even averaging over an interval of 60
minutes did not result in representative results. This is
illustrated in Figure 4 that shows the reduction of scatter
with extension of averaging time for a bimodal
distribution, Figure 5 shows again the probability
distributions of the short-term averaging intervals
plotted against time. Both plots indicate that for this
probability distribution the amount of scatter for an
averaging time of 30 minutes is very high, also for the
longer averaging interval of 60 minutes the scatter of
the short-term distributions is still significant. Figure 5
shows that for an averaging interval of 60 minutes the
maximum probability still shifts between the two peaks
of the distribution. How extreme the differences
between the short-term distributions can be is illustrated

360-390 Min

frequency [-]

0.8

0.6

0.4

0.2

0
-2

v [m/s]

frequency [-]

0.6

0.4

0.2

-1

0
-2

-1

v [m/s]

420-450 Min

0.8

450-480 Min

0.8

0.6

0.4

0.2

0
-2

390-420 Min

0.8

frequency [-]

-2

900s

0.8

0.2

0.2

0
-3

450s

0.8

frequency [-]

frequency [-]

frequency [-]

0.6

0.2

frequency [-]

900s

0.8

frequency [-]

450s

frequency [-]

0.8

in Figure 6 which shows the short-term distributions of

four directly succeeding 30-minute intervals at the same
measurement location. It is apparent that for this
measurement a 30 minute averaging time does not
produce a representative result.

0.6

0.4

0.2

-1

v [m/s]

0
-2

-1

v [m/s]

Figure 6: Probability distributions for succeeding 30minute averaging intervals for the v-component of the
wind in a street canyon

209

shows that at roof level for both wind components an

averaging time of 30 minutes produces results that still
show a significant amount of scatter.
Above roof level Above roof level the scatter for
both wind components reduces quickly and reaches
stable values for the variation range for a sufficient
height above the building. For the u-component these
values of the variation range are between 0.1 m/s and
0.2 m/s, for the v-component all values of the variation
range are < 0.1 m/s.
For the other statistical moments the results of the
analysis differed from those of the wind components.
The maximum variation range of the standard deviation
for a 30-minute averaging interval was 64% of the longterm average and therefore lower than for the mean
velocities. The standard deviation seems to react less
sensitive to different measurement locations. The
values for the skewness and flatness on the other hand
were significantly higher than for the mean velocities.
For the skewness the maximum value was 752 % of the
long-term average and for the flatness the maximum
was 12000 % of the long-term average. These two
statistical moments react extremely sensitive to
changes in averaging time and to different
measurement locations.

In the second part of the analysis of the flow

measurements the differences in temporal representativeness due to varying measurement locations was
investigated. The averaging interval chosen for this
analysis was always 30 minutes. The representativeness of the 30 minute-averages was judged with the
help of the variation range. A measurement with a
variation range < 0.1 m/s was considered highly
representative, because the instrumentation used in
field campaigns does have an accuracy of about +/0.05 m/s, measurements with a variation range < 0.2
m/s were considered sufficiently representative. Also for
each measurement the percentage of 30 minute interval
averages was determined which were within the range
of +/- 0.1 m/s. The results were grouped according to
the measurement location into four categories,
measurements in street canyons normal to the main
wind direction, on street crossings, on roof level and
above roof level.
Street canyons All street canyons analyzed were
oriented normal to the main wind direction, therefore the
v-component of the wind is the component along the
street canyon and the u-component is the one normal to
the street canyon.
For the u-component of the wind after an averaging
time of 30 minutes 98 % of the measurements showed
a variation range < 0.2 m/s which was considered
sufficiently representative. For an averaging interval of
15 minutes 80 % of the measurements lay within the
range that was considered sufficiently representative.
For the 30-minute average, the maximum variation
range was 0.22 m/s or 57 % of the long-term average,
which was calculated from the entire time series. The
scatter of the interval averages reduced with growing
distance from the crossings.
The interval averages of the v-component show a
quite distinct behavior. For this wind component less
then 20 % of the measurement averages reach a
variation range < 0.2 m/s, therefore for this wind
component significantly longer averaging times are
required to obtain results with high representativeness.
Here the maximum variation range was 0.56 m/s which
is 373 % of the long-term average. The scatter of the vcomponent decreased with increasing distance from the
crossings.
Street crossings The measurements on street
crossings were characterized by a high variability of the
representativeness within short distances. The values
of the variation range for the u- and v-component varied
from 0.1 m/s to 0.5 m/s within the same crossing. For
the u-component of the wind 10 % of the 30-minute
interval averages showed variation ranges < 0.2 m/s,
the maximum variation range was 0.54 m/s or 64 % of
the long-term average. For the v-component 45 %
showed a sufficient representativeness. The maximum
variation range had a value of 0.51 m/s which is 65 % of
the long-term average. No connection between the
distance of the measurement location from the buildings
and the representativeness was detected, as the
measurement locations with good representativeness
varied for different crossings.
Roof level - On roof level for the u-component 20 %
of the measurements showed variation ranges < 0.2
m/s using a 30-minute averaging interval. The highest
value in the variation range was 0.63 m/s or 788% of
the long-term average. For the v-component 35 % of
the short-term interval averages showed sufficient
representativeness for an averaging time of 30 minutes,
in this case the maximum variation range was 0.39 m/s
which equals 150 % of the long-term average. This

4.2 Dispersion measurements

For the dispersion measurements again two

different aspects were considered, first the change in
representativeness with changing averaging time and
second the differences in representativeness with
varying distance from the source for the same length of
the averaging interval.
Again an extension of averaging times lead to a
higher representativeness in all cases. To characterize
the reduction in scatter the ratio between the variation
range on the 7.5-minute average and the variation
range of the 30-minute average is used. Compared with
the flow measurements the dispersion measurements
show less differences between the different
measurement locations, for 97 % of the measurements
the ratio Vr30 min adopted values between 0.4 and 0.6.
Vr7.5 min
The second part of the analysis was concerned with
the differences in representativeness due to different
distances of the measurement locations from the
source. Therefore the results of the 30-minute averages
for different measurement locations were compared. In
general it was found that the representativeness of the
measurements increased with growing distance from
the source. This was especially true if the measurement
locations were located in a similar environment, for
example all within the same street canyon downwind of
the source. If the two measurements compared are set
up in a completely different environment the same
distance to the source can still result in differences in
the representativeness. In the immediate vicinity of the
source the 30-minute interval averages showed a
significant scatter of up to 265 % of the long-term
average. For measurement locations one or two blocks
downwind, the scatter of the 30-minute averages was
between 40 % and 60 % of the long-term average.

5 Conclusions
The study showed that the issue of temporal
representativeness of measurements within complex
urban areas should be treated with more care, because

210

differences of the temporal representativeness due to

varying averaging times and also due to different
measurement locations within an urban area can be
significant. Although due to the high complexity of the
model area it was not in all cases possible to establish
exact criteria to characterize and quantify the temporal
representativeness, the results provide an estimate on
how big the errors could be and on how long the
averaging intervals should ideally be. To further
investigate and understand the problem of differences
in temporal representativeness due to different
measurement locations a similar study within an
idealized urban setting of less complexity would be
useful. However in field campaigns it is usually not
feasible to use averaging intervals longer than 30
minutes as the synoptic conditions vary during longer
intervals. But results of such short-term measurements
should be treated with care especially when used for
model validation. Due to the lack of representativeness
the data might be highly uncertain which means a
repeat of the experiment under seemingly identical
conditions would deliver a different result. The degree
of uncertainty inherent to field data can not be
determined through field experiments alone. To get an
estimate of the uncertainty it appears to be advisable to
conduct field experiments in combination with wind
tunnel measurements. The common believe that the
uncertainty of field data can be assessed from the
accuracy of the instrumentation is surely a fiction.
Compared with the error due to a lack of
representativeness the instrumental error is usually
negligible.

211

How often is sufficient?

A program for the statistical analysis of puff dispersion in urban
environments
R. Fischer, M. Schatzmann, B. Leitl
Meteorological Institute,
University of Hamburg,
Hamburg, Germany
Email: Rasmus.Fischer@zmaw.de

Abstract The high variability inherent in the

results from urban puff dispersion experiments
requires measurement data to be processed and
analyzed in a statistical and probabalistic way.
Consequnetly, within the scope of a research
project focusing on validation data for emergency
response dispersion modeling a large number of
puff release experiments was carried out in the
Environmental Wind Tunnel Laboratory (EWTL) at
the University of Hamburg. The resulting database
is intended to be applied particilarly for the
validation of Large Eddy Simulation (LES)-based
models. In order to generate reliable reference
data from systematic wind tunnel experiments
efficiently a software tool was developed, which
is able to process long time series of puff
measurements
and
which
provides
a
comprehensive statistical analysis of puff
dispersion parameters.
Key words urban flow; urban dispersion; puff
dispersion; data analysis

Introduction
As a result of industrialisation, a large percentage
of the earths population lives in densely populated
cities. In a global view this process of urbanisation as
well as the industrialisation itself are still in progress.
As a result, the increasing likelihood of an accidental
or deliberate release of hazardous material in an
urban area is posing a threat to more and more
people. Consequently, reliable simulations of
accidental releases and the dispersion of puffs in
complex geometries attained a much higher
importance during the last years. A credible and
validated Computational Fluid Dynamics (CFD) based
plume prediction model of Contaminant Transport
(CT) in urban areas can be used in the licensing of
new industrial plants, in safety analysis studies for
accidental releases of hazardous materials in the
chemical industry, or in the context of a proper crisis
management after terrorist attacks in urban
environments.
The numerical simulations of cloud dispersion in
urban envirionments are very complex and they
require a comprehensive computational effort. In the
past it was common practice to use simpler models
as, e.g., analytical (Gaussian) models, diagnostic
models, based the mass conservation equation only
or CFD-models with full parameterization of
turbulence (RANS-models) for these complex tasks.
Such models are able to compute urban dispersion
within a reasonable time, but on the other hand, they

