Proceedings of the Institution of

Civil Engineers
Water & Maritime Engineering 156
??? 2003 Issue ?
Pages 1^8

Paper 12879
Received ??/??/2002
Accepted ??/??/2002

Keywords:
Sajjad Haider Andre¤ Paquier Robert Morel Jean-Yves Champagne
floods & floodworks/hydraulics &
Cemagref, Hydrology^ Cemagref, Hydrology^ Fluid Mechanics and Fluid Mechanics and
hydrodynamics/mathematical
Hydraulics Research Unit, Lyon, Hydraulics Research Unit, Acoustics Laboratory, INSA, Acoustics Laboratory, INSA,
modelling
France; Pakistan Water and Lyon, France Lyon, France Lyon, France
Power Development Authority

Urban flood modelling using computational fluid dynamics

S. Haider, A. Paquier, R. Morel and J.-Y. Champagne
The paper discusses the use of a code solving 2D shallow scenarios; crisis management personnel also stood to benefit
water equations in an urban area. Two application exam- from it, as they could devise their plans by taking into account
ples are described: a laboratory model representing low- a sample of likely flood events.
density housing, and the city of N|“ mes (France), which
suffered a catastrophic flood in October 1988. From these In the face of growing incidents of floods, and especially in
examples, the paper shows that such 2D models can pro- inhabited areas, a need was felt to apply the latest flow
vide general flow dynamics. Moreover, they can integrate computational methods to the problem of urban flooding in a
description of some details (such as obstacles and storage more accurate form. Thanks to the progress in information
areas), which can contribute to an understanding of some technology, the computing power available has considerably
flow features and thus provide accurate results locally. increased. In order to use this computing ability to better effect,
For engineering use, the limits are computational time one way forward is to choose a more refined and complete
and number of cells: the two examples described show description of the zone selected for modelling. Naturally, this
that a grid adapted to urban topography features is a way leads to the choice of a 2D code, which carries many
to reduce these disadvantages while providing reliable advantages.
results with a convenient numerical stability, even if the

DRAFT
size of the cells varies considerably. First, building the model is much easier in 2D than in 1D for
representing an essentially 2D situation. Typical examples are
NOTATION widening of a street (for instance, in a square) or a crossroads.
g gravitational acceleration In these situations, the use of a 1D model leads to the
h water depth introduction of certain approximations not wholly consistent
i index of cells with the physical reality: horizontal water level and homo-
n index of time geneous velocity inside one cross-section as shown, for
u velocity along x-axis instance, in experiments on junctions or bifurcations of three
v velocity along y-axis rectangular channels.5–8 The results of some of these experi-
K viscosity coefficient ments9,10 have shown that, although the flow is essentially 3D
Ks Manning–Strickler coefficient near the crossings, 2D shallow water equations provide very
S second member of equations accurate results for water levels and velocities. Similar results
W variable (h, hu, hv) on a four-channel crossing have been obtained for subcritical
Z bottom level flow in experiments recently performed in the Fluid Mechanics
Dt time space (¼ tnþ1 þ 1  tn ) Laboratory of INSA (National Institute of Applied Sciences) in
Lyon (France).11
1. INTRODUCTION
The application of computational fluid dynamic methods to the The second principal advantage stems from the fact that a 2D
propagation of a flood wave through an urban zone dates back model is equally efficient regardless of the structure of the
to the pioneering work of Gallati and Braschi.1 A simplified urban zone: that is, whether it is a city centre characterised by
form of the 2D depth-averaged equation and a rather coarse a dense street network, or a suburban zone with low-density
discretisation (grid size being of the order of hundreds of housing.
metres) were used to simulate the flood advance through the
city centre of Florence (Italy). The third improvement that can be brought to the flood
simulation computations by opting for a 2D code with a
Other methods have also been used (e.g. see references 2–4 for detailed topographical basis concerns the representation of the
the more recent and complete ones). All these results have water storage and the obstacles. A 2D model can be effectively
shown the promise that the emerging discipline of CFD holds in used to model large water storage areas, such as hospitals and
helping in the efforts to combat floods. A new computational lawns. Similarly, it is difficult to imagine a town centre without
tool was thus made available to urban planners to evaluate the the presence of obstacles (vehicles) that hinder the flow and
impact of an urban development measure on future flood lead to slowing down. An accurate simulation of the effects

