Computers & Geosciences 67 (2014) 1423

Contents lists available at ScienceDirect

Computers & Geosciences

journal homepage: www.elsevier.com/locate/cageo

DFLOWZ: A free program to evaluate the area potentially inundated

by a debris ow
M. Berti n, A. Simoni
Dipartimento di Scienze Biologiche, Geologiche e Ambientali Universit di Bologna, Via Zamboni 67, 40127 Bologna, Italy

art ic l e i nf o a b s t r a c t

Article history: The transport and deposition mechanisms of debris ows are still poorly understood due to the
Received 26 August 2013 complexity of the interactions governing the behavior of watersediment mixtures. Empiricalstatistical
Received in revised form methods can therefore be used, instead of more sophisticated numerical methods, to predict the
9 January 2014
depositional behavior of these highly dangerous gravitational movements. We use widely accepted semi-
Accepted 7 February 2014
Available online 18 February 2014
empirical scaling relations and propose an automated procedure (DFLOWZ) to estimate the area
potentially inundated by a debris ow event. Beside a digital elevation model (DEM), the procedure
Keywords: has only two input requirements: the debris ow volume and the possible ow-path. The procedure is
Debris ow implemented in Matlab and a Graphical User Interface helps to visualize initial conditions, ow
Flooding
propagation and nal results. Different hypothesis about the depositional behavior of an event can be
Hazard assessment
tested together with the possible effect of simple remedial measures. Uncertainties associated to scaling
Scaling relationships
relations can be treated and their impact on results evaluated. Our freeware application aims to facilitate
and speed up the process of susceptibility mapping. We discuss limits and advantages of the method in
order to inform inexperienced users.
& 2014 Elsevier Ltd. All rights reserved.

1. Introduction or, in case of simplied mono-phase models, from the extreme

Debris ows are rapid, gravity-driven mass movements made
of a mixture of soil, rock particles, and water (Cruden and Varnes,
1996; Iverson,1997). They occur in small, steep catchments where
loose debris is available and intense rainfall, possibly combined
with snowmelt, can be rapidly concentrated. Debris ows travel at
high velocity down the channel and usually start to deposit over a
fan where the slope angle progressively decreases. Debris ows
contribute to the build-up of the fan and their paths experience
periodic diversions. Due to the destructive potential of debris
ows and to the possible presence of vulnerable elements over a
mountain fan, the delineation of areas exposed to danger repre-
sents a key element for any valuable hazard assessment (Jakob and
Hungr, 2005).
Dynamic modeling of debris ow phenomena made substantial
progresses in the last two decades (Rickenmann, 1999; Naef et al.,
2006; Hurlimann et al., 2008; Hussin et al., 2012). Due to the
complexity of the phenomena, however, some limitations still exist
in numerical simulations. They derive either from the difculties of
describing the multi-phase physics of the ow (Iverson et al., 2010)
n
Corresponding author. Tel.: 39 051 2094541.
E-mail address: matteo.berti@unibo.it (M. Berti).
difculty to measure the relevant parameters (Rickenmann et al.,
2006; Sosio et al., 2007). Problems associated to the denition of
the frequencyvolume relationship (specic for each catchment)
add uncertainty to any prediction. In this context, simple, geo-
morphic-based, empiricalstatistical methods describing the mobi-
lity of debris ows (Corominas, 1996; Legros, 2002; Toyos et al.,
2007) maintain some appeal. They require little skill, produce
consistent results (limited user-dependency) and can be used for
testing multiple hypotheses with reasonable waste of time.
Originally, empirical relationships were developed to estimate
the maximum runout distance mostly as a function of the debris
ow volume. Some of them make use of the reach angle, dened
as the ratio between the vertical drop and the horizontal projec-
tion of the total travel distance (Corominas, 1996, Rickenmann,
1999). Others describe the runout distance on the fan (Crosta et al.,
2003; Rickenmann, 1999) or apply deposition criteria based on
slope gradient (Ikeya, 1989) and tributary junction angle (Benda
and Cundy, 1990; Bathurst et al., 1997).
Digital elevation models (DEMs) have increased the potential of
empirical approaches. They allow the implementation of simple
empirical models to delineate potentially threatened areas based
on minimum data requirements.
The approach proposed for lahars by Iverson et al. (1998) correlates
event volume (V) with inundated area (B) and cross-sectional ow

http://dx.doi.org/10.1016/j.cageo.2014.02.002
0098-3004 & 2014 Elsevier Ltd. All rights reserved.
 M. Berti, A. Simoni / Computers & Geosciences 67 (2014) 1423
area (A). It can be dened as semi-empirical in that the free
parameters of the theoretical scaling relationships between volume
and area are calibrated using historical debris ow-inundation data.
