Вы находитесь на странице: 1из 12

Journal of Hydrology 519 (2014) 32373248

Contents lists available at ScienceDirect

Journal of Hydrology
journal homepage: www.elsevier.com/locate/jhydrol

Modelling of green roof hydrological performance for urban drainage


applications
Luca Locatelli a,, Ole Mark b, Peter Steen Mikkelsen a, Karsten Arnbjerg-Nielsen a, Marina Bergen Jensen c,
Philip John Binning a
a
Technical University of Denmark, Dept. of Environmental Engineering (DTU Environment), Miljoevej, Building 113, 2800 Kgs Lyngby, Denmark
b
DHI, Agern All 5, 2970 Hrsholm, Denmark
c
University of Copenhagen, Rolighedsvej 23, 1958 Frederiksberg, Denmark

a r t i c l e i n f o s u m m a r y

Article history: Green roofs are being widely implemented for stormwater management and their impact on the urban
Received 20 December 2013 hydrological cycle can be evaluated by incorporating them into urban drainage models. This paper pre-
Received in revised form 1 October 2014 sents a model of green roof long term and single event hydrological performance. The model includes sur-
Accepted 11 October 2014
face and subsurface storage components representing the overall retention capacity of the green roof
Available online 17 October 2014
This manuscript was handled by
which is continuously re-established by evapotranspiration. The runoff from the model is described
Konstantine P. Georgakakos, Editor-in-Chief, through a non-linear reservoir approach. The model was calibrated and validated using measurement
with the assistance of Michael Bruen, data from 3 different extensive sedum roofs in Denmark. These data consist of high-resolution measure-
Associate Editor ments of runoff, precipitation and atmospheric variables in the period 20102012. The hydrological
response of green roofs was quantied based on statistical analysis of the results of a 22-year
Keywords: (19892010) continuous simulation with Danish climate data. The results show that during single events,
Green roofs the 10 min runoff intensities were reduced by 1036% for 510 years return period and 4078% for
Water Sensitive Urban Design 0.11 year return period; the runoff volumes were reduced by 25% for 510 years return period and
Modelling 1828% for 0.11 year return period. Annual runoff volumes were estimated to be 4368% of the total
Rainfallrunoff precipitation. The peak time delay was found to greatly vary from 0 to more than 40 min depending
Retention
on the type of event, and a general decrease in the time delay was observed for increasing rainfall inten-
Detention
sities. Furthermore, the model was used to evaluate the variation of the average annual runoff from green
roofs as a function of the total available storage and vegetation type. The results show that even a few
millimeters of storage can reduce the mean annual runoff by up to 20% when compared to a traditional
roof and that the mean annual runoff is not linearly related to the storage. Green roofs have therefore the
potential to be important parts of future urban stormwater management plans.
2014 Elsevier B.V. All rights reserved.

1. Introduction Green roofs are one of the many WSUD (Water Sensitive Urban
Design; the concept of water sensitive cities was presented by
Urbanization signicantly affects the natural landscape turning Wong and Brown, 2009), LID (Low Impact Development), SUDS
green areas into built environment. This process creates new (Sustainable Urban Drainage Systems), LIUDD (Low Impact Urban
impervious surfaces such as roofs, roads, cycling lanes, sidewalks, Design and Development) techniques aimed to improve storm-
public squares and parking areas which modify the natural water water management and address future climatic challenges. Green
cycle. Impervious areas increase stormwater runoff peaks and vol- roofs have the great advantage of not using new spaces; in fact
umes and reduce the time delay between peak rainfall and peak they can in some cases be retrotted onto existing traditional
runoff when compared to natural areas (Bengtsson, 2005). Current rooftops. The roof area in urban residential areas can be as high
urban drainage systems have limited capacity to deal with ooding as 4050% of the total impervious area (Palla et al., 2009;
and climate change will increase the risk of ooding from sewers in Lindblom et al., 2011; Vezzaro et al., 2012). Green roofs are claimed
urban areas (Larsen et al., 2009; Madsen et al., 2009). to reduce the risk of urban ooding and to re-establish a more
natural water balance (Bengtsson et al., 2005; VanWoert et al., 2005)
and to reduce water and contaminant loads to sewer systems
Corresponding author. Tel.: +45 4525 1432.
(Buccola and Spolek, 2011).
E-mail address: lulo@env.dtu.dk (L. Locatelli).

http://dx.doi.org/10.1016/j.jhydrol.2014.10.030
0022-1694/ 2014 Elsevier B.V. All rights reserved.
3238 L. Locatelli et al. / Journal of Hydrology 519 (2014) 32373248