213

are unable to acquire the inherently unsteady plume

dynamics driven by urban geometry. The highest
accuracy of numerical simulations in computational
fluid dynamics is provided by direct numerical
simulations (DNS). These are simulations in which
the Navier-Stokes equations are numerically solved
without any turbulence model. On this account DNSmodels are able to compute transient flow dynamics.
But unfortunately, they are prohibitively expensive for
most practical flows at moderate-to-high Reynolds
numbers, and especially so for urban CT studies. An
effective intermediate between DNS and RANS is
provided by the Large Eddy Simulation (LES) (Sagaut
2004). LES-models only simulate the large eddies,
while the effect of the smaller eddies is still
parameterized. For calculating the structure of the
large eddies, the Navier-Stokes equations are filtered
by a low-pass filter without Reynolds-averaging. So
LES-models are capable of simulating flow features
that cannot be handled with RANS such as significant
flow unsteadyness and local vortex shedding. LES
provides higher accuracy than the industrial methods
at much lower costs than a direct numerical
simulation. The increasing computer power enabled
the use of LES models for urban dispersion
simulations. However, particularly for the validation of
numerical models simulating transient atmospheric
dispersion phenomena (e.g., puff dispersion) reliable
experimental data are still required.
Although there have been numerous field campaigns of atmospheric puff dispersion, many were
done 20 or more years ago. Because of the
instrumentation available by that time, often the
experiments were used to analyse puff dispersion
over time periods of an hour or more and puff travel
distances of up to hundreds of kilometres (e.g.,
Gifford, 1957, 1977, 1995; Islitzer and Slade, 1968;
Draxler, 1984). In contrast to studies dealing with
plume dispersion, puff dispersion studies are also
rarer in general. Cionco et al. (1999) described puff
measurements taken over complex terrain, and
Mikkelsen et al. (1987) analysed smoke release
experiments over Denmark regarding a relative
dispersion estimated from one dimensional velocity
spectra. Hanna and Franzese (2000) analysed the
results from one wind tunnel study and eleven field
campaigns to test predictions for along-wind
dispersion of puffs derived from similarity theory. Min
et al. (2002) presented results from observations of
puffs above the atmospheric boundary layer as part
of a model validation exercise for Vandenberg Air
Force Base. More recently, Doran et al. (2007)
analysed puff releases from the field campaign Joint
Urban 2003, conducted in Oklahoma City, with

respect to certain puff characteristics like arrival- and

decay time.
Code validation based on experimental data
requires well-characterized datasets with information
adequate for defining and evaluating unsteady
simulations as they are provided specifically by LES
models. Unfortunately, this information cannot fully be
derived from full-scale field campaigns like they have
been mentioned above. The number of field trials is
usually limited and the data acquired is typically too
sparse
and/or
insufficient
for
a
proper
characterization of the flow and dispersion
phenomena driving puff dispersion. For the validation
of urban type LES-models, there is an urgent need
for reference data comprising flow and turbulence
fields measured with high resolution in space and
time within and above typical urban structures. In
general, such datasets can be generated under
carefully controlled and well-documented boundary
conditions in boundary layer wind tunnel.

being trapped in circulation zones between building

structures.
A simplified model representation of the average
flow field on the street scale is usually described as a
turbulent shear flow above a rectangular cavity
whereas the mean flow is perpendicular to the axis of
the street canyon. Provided that the wind speed at
roof -level exceeds 1.5-2 m/s and the aspect ratio of
the canyon is near unity, a re-circulating flow driven
by momentum transport from the shear layer above is
set up in the canyon. Although this re-circulating flow
patterns were found in time averaged results from
laboratory experiments (Hosker 1985) and field
campaigns (Nakamura and Oke 1988), it is neither
representative for nor applicable to puff dispersion
phenomena which happen at time scales of several
minutes or an hour.
A mean flow pattern as
described does not really exist within a complex
urban geometry for evaluation periods less than
several hours.

1 Urban flow and dispersion

2 Experimental setup

Concerning the wind flow within and above urban

areas driving the dispersion of passive tracers or
pollutants, it is worthwhile to adress these issues by
considering the problem at different scales (Britter
and Hanna, 2003). Field and laboratory experiments
at each of these scales provide observations that are
interpreted in terms of various physical processes.
The mathematical models into which these processes
can be combined establish a hierarchy of complexity
and sophistication. Because each model has its own
regime of applicability and accuracy, it is common
practice to parameterize detailed interpreta-tions from
one scale, to enhance interpretation at the next larger
scale.
For the Consideration of urban flow and
dispersion, Britter and Hanna (2003) use spatial
scales, whereas the corresponding time scales can
be obtained by dividing the spatial scales by an
adequate advection velocity. The urban area is
subdivided into four length scales, a regional scale
(up to 100 or 200 km), the city scale (up to 10 or 20
km), the neighborhood scale (up to 1 or 2 km) and
the street scale (less than ~100 to 200 m). Because
this paper deals with puff releases near the ground,
only neighborhood and street scale will be considered
in detail.
Studies on dispersion at the neighborhood scale require detailed knowledge of the flow within the urban
canopy as well as directly above. Increased
turbulence within the urban canopy causes larger
dispersion co-efficients that tend to reduce
concentrations. However, the reduced mean
advection velocity within the canopy tends to increase
concentrations. So it depends on the proportion of
these contrary effects whether the con-centrations
increase or decrease with increasing site density.
Clear evidence of substantial reductions of groundlevel concentrations from ground-level sources was
given by the Kit Fox experiments (Hanna and Chang,
2001). By contrast, other small-scale field
experiments (Davidson et al. 1995, Macdonald et al.
1997) and related laboratory experiments (Davidson
et al. 1996, Hall et al. 1996) found only little effect
from obstacle arrays on the maximum ground-level
concentrations. Furthermore concentration time
series from urban field campaigns (e.g., Birmingham,
UK, Britter et al. 2002) provide evidence for pollutants

In order to obtain a high quality reference dataset

for urban puff dispersion modeling, systematic model
experiments have been carried. Systematic repetitive
puff releases were simulated in a model of the central
business district of Oklahoma City, set up in the large
atmospheric boundary layer wind tunnel WOTAN at
Hamburg University.
To simulate different wind
directions, the model was mounted on one of two turn
tables of the wind tunnel. A detailed description of the
experimental campaign is given by Harms (2007).

Figure 1. Model of downtown Oklahoma City in the

boundary layer wind tunnel WOTAN
For all puff releases, ethane was used as tracer.
Two Flame Ionization Detectors (FIDs) were used for
concentration measurements, one recording the
background concentration of tracer in the flow
approaching the model and the second one
measuring turbulent concentration fluctuations at
different locations within the model area. The
background
concentration
measurements
are
necessary because ethane is released into the air
circulating in the laboratory and the wind tunnel and
the background concentration of ethane is slowly
increasing throughout a test series. In addition, a
Prandtl tube connected to a differential pressure
transducer was used for reference wind speed
measurements at the entrance of the test section of
the wind tunnel.

214

3 Puff measurements

Frame 001 29 M ar 2007

Puff-like clouds of tracer were released from ground

level sources located at positions corresponding to
the locations used during the Joint Urban 2003 field
experiment. For each source location, concentration
time series containing at least 200 individual puff
releases were recorded at about ten different points.
The acquired time series are designated by the
location of the source, the measurement location, the
mean wind speed and mean wind direction during the
test as well as the tracer release flow rate (~ source
strength) and the release duration. From the collected
puff dispersion data is obvious, that even under the
controlled mean boundary conditions in the wind
tunnel the variability in the individual puff signal
recorded is tremendous. Figure 2 gives just one
example for the scatter to be expected in the puff
dispersion signals, showing the concentration time
trace recorded for two consecutive puff signals under
exactly the same release conditions.

cef[ppm]

400

200

0
1414

1416

tt_abs[s]

1418

1420

Figure 2. Two consecutive measured puffs, released

under the same boundary conditions

EWTL@ZMAW.DE University of Hamburg

pc

60

cef[ppm]

40

at

: arrival time

pt

: peak time

lt

: leaving time

pc : peak concentration
do : dosage
asct : ascent time

20

dsct : descent time

do
threshold
0

asct
0

at

dsct
4

pt

tt_puff[s]

lt

Figure 3. Sketch to illustrate the characteristic parameters for dispersing puffs

reached at a certain receptor. Further parameters
characterize the local exposure to the pollutant at a
given location or the time until the puff is cleared out
4 Charactristics of puff dispersion
again. If the distance between the emission source
and the measurement location is known, the
A systematic processing of the recorded puff data
estimation of an effective advection speed can be
requires the definition of typical puff dispersion
based on the measurement of an advection time.
parameters to be defined. In Figure 3, the most
According to Doran et al. (2007), arrival time and
common parameters characterizing a typical puff
peak time can be used for estimating the transport
signal are illustrated. A very general dispersion
speed of a puff. The arrival time is defined as the
parameter in the case of an accidental release is the
point in time after the release when the concentration
effective advection speed, the released pollutant is
signal exceeds tirst time a certain threshold value and
traveling at as well as the peak concentration

215

interest are the ascent time and the descent time of

the recorded concentration signal. While the ascent
time is defined as the time period, the concentration
needs to reach its maximum value after the first puff
front arrived at the detector, the descent time is the
period of time the concentration signal needs to fall
from the maximum value back below the threshold
level.

remaines above this level for a specified period of

time. Since the arrival time can be interpreted as the
time when the edge of the puff reaches the detector,
it is also possible to obtain information on the slope of
the puffs front side. Applying different threshold
values when analyzing the same puff pattern
repetitively enables the structure of the puff to be
characterized based on the measured arrival time
and leaving time. The peak time is defining the point
in time after the release when the maximum
concentration signal is measured or detected. The
local exposure to a given contaminant can be
estimated as the integrated or accumulated
concentration during the presence of the puff and is
refered to as the dosage at a certain location. As
graphical interpretation, the dosage can be seen as
the area defined by the threshold concentration level
and the puff signal above the thershold (see Figure
3).
Calculating the dosage also implies the
knowledge of the duration of a puff event at a given
location. Accordingly, the leaving time of a puff is
defined as the point in time when the concentration
falls below the threshold without exceeding it for the
remaining time of the time trace again. The puff
duration is simply defined by the difference between
leaving time and arrival time. Other parameters of

5 Puff data processing

For an automated analysis of the extensive
amount of puff data acquired in the wind tunnel, an
extensive analysis software program was developed.
The code evaluates time series of puff releases with
respect to the mean characteristic parameters
mentioned above. In addition, puff parameters are
calculated for different numbers of puffs using an
ensemble averaging approach. The ensemble
averaging approach based on a systematically
changing ensemble size enables the minimum,
maximum, and mean values of every characteristic
parameter to be visualized and analyzed in
conjunction with the corresponding frequency
distributions.

F ra m e 0 0 1 1 9 M a y 2 0 0 7

100

b)

80

c[ppm]

60

F ra m e 0 0 1 1 9 M a y 2 0 0 7

40

20

200
0

a)

5 5 7 .0 2

150

5 5 7 .0 4

5 5 7 .0 6

tt[s ]

5 5 7 .0 8

5 5 7 .1

5 5 7 .1 2

c[ppm]

F ra m e 0 0 1 1 9 M a y 2 0 0 7

100

c)

60

cef[ppm]

50

40

20

0
555

556

557

tt[s ]

558

559

0
5 5 6 .9 8

557

5 5 7 .0 2

5 5 7 .0 4

5 5 7 .0 6

tt[s ]

5 5 7 .0 8

5 5 7 .1

5 5 7 .1 2

Figure 4. Principle of puff removing: a) unfiltered puff signal with spike; b) spike in detail, showing a damped
harmonic oscillation; c) same section as in b) after spike removal and subtraction of background concentration

216

Frame 001 19 M ay 2007

Frame 00 1 19 M ay 20 07

200

150

150

c[ppm]

cef[ppm]

200

100

100

50

50

0
545

0
550

555

560

tt[s]

565

570

575

580

545

550

555

560

tt[s]

565

570

575

580

Figure 5. Subsection of timeseries; left: unfiltered, with spikes; right: after spike removal and subtraction of
background concentration