Many researchers have demonstrated its applicability to debris ows
(Crosta et al., 2003; Berti and Simoni, 2007; Griswold and Iverson,
2008; D'Agostino et al., 2008; Scheidl and Rickenmann, 2010) and
other phenomena such as rock avalanches (Griswold and Iverson,
2008). The parameters of the equations regressing empirical data
varies between different phenomena, indicating that they possess
signicantly different mobility, as expected. Statistics conrm that
empirical data can be described by power-law equations with a x
slope, as originally envisioned by Iverson et al. (1998) on the basis of
scaling analyses. Therefore, the relationships A p V 2=3 and B p V 2=3
can be used to delineate the deposition zones for a range of probable
event volumes. Appropriate proportionality parameters can be
obtained by regressing V A and V B pairs describing an homo-
geneous set of natural mass ows (Griswold and Iverson, 2008;
Simoni et al., 2011) or can be expressed as a function of morphologic
parameters (Scheidl and Rickenmann, 2010). The zonation requires
a digital elevation model (DEM) and a ow-path. The ow-path
can be automatically identied by using ow-direction algorithms
(Hurlimann et al., 2008; Scheidl and Rickenmann, 2010) or manually
dened based on eld inspection and possible overow points (Berti
and Simoni, 2007). Automatic ow-path identication, beside obvious
convenience, can be associated to probabilistic signicance but the
results are inuenced by DEM resolution (Horton et al., 2013).
Berti and Simoni (2007) modied the original LAHARZ algo-
rithm developed by Schilling (1998) to include unconned ow
Fig. 1. Empirical scaling relationships for debris ows (from Simoni et al., 2011).
The dashed lines indicate the 95% prediction intervals.
15
conditions. The model was named DFLOWZ and tested against real
cases of alpine debris ows.
In this paper, we present a freeware version of DFLOWZ that is
written and implemented in Matlab. A Graphical User Interface
(GUI) is designed to facilitate the required preliminary operation
and to visualize the DEM, the ow path, and the cross-sections
together with the obtained zonation (inundated area). The code
includes the possibility to consider the uncertainties associated to
the empirical equations and to verify how these are reected in
the results. The user can also test different hypotheses about the
onset of the depositional process and possible ow paths. Topo-
graphy misrepresentations (e.g., coarse-resolution DEM, old topo-
graphy) can be adjusted by modifying the channel ow area.
Preliminary tests about the effectiveness of remedial measures,
such as retention basins or deection walls, can be done thanks to
the implementation of a DEM-modifying algorithm. Capabilities
and limitations of the model are critically discussed throughout
the text and by means of a sample application.
2. Theoretical background
A detailed description of the theory behind DFLOWZ can be
found in Berti and Simoni (2007). In this section we briey review
the general principles of the model and the modications made
for this work.
DFLOWZ extends to debris ows an empiricalstatistical model
originally proposed by Iverson et al. (1998) to predict the runout
length and the areas affected by lahars (LAHARZ model). Both
models are based on the simple observation that the larger the
volume of the ow (V), the larger the cross-sectional ow area (A)
and the inundate planimetric area (B). The dependence V A and
V  B has been documented for different types of ow (rock
avalanches, lahars, and debris ows) and reported in form of
two semi-empirical scaling relationships (Griswold and Iverson,
2008; Scheidl and Rickenmann, 2010):
A kA V 2=3 1a
B kB V 2=3 1b
where kA and kB are dimensionless coefcients (called mobility
coefcients) that vary with the ow type and 2/3 is a xed
exponent justied by geometric similarity (Iverson et al., 1998).
Simoni et al. (2011) combined the data of about 100 historical
debris ows obtaining kA  0:07 and kB  18 (Fig. 1). These values
can be considered of general validity for non-volcanic debris
ows since no systematic difference was found between datasets
collected in different geological environments. In reality, the
mobility of debris ows is certainly inuenced by water content,
grain-size distribution, or debris availability, but these differences
are masked by the scatter of the data due to measurement
uncertainties.
To account for the data dispersion around the regression lines,
Simoni et al. (2011) proposed to include two uncertainty factors
(a and b) in the scaling relations
A a0:07V 2=3 2a
B b18V 2=3 2b
The parameters a and b indicate how the expected values of A and
B differ from that predicted by the regression lines shown in Fig. 1.
The numerical values of a and b are plotted in the small insets of
Fig. 1 as a function of the condence level of the prediction, which
measures the distance from the regression based upon the sample
standard deviation and t-statistic (Weisberg, 1985). Depending on
the specic application, the uncertainty factors a and b allow the
16 M. Berti, A. Simoni / Computers & Geosciences 67 (2014) 1423
user to obtain predictions of the best/worst scenarios in terms of
inundated and cross-sectional areas as a function of the appro-
priate condence level. Also, different combinations of a and b
allow to reproduce different expected debris ow velocity and
mobility whenever information on the ow behavior is available.
For instance, one might expect that a dilute ow inundates a small
cross-sectional area (ao1) because of its velocity, and deposits
over a large planimetric (b 41) area because of its mobility.
Anyhow, the rationale of such observations is rarely solid enough
to base any prediction on, and the statistic approach is preferable.