The German Guidelines for the Planning, Construction and ow reduction of 60% for rain events of more than 22 mm accumu-
Maintenance of Green Roong (FLL, 2008) have suggested a classi- lated rain. Moran et al. (2005) reported 85% peak reductions.
cation of green roofs into two different categories: Several models have been developed for the hydrological per-
formance of green roofs. Hilten et al. (2008) used the soil moisture
 Intensive green roofs have thick soil layers (>15 cm) with large software HYDRUS-1D to simulate the single events runoff response
plants and moderate slopes, these are heavier and require reg- from green roofs and validated his results with data collected in
ular watering and fertilization. Georgia (USA). Palla et al. (2009) used SWMS_2D based on Richards
 Extensive green roofs can have soil covers with a thickness of as equation to simulate the variably saturated conditions within a
little as few cm and vegetation which requires hardly any main- typical green roof during individual storm events and validated
tenance. When the soil layer is thin and the weight of the green the model against data collected in Genoa (Italy). Zhang and Guo
roof is limited, extensive green roofs can often be placed on (2013) presented a physically based analytical probabilistic model
existing buildings without structural reinforcement and they to evaluate the average long term hydrological response of green
can reach slopes up to 4045% (FLL, 2008). roofs. Villarreal and Bengtsson (2005) analyzed single events run-
off from an extensive green roof in Sweden and derived a unit hyd-
From the 1990s green roofs have become common practice in rograph. Zimmer and Geiger (1997) proposed a linear/non-linear
several countries such as Austria, Switzerland and particularly Ger- reservoir model to compute the runoff from an experimental green
many where green roofs reached 13% of the at roofs in 2003 roof under constant rainfall intensities. Kasmin et al. (2010) used a
(Herman, 2003). Some countries have also released guidelines for simple conceptual model, with the runoff described by a non-linear
implementation of green roofs like the UK, Germany, United States, reservoir, in order to evaluate the performance of different meth-
New Zealand, Australia, etc. ods for modeling evapotranspiration. Sherrard and Jacobs (2012)
Green roof performance regarding stormwater management presented a ve parameter water balance model with a daily time
may vary among geographic regions due to varying climate, pre- step in order to reproduce the volume runoff on a daily and annual
cipitation patterns, building practices and green roof materials. basis. Jarret and Berghage (2008) presented one model for annual
During rainfall events the most important hydrologic mechanisms water balance and one for single event runoff. Stovin et al.
are the interception of rain by the vegetation layer, inltration and (2013) presented a model to quantify long term runoff using con-
retention/detention in the soil substrate, and retention/detention tinous simulation; results from the model showed volumetric
in the drainage layer. Any water in excess to the storage capacity retention values between 0.19 (cool, wet climate) and 0.59 (warm,
will be drained into an outlet and during non-rainy periods water dry climate) depending on local climate conditions in the UK.
stored in the green roof is lost through evapotranspiration. Other studies have derived empirical runoff relations from
Green roofs reduce stormwater runoff compared to conven- experimental data. Mentens et al. (2006) gathered experimental
tional roofs due to water retention and subsequent evapotranspira- data of green roof hydrological performance on seasonal and
tion. Volume retention depends on rainfall intensity distribution, annual bases from 18 different publications in Germany.
the initial moisture conditions and green roof characteristics (layer Bengtsson (2005) derived IntensityDurationFrequency curves
thickness, slope, materials, etc.), including the ability of the green for rainfallrunoff prediction from an extensive green roof in Mal-
roof to dry up. Volume retention also contributes to peak attenua- mo in Sweden. Moran et al. (2005) derived the rational coefcient
tion and delay. The following examples are based on observations based on green roof data from North Carolina (USA). Carter and
from experimental sites. Experimental sites in Germany showed Jackson (2007) used the Curve Number method for predicting
that the annual runoff from intensive green roofs was 1535% of green roof performances at large case studies in Georgia (USA)
the annual rainfall, whereas the performance of extensive roofs is and Getter et al. (2007) derived the curve number for the studied
reported to be 2075% of the annual rainfall (Mentens et al., green roof in Detroit (USA).
2006). In Sweden the runoff observed from thin extensive green A review of available WSUD models was provided by Elliott and
roofs was 46% of the annual precipitation (Bengtsson et al., Trowsdale (2007). The literature review shows that modeling
2005). In England Stovin et al. (2012) reported an overall retention approaches vary, from models for single event runoff to conceptual
capacity of green roofs of 50%. VanWoert et al. (2005) observed a models applied to simulate distributed urban runoff. None of these
60% retention in Detroit (Michigan), DeNardo et al. (2005) 45% in studies have presented a single and low computationally demand-
Pennsylvania, Voyde et al. (2010) 66% in Auckland (New Zealand), ing model that can continuously simulate both single events with
Monterusso et al. (2004) 49% and Carter and Rasmussen (2006) ne time resolution and the annual water balance of single green
78% in Athens (Georgia). roofs. The single event hydrological response of green roofs is
Volume detention, dened as temporary storage and subse- highly affected by initial moisture conditions which depends on
quent release, results in additional attenuation and time delay of evapotranspiration rates and can be estimated from a continuous
runoff peaks. Volume detention takes place when the initial mois- simulation; at the same time the long term performance is inu-
ture content is at eld capacity (micro- and mesosized pores are enced by single event runoff and evapotranspiration rates. Thus
water lled) and additional water percolates through macropores. it is relevant to develop a model that can continuously simulate
Peak attenuation and delay due to detention depend on the rainfall both single events and long term response. This paper is based
intensity distribution and the green roof characteristics, including on the preliminary model results presented in the conference
layer thickness, slope, poresize distribution, tortuosity, etc. of the paper of Locatelli et al. (2013).
materials. Observed data recorded from experimental green roofs The rst aim of this study is to present a simple conceptual
have been reported by several researchers. Peak runoff delay of model for the hydrological performance of green roofs that can
vegetated roofs was found to vary signicantly, for the majority be integrated into urban drainage models for evaluating the
of the observed rain events peak delay was found to be between impacts of implementing green roofs at a large scale (e.g. modeling
0 and 30 min when compared to traditional roofs (VanWoert of ood mitigation potential; modeling the potential for reduction
et al., 2005; Carter and Rasmussen, 2006; Simmons et al., 2008). of combined sewage volume and combined sewer overow). Low
Getter et al. (2007) observed only minimal runoff delay and computational costs and ne time resolutions are thus a require-
Villarreal and Bengtsson (2005) found peak delays to be around ment. Many of the previously developed green roof models such
1 min. DeNardo et al. (2005) showed considerable peak reductions as SWMS-2D and HYDRUS have a high computational cost and so
for low intensity rains. Stovin et al. (2012) observed a mean peak are not suitable for large scale urban water modelling.
L. Locatelli et al. / Journal of Hydrology 519 (2014) 32373248 3239

The approach used here is based on a deterministic lumped It has a 3  3 m2 surface, east oriented, a slope of 10 and sedum
rainfallrunoff conceptual model applied to a single green roof. vegetation growing on a 40 mm layer substrate drained by a
The model was calibrated and validated against observed runoff 40 mm of mineral-wool. The roof has a maximum fully wet weight
from three different extensive green roofs installed in Denmark. between 80 and 130 kg/m2. Runoff data from this roof are available
The second aim of the study is to simulate the hydrological perfor- from August 2010 to May 2012. The second green roof (GR2, Fig. 1
mance of green roofs in the long term. Statistical analysis of contin- right) has Sedum vegetation growing on a 30 mm substrate
uous simulation results was applied to determine the single event drained by a 40 mm drainage layer made of high porosity oxy-
peak time delay, peak and volume reductions and mean annual foam mats. This roof has a water saturated weight of 55 kg/m2.
runoff for a 22 year simulation period exposed to Danish rainfall The third green roof (GR3, Fig. 1 right) has sedum growing on
and climate data. The third aim of the study is to analyze the var- 60 mm high plastic trays lled with substrate soil and drained
iation of the mean annual runoff as a function of storage capacity through raised holes, placed approximately 10 mm above the bot-
and crop coefcient. tom. This has a water saturated weight of 45 kg/m2. A reference
The modeling approach developed in this paper is similar to traditional roof was also installed. The latter 3 roofs are connected
that employed in several other classical hydrological models, such to each other and have a surface area of 20 m2 each and slope of
as the NAM model of DHI (1999) and Gustafsson et al. (1999). The 4.3 oriented to the south-east and they are located in Copenhagen.
emphasis here, is therefore not on presenting a novel hydrological Runoff data are available from March 2011 to February 2012. The
model. Rather, it is on presenting and testing a modeling approach characteristics of the different roofs are summarized in Table 1.
for green roofs that is simple, computationally effective and capa- Each roof has a distinctly separate drainage and runoff cannot
ble of simulating the response to individual rainfall events and migrate among them. Runoff from the roofs was collected in sepa-
long term runoff. A comparison between the actual model and an rated tanks where the water depth was measured by pressure
existing rainfallrunoff model included in the commercial software transducers to provide a continuous record of runoff. Rainfall was
MIKE-URBAN was made. monitored using tipping buckets installed at the test sites at a dis-
tance of <50 m from the roofs. Meteorological data (wind speed, air
temperature, relative humidity, global radiation and atmospheric
2. Materials and methods pressure) were collected from the nearest available meteorological
stations which were located 10 km away for one green roof (GR1)
2.1. The experimental data and 3 km away for the other two green roofs (GR2 and GR3). The
rainfall data were logged by tipping bucket rain gauges with a
Three different extensive green roofs were installed in Denmark bucket volume of 0.2 mm and runoff data were recorded at
and runoff from each test setup was measured. Data from these 1-min time intervals while meteorological data were recorded
installations were provided by the different partners. Denmark every 1-h. Usually small distances (<1 km) are considered to be
has a cold climate, without a dry season (Peel et al., 2007) and with acceptable for weather stations; however given the lower spatial
an average precipitation of 600700 mm/year (Frich et al., 1997). variation of hourly averaged climate variables we assume that
The rst green roof (GR1, Fig. 1 left) is located in the city of Odense. the actual distances are acceptable. Single rain events were dened

GR1

GR3

GR2

Fig. 1. The 3 experimental extensive green roofs (GR1, GR2 and GR3). GR2 and GR3 are covered by the same vegetation but have different substrate thickness and drainage
media.