5.1 Data preprocessing

The measured data file is containing time series of

the reference wind velocity, the locally measured
tracer concentration, the background concentration
measured, a trigger signal indicating the open/closedstatus of the emission source and the measured
release flow rate present during the discharge. While
the reference wind velocity as well as the by-passed
source flow rate are assumed to be constant during a
recorded timeseries, the trigger signal is changing
periodically between 0 and 5V, whereas sequences
of 5V indicate the release duration of the source.
Due to the periodic puff releases there is a
continuous increase of the ethane background concentration. Hence, the background concentration has
to be subtracted from the actual puff concentration
signal before the parameters of the puff are
calculated.
In the wind tunnel, the occurance of spikes in the
recorded concentration signal, caused by dust
particles sampled by the fast FID cannot entirely be
prevented. In order to avoid errors in the calculation
of characteristic parameters due to spikes, the
program provides an option for detecting and
removing unwanted spikes. The spike detection
algorithm is based on the detection of a highfrequency harmonic oszillation present in the signal
whenever a spike is generated by dust. The
sensitivity of the filter can be adjusted in order to
prevent the code eliminating effectively measured
concentration peaks instead of spikes. A block wise
systematic Fourier analysis applied to the continuous
concentration signal recorded and if the Fourier
coefficient exceeds an adjustable threshold, the
concentration values are replaced by the last known
measured concentration (see Figure 4).
After subtracting the background concentration
from puff concentration and an optional spike
removal, the program is extracting the concentration
signals from individual puff releases. The separate
releases are detected by searching for the rising
edge of the source trigger signal. Every time, the
trigger signal changes from 0V to 5V, the source is
opened again and the algorithm detects a new
release.
In the wind tunnel, the puff concentration signal is
sampled with a frequency much higher than the

217

frequency response of the fast FID system used for

instantaneous concentration measurements. The
oversampling applied during data acquisition enables
a block-filtering to be applied before the concentration
signal is analyzed in order to eliminate possible noise
contained in the raw measurement signal.
In a final step of pre-processing, the program
detects whether a puff release is resulting in a
concentration signal for the given measurement
location or whether the cloud of tracer was flushed by
without a detectable concentration signal. The
presence of a signal is detected simply by searching
for an exceedance of the the given detection
thershold value within a puff period at all. For further
puff analysis, only the puffs arrived at the
measurement are considered, but for a complete
statistics the puffs flashed by are counted as well.

5.2 Detection of puff charcteristics

After pre-processing the acquired puff data, the

algorithm is analyzing puff characteristics for every
arrived puff within a time series with respect to a
certain concentration threshold value specified when
the program is started.
Typical values for the
threshold level and the remaining time or persistency
of the signal are 5 ppm (approx. twice the
measurement accuracy of recorded concentration
values) and 0.05s minimum remaining time of the
concentration signal.
The first detected parameters are the peak
concentration and peak time. While the peak of the
puff is detected as the maximum sampler concentration within a release period, the peak time is simply
the point in time at which this maximum occurs.
Arrival time and leaving time are detected similarly
by searching for an exceedance of the threshold for
the defined remaining time. Whereas for the arrival
time the program starts searching from the beginning
of the puff release, for the leaving time the detection
starts from the end of a release period counting
backwards in time.
Based on the detected arrival and leaving time,
the puff duration is calculated. Subsequently, ascent
time and descent time are calculated as the
difference between peak time and arrival time, and
leaving time and arrival time respectively.
After calculating the puff duration, the algorithm is
defining the dosage of the puff calculating the

concentration accumulated during the presence of

the puff signal. The discrete concentration values are
integrated numerically by a simple trapezoidal

technique, assuming a sufficiently high rate present in

the signal.
EWTL@ZMAW.DE University of Hamburg

EWTL@ZMAW.DE University of Hamburg

60

frequency of occurence

frequency of occurence

50

40

30

20

10
0

0
3

leaving time [s]

leaving time [s]

Figure 6. Frequency distributions of the leaving time; left: for 215 puffs; right: for 10 puffs
probabilistic approach to the prediction of possible
extreme values for example from a limited number of
field trails. Furthermore, the frequency distributions of
observed puff parameters are calculated for different
ensemble sizes. Figure 6 shows a comparison
between frequency distributions of the leaving time
for 10 and 215 puffs. The frequency distribution
based on 215 puffs shows an almost symmetric nearGaussian. Similar distributions of the same ensemble
size have been obtained for peak time, descent time
and duration of the puff, whereas peak concentration,
dosage, arrival time and ascent time mostly show a
significantly positive skewness. However, for all puff
parameters calculated the frequency distributions
based on ten puffs only do not give any indication for
the shape of a statistically safe ensemble of, for
example, 215 puffs. Consequently, a very limited
amount of puff samples cannot be used to predict
puff exposure in a probabilistic way or in other words,
a limited number of releases do not give a realistic
impression of what the threat at a certain location
would be, if a cloud of hazardous material passes by.

5.3 Statistical analysis

In order to obtain information about the representativeness of results calculated from a limited
number of individual releases the detected puff
signals are subdivided into ensembles of different
sizes. Since all puff signals collected within a test
series were recorded with exactly the same mean
boundary conditions present in the wind tunnel, the
data is qualified for puff ensemble analysis with
systematically changing ensemble sizes. Typically,
the size of the averaged puff ensembles was
increased in steps of 5 ranging from 5 puffs per group
up to 215 puffs. The smallest ensemble size analyzed
is in the order of the number of puff releases during
the Joint Urban 2003 field campaign repeated for
similar ambient conditions. Only three to six puff
releases could be carried out under more or less
similar boundary conditions in the field.
Every ensemble is examined with respect to the
minimum, maximum and mean value of every puff
parameter. The extreme values of puff parameters
observed for a certain ensemble were also
determined as they are of importance for a

218

Figure 7. Scatterplot on the mean value of the dosage for several ensemblesizes (every dot is assigned to one
ensembele, and all ensemblesizes are represented equal )

Exemplarly, Figure 7 shows a scatterplot of the mean

value of the dosage as a function of the ensemble
size. In the diagram, each dot is representing one
calculated mean dosage value for per ensemble. The
ensemblesize is ranging from 2 to 216 puffs in steps
of one puff. All ensemble sizes are represented by
the same number of individual ensembles, ensuring
the same statistical reliability for all ensemble sizes.
The plot indicates a huge scatter in the data or large
variation respectively for ensmble sizes smaller than
about 50 puffs. For ensemble sizes larger than 100
individual releases, the
gain
in
statistical
representativity is small for the given example of a
locally observed mean dosage.

6 Conclusions
For the validation of micro-scale LES based C&T
models, qualified experimental datasets are re-quired.
For an adequate description of puff disper-sion, puffs
must be characterized by well defined parameters
such as arrival time, leaving time, peak time, peak
concentration, puff duration or dosage, for example.
For statistical analysis and description of puff
dispersion phenomena, a large number of puff
releases under similar environmental is required.
Data qualified for a safe statistical and probabilistic

219

approach to the problem can be measured

systematically in boundary layer windtunnels.
In order to process the huge amount of data, a
software tool for systematic puff analysis was
developed. The code enables wind tunnel data to be
evaluated in a comprehensive way right after a test.
Based of the results, the investigator can decide
whether a specific test delivers the expected
statistically representative data or whether the test
hast to be repeated or extended. Besides the
analytical capabilities, in this regard, the software tool
also provides substantial support in order to produce
quality assured wind tunnel data efficiently.
As a first application of the software, the tool was
used during puff dispersion measurements in a
scaled model of the Joint Urban 2003 field test. With
the help of the code developed, it became instantly
clear, that a very limited number of puff releases do
not provide sufficiently reliable information for
characterizing the dispersion of puffs in a complex
environment. Hence, the limited number of individual
releases feasible in a typical field test might be
counted as a major limiting factor in the use of field
data for example for validation purposes.

References
Britter, R.E., Disabatino S., Caton, F., Cooke, K.,
Simmonds, P., Nickless, G., (2002). Results
from three field tracer experiments at the
neighborhood scalein the city of Birmingham,

Min, I.A., Abernathy, R.N., Lundblad, H.L., (2002).

Measurement and analysis of puff dispersion
above the atmospheric boundary layer using
quantitative imagery, Journal of Applied
Meteorology, 41, 1027-1041.
Nakamura, Y., Oke, T.R., (1988). Wind, temperature
and stability conditions in an E-W oriented
canyon, Atmospheric Environment, 22, 26912700.
Robins, A.G., Scarperdas A., Grigoriadis, D., (2001).
Spatial variability and source-receptor relations
at a street intersection. 2001, Water Air Soil
Pollut.-Focus, 2, 5-6, 381-393.
Sagaut, P., (2004). Large Eddy Simulation for
nd
Incompressible Flows, 2 Edition, Springer.

UK, Water Air Soil Pollut.-Focus, 2, no. 5-6, 7990.

Britter, R.E. and Hanna, S.R., (2003). Flow and
Dispersion in Urban Areas, Annu. Rev. Fluid
Mech. 2003. 35:469-96.
Cionco, R.M., et al., 1999. An overview of Madona: a
multinational field study of high-resolution
meteorology and diffusion over complex terrain,
Bulletin of the American Meteorological Society,
80, 5-19.
Davidson, M.J., Myline, K.R., Jones, C.D., Phillips,
J.C., Perkins, R.J. et al., (1995). Plume
dispersion through large groups of obstacles a
field investigation, Atmospheric Environment,
29, 3245-3256.
Davidson, M.J., Snyder, W.H., Lawson, R.E., Hunt,
J.C.R., (1996). Plume dispersion from point
sources upwind of groups of obstacles wind
tunnel simulations, Atmospheric Environment,
30, 3715-3725.
Doran, J., et al., (2007). Characteristics of puff
dispersion
in
an
urban
environment,
Atmospheric Environment, 41, issue 16, 34403452.
Draxler, R.R., (1984). Diffusion and transport
experiments, Atmospheric Science and Power
Production, DOE/TIC-27601, OSTI, US DOE,
367-422.
Gifford, F.A., (1957). Relative atmospheric diffusion
of smoke puffs. Journal of Meteorology, 14, 410414.
Gifford, F.A., (1977). Tropospheric relative diffusion
observations, Journal of Applied Meteorology,
16, 311-313.
Gifford, F.A., (1995). Some recent long-range
diffusion observations, Journal of applied
Meteorology, 34, 1727-1730.
Hall, D.J., Macdonald, R., Walker, S., Spanton, A.M.,
(1996). Measurement of dispersion within
simulated urban arrays a small scale wind
tunnel study, BRE Client Rep. 178/96, Build.
Res. Establ., Garston, Watford, UK.
Hanna, S.R., Franzese, P., (2000). Along-wind
disperrsion a simple similarity formula
compared with observations at 11 field sites and
in one wind tunnel, Journal of Applied
Meteorology, 39, 1700-1714.
Hanna, S.R., Chang, J.C., (2001). Use of the Kit Fox
field data to analyze dense gas dispersion
modeling issues, Atmospheric Environment, 35,
2231-2241.
Harms, F., Schatzmann, M., Leitl, B., (2007). How
comprehensive is comprehensive enough
Model-specific reference data for the validation of
micro-scale LES flow and dispersion models,
Physmod 2007, ibid.
Hosker, R.P., (1985). Flow around isolated
structures and building clusters: a review,
ASHRAE Trans., 91, 1671-1692.
Islitzer, N.F., Slade, D.H., (1968). Diffusion and
transport experiments, Meteorology and Atomic
Energy 1968, D.H. Slade (Ed), USAEC Report
TID-24190, NTIS, Washington, D.C.
Macdonald, R.W., Griffiths, R.F., Cheah, S.C., (1997).
Field experiments of dispersion through regular
arrays of cubic structures, Atmospheric
Environment, 31, 783-795.
Mikkelsen, T., Larsen, S.E., Pecseli, H.L., (1987).
Diffusion of Gaussian puffs, Quarterly Journal
of the Royal Meteorological Society, 113, 81-105.