Eqs. (2a) and (2b) are implemented in DFLOWZ to predict the
inundated area on a debris ow fan. Input data are the debris ow
volume V, the uncertainty factors a and b, the Digital Elevation
Model (DEM) of the fan area, the path of the ow channel, and the
traces of several representative cross-sections. For a given volume,
the model rst computes the expected value of cross-sectional (A)
and planimetric (B) inundated area using Eqs. (2a) and (2b). The
ow area A is assumed to be constant for any location along the
depositional reach and it is used to compute the inundated width
along the proles extracted by the DEM (Fig. 2). The calculation
starts with the rst section upstream and proceeds downstream
determining the inundation area between the sections. The
inundated area is progressively summed until it equals the
expected value of B given by Eq. (2b).
The main difference between DFLOWZ and LAHARZ (Iverson
et al., 1998; Schilling, 1998) is that our model is designed to
(roughly) simulate unconned ow deposition. When the ow
exceeds the available channel area it is assumed that the debris
deposits over the ground surface with a constant thickness h. The
thickness of unconned ow is a function of debris ow volume
and it is simply obtained by dividing V by the expected B
(h 0:06V 1=3 ; Fig. 2). The owing material rst lls the ow
channel and any nearby surface depressions, then spreads laterally
until the constant ow area A is reached. Scheidl and Rickenmann
(2010) relaxed the assumption of a constant ow area by simulat-
ing the potential ow spreading on the basis of a Monte-Carlo
analysis of multiple ow pathways. Their model is conceptually
similar to DFLOWZ but assumes a progressive increase of cross-
sectional ow area along the depositional zone.
The practical use of DFLOWZ suggested several improvements
to the model that are presented in this work. The most relevant are
the automatic generation of the computational cross-sections
using different interpolation algorithms, the possibility to account
Fig. 2. The planimetric inundated area (a) is estimated by interpolating the cross-sectional inundation widths (b) computed on several cross-sectional channel proles.
The expected values of cross-sectional ow area A and planimetric inundated area B are given by the two scaling relationships shown in Fig. 1.
for a channel geometry not represented in the DEM, and the
capability to modify the DEM to simulate the construction of
debris ow mitigation measures. All these changes are described
in the next section.
3. Program description
DFLOWZ is written and implemented in Matlab 7.11 with a
Graphical User Interface (GUI). Matlab was chosen because of its
powerful data analysis capabilities and visualization features. The
program is available as a standalone application that can run on
Windows systems without requiring Matlab or its auxiliary tool-
boxes. The software may be downloaded from the Internet for free
via http://www.bigea.unibo.it/it/ricerca/dowz together with the
original source code. The code is portable between operating
systems (e.g. Unix, PC, and Macintosh) and it was designed to
allow easy extensibility. New functionality can be added through
modication of existing functions or adding new ones according to
the users' goals. Compared to other commercial (FLO-2D, OBrien
et al., 1993; DAN-W, Hungr, 1995) or free (LAHARZ, Schilling, 1998)
software for debris ow routing, DFLOWZ is the only one that
combine a user-friendly graphical interface with the possibility to
modify and enhance the source code.
The main window of the GUI (Fig. 3) is divided into ve tasks
which allow for the input of physical parameters and selection of
run-time options. Important features of each component in the
GUI are discussed below.
3.1. Digital elevation model
The DEM must include the active part of the fan, that is the area
where relatively recent deposition, erosion, and alluvial-fan-
ooding have occurred. This area typically contains the active
debris ow channel, which is the most probable path for future
events (Glade, 2005). When the identication of active zones is
difcult, or when it is likely to have ow diversion from the active
channel, the entire fan area should be included in the DEM and
different potential debris ow paths must be analyzed.
DFLOWZ is adapted to run with high resolution digital eleva-
tion models (1 to 3 m cell size) such as those derived from
airborne LiDAR data. Care should be taken when using coarse
resolution DEMs (10 m cell size or larger) since the ow channel
 M. Berti, A. Simoni / Computers & Geosciences 67 (2014) 1423
Fig. 3. The DFLOWZ Graphical User Interface.
can be poorly represented and this will have important conse-
quences on simulation results. In general, a smoothed topography
will produce an inundation area which is wider and shorter than
expected because a low resolution DEM underestimates the
available area of the ow channel and the debris ow will
inundate a large width to accomplish the theoretical value of A
(Fig. 4). The main risk in these cases is to underestimate the runout
distance and to spread the ow too much sideways.
DEMs often contain erroneous elevations referred to as sinks
that are usually lled before hydrological modeling. However,
some sinks may represent real surface depressions where the
debris can be trapped reducing the ooded area. The user must be
aware of real topography to determine whether sinks in the DEM
are errors or not.
The largest DEM array that can be loaded mainly depends on
the system memory (RAM plus swap le) and operating system. By
assuming that no major processing are launched and that an
unlimited memory is available, the largest matrix size is about
1500 MB for 32-bit platforms and 16 GB for 64-bit platforms. This
roughly corresponds to a maximum DEM size of 14,000  14,000
cells or 46,000  46,000 cells. Input data le is an ASCII Grid (6
lines of header info followed by the elevation data) compatible
with most GIS software.