Table 1
Roofs characteristics.

GR1 GR2 GR3 Reference roof


Location Odense Copenhagen Copenhagen Copenhagen
Area (m2) 9 20 20 20
Slope () 10 4.3 4.3 4.3
Overall thickness (mm) 80 70 60
Vegetation type Sedum and herbs Sedum and herbs Sedum and herbs
Saturated weight (kg/m2) 130 55 45
a
Drainage (mm) 40 40
Drainage material Mineral wool Exapanded foam Holes in plastic trays
Substrate depth (mm) 40 30 60
Substrate material Soil Crushed clay, medium-ne peat, compost Crushed clay, medium-ne peat, compost
Water holding capacity (l/m2) 53 31 No data
a
Drainage holes are placed 1 cm above the bottom.
3240 L. Locatelli et al. / Journal of Hydrology 519 (2014) 32373248

Soil moisture in P
the porous media
Pe ET
max
Smax ET Overland ow
Surface storage S

Dmax
k1(h-Dmax)n1
FC
ET Dmax
Rmax Qss Detenon storage h

khn
Rmax
Runo
R
WP Subsurface

aRmax
storage

Fig. 2. Conceptual mass balance model of a green roof.

Fig. 3. Composite scaled sensitivities of the model parameters for the water balance and peak runoff. Long term simulation (left) and a selected single event simulation
(right).

as rainfall periods separated by more than an hour. One-hour inter- based on the mass balance equation for the system was imple-
event period is the denition adopted in Denmark for single rain mented in MATLAB. Some of the model parameters can be esti-
events based on the considerations of Arnell et al. (1984). Single mated from physical data (Rmax, Smax, Kc), but nal calibration to
event runoff was dened to begin with the corresponding rain hydrological data is needed.
event and to end at the beginning of the following rain event. This The mass balance for each of the storages is described as:
denition might cause some runoff events starting with non-zero
runoff values resulting in single event runoff volume > rainfall vol- ds
qin  qout 1
ume. This is not considered a problem in the present analysis since dt
the observed green roofs cease runoff within 12 h meaning that
where s is the storage depth [mm], qin and qout are the inow and
after 1 h most of the water is drained. Results from Fig. 8 (Right)
outow intensities normalized by the roof area [mm/min] and t is
show only couple of events with runoff volume > rainfall volume
time [min].
and the difference in volume is approximately 1 mm.
Rainfall intercepted by the vegetation layer of the green roof is
Rainfall data were corrected according to Allerup et al. (1998) to
referred to as surface storage and Smax (see Fig. 2) indicates its max-
take into consideration the inuence of wind speed, rain intensity,
imum capacity. The capacity of the surface storage is continuously
wetting and sheltering effects of the rain gauge.
re-established through evaporation. When the maximum surface
storage Smax is exceeded, the effective precipitation Peff is driven
2.2. The model as inltration into the subsurface storage.
The volume of water that can be stored (retained) both in the
A deterministic lumped rainfallrunoff conceptual model was green roof substrate and in the drainage layer (and eventually
used. Lumped conceptual models (grey box models), rather than other built-in layers) is referred to as subsurface storage R. The
physically distributed (white box models) or analytical models drainage layer is normally constructed to ensure that the roof
(black box models), require lower computational costs and can has sufcient underdrain and only few commercial products
have the advantage of easier interpretation of model parameters. introduce storage (retention) elements into the drainage board.
The conceptual model divides the green roof into 3 different water The drainage layer of one green roof (GR1) in this study is made
storages connected to each other and is shown in Fig. 2. A model of mineral wool, which has a high retention capacity, and for
L. Locatelli et al. / Journal of Hydrology 519 (2014) 32373248 3241

Fig. 4. Rainfallrunoff intensity from the 3 extensive green roofs during three different rain events within the calibration period.

Fig. 5. Simulated and observed accumulated rainfallrunoff depth for the summer period (left) and for the winter period (right). Data from GR1.

Fig. 6. Simulated 10 min peak intensities per event as a function of the return period (left). Simulated 10 min peak reductions per event as a function of different return
periods (right).

this reason we decided to keep the drainage layer to account for The detention storage temporarily stores water up to a maxi-
the subsurface storage. Rmax represents the maximum capacity of mum volume Dmax. The detention storage represents the excess
the substrate and drainage layer (and other built-in layers) water that cannot be held against gravity by the green roof and
together; this can be estimated as the difference between the therefore will runoff as subsurface ow through the drainage layer.
water content at eld capacity hFC (after drainage, macropores Theoretically the maximum capacity of the detention storage Dmax
empty) and the permanently retained water content hWP cannot exceed the difference between the saturated water content
(comparable to permanent wilting point, mesopores empty). and the water content at eld capacity. When the maximum capac-
R = 0 represents the permanently retained moisture in the green ity Dmax is exceeded saturation overland ow runoff occurs. The
roof (it does not correspond to oven dry conditions) and may runoff from the detention storage is described by the non-linear
vary in response to atmospheric variables. The water content reservoir method:
in the subsurface storage is continuously reduced by evapotrans- Runoff k  h ;
n
h  Dmax
piration. From the subsurface storage Qss is diverted into the n1
2
detention storage. Runoff k  Dnmax k1  h  Dmax ; h > Dmax
3242 L. Locatelli et al. / Journal of Hydrology 519 (2014) 32373248

Fig. 7. Simulated single event runoff volume as a function of the return period (left). Simulated single event runoff volume reductions as a function of different return periods
(right).

Fig. 8. Simulated single event intensity as a function of the 10 min rainfall return period (left). Simulated single event runoff volume as a function of the rainfall volume
return period (right).

where n, k are the routing parameters for subsurface runoff through Dingman (2002) in order to take the roof slope and orientation into
the drainage layer and n1, k1 are the routing parameters for satu- consideration.
rated overland ow. This method was proposed by Zimmer and The crop coefcient Kc depends on the type of vegetation and
Geiger (1997) for the design of multilayered inltration systems. takes into account the differences in stomatal resistance of the leaf
Due to the high conductivity of the subsurface materials all the rain and aerodynamic resistance through the body of the plant (Allen,
is assumed to inltrate and percolate vertically to the drainage 1998). The reduction factor F accounts for environmental stress.
layer. However the water can accumulate in the system up to a cer- In the actual model evapotranspiration rates are reduced due to
tain volume beyond which saturated overland ow occurs. water shortage. The reduction factor F is equal to 1 for both the
Evaporation and transpiration continuously restore the capacity surface and detention storages, whereas for the subsurface storage
of all the storages. Evaporation rstly acts in the surface storage, if F = 1 for a water content R > w Rmax and then decreases linearly to
the surface storage falls below zero then water is withdrawn from zero between R = w Rmax and the minimum water content R = 0
the detention storage and when the detention storage is empty (0 6 F 6 1). The quantity w Rmax can be interpreted as the water
evapotranspiration takes place in the subsurface storage. The model content at the stomatal closure point.
assumes no evapotranspiration during rain events and it does not The ow recharging the detention storage depends on the lling
distinguish between evaporation and transpiration. ratio of the subsurface storage.
The actual ET rates are calculated based on the FAO Penman
(
Monteith equation (Allen, 1998): Peff R=Rmax a
; R=Rmax > a
Qss 1a 4
ET ET 0  K c  F 3 0; R=Rmax  a