220

Issues concerning wind tunnel modelling of heavy gas dispersion with

focus on risk assessment
1

Klara Bezpalcova1,2, Masaaki Ohba1

Department of Architecture, Faculty of Engineering, Tokyo Polytechnic University, Atsugi, Japan

2
Institute of Thermomechnics, Czech Academy of Sciences, Prague, Czech Republic
bezpalcova@it.cas.cz
and Walker (1997), Robins et al. (2001), and Zhu et al.
(1998). Among the normal modelling criteria like the
similarity of geometric conditions, properties of the model
and prototype boundary layer structures, etc., the
similarity of the dense gas cloud is required.
Two separate criteria were proposed to ensure
suitable dense gas plume behavior. The first is describing
the plume stability condition and it is measured by
Richardson number, Ri. Robins et al. (2001) defined the
local Ri* as

Abstract The presented paper summarized

modelling requirements for reduce-scale modelling of
heavy gas releases. Heavy gases are of interest for
risk assessment purposes since most of the very
toxic chemical compounds are heavy than air. The
wind tunnel modelling can provide concentration data
with high temporal and spatial resolution, where the
fluctuation nature of dispersion can be observed. It is
shown how smoothing of the fluctuation amplitude
can influence the risk assessment results.

Ri*

Key words atmospheric boundary layer, urban canopy,

concentration fluctuation.

g ' h / u *2 ,

g ' g (U p  U ) / U ,

where g is the modified gravitational acceleration, p is

the density of the plume, the ambient density, h is the
depth of the plume, and u* is the friction velocity. On the
other hand Hall and Walker (1997) introduced more
universal Richardson number

Introduction
It is well known that the thermal stratification as well as
density of pollutants influences turbulence properties of
the ABL, i.e. dispersion characteristics. Dispersion of
different pollutants, including the heavy gas dispersion
problems, came forward in eighties of 20th century with
the expansion of chemical industries. Since many new
instrument equipments and methods were developed from
those times, it is highly interesting to return to this topic
and conducted systematic study. The study of heavy
gases dispersion was firstly arouse on a basis of chemical
factory accidents, where very dangerous substances like
chlorine or ammonium were released. Nowadays this
topic is becoming actual again with regard to terrorist
attack with aim to harm citizens by chemical or biological
weapons (e.g. the Sarine accident in Tokyo underground).
Dispersion of toxic pollutants emitted accidentally or
intentionally from a ground level in an urban area under
various atmospheric thermal stratifications is scientifically
important. Obtained data will be used to design a risk
assessment guideline that considers thermal stability
effects and physical and chemical properties of released
gases.
The wind tunnel modelling offers the possibilities of
conducting experiment with high temporal and spatial
resolution. Well-equipped wind-tunnel laboratory of the
Wind Engineering Center of Tokyo Polytechnic University
has several wind tunnels including one allowing modelling
a thermal stratification of ABL. There is also available a
new method for simultaneous measurement of velocity,
concentration and temperature in one place using hot wire
and hot film probes together with fast flame ionization
detector. This set-up allows deriving turbulent fluxes
related to heat, momentum, and concentration, which are
of high interest for numerical modellers using advanced
Computational Fluid Dynamics (CFD) models, i.e. Large
Eddy Simulation (LES).

Ri

g' L / U 2 ,

g' g(U p  U ) / U,

where L is characteristic length scale for dispersing

releases, and U is reference wind speed. Zhu et al. (1998)
considered only initial or source Ri0:

Ri 0

g 0 ' Q / D S u *3 ,

g 0 ' g (U 0  U ) / U ,

where Q is volume flow rate of the dense gas, DS is

diameter of the dense gas source, and 0 is the intial
density of the dense gas.
The modelling criteria has to be carefully adjusted to
keep all the similarity requirements. Since the Ri is
function of the squared velocity and scale, it is obvious
that the obtainable values of Ri in wind tunnel are smaller
than those in the field.
The second condition is a limit on the emission velocity
relative to the friction velocity, introduced to minimize
interactions between the emission and the cross-flow. The
consensus based on many works was that acceptable
values of outflow velocity are smaller than actual friction
velocity.
Briggs et al. (2001) also mentioned the problem of flow
laminarisation after releasing big amount of heavy gas
into wind tunnel boundary layer, i.e. during the cases with
high Ri. There is evidence of turbulent difusion coefficient
decrease with increasing Ri. The very low Reynolds
number connected with laminar flow can be observed in
the case of high Ri Plumes with small values of Ri are
dominated by ambient turbulence and dense gas effects
are insignificant.

2 Risk Assessment Procedures

The risk assessment is important policy tool, which
evaluates acceptance of the risk and it is also the basis
for emergency planning and evacuation plans. There are
many guidelines introduced by official bodies e.g. EPA
(1986) in U.S. and CPR18E (2005) in Europe. The risk
assessment methodologies are usually based on mean
concentration spread obtaining from Gaussian type
models Gaussian type of models showed to be
appropriate for using within the urban canopy, but in the

1 Heavy Gas Dispersion

The comprehensive overview of scaling rules for
reduced-scale heavy gas dispersion was given e.g. in Hall

221

field of risk assessment only the parametric models are

used.
There is big issue for numerical as well as physical
modellers to improve the methodology used in the field of
risk assessment.
Not only the mean concentration, but also the
concentration fluctuation should be taken into an account
since the hazard posed by an acutely toxic gas release
depends non-linearly upon exposure concentration and
exposure time (Hilderman et al., 1999). Acutely toxic
chemicals cause effects ranging from annoyance caused
by an offensive odor to fatality.
The mechanism of toxic gas influence on human
health is very complicated and differ from one individual to
another. At low doses only sensitive individuals respond
while much higher doses are required to affects the
resistant individuals. The limited information available for
creating acute toxicity models consists mostly of
experiments in which laboratory animals were exposed to
constant concentrations for a fixed period of time and the
number of fatalities were recorded.
The toxic load equation can be derived from
experimental data that are fit using probit method of
Finney, 1971.

(3)
where n is the toxic load exponent equal to r/s from Eq.2
and the toxic load L is defined as L = cnt. In terms of the
toxic load L, Eq.3 can be rewritten as:
(4)
P q  s ln L ,
To determine the proportion of a population
responding to a release, the toxic load L is calculated and
then Eq.5 is used to determine the probit value Pr. The
percentage fatalities is obtained from the Pr value and
Eq.1.
In animal experiments, the exposure is in controlled
conditions at a constant concentration for a set period of
time. In this case, there are no fluctuations in
concentration
and
the
instantaneous
exposure
concentration c is constant with time over the entire
exposure duration te. With instantaneous concentration c
equal to the mean concentration the toxic load Lmean is
Lmean = Cnmean te. This is the original definition of toxic
load. Note that Lmean is not the mean toxic mean load, but
rather is a representative toxic load based on the mean
concentration.
The toxic load of a fluctuating exposure concentration
could also be calculated with the mean concentration and
the exposure duration te. If n = 1, the effect of e
concentration is linear and this is a reasonable approach,
but it still does not take into account any uptake, recovery
or saturation processes. For most chemicals, where n > 1,
the mean concentration toxic load Lmean misses the
important non-linear effects of the mean concentration
fluctuations as well as any limitations on receptor
response.
In the risk assessment literature, the definition of toxic
load has been extended, without any toxicological
justification, to include time varying exposure
concentrations, see Ride (1984) and ten Berge et al.
(1986)

3 Probit method
The probit method is a way of linearizing a cumulative
normal distribution of population response to some toxic
exposure variable. One probit unit, P, is equal to one
standard deviation of the normal distribution. The median
or 50th percentile response was defined arbitrarily as P =
5.0. A probit value of P = 4.0 is one standard deviation
below the median at a cumulative probability of 16%. That
is, it is expected that 16% of the population responds to a
toxic load that produces a probit value of 4.0. Similarly,
84% of the population would be expected to respond to a
toxic load that produces a probit of P = 6.0, one standard
deviation above the mean. The fraction F of the
population responding to a toxic exposure can be
calculated from the probit value P using the following
relationship

1 P 5
efr
 1,
2 2

te

c n dt ,

(5)

where c is the exposure concentration as a function of

time. This definition of the toxic load L is the total
fluctuating exposure toxic load, and is the most useful
toxic load for real exposure scenarios. If the toxic load
exponent n is greater than 1 then the exposure toxic load
L will be larger than the toxic load calculated with the
mean exposure concentration Lmean The exposure toxic
load does not take into account any physically realistic
limitations on the fluctuations that will determines the
effective toxic load that produces fatalities. In Eq.5 it is
implicitly assumed that the uptake of any exposure
concentration is instantaneous; recovery does not occur,
so toxic load increases indefinitely with time and repeated
exposures; and saturation of biological uptake pathways
does not occur. None of these assumptions are justifiable
for real exposures and responses and the way how these
processes can be treated is shown e.g. in (Hilderman et
al., 1999).
For about 20 commonly used substances, there is
some information on dose-response relationship that can
be applied to a probit function to quantify the number of
fatalities that are likely to occur with a given exposure.
The tables of parameters q, s, and n in Eq.3 can be found
e.g. in CCPS (1999), or CPR (2005) .
For substances that are not listed in the references
mentioned above, Levels of Concern (LOCs) are
estimated from median lethal concentration (LC50) levels
reported for mammalian species. LC50s are concentration
levels that kill 50% of exposed laboratory animals in

(1)

where erf is the error function

erf ( x)

q  s ln c n t ,

e t dt .

For acutely toxic gases it is observed that the

logarithm of the toxic load L follows a normal distribution.
This implies that the population response level follows a
log-normal distribution with L. Toxic load L = cnt is the
combination of two variables, the concentration c and the
exposure duration t. To find the value of n for a particular
chemical both variables must be considered using a two
dimensional probit relationship:
(2)
P q  r ln c  s ln t ,
where q is the offset from zero, r is the coefficient of the
logarithm of concentration, and s is the coefficient of the
logarithm of time. The logarithms in Eq.2 produce the
required lognormal distribution of response with toxic load.
For each experiment the probit of response is recorded
along with the logarithm of the concentration c and the
logarithm of exposure duration t. The linear twodimensional relationship is solved to give the coefficients
q, r, and s. The toxic load relationship is obtained by
combining the last two terms of Eq.2:

222

Fractions of population which would response on such

a dose, F, can be estimated from probit function using
Eq.1 for each situation.

controlled experiments. Using these data, the level of

concern is estimated as follows (EPA, 1986)
LOC = 0.1 LC50
(6)
If the concentration at any place reaches the LOC
concentration, the site is consider to be dangerous and an
appropriate precautions should be done.