17
3.2. Flow channel
DFLOWZ calculates the deposition area by assuming a constant
cross-sectional ow area A which extends across a user-specied
channel path. The path of the ow channel must be specied by a
polyline shape le (one single feature, no attributes required). The
channel path has a profound effect on the results because the
ooded area always wanders around the predened channel. A
proper denition of the potential ow path is therefore essential
for meaningful results. In most cases, it is quite easy to identify the
active debris ow channel on the fan by means of aerial photo-
graphs and eld surveys. The active channel is directly connected
with the main feeder channel at the fan apex and it is usually
characterized by fresh deposits, scouring, bare soil or scattered
vegetation (Hungr et al., 2005).
However, when there is no morphological evidence of recent
activity, or there are no historical records of past events, the choice
of a likely debris ow pathway can be difcult. In this case it is
necessary to perform a sensitivity analysis to evaluate the effect of
different ow paths on the inundated area. The method proposed
by Hurlimann et al. (2008) is suitable for this purpose. The method
combines the D8 ow routing algorithm with a Monte Carlo
random walk model to generate trajectories of debris ows that
18 M. Berti, A. Simoni / Computers & Geosciences 67 (2014) 1423
include the spreading effect around the steepest path. Scheidl and
Rickenmann (2010) implemented this method in their TopRunDF
model. The computed inundated area for each potential debris
ow path can be then combined to obtain a hazard map showing
the probability of each cell to be affected by a future debris ow
(given for example by the ratio of the times the cell has been
ooded divided by the total number of simulations). Such an
analysis is also advisable when it is likely that the debris would
divert from the active channel (as a consequence of channel
blockage or overbanking ow) and ow downhill on the fan.
The probabilistic analysis of the multiple channel paths is not
Fig. 4. (a) Effect of the DEM resolution of the inundated cross-sectional width.
The inundated width tends to be larger for a low-resolution channel prole (b).
Fig. 5. Empirical scaling relationships implemented in DFLOWZ. The red point indicates the expected value of cross-sectional ow area A (left) and planimetric inundated
area B (right) for the selected debris ow volume. (For interpretation of the references to color in this gure legend, the reader is referred to the web version of this article.)
implemented in the current version of the program, but it can be
easily done by combining the results obtained with different
channel paths
3.3. Design debris ow
The design debris ow should indicate the largest probable
debris ow generated by hydrological events in the basin. It is
dened by a design volume (V) and by two uncertainty factors (a
and b, see above). Debris ow volume is here intended as the total
amount of sediment, organic material, and water reaching the fan
apex. This volume is a function of the volume of the initiating
failure (or failures) plus the volume entrained along the transport
reach. The estimate of debris ow magnitude is an essential step in
the analysis since the extent of the ooded area mainly depends
on the input volume. Although different methods have been
proposed in the literature to guess the potential debris ow
volume (Kronfellner-Kraus, 1985; Bianco and Franzi, 2000;
Ceriani et al., 2000; D'Agostino and Marchi, 2001; Marchi and
D'Agostino, 2004; Jakob and Hungr, 2005) this estimate still
remains a difcult task. The main uncertainties are related to
sediment availability, expected water inow, effect of vegetation,
amount of sediment entrained along the channel (bulking) and to
the fact that debris ow material deposited in previous surges (or
delivered by bank failure) can be remobilized in subsequent
surges. In practice, the estimate of the debris ow volume heavily
relies on past observations and it is very difcult if historical data
are not available.
DFLOWZ is computationally simple and it allows to perform a
sensitivity analysis on the input volume quickly and easily. Such a
sensitivity analysis will detect the most signicant zones of
inundation from debris ows of various volumes, thus providing
a scenario map of ooding likelihood. Of course, additional
information are required to convert any likelihood (susceptibility)
map into a hazard map of debris ow ooding since ows with
different volumes are characterized by different return periods
which make larger ows less probable.
 M. Berti, A. Simoni / Computers & Geosciences 67 (2014) 1423
Once a design debris ow volume is selected, the uncertainty
factors a and b (Eqs. (2a) and (2b)) can be varied to calibrate the
expected ow area A and inundated area B according to the
specic eld conditions (see Section 2). For sake of clarity, the
possible values of a and b are reported as percentages in the two
pull-down menus Condence interval for the predictions of the
GUI (Fig. 3). The condence intervals can be varied from 90% to
90% and the corresponding values of a and b are computed using
the curves shown in the small insets in Fig. 1. The default values
are 0% which correspond to a 1 and b 1. In this case, A and B are
estimated from the input volume V using the mean regression
lines shown in Fig. 1. The implicit assumption is that the mobility
of debris ows in the study area is similar to that of typically
observed for subaerial debris ows. By selecting for instance a
condence interval for b 90% we assume that the ow can be
more mobile than the average, and that the inundated area will be
larger than usually is. In the example shown in Fig. 5, the user
selected a 0% and b 60%, then the expected values of A and B fall
respectively on and above the corresponding regression lines.
Simoni et al. (2011) provide several indications on the selection
of the uncertainty parameters a and b.