where Kc is the crop coefcient (vegetation coefcient), F is the where a Rmax is the threshold water content above which water
reduction factor due to water shortage and ET0 is the reference starts draining from the subsurface storage (0 6 a < 1). When the
evapotranspiration. The input parameters for ET0 are the observed subsurface storage is full then Qss = Peff. The parameter a is a thresh-
wind speed, atmospheric pressure, relative humidity, global radia- old value that changes the hydrograph at the beginning of the rain
tion and air temperature. The clear sky solar radiation used in the event by allowing runoff from the green roof even if the subsurface
calculation of ET0 (Allen, 1998) was corrected according to storage still has some capacity.
L. Locatelli et al. / Journal of Hydrology 519 (2014) 32373248 3243

2.3. Calibration, validation and sensitivity analysis The return period T [years] was calculated using the Weibull
(1939) plotting position formula:
The model parameters for the green roofs were determined by
y1
inverse modeling with a calibration period of approximately T 6
r
23 months during the summer period. The calibration procedure
used the Shufed Complex Evolution Metropolis algorithm (Vrugt where y is the duration of the time series in years and r is the rank
et al., 2003). of the observed event.
Each of the three extensive green roofs was calibrated according
to three different objectives: 2.5. Annual runoff as a function of storage capacity and crop
coefcient
(1) Peak runoff. The objective function to be minimized was the
sum of the squared peak errors. The peak error was calcu- The variation of the annual runoff as a function of the total
lated for each event as the difference between the observed green roof retention capacity (total storage capacity = surface stor-
and the computed peak runoff. Only runoff peaks above age + subsurface storage; detention storage is not considered since
4 mm/h were taken into account. The threshold of 4 mm/h it is low sensitive parameter (see Fig. 3) for annual runoff) and
was chosen in order to have at least 46 peaks to be cali- for different crop coefcients Kc was simulated. The total storage
brated and still avoiding the inuence of small peaks. capacity was assumed to vary between 2.8 and 40 mm and the crop
(2) Water balance. The objective function to be minimized was coefcient Kc = 0.60; 0.89; 1.20; 1.50.
the sum of the squared water balance error where the error
was dened to be the difference between the accumulated 2.6. Model with MIKE URBAN
observed and computed runoff. The water balance error
was calculated for periods with durations of approximately The single event and long term hydrological response from the
7 days each. green roof (GR1) was also computed using the commercial soft-
(3) Peak runoff and water balance. The objective function to be ware MIKE URBAN (Andersen et al., 2004; DHI, 2009). Mike Urban
minimized included peak runoff and water balance together includes a model referred to as NAM (Nedbr-Afstrmnings-Model.
with different weights and was dened as: In English: rainfallrunoff model), which was developed to compute
the runoff from pervious areas and it is furthermore used to simu-
" # late inltration-inow to urban drainage systems. It is a determin-
XNc
wi Xm
C obsi;t  C simi;t 2
5 istic lumped model with linear reservoirs and threshold functions
i1
m t1 r2i;t that allow to simulate a range of phenomena. A detailed descrip-
tion of the NAM model and its parameters was given by Nielsen
where Nc is the number of objectives considered (in this case Nc = 2, and Hansen (1973) and in the manual DHI (1999). The input data
peak runoff and water balance), m is the number of observations for required for the NAM model are ET rates [mm/day], rainfall inten-
each objective, Cobs are the observed values, Csim are the simulated sities [mm/min] and the horizontal projected area of the green roof
values, r is the standard deviation of observations and wi are the [m2]. The input ET rates had an hourly time step and were obtained
weighting parameters. from Equation 3. The time step for the NAM model was set to 1 min
The reference hard roof was also calibrated using the green roof in order to have a direct comparison to the actual model approach.
model. Subsurface and detention storage capacities were set to The chosen simulation period was the month of June 2011. The
zero (Rmax = Dmax = 0). NAM model input parameters were calibrated by trial and error.
The model was validated for each of the three extensive green
roofs and the validation period (35 months) made use of all the 3. Results and discussion
remaining observation data (data out of the calibration and winter
periods). The standard deviation of the peak error and water bal- 3.1. Calibration, validation and sensitivity analysis
ance error during the validation period was calculated.
Sensitivity of the model parameters was analyzed using Com- The calibrated values for the reference roof are Smax = 0.5 mm,
posite Scaled Sensitivities (CSS) (Hill, 1998). The CSS were calcu- k1 = 0.215 mm1min1 and n1 = 1.67. The results of the calibration
lated considering four different objectives: (1) peak ows during for the three extensive green roofs are given in Table 2. The over-
a single event simulation; (2) runoff volume during a single event land ow from the green roofs was assumed to have the same
simulation; (3) peak ows during a long term simulation; (4) run-
off volume during a long term simulation. Table 2
Calibration results.

Calibration objective k Rmax Kc a a Dmax


2.4. Long term simulations
(mm min)1 (mm) (mm)
GR1
The long term simulations were based on rainfall and climate
Peaks 0.0038 21.2 0.93 0.65 10
data from Copenhagen in the period 19892010. The rainfall data Water Balance 0.0038 20.4 0.88 0.65 10
were taken from a tipping bucket rain gauge in Copenhagen and 0.60 Peaks; 0.40 WB 0.0040 21.5 0.89 0.53 10
corrected as described in Section 2.1. The rainfall data were GR2
recorded for every 0.2 mm of rain. Global radiation, wind speed Peaks 0.018 11.0 1.44 0.85 4.2
and temperature were obtained with a daily time step from the Water Balance 0.018 8.8 0.80 0.85 4.2
Danish national climate grid data (Scharling, 1999,2001). Atmo- 0.60 Peaks; 0.40 WB 0.022 9.7 0.85 0.85 4.2

spheric pressure and relative humidity were taken from a measur- GR3
ing station in Copenhagen. The clear sky solar radiation in the Peaks 0.065 9.7 0.83 0.90 3.5
Water Balance 0.065 10.6 0.90 0.90 3.5
calculation of ET0 was corrected as explained in Section 2.2. It
0.60 Peaks; 0.40 WB 0.075 10.8 0.87 0.90 3.5
was assumed that the uncertainty in the model predictions is the
a
same as the uncertainty calculated for the validation period. The crop coefcient was calibrated within the period AprilOctober.
3244 L. Locatelli et al. / Journal of Hydrology 519 (2014) 32373248