4 Application
The wind tunnel modelling offer an opportunity to
measure concentration time series with high spatial and
temporal resolution. This concentration time series can be
used for better estimating of the personal risk
assessment.
Examples of concentration time series are shown in
Fig.1 and Fig.2, respectively. The time series in Fig.1 are
example of intermittent plume behavior. On the other
hands the time series in Fig.2 are concentration time
series taken in the middle of thr plume, where the passive
tracer is relatively well mixed within the plume. These time
series were measured in the wind tunnel within the
idealized urban roughness (for more details see
Bezpalcova, 2007). The wind-tunnel based dimensionless
data were adapted to the full scale situation under these
conditions:
x
x
x

time series 1

[s]

0.5
25
250
1000

2
1
0
0

41
38
33
30

time series 2

Due to non linearity of the calculation (logarithm, error

function, etc.) the final results show bigger differences
between different averaging times in fraction of population
which would response on such a toxic load for well mixed
case (second case) although the differences in probit
functions are smaller than in the intermittent case.
However, the observed differences are not negligible and
therefore the fluctuation characteristics are important for
the risk assessment issues.
Moreover, Hilderman et al. (1999) showed that even
bigger differences between mean load and fluctuatingtime-series based load can be observed when considering
such mechanisms as recovery time, uptake time, and
saturation concentration.

Escape of ammonia, source strength Q = 20 kg/s;

Average building height H = 10 m;
Reference wind speed at the average building height
UH = 5 m/s.

5 Conclusions
The risk assessment issue is quite new discipline and
it is based on predictions. In the case of sudden or
delivery release of toxic gas the prediction of dispersion is
needed. However, it is common practice to use simple
parametric models, which often failed with the urban or
industrial areas. Therefore the wind tunnel modelling
should be consider as a tool for obtaining more trustful
prediction of actual concentration. Most of the toxic gases
are heavier than air, therefore more advanced simulation
technique is required.

There are for different lines in Figures 1 and 2. The

solid line is the measurement with frequency of 2 Hz, i.e. it
is assumed that the averaging time was 0.5 s. The bold,
dashed-dot, and dashed lines represent measurement
with 25, 250, and 1000 s averaging time, respectively.
According CPR18E (2005) the constants q, s, and n,
which appear in Eqs.3-5, for ammonia are following:
x
x
x

averaging
time

q = -15.6;
s = 1;
n = 2;

Acknowledgments

If the Eq.5 is aplied on this concentration curves,

following values for toxic loads can be obtained.
averaging
time

time series 1

time series 2

[s]

dose

dose

0.5
25
250
1000

30 000 000
23 000 000
15 000 000
11 000 000

178 000 000

164 000 000
142 000 000
130 000 000

Authors would like to acknowledge Japan Society for

the Promotion in Science for supporting the research.

References
Bezpalcova, K. Physical Modelling of Flow and Diffusion
in Urban Canopy, Ph.D. thesis, Charles University in
Prague, 2007.
Briggs, G. A., Britter, R. E., Hanna, S. R., Havens, J. A.,
Robins, A. G., and Snyder, W. H., (2001). Dense
gas vertical diffusion over rough surfaces: results of
wind-tunnel studies, Atmospheric Environment 35,
pp. 2265-2284.
Center for Chemical Process Safety (CCPS), Guidelines
for Chemical Process Quantitative Risk Analysis,
Wiley-AIChE, 1999.
Committee for the Prevention of Disasters (CPR),
Guidelines for quantitative risk assessment Purple
Book CPR18E, SDU, The Hague, 2005.
EPA, (1986). Emergency Planning and Community Right
to Know Program, 40 CFR 300. Washington, DC:
U.S. Environmental Protection Agency.
Finney, D.J., Probit Analysis, 3rd edition, Cambridge
University Press, 1971.

Then the probit values P can be estimated from Eq.4

averaging
time

time series 1

time series 2

[s]

probit

probit

0.5
25
250
1000

1.61
1.37
0.91
0.61

3.4
3.31
3.17
3.08

223

gas dispersion in a neutral boundary layer over a

rough surface, Atmospheric Environment 35, pp.
2243-2252.
ten Berge, W. F., Zwart, A., and Appelman, L. M., (1986).
Concentrationtime mortality response relationship
of irritant and systemically acting vapours and gases,
Journal of Hazardous Materials 13, pp. 301-309
Zhu, G., Arya, S.P., and Snyder W., (1998). An
experimental study of the flow structure within a
dense gas plume, Journal of Hazardous Materials
62, pp. 161-186.

Hall, D.J., Walker, S., (1997). Scaling rules for reducedscale field releases of hydrogen fluoride, Journal of
Hazardous Materials 54, pp. 89-111.
Hilderman, T.L., Hrudey, S.E., and Wilson, D.J., (1999).
A model for effective toxic load from fluctuating gas
concentration, Journal of Hazardous Materials A 64,
pp. 115-134.
Ride, D. J., (1984). An assessment of the effects of
fluctuations on the severity of poisoning by toxic
vapours, Journal of Hazardous Materials 9, pp. 235240.
Robins, A., Castro, I., Hayden, P., Steggel, N., Contini, D.,
and Heist, D., (2001). A wind tunnel study of dense

averaging time 0.5 s

averaging time 25 s
averaging time 250 s
averaging time 1000 s

2000

c [mg/m3]

1500

1000

500

0
2000

2100

2200

2300

2400

2500

2600

2700

2800

2900

3000

t [s]

Figure 1: An example of intermittent concentration time series. Different lines show diffrent averaging time of
measurements.

2500

averaging time 0.5 s

averaging time 25 s
averaging time 250 s
averaging time 1000 s

c [mg/m3]

2000

1500

1000

500

0
2000

2100

2200

2300

2400

2500

2600

2700

2800

2900

3000

t [s]

Figure 2: An example of well mixed pasive tracer in the middle of the plume. Different lines show different
averaging time of measurements.

224

Visualisation of dispersion processes

in the surrounding of livestock buildings
K. von Bobrutzki, H.-J. Mller

Leibniz-Institute for Agricultural Engineering e.V. (ATB),

Department of Engineering for Livestock Management,
Max-Eyth-Allee 100, D-14469 Potsdam, Germany
kvbobrutzki@atb-potsdam.de; hmueller@atb-potsdam.de
Abstract - The Leibniz-Institute for Agricultural Engineering Potsdam-Bornim (ATB) engages in the
investigation of the dispersion of emissions out of
livestock buildings for a long time. In this connection the measurements in and around real stables
are linked with wind tunnel experiments. These
physical tests of the airflow are very useful to get a
better understanding towards the difficult theoretical background. It is possible to investigate special
situations of dispersion of substances in a qualitative and quantitative way.
Key words: wind tunnel, dispersion, livestock building,
emission

1 Introduction
In the surrounding of livestock buildings odours can
cause annoyances and ammonia can lead to damages
on plants and ecosystem. There exist German guidelines and standards for example minimum distances
between animal houses and residential area. To control
the environmental laws, measurements in and around
stables are necessary.
The ATB possess a large equipment of climate and
emissions technology. These are used for measure
dates of ammonia, odours and climate parameters. So it
is possible to get information about the high of disturbance and the over or under usage of critical value.
Furthermore in the boundary layer wind tunnel of the
institute there are realise systematically measurements
regarding to the dispersion of gases. The physical surrounding of the investigation area can be modelled and
analysis of emission can be carried out. The airflow can
be descripted by quantitative and qualitative measurement. One of the qualitative investigations represents
the laser-light-sheet technique, which is used to illuminate the flow in the working section of the wind tunnel. It
is possible to explain complex questions of the flow and
the dispersion of substances in a clear and visually way.

The wind tunnel investigation is an economic method for

testing different methods towards the dispersion of contaminated material, which pass off from animal stables.
In a practical example shall be shown a livestock
building which leads to odour disturbances in the
neighbour residential area. There are to find features
which can possible reduce this negative appearance,
but not by decrease the existing livestock.
A comparison between different structural modifications provides opportunities of variances. All shown
examples are produced by physical modelling in the
wind tunnel.

2 Materials and methods

2.1 Wind tunnel
The ATB is able to investigate the dispersion of substances with a boundary layer wind tunnel. Inside the
tunnel the turbulent boundary layer is generated by
additional fittings like eddy generators and roughness
elements. These fixtures form a wind profile which is
comparable with a naturally one in a suburban area
(VDI 3783). So the results of the investigations represent the nature and a directly comparison gets possible.
The length of the wind tunnel is about 12 m and the
cross section possesses 2.00 x 1.25 m (figure 1). In the
space of the working section the maximum velocity of 5
m/s can be reach. The physical surrounding of the investigation area can be modelled and the associated
analysis of emission can be carried out.
2.2 Laser-light-sheet technique
The airflow can be descripted by quantitative and
qualitative measurement. One of the qualitative investigations represents the laser-light-sheet technique,
which is used to illuminate the flow in the working section of the wind tunnel. Particles of fume are admixed to
the flow and through a laser light the airflow becomes
visible. This process is captured by a CCD camera. In a
next step the film is split up into single frames. As a

Figure 1. Atmospheric boundary layer wind tunnel

225

Figure 2. Animal livestock building in eastern Germany

result of this procedure, an average image from one
special situation is developed. So it is possible to explain complex questions of the attitude of the flow and
the dispersion of substances in a clear and visually way.
2.3 Measurement of velocity and gas concentration
The speed of the flow and the wind field is detected by a
hot-wire anemometer. In addition to a mean velocity this
device is able to acquire the fluctuation of the velocity.
Further the concentration field is taking up by a detector.
The physical background of the Flame Ionisation Detector (FID) is based upon the burning of a carbon gas
which sent a signal to an analyser. We use Ethan as
tracer gas for investigations of dispersion and get a
mean concentration and a fluctuation part, too.

3 Results
In the surrounding of a big livestock building for pigs,
disturbances appear because of odour (figure 2.). The
task was to reduce these negative effects by structural
arrangements as well as inside and outside. In this
article only the possible outdoor changes are mentioned. For further details towards different variances of
the ventilation system see Krause, 2006.

3.1 Pictures by the laser-light-sheet technique

Figure 3 is an average picture which is a result of the
laser-light-sheet technique. It is a situation which shows
the dispersion of plumes form three stacks out of a
stable. It represents a detail in a scale 1:300 of the big
pig farm (figure 2.).
In 150 mm which is equivalent to 45 m in nature the
velocity of the air flow is 0.46 m/s and goes from the
right to the left site (Lee). The exhaust air velocity of the
fume is 0.47 m/s.
Figure 4 shows the same situation like figure 3 but
with the difference of an obstacle on the Lee site in a
distance of 100 mm (30 m in nature). So a structural
changing toward the surrounding of the stable exists
and is represented by a wall with trees or copses.
The positive effect of obstacles on the Lee site of an
air flow is higher than on the Luv site (Blenk & Trienes,
1956).
3.2 Measurement of gas concentration
The following diagrams are emerged by tracer gas
(Ethan) dispersion measurements and were detected by
the FID within a wind tunnel experiment. Figure 5 and 6
are associated with the average pictures (figure 3., 4.)
and represent the same dispersion situations. The dia-

wind

Figure 3. Average picture of a stable with 3 stacks and without a wall

wind

Figure 4. Average picture of a stable with 3 stacks and a wall on the Lee site

226

160
1200 mm (360 m)
800 mm (240 m)
400 mm (120 m)

140

high [mm]

120
wind velocity in a high of 150 mm: 0.95 m/s
exit velocity: 0.47 m/s

100
80
60
40
20
0
0

500

1000

1500

2000

concentration [ppm]

2500

Figure 5. Gas measurement of a stable with 3 stacks and without a wall

160
1200 mm (360 m)
140

800 mm (240 m)
400 mm (120 m)

120
wind velocity in a high of 150 mm: 0.95 m/s
exit velocity: 0.47 m/s

high [mm]