3.4. Cross-sections
DFLOWZ evaluates the inundated area B by connecting the
inundation width Wi computed at each cross-sectional prole
(Fig. 2). The traces of the proles can be dened manually or
generated automatically by DFLOWZ (Fig. 3). In the rst case the
traces are manually drawn on a GIS software and imported in
DFLOWZ as polyline shape le. This option is useful when the
topography is complex or irregular and a close control on the
computational cross-sections is needed. More commonly the
traces are generated automatically by specifying the number of
sections and their direction (normal to the ow channel or parallel
to the slope). Normal to ow channel means that each trace is
drawn perpendicularly to the local direction of the ow channel;
parallel to the slope means that each trace is parallel to the local
direction of the slope, evaluated as the mean aspect of a 3  3
kernel centered at the intersection point between the channel and
the prole. Unless the ow channel cuts transversally the slope
and the differences between these two options are usually minor.
The traces are numbered consecutively from the rst section
upstream. Before running the analysis it is important to check the
Fig. 6. In case of intersecting cross-sections, DFLOWZ checks if the intersection occurs along the inundated width or not. In rst case (b), the downstream section is omitted
and the computation proceeds with the section further downstream. In the second case (a) both the sections are maintained.
19
location of the traces (button Map on the GUI; Fig. 3) to ensure
the correctness of the result. Moreover, since the number, location,
and direction of the proles affect to some extent the ooded area,
it is also important to evaluate the sensitivity of the analysis to
such a factor.
Two methods of surface interpolation can be selected to
generate the cross-sectional proles, referred to as IDW and
Planar in the GUI. The Inverse Weighted Distance method (IDW
method; Watson and Philip, 1985) calculates the unknown eleva-
tion at each point of the prole as the average of the ve neighbor
cells weighted by their distance. The Planar method approximates
the ground surface of the nine cells around the point with a plane
and computes the unknown elevation by planar tting. Planar
method typically provides a smoother topographic prole.
It is common that two cross sections intersect near bends in the
channel. In this case DFLOWZ identies the intersection point
between the two lines and compares it with the inundated width
computed for the downstream section (Fig. 6): if the inundated
width exceeds the intersection point (that is the debris ow would
inundate an area already ooded) the cross-section is excluded by
the analysis and the computation continues with the section
downstream (Fig. 6b); otherwise, both sections are considered
(Fig. 6a). This control is necessary because the model assumes
independent ooding of 2D proles without interaction effects.
The rst cross section upstream must be located where the
beginning of deposition is expected to take place, which is
generally the fan apex. However, if the ow channel is deeply
incised in the fan, or if it is blocked by debris along its path, the
deposition may start further downstream or upstream. The start-
ing point of deposition is therefore an important input parameter
that should be chosen on the basis of a detailed geomorphological
survey. The effect of this parameter on model results can be easily
evaluated by starting the analysis at different cross-sections (see
the input box Start deposition from section in Fig. 3).
3.5. DEM resolution and change in surface topography
The inuence of DEM resolution on debris ow routing is a
well-known problem (e.g. Stolz and Huggel, 2008). All the existing
models are highly sensitive to the accuracy of topographic data,
and this also applies to simple methods like LAHARZ and DFLOWZ
(Stevens et al., 2002; Berti and Simoni, 2007). In particular, results
can be very inaccurate when the channel morphology is poorly
20 M. Berti, A. Simoni / Computers & Geosciences 67 (2014) 1423
represented in the DEM, for instance because of a low spatial
resolution (Fig. 7a) or because the channel has been lled or
excavated (Fig. 7b).
To partially overcome this problem, DFLOWZ allows the user to
import an external text le which contains, for each channel reach
between points i and i1, an extra ow area AE (in m2) not
represented in the DEM (Fig. 7). A positive value of AE will reduce
the theoretical value of A on the selected channel reach, while a
negative value will increase it. In the rst case the inundated
width will be smaller, simulating the presence of a larger channel;
in the second case the inundated width will be larger, simulating
the presence of a smaller channel. It is important to realize that AE
only affects the theoretical ow area A while B does not vary. This
feature can be also used for a preliminary evaluation of debris ow
countermeasures such as channel excavation or widening.
Another way to account for topographic effects is to modify the
elevation grid. The DEM can be modied to improve the accuracy of
the topography (for example by generating 3D structures such as
buildings or houses) or to simulate the construction of channel
levees or deection walls. DFLOWZ takes advantage of the Matlab's
capability to read geodata les in order to perform this analysis
easily. The areas to modify can be digitized as a polygon layer on a
GIS and saved as a shape le, with the required change in elevation
Fig. 7. The user can specify an extra ow area AE along some stretches of the
channel in order to account for a poor resolution of the DEM (a) or to roughly
simulate a channel excavation (b). The extra ow values must be listed on a
separated text le together with the channel coordinates.
Fig. 8. The Modify the DEM option allows the user to simulate a positive
(ChangeEl40) or negative (ChangeEl o 0) variation of the ground topography.
DFLOWZ reads the areas to modify from a polygons shape le where ChangeEl is
an attribute eld.
stored in an attribute eld named ChangeEl (number data format,
values in meters). DFLOWZ loads the polygon shape le and
changes the elevations of grid cells inside the polygon according
to the specied value (Fig. 8). A positive value will increase the
elevation, a negative value will lower the topography.