routing parameters (k1 and n1) as the reference hard roof. During to the drainage layer (unless maximum detention storage Dmax
the observation period few events with overland ow occurred was exceeded) and that interow through the substrate layer is
from GR2 and GR3, and no overland ow was registered from negligible (in GR3 interow is not possible due to trays forming
GR1. Overland ow was assumed to occur when the observed run- vertical barriers). The latter assumption might not be valid when
off hydrograph suddenly jumped to the same intensity as the substrate thicknesses become larger than those of the tested roofs
incoming rainfall. The value of Dmax for GR1 was thus assumed to (>40 mm) and the interow through the substrate layer might
be just above the maximum simulated detention height during become larger.
the calibration period; this assumption was made in order to avoid The time delay due to percolation processes was not considered
overestimation of the parameter Dmax which could result in higher in the model, nevertheless the time of the observed peaks matched
peak reductions (simulations will be in the safe side). The Dmax the computed one (12 min). This means that percolation is fast
parameter for roofs GR2 and GR3, was calibrated for a couple of through the substrate layer due to the highly porous and granular
events. The parameters k = 2, Smax = 0.8 and w = 0.5 were xed materials (presence of macropores, low tortuosity).
due to high parameter correlation. The retention storages Smax Accumulated rainfall and observed and simulated runoff were
and Rmax were found to be highly correlated meaning that they computed for GR1 and shown in Fig. 5. Results show good match
could be simplied into a single storage compartment; however between observed and simulated values. Two different graphs
for a more exibility of the model (i.e. one could have a function are shown in Fig. 5, one for the summer period (April 2011Octo-
describing the seasonal variation of the interception storage) they ber 2011) and one for winter (October 2011April 2012). The crop
were kept separated. The calibrated values of Dmax are in the same coefcient for the winter period was assumed to be the same as the
range as those reported by Villarreal (2007) for green roofs with calibrated Kc from Table 2. Proper calibration for the winter period
30 mm substrates. The sensitivity analysis (Fig. 3) shows that in was not attempted due to the fact that precipitation as snow was
the long term hydrological performance (continuous simulations) not accounted; nevertheless the modelled runoff is shown to
the most sensitive parameters for the water balance are the crop approximate the observed data well (Fig. 5). The annual runoff of
coefcient Kc and the subsurface storage Rmax, whereas for individ- the green roof GR1 was observed to be 0.47 whereas observations
ual peaks in the long time series the most sensitive parameters are from GR2 and GR3 were not available due to the lack of continuous
k, n and Dmax which control the slope of the hydrograph and Kc and data set for a whole year. Fig. 5 shows that there is a seasonal var-
Rmax, which determine the initial moisture conditions and available iation in retention. Retention is 53% for the period AprilOctober
storage. For the single events (without continuous simulations), and 35% for the period NovemberMarch.
the most sensitive parameters to the water balance were the sub- The calibrated crop coefcients for the observed green roofs are
surface storage Rmax and the initial moisture conditions whereas in agreement with the range of values found in the literature.
for the peaks the most sensitive parameters are k, n and Dmax which Lazzarin et al. (2005) experimentally found crop coefcients for
control the hydrograph shape, and Rmax and the initial moisture sedum to decrease with decreasing soil relative humidity. For
conditions which affect the initial loss. well-watered conditions they observed crop coefcients to vary
The non-linear reservoir method was also employed by Kasmin between 0.42 and 0.65 and for stressed water conditions between
et al. (2010) to simulate the runoff from an observed extensive 0.16 and 0.39. Sherrard and Jacobs (2012) calibrated their model
green roof and they found routing parameters n = 2 and with experimental data and obtained a crop coefcient for sedum
k = 0.030 mm1 min1 which are of the same order of magnitude of 0.53. Rezaei et al. (2005) conducted green-house experiments
as those found in our calibration. and found crop coefcients of 0.74 in winter and 1.97 in fall/spring.
Table 2 shows that peaks had 0.6 weight and water balance 0.4. Schneider et al. (2011) presented average sedum crop coefcients
Several optimizations with different combinations of weights were between 1.1 and 1.9 based on experiments. Seasonal variation of
carried out and it was observed that reducing the weight on the the crop coefcient in the present study was assumed not to be sig-
peaks would signicantly worsen the modeled peaks, whereas nicant due to the limited transpiration of Sedum crop. Actual
the water balance would keep more stable. This is why peaks were model results (see Fig. 5) showed that using the same crop coef-
given a higher weight. cient for the whole year did reproduce observed runoff for the
The model validation results are reported in Table 3. It should 12 month continuously; this supported the assumption of no sea-
be noticed that the retention component was not validated specif- sonal effects on the crop coefcient.
ically for single events, however the observed runoff was shown to Theoretically the crop coefcient should only account for the
be continuously reproduced by the model (Fig. 5) and the objective differences in crop canopy and aerodynamic resistance (Allen,
function for the water balance considered the runoff volume every 1998). However the crop coefcient in this study might be overes-
710 days. timated due to the effect of roof elevation into the atmospheric
Examples of the computed and observed runoff hydrographs boundary layer and the enhanced turbulence generated by the roof
and the corresponding hyetograph for the 3 extensive green roofs slope and neighbouring structures. GR1 might also include the
are illustrated in Fig. 4. These examples show that the non-linear effect of a possible heat ux from the wooden house below the
reservoir routing method adopted in the model suitably describes green roof (see Fig. 1).
the runoff from the 3 observed green roofs. In the present model it The parameters presented above are case-specic. In order to
was assumed that, due to the high hydraulic conductivity of the estimate some of the parameter values for a previously-unmoni-
substrate materials, all the water inltrates and percolates down tored roof the authors suggest the following. The maximum capac-

Table 3
Validation results.

GR1 GR2 GR3


Standard deviation Standard deviation Standard deviation
Calibration target WB (mm) Peaks (mm/min) WB (mm) Peaks (mm/min) WB (mm) Peaks (mm/min)
Peaks 2.4 0.024 3.5 0.028 2.9 0.043
Water Balance 1.6 0.027 1.8 0.041 2.2 0.056
Peaks and WB 1.9 0.026 2.3 0.035 2.6 0.051
L. Locatelli et al. / Journal of Hydrology 519 (2014) 32373248 3245