100
80
60
40
20
0
0

200

400

600

800

1000

1200

concentration [ppm]

Figure 6. Gas measurement of a stable with 3 stacks and a wall on the Lee site
grams in figure 5 and 6 are results from model measurements in the scale of 1:300. But the velocity of the air
reaches 0.95 m/s in a high of 150 mm. As an outputsignal we used a mixture of an Ethan / Air volume
stream with a ratio of 1:10. The diagrams 5 and 6 show
the relations between high (mm) and concentration
(ppm) in different distances from the emission source
(1200 mm, 800 mm, 400 mm). Thereby the blue graph
is in the distance of 1200 mm, the red one is located in
800 mm and the green graph stands for the distance of
400 mm from the font. These curves represent 360 m,
240 m and 120 m in nature at a scale of 1:300. Figure 5
shows the dispersion situation of a stable without an
obstacle. The high is equivalent to our studies in the
wind tunnel and so the y-coordinate is determined in the
units of mm.
In a scale of 1:300 the maximum of the high reaches
50 m in nature.
Contrary to figure 5 represents image 6 the influence
of a wall on the Lee site as a disturbance of the dispersion process. As a commonality of both situations the
concentration is sinking with the high and the distance

227

to the emission font. But with a wall on the Lee site the
distinction of concentration is nearly the half. Without
the wall the concentration in 3 m high detracts from
1800 ppm in a distance of 120 m to 1100 ppm in a
space of 360 m. This responds 38 % minimization. In
comparison to the situation with a wall the reduction
reaches 80 % in 360 m distance toward the value in 120
m distance (figure 7).
Within the high from 3 to 30 m in space of 120 m to
the emission font in case of an obstacle, the rising of
reduce of concentration is the highest. In the distance of
240 and 360 m the decrease is much more less.

4 Conclusions
In the wind tunnel it is possible to create the surrounding of a special situation. Thereby complex queries toward the attitude of flow and dispersion of substances can be explained in an easy manner. It is simple to change the structural surrounding of an area and
find out an optimal set-up for a special problem. The
dispersion of emissions can research in a qualitative

120
1200 mm (360 m)

110

800 mm (240 m)

100

400 mm (120 m)

90

high [mm]

80
wind velocity in a high of 150 mm: 0.95 m/s
exit velocity: 0.47 m/s

70
60
50
40
30
20
10
0
30

40

50

60

70

80

90

100

reduction [%]

Figure 7. Comparison of concentration reduction [%] without and under the influence
way by the laser-light-sheet technique and through a
quantitative method by gas measurements. Both investigations are complementing each other. Obstacles at
the Lee site (Figure 4) visualise a better deflection of the
plumes in higher air levels and a better admixing with
fresh and waste air. The combined measurements (Figure 6) confirm this receiving by a high decrease of the
concentration mainly in the close-up range of 120 m
transferred to nature. Because of the similarity of the
wind profile inside the tunnel with a naturally one in a
suburban area a portability to practise gets possible. So
an obstacle like a wall can improving the ambient air of
a livestock building and brings more comfort to the environment and the people.

228

References
Krause, K.-H. , Linke, S., Mulick, M., Mller, H.-J.,
Weihs, C. 2006. A New Ventilation System for Reduction of Odour and Ammonia Emissions from Pig
Houses. In: VDI.-Berichte Nr.1958: World Congress
Agricultural Engineering for a Better World, VDI
Verlag GmbH, Dsseldorf, pp. 447 448.
Blenk, H., Trienes, H. 1956. Strmungstechnische Beitrge zum Windschutz./An article to the topic windbreak in fluid mechanics. Grundlagen der Landtechnik. Heft 8. VDI-Verlag GmbH, Dsseldorf.
VDI 3783, Part 12. 2000. Environmental meteorology.
Physical modelling of flow and dispersion processes in the atmospheric boundary layer. Application of wind tunnels. VDI-Verlag GmbH, Dsseldorf.

Wake development and interactions within an array of large wind

turbines
Frauke Pascheke and Philip E. Hancock

School of Engineering, University of Surrey, Guildford, GU2 7XH, UK

E-mail: f.pascheke@surrey.ac.uk
Abstract -The investigation of wake development and
wake interactions within an array of large wind turbines is a key objective in the EnFlo Wind tunnel laboratory contribution to the SUPERGEN V - Wind Energy Technologies project. Extracting energy from the
wind, the turbine wake is generally characterised by
reduced wind speeds and increased levels of turbulence. Within a wind farm array, the effects of several wakes interact. Thus, the turbines produce less
energy and stand greater structural loads than single
turbines placed in the free stream (see e.g. Burton
et al. (2002); Vermeer et al. (2003)). Current prediction
models need improving for these large machines.
Two primary mechanisms determine the decrease
of momentum deficit in the wake of a wind turbine:
mechanical turbulence generated by the turbine itself,
which is controlled by the turbine design and performance and usually is of a relatively high frequency
and small scale, and the turbulence level in the ambient atmospheric boundary layer (ABL) flow. With
the latter being controlled by terrain roughness, topography, stratification, etc., the characteristic length
scales as well as spectral characteristics of these two
interacting mechanisms are quite different. Moreover,
the quantitative effect of the interactions change as
rotor size increases.
Wake characteristics, development and interactions will be studied in the large EnFlo wind tunnel
(L W H : 20 3.5 1.5 m) for two different machine
sizes: 5MW with a rotor diameter of 126 m and a hub
height of 90 m and 2MW with a rotor diameter of 75 m
and a hub height of 65 m. At a model scale of 1 : 300
up to five 5MW model turbines (eight 2MW machines
respectively) will be arranged successively in the test
section as shown in Figure 1. The planned case
studies comprise off-shore ABLs for neutral and nonneutral conditions as well as a rural neutrally stratified ABL over different terrain (from flat to steep). The
wake measurements will be made using LDA, including phase-locked measurements to separate ordered
motion from genuine turbulence.
First wind tunnel experiments are planned for Summer 2007. Results expected in August 2007 comprise
the documentation of neutral and non-neutral boundary layer characteristics and verification of Reynolds
number independence.

References
Burton, T., D. Sharpe, N. Jenkins, and E. Bossanyi,
2002: Wind Energy Handbook. John Wiley &
Sons, Ltd.
Vermeer, L., J. Sorensen, and A. Crespo, 2003:
Wind turbine wake aerodynamics. Progress in
Aerospace Sciences, 39, 467510.
Figure 1: Model set-up for the 5MW turbine array (topview). Flow from right
to left, shaded area indicates the probe traversing range.

229

Wind-tunnel modelling and numerical simulation within urban

intersection
Radka Kellnerov
a1,2
2

Mohamed F. Yassin1,3

1
Institute of Termomechanics, Czech Academy Science, Prague, Czech Republic
Department of Meteorology and Environment Protection, Charles University, Prague, Czech Republic
3
Faculty of Engineering, Assiut University, Assiut, Egypt
radka.kellnerova@email.cz

Abstract - The inuence of street canyon

intersection setting on contaminant spreading from the vehicle trac has been studied
in wind tunnel. For street canyon intersection with layouts of X-shaped crossing, completely encircled by buildings we have measured the mean velocity and concentration
eld. The ow inside the canopy has strong
three-dimensionality, so the recirculation zone
is not created immediately behind the entry
of the canyon but further inside. The ow
investigation has conrmed that within the
area of corner vortexes occurs the largest accumulation of contaminants. Further in the
canyon the concentration eld is aected by
developed vertical vortex and accumulation
appears mostly on the leeward side.
Key words: Canopy layer, canyon intersection,
trac pollution.

Zbyn
ek Ja
nour1

Introduction

The project has focused on spreading of air pollution

inside the idealised urban area. The idealised model
was designed after the typical inner-city area with
20m high apartment houses with pitched roof. The
houses form regular blocks divided by intersections
(see scheme on Figure 1). Model has been scaled
down to 1 : 200. The dispersion and transportation
of trac emission inside the intersections area is
main objective of this work.

Figure 1: Scheme of building, X-shaped intersection,

and photograph from experiment. Street aspect ratio
was S/H =1, where S is width of the street and H
is the height of buildings. The depth of buildings is
B/H=1/2.

laser. Droplets with mean radius approximately

1m passively ow in the air with suciently low
sedimentation velocity (approximately 0.01 m.s1 ).
Concentration was measured using slow-response
ame ionisation detector (FID) with ethane used
as sample gas. Ethane is passive and non-reactive
gas with its own density close to density of the
air. Trac exhalations were simulated by a continual double line source. The source was constituted
by 2 500 parallel needles with inner diameter 0.1 mm.

At rst, it had to be veried that internal roughness layer is fully developed above measuring position. We checked the sucient distance from the
model front that is the required fetch for ow inside
the roughness layer to be in equilibrium with the surface beneath.
Under neutrally stratied conditions, vertical proles of turbulent characteristics were measured in the
central and the symmetrically upstream and downstream intersection (see Figure 2).
Strong similarity of the proles suggests that inFlow measurements were carried out using twodimensional bre-optic laser Doppler anemometry ternal boundary layer above these positions has very
(LDA). In order to measure air-particle velocity it close turbulent characteristics.
Requirements for equilibrium of canopy sublayer,
was necessary to add the glycerine droplets in the
main ow , because the droplets can be traced by e.g. layer below the roof level, are weaker. Hence
The experiment was conducted in low-speed
open-circuit aerodynamic tunnel IT AS CR in Nov
y
Knn. Tunnel had 20.5m long generating section
with roughness elements and spires. With the air
owing over the section, fully turbulent boundary
layer can be developed. On the beginning of working
section the thickness of turbulent layer attained
more than 50 cm.

231

positions have dierent behaviour. Up to the 1.5 H

level, Reynolds stress values are higher at the frontal
position than at other ones. It can be explained by
transformation of ow characteristics after reaching
the roughness change on its way. The shear stress
prole that is generated over the roughness elements
has typically enhanced values at the elements height
HE . When ow reaches the roughness step change
and enters into central street, Reynolds stress begins
to decrease. The change comes from surface level
upward.