In both cases it is essential to bear in mind that DFLOWZ is a
simple geometrical model that does not reproduce the complex
dynamics of debris owchannel interaction. Therefore, any simu-
lation of debris ow countermeasures has to be carefully evalu-
ated, and the nal design must always rely on detailed hydraulic
modeling that account for ow velocity, shear stress, and pressure
gradient. For instance, the possibility of structure damage, failure,
and subsequent ow diversion is not considered in DFLOWZ.
3.6. Results
The Calculate button runs the analysis and shows the results
on a separate window (Fig. 9). The planimetric inundated area and
the ow channel path are drawn on a low-resolution contour map
(max 100  100 cells) to speed up the display. The window
provides a quick view of the results. For a more accurate display,
click the Export grid button to save the inundated area in ASCII
Grid format with the original DEM resolution. The grid can be then
opened using most GIS software and combined with other digital
maps and georeferenced data. The box Show cross-section traces
shows the proles used in the analysis (as said above, the
intersecting cross-sections could be excluded by the analysis)
while the button Edit Figure allows users to customize the
appearance of the map and to save it as image using the built-in
editing functions of Matlab.
Click on the listbox View section to see the inundated ow
area in each cross-section. The cross-sectional prole is shown on
a separate chart with the ow area colored in red (Fig. 9).
Interactive zoom and pan (available as built-in Matlab functions)
allow to check whether the model provides realistic results. A
careful check of each cross-section is recommended in order to
detect unexpected results related to particular topographic
conditions.
4. Sample application
Simoni et al. (2011) tested DFLOWZ against a variety of eld
conditions. In their paper the authors discussed the main cap-
abilities of the model, as well as its limitations, by comparing the
inundated area predicted by DFLOWZ with that observed in the
eld for 25 historical debris ow events in the Italian Alps. In this
section we provide a sample application of the model to illustrate
some relevant points for a correct analysis.
The Rio Valburga basin is located in the Italian Alps about 70 km
to the north of the city of Bolzano (Fig. 10a). The basin has an area of
4.5 km2, relief of 2100 m and an average slope of 421. It is extensively
gullied in the upper part (Fig. 10b) and the overall drainage density is
6.4 km  1. The deposition fan is 1.2 km2 with an average slope of 91.
Bedrock geology mostly consists of metamorphic rocks (biotite
paragneiss) associated to tonalitie and pegmatite in the lower part.
Most of the basin, however, is covered by debris deposits of varying
thickness, which are loose and unvegetated in the upper part. Debris
ows are initiated in this area by surcial sliding and progressive
erosion of the loose debris, triggered by high-intensity short-duration
thunderstorms. A rather large event occurred on the 11th of July
2007. The debris ow reached the lower part of the fan covering
an area B32,000 m2 (Fig. 11) with an estimated volume of about
35,000 m3. The cross-sectional ow area in the lower reach of the
channel was A55 m2.
 M. Berti, A. Simoni / Computers & Geosciences 67 (2014) 1423 21

Fig. 9. Sample results of DFLOWZ. Planimetric inundated area (left) and cross-sectional ow area for two selected proles (right).

Fig. 10. Aerial photograph (a) and hillshade relief and (b) of the Valburga debris ow basin (Bolzano Province, Italian Alps).

To replicate this event with DFLOWZ we may: (i) use the
pre-event DEM of the fan area; (ii) consider the measured
debris ow volume as input value; (iii) place the rst analysis
section at the fan apex. Since there is limited uncertainty in the
data we expect good results from the model. Instead, as shown
in Fig. 12 (case 1), the inundated area predicted by DFLOWZ is
remarkably different from that observed in the eld. The rst
reason for this discrepancy is that the typical scaling rela-
tionships obtained for debris ows (the two regression lines
shown in Fig. 1) do not apply exactly to our case. In fact, the
inundated area predicted by Eq. (2b) for a volume
V 35,000 m3 (and b 1) is B 17000 m 2 , while the real deposi-
tion area is about two times larger (32,000 m 2 ). To get a better
estimate we should use b E 2, which indicates a ooded area
well above the regression line (about the 60% condence level
of the prediction). The July 2007 debris ow was evidently
more dilute and mobile than usually observed for debris ows
of the same volume. Whether or not this behavior is a
22 M. Berti, A. Simoni / Computers & Geosciences 67 (2014) 1423
Fig. 11. Area inundated by the 11th July debris ow event on the Valburga fan.
Fig. 12. Comparison between the area inundated by the 11th July debris ow event
and the area computed by DLOWZ using two different combinations of the input
parameters (see text).
systematic characteristic of the ows in the Valburga basin
cannot be established from a single event.
The second (and in this case more important) source of error is
the starting point of the deposition. Although the overall deposi-
tion zone is clearly marked by the fan apex, the July 2007 debris
ow bypassed the apex and ooded the lower part of the fan
(Fig. 11). This introduces a big error in the model prediction.
A closer view of fan morphology, however, would reveal that the
ow channel is steep and deeply incised in the upper part of the
fan (where the debris ow propagated with little or no deposition)
and suddenly decreases in slope and depth at in the lower part.