ity of the substrate and drainage layer Rmax could be calculated as observed time series) the maximum achievable volume reduction
the difference between a physical measure of eld capacity and the would be 40%.
permanently retained water content; the crop coefcient Kc could Fig. 8 shows the single event 10-min runoff intensities and run-
be determined from literature. For the parameters a, Dmax and k, off volumes as a function of the rainfall return period (runoff inten-
calibration is needed, which could be achieved by use of rainfall, sities and volumes are associated with the corresponding rain
runoff and moisture data collected during controlled experiments. event). Results show that the green roof response to single events
has a high variability. This highlights the importance of continuous
modelling when forecasting the response of green roofs. For
3.2. Long term simulations instance if we look at the intensity for return periods between 1
and 10 years (Fig. 8, left) there is a great variation in the green roof
The simulated 10 min runoff intensity as a function of the response with some of the events that show similar runoff inten-
return period is illustrated in Fig. 6 (left). The 10 min intensity is sity as the incoming rainfall and others where the runoff intensity
especially relevant for local sewer surcharge problems, i.e. for is almost nil.
small catchments with short time of concentration. Extensive The peak time delay dened as the time lag between the com-
green roofs are most effective for short duration storms because puted runoff peak from the green roof and the one from the refer-
they have a small detention capacity; shorter duration storms ence roof is shown in Fig. 9 as a function of the 10 min rainfall
(i.e. 1 min) result in higher peak ow reductions and vice versa. return period. The peak time delay is relevant for calculating
The results show that the observed extensive green roofs reduce reductions in sewer surcharges. Results show that green roofs
single event 10 min peak runoff ow and volume when compared can delay peak ows from as little as 0 min up to more than
to the reference traditional roof. Peak runoff ows from GR2 and 60 min (peak delays higher than 40 min were simply shown as
GR3 approach those of traditional roofs for return periods above 40 min, this was done to simplify the gure and to avoid mislead-
710 years whereas GR1 has higher peak reductions. Fig. 6 (right) ing values. Some events where no runoff is observed actually have
shows the peak reduction for 4 different ranges of return periods. It no time delay by denition. Other events resulted in more than
can be seen that peak ow reductions decrease with increasing the 200 min delay, which is a misleading result because it is an artefact
event return period. The 10 min peak reductions are 4078% for of the method used to calculate peak time delay). Peak time delay
return periods between 0.1 and 1 and 1036% for return periods variability depends on initial available storage, rainfall patterns,
between 5 and 10. The results show that the best performing green detention storage capacity and material hydraulic conductivity.
roof is GR1, which has the thickest substrate and drainage layer. An overall trend of decreasing time delay with increasing the rain-
Overland ow during the simulation period was observed twice fall return period is seen. Nevertheless for some events with return
for GR1, 9 times for GR2 and 22 times for GR3. Villarreal (2007) periods above 1 year the model indicates delays above 20 min.
observed the 1 min peak intensity runoff to be on average 47% of The annual runoff dened as the ratio between the runoff and
that of a reference hard roof for a real rain event of 2 years return the incoming rainfall in 1 year period is shown with a box plot in
period in Sweden. He also estimated, using the unit hydrograph Fig. 10. The annual runoff reduction is useful for estimating dis-
method, 1 min peak runoff reductions to be 64% for a 5 years charge reductions to the sewer system. Results show that annual
return period event and 63% for a 10 years events. Peak reduction runoff ratios vary between 43% and 68% for the given extensive
depends on many factors such as the initial available storage, rain- green roofs. Fig. 10 also reports the values presented by Mentens
fall patterns and the detention capacity of the roof, which depends et al. (2006) which are based on observations from 121 different
on the conductivity and thickness of the layers. In general, the extensive green roofs in Germany. The annual runoff from green
higher the detention capacity the higher the peak intensity reduc- roofs is considerably lower compared to that of traditional roofs.
tion and the higher the hydraulic conductivity of the materials the The annual runoff was computed assuming a constant crop coef-
lower the peak reduction. Initial available storage and rainfall pat- cient throughout the year even though literature values suggest
terns are more correlated when a rainfall peak occurs at the begin- that winter crop coefcients might be 050% smaller. Nevertheless
ning of the event, where initially low moisture content in the green results are not expected to vary much since the rainfall depth dur-
roof helps reducing the peak; conversely if the rainfall peak ing the winter months in Denmark is only 23% of the whole year
occurred at the end of the event when the green roof had been sat- (Thomsen, 2011) and evapotranspiration rates in this period are
urated, initial moisture content would not matter.
The simulated single event rainfallrunoff volumes as a func-
tion of the return period are illustrated in Fig. 7 (left). These gures
can be used to estimate the stormwater volume reduction during
single events. Fig. 7 (right) shows the volume reductions for 4
ranges of return period. The single event volume reductions
decrease with increasing the return period. The reductions are
1828% for 0.11 year return period and 25% for 510 year return
period. Similar conclusions were found by Stovin et al. (2012) who
showed that retention during single 10 year return period events
could vary between 0 and 80% depending on initial moisture con-
ditions. Voyde et al. (2010) also concluded that the single event
retention is likely to decrease with increasing the event return per-
iod. The simulated runoff volume reductions only depend on the
available storage at the beginning of the rain event. Nevertheless
Getter et al. (2007) and Sherrard and Jacobs (2012) observed reten-
tion to change depending on the rainfall intensity. It is important
to notice that the higher the rainfall depth the lower the potential
for runoff volume reduction. In fact if we assumed to have a max- Fig. 9. Single event peak time delay as a function of the 10 min rainfall return
imum storage available of 20 mm (similar to GR1) and a rainfall period. Peak time delay above 40 min is simply shown as 40 min value (for
volume of 50 mm (710 years return period according to the illustration purposes).
3246 L. Locatelli et al. / Journal of Hydrology 519 (2014) 32373248

GR2 and GR3 have very similar annual runoff volumes. This is
because the most sensitive parameters to the long term water bal-
ance (Rmax and Kc; see Fig. 3) are similar for the two roofs. This is
reasonable since these two green roofs have similar substrate
material and vegetation, and are placed on the same structure.
The model results presented in this paper were based on
extrapolation of a model calibrated with limited eld data. During
the observation period the highest 10 min rainfall intensity regis-
tered had a return period of T  2 years whereas the single event
rainfall volume recorded had a return period in the range of
T  8 years. This means that the modeled runoff associated with
higher return period events might be especially uncertain (a higher
uncertainty compared to the one presented for the validation per-
iod in Table 3). The modeled single event runoff volume was shown
to be only function of the initial moisture content and therefore
results can be considered reliable even at higher return periods.
Fig. 10. Simulated average annual runoff over 22 years. The box-whiskers plot The modeled single event peak runoff associated with high return
indicate inter-annual variation. The gure also includes the results from Mentens period events likely occurs as overland ow; since the model
(2006) derived from observation of 121 different extensive green roofs.
accounts for overland ow which was shown to happen even dur-
ing the observation period, results can be assumed to be reliable
and most likely not overestimating peak reductions.

3.3. Annual runoff as a function of storage capacity and crop


coefcient

Fig. 11 shows the variation of the mean annual runoff as a func-


tion of the total green roof retention capacity and for different crop
coefcients. It can be seen that increasing the retention capacity
reduces the annual runoff and the relation between the two is
not linear. Even few mm of storage can result in signicant reduc-
tions in annual runoff as shown by Mentens et al. (2006) who
experimentally observed annual runoff from gravel-covered roofs
to be 7085% of the total annual rainfall. Similar conclusions were
also found by Berghage and Jarret (2007). Lastly it can be seen that
increasing the crop coefcient by changing the type of vegetation
reduces the mean annual runoff. Changing the crop species might
require different substrate depth and composition which could not
Fig. 11. Simulated mean annual runoff as a function of the green roofs total storage
be taken into account into the actual simulation.
capacity (surface storage + subsurface storage).