Figure 2: Vertical velocity proles in the intersections. The positions of prole are shown in the
sketch. U0 means the reference velocity.

the requirements for fully turbulent development of

urban canopy is satised.
Spatially averaged velocity data were tted by the
logarithmic and power law. Roughness length and
displacement were obtained from shear stress values
within the inertial sublayer (for more information Figure 3: Vertical velocity proles in the central
see [1]). The exponent was determined by tting street. The positions of prole are marked in the
the power law:
sketch.
Roughness length z0 = 3,96 mm
Displacement d0 = 17 mm
Power law exponent = 0,24
For comparison purpose we have measured the proles above street canyons corresponding to the inThe roughness length equates to z0 =0.8 m in a tersections. The intersections are open to incoming
full scale. This corresponds according to Britter ow, whilst canyons are perpendicular to free ow
and Hanna (2003) [2] with parameters for skimming (see sketch in Figure 4). In accordance with varregime z0 1.0 m in a densely built-up area without ious authors, two eects at roof level were found
much obstacle height variation. On the other hand, in a street canyon in comparison with the situation
parameters normalised by the height of building in the intersection: signicant velocity retardation
HB are lower than expected range proposed by accompanied by increase of vertical Reynolds stress
Grimmond and Oke (1999) :
< u w >.
z0 /HB =0.04 (range from 0.06 to 0.2)
High absolute stress values indicate the region of
d0 /HB =0.17 (range from 0.35 to 0.85)
enhanced mixing and turbulent kinetic energy, where
intensive mixture of canyon air with upper air occurs.
Better agreement has been found for parameter of Based on theory, large-scale eddies are disturbed to
dimensionless frontal area f , which provides infor- small-scale eddies and high momentum ux causes
mation about building dimension and spacing. f is the fast concentration exchange. As a result, trac
ratio between frontal surface area Af and total sur- pollution quits the canyon area in this important reface area AT . In our case f = Af /AT equals 0.3 and gion. According to Rafailidis (1997) [3], these eects
lies in the lower edge of interval for downtown areas. are strengthened by the presence of pitched roofs.
Measurement along the central street, parallel with
Within the urban canopy, inside the central street
incoming ow, shows the uniformity of Reynolds parallel with incoming ow the velocity gradually instress behaviour with increasing height for positions creases with longitudinal distance (Figure 5). On
in the central part of the model (Figure 3). Outer the entry of the model, the velocity grows explo2
232

2
2.1

Results
Velocity eld

The ow inside the canopy is strongly threedimensional. Velocity in canyons has both horizontal
and vertical components and spiral vortex is formed.
An almost symmetrical velocity eld with corner
vortexes and slow moving contra-ow in the street
canyon was found (Figure 6 - up). Within an intersection, area velocity drops due to divergence. Figure
6 - down shows symmetrical area of inverse values of
Reynolds stress < u v >. High values of horizontal
Reynolds stress occurs close to corner areas. Further
in the canyon, the sign of Reynolds stress reverses.

Figure 4: Vertical velocity prole in the street

canyons. The positions of prole are marked in the
sketch.

sively due to ow convergence and immediately drops

down. In agreement with Cheng and Castro (2002)
[4] the stress < u w > varies with longitudinal fetch.
The variation has wave character with wave length
equivalent to the pattern length X/H=3.5 . Over
the house block the stress descends to minimal value,
whereas it ascends to its maximum above the canyon.

Figure 5:

Wave variation of vertical Reynolds stress with

longitudinal distance. Z/H=0.75

3
233

Figure 6: Horizontal cross-section at Z/H=0.3.

Up: Velocity eld with vectors and streamlines.
UH means velocity at height of building. Down:
Reynolds stress < u v > eld, black points mark
measuring positions.

The absolute value of stress < u v > decreases

with increasing height (see Figure 7). Above the roof
level, Reynolds stress reaches small positive values
that are spatially independent.
From quadrant analysis can be concluded that
contributions of Reynolds components (mean stress
over the members in each quadrant multiplied by
percentual representation of this quadrant from total count) from rst quadrant (outward interaction,
u > 0, v > 0) and third quadrant (inward interaction, u < 0, v < 0) are nearly equal. The same
notice is valid for second and fourth quadrant. At
the lower level, mean values from particular quadrants are strongly depended on horizontal position.
With height the values increase and start to be constant with horizontal coordinate Y in all quadrants.
At the same time their global sum decreases. Above
the roof level data from all quadrants do not vary
with space and their total sum falls to low positive
values.

Figure 7: Horizontal Reynolds stress along the

canyon. X/H=0
The vertical velocity eld inside the street canyon
depends on the distance from the intersection. At
specic fetch from the intersection (Y/H=1.5) the
recirculation zone emerges (Figure 9 - third from the
top). With increasing distance the zone center moves
upwards. Inside the lower part of canyon, large absolute values of Reynolds stress rapidly decrease with
distance from the corner, whereas the upper canyon
stress values are spatially independent (Figure 9 fourth from the top).

Figure 8: Scheme of cross-sections.

Figure 9: Vertical cross-section in Y/H=0.5 (two up)

and Y/H=1.5 (two down) - marked in sketch in gure
8. Upper: velocity eld with streamlines. Lower:
Reynolds stress < u w > eld.
4
234

2.2

Concentration eld

We have found the critical Reynolds building number ReB = uH H/, where is kinematic viscosity,
and veried the requirements for Towsend hypothesis. The experiment was done with Reynolds building
number ReB =15 200 that lies on the lower edge of
interval for valid Towsend hypothesis. Relevant free
stream velocity was 4 m.s1 . The vehicle trac was
simulated by continual double line source situated
on the oor, therefore the height of the source equals
zero. The dimensionless concentration was obtained
from formula 1 :
K =

CuH HLQ
.
Q

(1)

Figure 11: Concentration eld, vertical cross-section

at Y/H=0.75 (up) and Y/H=1.5 (down) - positions
are shown in the gure above.
Figure 10: The concentration eld, horizontal cross- thereupon the recirculation zone is not created imsection at Z/H=0.3. Black points mark measurement mediately behind the entry of the canyon but further
points.
inside. Strong accumulation of trac pollution occurs inside the corner vortexes created in the street
Signicant accumulation of trac pollution oc- canyons. Similar accumulation of gas, but with lower
curs inside the corner vortexes created in the street concentration, was found near to the leeward side of
canyons (Figure 10). Similar eect - accumulation of the buildings.
gas - was found near the leeward side of the buildings due to developing of vertical vortex. Windward
side concentrations depend complexly on model geometry.
Concentration of contaminants achieved maximum
in the center of the corner vortex (Y/H=0.75), therefore it reaches extremely high values up to the roof
level (Figure 11 - up). Inside the canyon, vertical vortex is developed and aects spreading of pollutants.
As a result, the concentration with regard to the latter is much lower, nevertheless the leeward side shows
notable accumulation of dangerous gas and particles
(Figure 11 - down).

Conclusion

Acknowledgments
This project was supported by Institutional Research
Plan AVOZ20760514 and COST 732.

References
[1] Cheng H., Castro I.P.: (2002) Near Wall Flow
Over Urban-Like Roughness. Boundary-Layer
Meteorology 104: 229259.
[2] Britter R.E., Hanna S.R.: (2003) Flow and Dispersion in Urban Areas. Annu. Rev. Fluid Mech.
35:46996
[3] Rafailidis S.: (1997) Inuence of Building Areal
Density and Roof Shape on the Wind Characteristics Above a Town. Boundary-Layer Meteorology 85, 255-271.

The wind tunnel experiments described in this paper have provided signicant data sets about pollutant diusion in an urban intersection. The ow
inside the canopy has strong three-dimensionality, [4] Cheng H., Castro I. P.: (2002) Near-Wall Development After a Step Change in Surface Rough1 C means measured concentration, u
H means velocity at
ness. Boundar- Layer Meteorology 105, , 411the height of buildings H, LQ is length of the line source, Q is
432.
source ux.
5
235

Adaptation of the Lucien Malavard Wind Tunnel (L.M.E. Orlans,

France) as an Atmospheric Boundary Layer Wind Tunnel
S. Aubrun, G. Espaa, P. Devinant
Laboratoire de Mcanique et dEnergtique
8, rue Lonard de Vinci
F-45072 Orlans cedex 2, France
Emaill: sandrine.aubrun@univ-orleans.fr

Abstract This paper presents the modifications

that have been performed in the Lucien Malavard
wind tunnel to enable to physical modelling of
atmospheric boundary layers. The first modelled
ABL is characteristic of a rough-type terrain. Its
properties are shown.
Key words modelled ABL, experimental set-up,
turbulence, Integral length scale.

1 Introduction
The major activity of the Fluid dynamics team
from the LME is focused on experimental aerodynamics.
Fine analysis of separated flows, flow control in order to
reduce drag on vehicles, aerodynamics of rotating
systems (helicopter rotors and horizontal-axis wind
turbines) are its priority issues. After several
contributions to the wind turbine aerodynamics and
regarding the needs expressed by industrial partners,
the LME focused on the wind turbine wake, and
particularly on the far-wake interactions with other wind
turbines in order to minimize it. Indeed, the wind
turbines arrangement in a wind farm should be well
predicted before its implantation. It is therefore of great
interest to correctly assess these interactions. Since the
field measurements are rare and difficult to interpret,
wind turbine wakes and their interactions are usually
treated with numerical models. Though numerical wake
models are still very simple and they cannot take
correctly into account complex terrain configurations. An
alternative is therefore the physical modelling in wind
tunnels. This issue motivates the LME to modify the
return test section of its wind tunnel to model
atmospheric boundary layers at a scale of around 1:400.
The presentation will focus on the validation of the
obtained boundary layers using standard litterature as
VDI (2000), ESDU (1985), Snyder (1981). The first
application of this set-up is the study of wind turbine
wake and that will be presented in the paper Properties
of the far wake of a wind turbine in an atmospheric
boundary layer, G. Espaa, S. Aubrun, P. Devinant, L.
Laporte-daube, E. Dupont.

in order to improve the flow homogeneity. In that case,

the freestream flow is a jet-type flow. One great interest
of this second facility is the large dimensions of the
return circuit, which enables to get rid of blockage
considerations in our applications.
For one year, this second test section has been
adapted to model at a geometric scale of 1:400 neutral
atmospheric boundary layers representative of slightly
rough to very rough terrains. The used set-up (Fig. 2) is
strongly inspired from wind tunnels of the Environmental
Wind Tunnel Laboratory, University of Hamburg. A 16m
flat plate is mounted in the return circuit and is equipped
with mobile roughness elements made of 3cm-high Lshape metallic pieces. The 12 first meters of the plate
are in the divergent part of the return circuit (divergence
angle 3.2). Turbulence generators made of wooden
plates with a triangular shape, are fixed to the
turbulence grid in the second convergent. The
combination of these both setups enables to replicate
required atmospheric boundary layers.
Return
section
Return
testtest
section
5 x 5 m
4u4m
Vmax
= 22 m/s
Vmax = 12.5 m/s

Main test section

2u2 m 2
Vmax = 50 m/s

Figure 1 : Malavard wind tunnel

Velocity measurements are performed with a Dantec
2D Laser Doppler Velocimetry. The transmitting of
receiving optics head is located in the test section and it
is moved in the three directions of space with an
automated transverse. The measurement volume is
located at 50cm from the head. The seeding is made of
olive oil with a PIVTEC system, generating micro-size
droplets. All the test section is seeded.