The break in slope closely corresponds to the start of the July 2007
deposit (Fig. 11). Running a new analysis with b 2 and locating
the rst section in the right place we obtain a reasonable
agreement between predicted and observed ooded area (Fig. 12,
case 2).
Potentials and limitations of DFLOWZ clearly emerge from this
sample application. The modeling accuracy is adversely affected by
the uncertainties in the input parameters, in particular the starting
point of the deposition, the parameters a and b of the scaling
relationships, and the design volume. Different combinations of
these parameters may result in very different inundated areas. To
the other hand DFLOWZ is simple and exible, and the effect of
input parameters can be easily investigated and understood. The
model allows to identify the most signicant zones of inundation
by varying these relevant parameters within a reasonable range,
thus obtaining scenario maps with a probabilistic view.
5. Conclusions
DFLOWZ is a free software for a preliminary evaluation of
debris ows inundation areas. Although the model is based on
empirical scaling relationships which do not describe the complex
dynamics of debris ows, it can be a valuable alternative to more
comprehensive numerical models when calibration data are not
available or when a preliminary hazard analysis is required. The
Graphical User Interface of DLOWZ allows easy data input and
graphical representation of the results. The user can quickly
perform a sensitivity analysis to see how the most inuencing
parameters (such as the design debris ow volume, the starting
point of the deposition, or the ow channel path) affects the debris
ow inundated area. Given the uncertainties typically involved in
such a predictions, this is an important advantage of the model.
Another benet of DFLOWZ is the possibility to quickly evaluate
the impact of DEM modications on the ooded zones. The main
limitations of the model come from the use of a constant thickness
for the unconned deposits (Berti and Simoni, 2007), from the
hypothesis of a constant ow area (Scheidl and Rickenmann, 2010)
and, in general, from the sensitivity of the results on DEM
resolution. The authors warmly welcome any feedback, criticism,
or comment arising from the use of DLOWZ.
Acknowledgments
This work was supported by the Italian Ministry of University and
Scientic Research (PRIN 20102011, Ref. 2010E89BPY_005, Time-
space prediction of high impact landslides under the changing
precipitation regimes).
References
Bathurst, J., Burton, A., Ward, T., 1997. Debris ow run-out and landslide sediment
delivery model tests. J. Hydraul. Eng. 123 (5), 410419.
Benda, L.E., Cundy, T.W., 1990. Predicting deposition of debris ows in mountain
channels. Can. Geotech. J. 27 (4), 409417.
Berti, M., Simoni, A., 2007. Prediction of debris ow inundation areas using
empirical mobility relationships. Geomorphology 90, 144161, http://dx.doi.
org/10.1016/j.geomorph.2007.01.014.
Bianco, G, Franzi, L., 2000. Estimation of debris ow volumes from storm eventsIn:
Wieczorek, G.F., Naeser, N.D. (Eds.), Debris-Flow Hazards Mitigation:
Mechanics, Prediction, and Assessment. Balkema, Rotterdam, pp. 441448
Ceriani M., Crosta G., Frattini P.E. Quattrini S., 2000. Evaluation of hydrogeological
hazard on alluvial fan. In: Proceeedings of the International Symposium
Interpraevent 2000, vol. 2, Villach, Band, pp. 213225.
Corominas, J., 1996. The angle of reach as a mobility index for small and large
landslides. Can. Geotech. J. 33, 260271.
Cruden, D.M., Varnes, D.J., 1996. Landslides types and processes. In: Turner, A.K.,
Schuster, R.L. (Eds.), Landslides: Investigation and Mitigation. Transportation
Research Board Special Report 247. National Academy Press, WA, pp. 3675
Crosta, G.B., Cucchiaro, S., Frattini, P., 2003. Validation of semi-empirical relation-
ships for the denition of debris-ow behavior in granular materials. In:
Proceedings of the 3rd International Conference on Debris-Flow Hazard
Mitigation, vol. 2, Davos, Rotterdam, Millpress, pp. 821831.
D'Agostino, V, Marchi, L., 2001. Debris ow magnitude in the Eastern Italian Alps:
data collection and analysis. Phys. Chem. Earth, Part C 26 (9), 657663.
D'Agostino, V., Cesca, M., Marchi, L., 2008. Field and laboratory investigations of
runout distances of debris ows in the Dolomites (Eastern Italian Alps).
Geomorphology 115, 294304.
Glade, T., 2005. Linking debris-ow hazard assessments with geomorphology.
Geomorphology 66, 189213.
Griswold, J.P., Iverson, R.M., 2008. Mobility statistics and automated hazard
mapping for debris ows and rock avalanches, U.S. Geological Survey Scientic
Investigations Report, 2007-5276 p. 59 pp.
Horton, P., Jaboyedoff, M., Rudaz, Zimmermann M., 2013. Flow-R, a model for
susceptibility mapping of debris ows and other gravitational hazards at a
regional scale. Nat. Hazards Earth Syst. Sci. 13, 869885, http://dx.doi.org/
10.5194/nhess-13869-2013.