3.4. Model with MIKE URBAN


about 10% of the ones in summer. Moreover Fig. 5 showed that
A comparison between the computed single event runoff with
using the same crop coefcient for the summer and winter period
the model and the Mike Urban software is shown in Fig. 12 (left).
was a reasonable assumption. It should be noted that crop transpi-
The results show that both models are suitable for reproducing
ration is not that important for Sedum species; nevertheless it
the observed runoff from green roofs during single rain events.
would become relevant when non-succulent species are used.
The NAM model integrated in MIKE URBAN routs runoff with the

Fig. 12. Comparison between the actual model and the NAM model integrated into MIKE URBAN. 10 min rainfall-runoff intensity from a single rain event (left). Accumulated
rainfall-runoff for the month of June 2011 (right). Data from GR1.
L. Locatelli et al. / Journal of Hydrology 519 (2014) 32373248 3247

linear-reservoir method whereas the present model uses the non- 40 min, while for 510 year return period events the simulated
linear reservoir. The parameter n > 1 changes the curvature of the time delay was less than 10 min. The mean annual runoff was
computed hydrograph, during peak ows it gives a steeper shape found to decrease with increasing the roof retention capacity and
and during the descending hydrograph a milder one. Long term crop coefcient. Finally a comparison between the developed
accumulated rainfallrunoff was also compared and shown in model and the rainfallrunoff model (NAM) integrated into the
Fig. 12 (right). Results show that the two models behave similarly. software MIKE URBAN showed similar results both during single
The NAM has a similar conceptual model based on several storages events and long term hydrological response of green roofs.
which are continuously restored through evapotranspiration. The The long-term continuous simulation allowed us to classify the
actual evapotranspiration rates in the NAM model vary linearly performance of green roofs for both annual water balance and for
from 100% at R = Rmax to 0% when R = 0. This is the equivalent to single events. Annual runoff coefcients were introduced for green
setting w = 1 in the actual model. roofs. For single events, variables like initial moisture content and
rainfall patterns make runoff analysis quite uncertain and so con-
tinuous long term simulations are recommended.
4. Future research prospects

The model is designed to be simple so that it can be incorpo- Acknowledgments


rated into large scale simulations of whole cities. Future research
could model the impact of retrotting green roofs at a large scale. The authors thank the Danish Council for Research that
The presented model could also be validated for intensive green nanced the present research through the project BIV (Byer I Vand-
roofs. The model is uncertain particularly for peaks resulting from balance/Cities in water balance), Annette Brink-Kjr from Vand-
high return period events where data were not available; the center Syd who provided the green roof (GR1) runoff data, and
model should be tested when such data becomes available. It Nykilde APS and HOFOR for nancing green roof experiments in
would also be useful to determine how the model parameters (a, Copenhagen (GR2 and GR3).
Dmax and k) can be related to roof characteristics such as the slope,
thickness and material, pore size distribution, length and width. References

Allen, R.G., 1998. Crop evapotranspiration: guidelines for computing crop water
5. Conclusion requirements.
Allerup, P., Madsen, H., Vejen, F., 1998. Standardvrdier (196190) af
nedbrkorrektioner (Standard values of rainfall corrections). Danish
The presented model for the hydrological response of green
Meteorological Institute, Copenhagen.
roofs is demonstrated to reproduce runoff data from 3 different Andersen, H.S., Tamaauskas, H., Mark, O., 2004. The full urban water cycle
extensive green roofs in Denmark. Both the single event and the modeling with MIKE URBAN. 7th Urban Drainage Modelling, Dresden, Germany.
long term hydrological performance can be continuously simu- Arnell, V., Harremos, P., Jensen, M., Johansen, N.B., Niemczynowicz, J., 1984. Review
of rainfall data application for design and analysis. Water Sci. Technol. 16 (89),
lated. Model results for runoff peak and volume reductions, and 145.
peak time delay associated with high return period events were Bengtsson, L., 2005. Peak ows from thin sedum-moss roof. Nord. Hydrol. 36 (3),
shown. The model is divided into surface, subsurface and detention 269280.
Bengtsson, L., Grahn, L., Olsson, J., 2005. Hydrological function of a thin extensive
storage. The surface and subsurface storages represent the overall green roof in southern Sweden. Nord. Hydrol. 36 (3), 259268.
retention capacity of the green roof whereas the detention storage Berghage, R., Jarret, A., 2007. Quantifying Evaporation and Transpirational Water
represent the amount of water that can be temporarily stored in Losses from Green Roofs and Green Roof Media Capacity for Neutralizing Acid
Rain. The Pennsylvania State University and North Carolina State University.
the system. All the rainfall is assumed to inltrate into the sub- Buccola, N., Spolek, G., 2011. A pilot-scale evaluation of greenroof runoff retention,
strate, unless the maximum detention capacity is exceeded. Runoff detention, and quality. Water Air Soil Pollut. 216 (14), 8392.
from the model is assumed to occur either as subsurface ow or, in Carter, T., Jackson, C.R., 2007. Vegetated roofs for stormwater management at
multiple spatial scales. Landscape Urban Plann. 80 (12), 8494.
case the maximum detention storage capacity is exceeded, as sat- Carter, T.L., Rasmussen, T.C., 2006. Hydrologic behavior of vegetated roofs. J. Am.
urated overland ow. Runoff time delay due to percolation through Water Resour. Assoc. 42 (5), 12611274.
the substrate layer and saturated ow in the drainage layer was DeNardo, J.C., Jarrett, A.R., Manbeck, H.B., Beattie, D.J., Berghage, R.D., 2005.
Stormwater mitigation and surface temperature reduction by green roofs.
not considered in the model since the simulated timing of peak
Trans. ASAE 48 (4), 14911496.
ows matched the observed ones, suggesting that for such thin lay- DHI, 1999. NAM Model Documentation.
ers and high conductivity materials this process is fast. Evapotrans- DHI, 2009. Collection System. User Manual, Dhi Water & Environment. In: Agern
piration continuously regenerates the capacity of the different All 5, Hrsholm, Denmark.
Dingman, S.L., 2002. Physical Hydrology. Prentice-Hall Inc., ISBN 1-57766-561-9.
storages between rainfall events. The model has a ne temporal Elliott, A.H., Trowsdale, S.A., 2007. A review of models for low impact urban
resolution and low computational costs so that it can be integrated stormwater drainage. Environ. Modell. Software 22 (3), 394405.
into urban drainage models to simulate the impact of city-wide FLL, F.L.L., 2008. Guidelines for the planning, Construction and Maintenance of
Green Roong. Green Roong Guideline. Germany.
implementation of green roofs. Frich, P., Rosenrn, S., Madsen, H., Jensen, J.J., 1997. Observed Precipitation in
Intensity-frequency and volume-frequency curves were calcu- Denmark, 196190. Danish Meteorological Institute, Copenhagen.
lated using a statistical analysis of a 22 year continuous simulation Getter, K.L., Rowe, D.B., Andresen, J.A., 2007. Quantifying the effect of slope on
extensive green roof stormwater retention. Ecol. Eng. 31 (4), 225231.
results with Danish climate inputs. Results showed peak ows and Gustafsson, L.-G., Hernebring, C., Hammarlund, H., 1999. Continuous Modelling of
single event runoff volumes reductions to decrease with increasing Inow/Inltration in Sewers with MouseNAM 10 years of experience. In: 3rd
return period of the event. Single events 10 min runoff intensities DHI Software Conference, 1999.
Herman, R., 2003. Green roofs in Germany: yesterday, today and tomorrow. In: 1st
were reduced by 1036% for 510 years return period and 40 North American Green Roof Conference: Greening Rooftops for Sustainable
78% for 0.11 year return period; the runoff volumes were reduced Communities, Chicago.
by 25% for 510 years return period and 1828% for 0.11 year Hill, M.C., 1998. Methods and guidelines for effective model calibration. Water-
resources investigation report 984005, U.S. Geological Survey.
return period. The simulated average annual runoff was found to
Hilten, R.N., Lawrence, T.M., Tollner, E.W., 2008. Modeling stormwater runoff from
be 4368% of the one from reference roofs (runoff coef- green roofs with HYDRUS-1D. J. Hydrol. 358 (34), 288293.
cient = 0.430.68). Retention was shown to be higher in the period Jarret, A., Berghage, R., 2008. Annual and individual green roofs stormwater
AprilOctober than NovemberMarch and seasonal variation of the response models. In: 6th Greening Rooftops for Sustainable Communities
Conference. Baltimore, MD.
crop coefcient was found to be negligible. The peak time delay Kasmin, H., Stovin, V.R., Hathway, E.A., 2010. Towards a generic rainfall-runoff
was shown to have a great variability ranging from 0 to more than model for green roofs. Water Sci. Technol. 62 (4), 898905.
3248 L. Locatelli et al. / Journal of Hydrology 519 (2014) 32373248