2 Experimental set-up
The Lucien Malavard wind tunnel is a closed-circuit
wind tunnel (Fig. 1). It has two test sections V1 and V2.
The first one has a square section of 2m times 2m and
is 5m long. The maximum velocity is 60 m/s and the
turbulence intensity is lower than 0.4%. The second test
section V2 is located in the return circuit of the wind
tunnel. It has a square section of 5m times 5m and is
20m long. The maximum velocity is 10 m/s. The input
section can be reduced up to 3m times 3m with a
second convergent and a turbulence grid can be added

237

Figure 2 : Experimental set-up in Malavard wind

tunnel

4 Mean profiles
To determine the properties of the modelled
boundary layer, velocity profiles were measured at 13m
from the entrance of the wind tunnel. Fig. 5 presents the
mean streamwise velocity in a linear and in a semilogarithmic scale, at a geometric scale of 1:400. Since it
was the first attempt to model an ABL in this facility, a lot
of care was taken to properly characterise the modelled
boundary layer .Consequently, the velocity profiles were
measured up to 360m F.S. from the ground.
These data enables us to determine the roughness
length and the power law exponent of the boundary
layer. Fig. 5 shows that the wind profile fits to a
logarithmic function from 12m d z d 100m full scale.
The extrapolation of the logarithmic law towards the
zero-value of velocity gives the associated roughness
Figure 3 : Transformations in the Malavard return
section

3 Longitudinal pressure gradient

The horizontal static pressure distribution in the wind
tunnel was measured with a pressure transducer
through the difference of static pressure between the
static pressure tappings of a Pitot tube fixed at the
entrance of the test section and a mobile one at a height
of 0.6m above ground (Fig 4). The absolute value on the
graph has no significance, only the slope is meaningful
since it gives the static pressure gradient along the
development section. The guideline VDI 3783/12 (2000)
recommends a tolerance threshold of 0.05 for the static
pressure distribution criterion:

wp 1
2
G U uG d 0.05
wx 2

Where

wp
is the longitudinal pressure gradient, G is
wx

the boundary layer thickness, U the density of air and

uG

the flow velocity at the top edge of the modelled

relative static pressure [Pa]

boundary layer. The slope 1 gives a value of 0.023 and

the slope 2 gives a value of 0.009. The criterion is
respected from 12 to 16m, along the development plate
and the pressure gradient could be considered as
negligible from 14 to 16m.
5

slope 2 = 0.38 Pa/m

4
3

slope 1 = 1.01 Pa/m

0.3m (in full scale F.S.). It is characteristic

It was then noticed that the power law exponent D

was very dependent of the vertical region used to its
determination. Table 1 summarises this problem. It
shows the best fit to the measured wind profile with an
exponential function obtained for different vertical
regions. The modelled ABL can be characteristic of a
rough to very rough terrain. The combination of this
information and the obtained roughness length tend to
prove that the modelled ABL is of rough type. It is
nevertheless necessary to check the turbulence
properties of the boundary layer to definitely determine
the type of the modelled ABL.
Reynolds stresses are presented also on Fig. 5.
They are non-dimensioned with a friction velocity u* of
0.45. One can notice that Reynolds stresses are nearly
constant for an altitude from 60m to 200m. It indicates
that the surface layer is 200m high.
12m d

zd

100

160

200

260

360

0.2

0.22

0.23

0.25

0.27

Table 1: Dependence of the power law exponent to

the vertical region used to calculate it.
Fig. 6 presents the turbulence intensity profiles for
the three velocity components I u , I v and I w . All of
them are located at the upper limit of the ranges advised
by VDI 3783/12 (2000) for a rough terrain.
All these results contribute to show that the modelled
ABL is representative of a terrain in the upper limit of the
rough class.

2
1
0
12

z0

length,

of a rough terrain.

13

14

15

Distance of CLA development [m]

16

Figure 4: static pressure evolution along the

longitudinal direction, measured at 0.6m (model scale)
above the ground

238

400

10

a)

400

b)

350

c)

350
102

300

300
250

Z [m]

250
101

200

200

150

150
10

100

100

Z0 = 0.3m

50
0
0

U mean [m/s]

10

-1

50

U mean [m/s]

Figure 5 : a) Vertical profile of the time-mean velocity

0
0

0.5

-u'w'/u2*

1.5

b) Same profile on a semi-logarithmic scale. c) Profile of the

shear stresses - u cwc non-dimensioned with the friction velocity u * . Measured in wind tunnel at 13m (model scale)
from the entrance of the wind tunnel

Z [m]

400

400

a)

400

b)

350

350

350

300

300

300

250

250

250

200

200

200

150

150

150

100

100

100

50

50

50

0
0

10

20

Iu [%]

30

40

0
0

10

20

Iv [%]

30

40

0
0

c)

10

20

Iw [%]

30

40

Figure 6 : a) Longitudinal turbulence intensity profile. b) Lateral turbulence intensity profile. c) Vertical turbulence
intensity profile. Lines delimit the ranges authorized by VDI 3783/12 (2000) for the rough class. Measured in wind
tunnel at 13m (model scale) from the entrance of the wind tunnel

239

5 Spectral content and integral length

scales
The spectra of turbulent fluctuations measured in the
wind tunnel corresponding to 50m full scale above
ground are presented on Fig. 7. They are compared to
the empirical laws based on field measurements over a
flat uniform terrain under neutral stability conditions and
published in Kaimal and Finnigan (1994):

f Svv  f 
u*2
f S ww  f 
u*2

1  33n 
17 n

1  9.5n 

5/ 3

U mean the mean

xx = uu
xx = vv
xx = ww

102

103

Lux [m]

Figure 8 : the longitudinal integral length scale of the

approach flow compared with the empirical law from
Counihan (1975) for z0  0.3m .

10

-1

6 Conclusions

Sxx u f / u *

10

-2

10

-2

-1

10
10
f u z / Umean

10

10

Figure 7 : Spectral density distributions of turbulence

for the three velocity components of velocity

Suu , Svv , S ww

, compared with Kaimal et al. (1972).

Fluctuations for the three components are slightly

larger than those expected from the empirical laws. It
was expected since field measurements were obtained
over a flat and moderately rough terrain. On the other
hand, the maxima of energy are located on lower
frequencies than expected. This feature is confirmed by
the values of the longitudinal integral length scale
(Fig. 8).

Lux

Lux

was calculated applying the autocorrelation

method on the time series of velocity fluctuations at

different heights above ground. The vertical distribution

Lux

is compared to the empirical law proposed by

Counihan (1975) :

Lu  z  C  z0  z

with

z0

101

10 -3
10

of

40

20

2.1n
1  5.3n5/ 3

velocity spectra, respectively and

velocity measured at z .

Wind tunnel
Counihan (1975)

60

5/3

n f z / U mean , with f the frequency [Hz],

S uu , S vv , S ww the longitudinal, lateral and vertical

of

80

102n

where

10

100

Z [m]

f Suu  f 
u*2

The experimental values are twice larger than

expected. No explanation was found so far since the
correlation curves did show a classical shape, with an
unambiguous zero-crossing behaviour. The exponential
fit technique (Teunissen, 1980) was also tested and
gave the same results.

1 n z0 

C 30 and 1 n 0.4 for a roughness length

0.3 m.

240

An atmospheric boundary layer over rough-type

terrain was modelled at a geometric scale of 1:400 on
the return circuit of the Malavard wind tunnel of LME.
The roughness length is assessed to 0.3m and a
power law exponent D of 0.23 is selected since it is
obtained in the portion which is the best adapted to our
applications to wind farms : 12m d z d 200m (1 turbine
diameter above the upper limit of the actuator disc).
Vertical profiles of turbulence intensities and the spectral
content of turbulence are globally in agreement with the
literature. Only integral length scale are twice larger than
expected for a rough terrain. No explanation was found
so far.

References
Counihan J. (1975). Adiabatic atmospheric boundary
layers: a review and analysis of data from the
period 1880-1972. Atmospheric Environment 9, pp.
871-905.
Kaimal, J.C., Finnigan, J. (1994). Atmospheric Boundary
Layer Flows. Oxford University Press.
Snyder, W.H. (1981). Guideline for fluid modelling of
atmospheric diffusion. US Environmental Protection
Agency. EPA-600/8-81-009, p. 185.
VDI-guideline 3783/12 (2000). Physical modelling of flow
and dispersion processes in the atmospheric
boundary layer Application of wind tunnels. Beuth
Verlag, Berlin
ESDU Engineering Sciences Data Unit, (1985).
Characteristics of atmospheric turbulence near the
ground. Part II: Single point data for strong winds
(neutral atmosphere). Item no 85020. ESDU
International, London.
Teunissen H.W. (1980) Structure of mean winds and
turbulence in the planetary boundaryr layer over
rural terrain. Boundary-Layer Meteorology 19, pp.
187-221.

Numerical Simulation of Room Air Motion, Physics and CFD Modeling

V. Esfahanian E. Moallem

Mechanical Engineering Faculty,

Tehran University, Tehran, Iran
Email : evahid@ut.ac.ir, Emoallem@yahoo.com

Abstract - Considering the recent works of

numerical
simulation
of
ventilation
flows,
Computational Fluid Dynamics (CFD) has gained
widespread applications in modeling indoor air flow
and heat and mass transfer in buildings. With
correct simulation of air flow in a ventilated space
there would be possible to predict a valid locally
heat transfer for each part of the walls in rooms in
order to have a correct prediction of whole building
heat dispersal.
Since the pioneering work of numerical simulation
by International Energy Agency (IEA) Annex 20
project and Luo investigation in comparing
numerical and experimental results, it was found
that CFD prediction can provide detailed
information on air flow velocity, turbulent intensity
and temperature distribution in ventilated spaces.

241

In this paper, Annex test room has been modeled

numerically and result has been compared to Luo
results and also against experimental data. It is
found that with ReNormalization Group (RNG) k-
turbulence model and local mesh refinement, this
model can give a good prediction of wall jet flow
issued from Hesco nozzle diffuser which was used.
A good agreement was found between obtained
results and Luo results and also experimental data
showing the validity of used model which could be
used for future applications.
The numerical simulation results of air flow pattern
are used to predict heat transfer coefficient between
air and walls and also satisfactory predicting of
temperature distribution of outside walls that
results in valid prediction of flow outside the
buildings.

Authors Index
Author

Pages

Author

Pages

Addepalli

79

Herbst

41

Aubrun

47, 237

Holtslag

175

Balcz

185

Izarra-Garcia

201

Balogh

93

Jaour

129, 231

Barlow

75, 113

Jonker

175

Bartoli

59

Kellnerov

231

Bezpalcov

221

Kristf

93

Borri

59

Kukadia

17

Borsani

59

Lajos

185

Bowker

63

Laporte-Daube

47

Brixey

63

Lawton

203

Builtjes

175

Leitl

1, 23, 27, 33, 41, 69, 75, 207, 213

Carter

105, 143

Lin

135

Cesari

157

Loredo-Souza

99

Contini

157

Makita

179

Czder

185

Mapurisa

87

De Paoli

99

Marsland

17

Denev

121

Miyagi

177

Devinant

47, 237

Moallem

241

Donateo

157

Mueller

225

Duan

149

Nagai

167

Dupont

47

Nakayama

165

Esfahanian

241

Ohba K.

179

Espaa

47, 237

Ohba M.

221

Fischer

213

Ohba R.

167

Franke

201

Ohta

165

Goricsn

185

Okabayashi

167

Gorle

53

Pardyjak

79

Gromke

121

Pascheke

113, 229

Gupta

111

Perret

197

Hall

17

Perry

63

Hancock

229

Petersen

105, 143

Hara

167

Planquart

53

Harms

27

Procino

59

Havens

15

Rcz

93

Hayashi

167

Rgert

185

Hayden

87

Repschies

33

Heist

63

Restorick

149

243

Author

Pages

Rix

207

Robins

87, 113, 157, 203

Ruck

121

Saathoff

111

Sanz Rodrigo

53

Sassa

177

Savory

149

Schatzmann

1, 23, 27, 33, 41, 69, 207, 213

Schettini

99

Schultz

69

Sedenkova

129

Sheppard

15

Shiau

7, 135

Spicer

15

Stathopoulos

111

Tamura

165

Tily

17

Van Beeck

53

Van Dop

175

Van Rooij

191

Vanweert

191

Von Bobrutzki

225

Walker

17

Wittwer

99

Wu

Yassin

231

Yoneda

167

244