Hungr, O., 1995. A model for the runout analysis of rapid ow slides, debris ows
and avalanches. Can. Geotech. J. 32 (4), 610623.
Hungr, O., Fell, R., Couture, R., Eberhardt, E., 2005. Landslide risk management. In:
Proceedings of the International Conference on Landslide Risk Management,
Vancouver, Canada, Taylor & Francis, 31 May3 June 2005, 776 pp.
 M. Berti, A. Simoni / Computers & Geosciences 67 (2014) 1423
Hurlimann, M., Rickenmann, D., Medina, V., Bateman, A., 2008. Evaluation of
approaches to calculate debris-ow parameters for hazard assessment. Eng.
Geol. 102, 152163.
Hussin, H.Y., Quan, Luna, B., van Westen, C.J., Christen, M., Malet, J.P., van Asch, T.,
2012. Parameterization of a numerical 2-D debris ow model with entrain-
ment: a case study of the Faucon catchment, Southern French Alps. Nat.
Hazards Earth Syst. Sci. 12, 30753090.
Ikeya, H., 1989. Debris ow and its countermeasures in Japan. Bull. Int. Assoc. Eng.
Geol. 40, 1533.
Iverson, R.M., 1997. Physics of debris ow. Rev. Geophys. 35, 245296.
Iverson, R.M., Schilling, S.P., Vallance, J.W., 1998. Objective delineation of lahar-
inundation hazard zones. Geol. Soc. Am. Bull. 110, 972984.
Iverson, R.M., Logan, M., LaHusen, R.G., Berti, M., 2010. The perfect debris ow?
Aggregated results from 28 large-scale experiments. J. Geophys. Res., 115, http:
//dx.doi.org/10.1029/2009JF001514
Jakob, M., Hungr, O., 2005. Debris-Flow Hazards and Related Phenomena. Springer,
Chichester, UK p. 798
Kronfellner-Kraus G., 1985. Quantitative estimation of torrent erosion. In: Takei, A.,
(Ed.), Proceedings of the International Symposium on Erosion, Debris Flow and
Disaster Prevention, Tsukuba, Japan, pp. 107110.
Legros, F., 2002. The mobility of long-runout landslides. Eng. Geol. 63, 301331.
Marchi, L, D'Agostino, V., 2004. Estimation of debris-ow magnitude in the Eastern
Italian Alps. Earth Surf. Process. Landf. 29, 207220.
Naef, D., Rickenmann, D., Rutschmann, P., McArdell, B.W., 2006. Comparison of ow
resistance relations for debris ows using a one-dimensional nite element
simulation model. Nat. Hazards Earth Syst. Sci. 6, 155165.
OBrien, J.S., Julien, P.Y., Fullerton, W.T., 1993. Two-dimensional water ood and
mudow simulation. J. Hydraul. Eng. 119, 244259.
Rickenmann, D., 1999. Empirical relationships for debris ows. Nat. Hazards 19,
4777.
23
Rickenmann, D., Laigle, D., McArdell, B.W., Hubl, J., 2006. Comparison of 2D debris-
ow simulation models with eld events. Comput. Geosci. 10, 241264.
Scheidl, C., Rickenmann, D., 2010. Empirical relationships for debris ow mobility
and deposition on fans. Earth Surf. Process. Landf. 35, 157173.
Schilling, S.P., 1998. LAHARZ; GIS programs for automated mapping of lahar-
inundation hazard zones, U.S. Geological Survey Open-File Report 98-638 p.
80 pp.
Simoni, A., Mammoliti, M., Berti, M., 2011. Uncertainty of debris ow mobility
relationships and its inuence on the prediction of inundated areas. Geomor-
phology 132, 249259, http://dx.doi.org/10.1016/j.geomorph.2011.05.013.
Sosio, R., Crosta, G.B., Frattini, P., 2007. Field observations, rheological testing and
numerical modeling of a debris-ow event. Earth Surf. Process. Landf. 32,
290306.
Stevens, N.F., Manville, V., Heron, D.W., 2002. The sensitivity of a volcanic ow
model to digital elevation model accuracy: experiments with digitised map
contours and interferometric SAR at Ruapehu and Taranaki volcanoes. N. Z. J.
Volcanol. Geotherm. Res. 119, 89105.
Stolz, A., Huggel, C., 2008. Debris ows in the Swiss National Park: the inuence of
different ow models and varying DEM grid size in modeling results. Land-
slides 5, 311319.
Toyos, G., Oramas Dorta, D., Oppenheimer, C., Pareschi, M.T., Sulpizio, R., Zanchetta,
G., 2007. GIS-assisted modeling for debris-ow hazard assessment based on
the events of May 1998 in the area of Sarno, southern Italy: part I. Maximum
run-out. Earth Surf. Process. Landf. 32, 14911502, http://dx.doi.org/10.1002/
esp.1472.
Watson, D., Philip, G., 1985. A renement of inverse distance weighted interpola-
tion. Geo-Processing 2, 315327.
Weisberg, S., 1985. Applied Linear Regression. John Wiley & Sons, New York p. 344