Larsen, A.N., Gregersen, I.B., Christensen, O.B., Linde, J.J., Mikkelsen, P.S., 2009. Schneider, D., Wadzuk, B.M., Traver, R.G., 2011. Using a weighing lysimeter to
Potential future increase in extreme one-hour precipitation events over Europe determine a crop coefcient for a green roof to predict evapotranspiration with
due to climate change. Water Sci. Technol. 60 (9), 22052216. the FAO standardized PenmanMonteith equation. World Environ. Water
Lazzarin, R.A., Castellotti, F., Busato, F., 2005. Experimental measurements and Resour. Cong. 2011, 36293638.
numerical modelling of a green roof. Energy Build. 37 (12), 12601267. Sherrard, J., Jacobs, J., 2012. Vegetated roof water-balance model: experimental and
Lindblom, E., Ahlman, S., Mikkelsen, P.S., 2011. Uncertainty-based calibration and model results. J. Hydrol. Eng. 17 (8), 858868.
prediction with a stormwater surface accumulation-washoff model based on Simmons, M.T., Gardiner, B., Windhager, S., Tinsley, J., 2008. Green roofs are not
coverage of sampled Zn, Cu, Pb and Cd eld data. Water Res. 45 (13), 3823 created equal: the hydrologic and thermal performance of six different
3835. extensive green roofs and reective and non-reective roofs in a sub-tropical
Locatelli, L., Mark, O., Mikkelsen, P.S., Arnbjerg-Nielsen, K., Kleinlercher, B., Roldin, climate. Urban Ecosyst. 11 (4), 339348.
M., Binning, P.J., 2013. A simple rainfallrunoff model for the single and long Stovin, V., Vesuviano, G., Kasmin, H., 2012. The hydrological performance of a green
term hydrological performance of green roofs, Novatech 2013. Lyone (France). roof test bed under UK climatic conditions. J. Hydrol. 414, 148161.
Madsen, H., Arnbjerg-Nielsen, K., Mikkelsen, P.S., 2009. Update of regional intensity- Stovin, V., Poe, S., Berretta, C., 2013. A modelling study of long term green roof
duration-frequency curves in Denmark: tendency towards increased storm retention performance. J. Environ. Manage. 131, 206215.
intensities. Atmos. Res. 92 (3), 343349. Thomsen, R.S., 2011. Drift af Spildevandskomitens Regnmlersystem. Teknisk
Mentens, J., Raes, D., Hermy, M., 2006. Green roofs as a tool for solving the rainwater rapport Danish Meteorological Institute (DMI). 12-03.
runoff problem in the urbanized 21st century? Landscape Urban Plann. 77 (3), VanWoert, N.D., Rowe, D.B., Andresen, J.A., Rugh, C.L., Fernandez, R.T., Xiao, L., 2005.
217226. Green roof stormwater retention: effects of roof surface, slope, and media
Monterusso, M.A., Rowe, D.B., Rugh, C.L., Russell, D.K., 2004. Runoff water quantity depth. J. Environ. Qual. 34 (3), 10361044.
and quality from green roof systems. ACTA Horticult. (639), 369376. Vezzaro, L., Ledin, A., Mikkelsen, P.S., 2012. Integrated modelling of priority
Moran, A., Hunt, B., Snith, J., 2005. Hydrologic and water quality performance of pollutants in stormwater systems. Phys. Chem. Earth, Parts A/B/C 4244, 4251.
green roofs in Goldsboro and Raleigh, North Carolina. Greening Rooftops for Villarreal, E.L., 2007. Runoff detention effect of a sedum green-roof. Nord. Hydrol. 38
Sustainable Communities, Washington. (1), 99105.
Nielsen, S.A., Hansen, E., 1973. Numerical simulation of the rainfallrunoff process Villarreal, E.L., Bengtsson, L., 2005. Response of a Sedum green-roof to individual
on a daily basis. Nord. Hydrol. 4, 171190. rain events. Ecol. Eng. 25 (1), 17.
Palla, A., Gnecco, I., Lanza, L.G., 2009. Unsaturated 2D modelling of subsurface water Vrugt, J.A., Gupta, H.V., Bouten, W., Sorooshian, S., 2003. A shufed complex
ow in the coarse-grained porous matrix of a green roof. J. Hydrol. 379 (12), evolution metropolis algorithm for optimization and uncertainty assessment of
193204. hydrologic model parameters. Water Resour. Res. 39 (8).
Peel, M.C., Finlayson, B.L., McMahon, T.A., 2007. Updated world map of the Kppen- Voyde, E., Fassman, E., Simcock, R., 2010. Hydrology of an extensive living roof
Geiger climate classication. Hydrol. Earth Syst. Sci. 11 (5), 16331644. under sub-tropical climate conditions in Auckland, New Zealand. J. Hydrol. 394
Rezaei, F., Jarrett, A., Berghage, R., Beattie, D., 2005. Evaporation Rates from (34), 384395.
Extensive Green Roof Plant Species. American Society of Agricultural Engineers Weibull, W., 1939. A statistical theory of strength of materials. Ing. Vetensk. Akad
International Meeting, Tampa, Florida. Handl. 151, 145.
Scharling, M., 1999. Klimagrid - Danmark Nedbr, lufttemperatur og potentiel Wong, T.H.F., Brown, R.R., 2009. The water sensitive city: principles for practice.
fordampning (Climate grid Denmark, air temperature and potential Water Sci. Technol. 60 (3), 673682.
evaporation) 20 20 & 40 40 km, Danish Meteorological Institute. Zhang, S., Guo, Y., 2013. Analytical probabilistic model for evaluating the hydrologic
Scharling, M., 2001. Klimagrid Danmark Sammenligning af potentiel fordampning performance of green roofs. J. Hydrol. Eng. 18 (1), 1928.
beregnet ud fra Makkinks formel og den modicerede Penman formel (Climate Zimmer, U., Geiger, W.F., 1997. Model for the design of multilayered inltration
grid Denmark, comparison of potential evaporation calculated with Makkinks systems. Water Sci. Technol. 36 (89), 301306.
method and the modied Penman method). Danish Meteorological Institute,
Copenhagen.