Water & Maritime Engineering 156 Issue ? Urban flood modelling Haider et al. 1

FIRST PROOFS TH i:/Thomas Telford/Wme/WME-12879.3d WME-12879 Page: 1 KEYWORD

 created is much easier with a topography representing a lot of by the method of least squares and applying limitations of
details such as that which can be taken into account by a 2D slopes following reference 14 in order to obtain a TVD
code. (total variation diminishing) scheme.
(b) Compute the values of W ¼ ðh; hu; hvÞ at intermediate time
Another feature that distinguishes the computations described tnþ1=2 in the middle of the edge of cell Mi to obtain a
in this paper is the use of the depth-averaged equations in their second-order scheme.
complete form. This is an important consideration because the (c) Solve a 1D Riemann problem in the direction normal to the
simplified versions of these equations (diffusive or cinematic edge at tnþ1=2 in order to estimate the fluxes through edges
waves) cannot correctly represent the flow in the case of for the conservative part of the equations. It is possible to
irregular topography and of a sudden change of the flow use a Roe-type linearization,15 which provides an estimate
profile.12 of the fluxes directly.
(d ) Integrate the terms of the second member of the set of
Here, both subcritical and supercritical flows across complex equations (1)–(3) on the surface of the cell in order to add
topography are modelled, and in particular the change in the the corresponding contribution, and obtain the final value
results due to the introduction of details is analysed. Two cases of the 3D variable Winþ1 by adding a second member
are considered. One corresponds to a laboratory model estimated at an intermediate time.
incorporating scattered obstacles. These obstacles represent
villages in a floodplain submerged by a dam-break wave. The Gravity or slope terms (gh(qZ=qx) or gh(qZ=qy)) are treated
results of modelling with and without the obstacles are as fluxes in such a way that a horizontal water surface remains
compared with the water depth measurements. The second case strictly horizontal. Bottom friction terms are more simply
concerns the simulation of the 1988 flood in the southern assessed at the centre of the cell; their computation uses an
French city of Nı̂mes. An area comprising 50 streets is chosen implicitation in time in order to avoid numerical instabilities
to create a very detailed computation grid. A sensitivity when a rapid change in water depth or velocity occurs.
analysis is performed on two topics: introduction of storage Diffusion terms are treated as fluxes: first the derivatives are
areas and representation of parked cars. computed at the middle of the edges; then, these derivatives are
treated in the same way as the original variables.
2. EQUATIONS AND NUMERICAL SCHEME
Code Rubar 20, developed by Cemagref, was used. Its explicit Rubar 20 was used because it easily enables modelling of
second-order finite volume scheme solves the following set of hydraulic jump and any change from subcritical to supercritical
equations, which is one of the simplest ways to write 2D de flow.13 Moreover, drying and wetting of large areas are

DRAFT
Saint Venant equations: conveniently modelled.16 Consequently, in the examples
selected here below, the initial conditions were simply con-
qh qhu qhv stituted from a completely dry model.
1 þ þ ¼0
qt qx qy

 3. PHYSICAL MODEL OF THE TOCE RIVER VALLEY

h2 ENEL (Italian Agency for Electricity and Dams) constructed a
q hu2 þ g
qðhuÞ 2 qðhuvÞ
þ þ physical model at its Milan hydraulic laboratory in order to
2 qt qx qy
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
reproduce a 5 km reach of the Toce river in the northern Alps.



qZ u u2 þ v2 q qu q qu The model took into account houses, a reservoir, a dam and two
¼ gh g 2 1=3
þK h þ h
qx Ks h qx qx qy qy bridges. The model was then subjected to flooding and depths

 of water measured at 28 selected points in the model.
h2
q hv2 þ g
qðhvÞ qðhuvÞ 2 In the framework of the European CADAM (Concerted Action
þ þ
3 qt qx qy
for Dam Break wave), these experiments were used to evaluate
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 


qZ v u2 þ v2 q qv q qv the performance of the computer simulation methods. The
¼ gh g 2 1=3
þK h þ h
qy Ks h qx qx qy qy calculated water depths at these 28 stations were compared
with the measured values in order to assess the amount of
in which u and v are the velocities along the x-axis and y-axis confidence that can be put on the predictive capacity of the
respectively, h is the water depth, Z is the bottom level, g is the simulation methods.17
gravitational acceleration, Ks is the Manning–Strickler coeffi-
cient, and K the viscosity coefficient, which is estimated as a 3.1. Model details and information supplied
constant. The scale of the model was 1:100, and the total dimensions of
the experimental facility were 55 m by 13 m.
The mesh is constituted of quadrilaterals or triangles that have
0 or 1 (full) common edge. The mixing of these two kinds of Two floods were modelled, corresponding respectively to the
cell provides various possibilities for adapting to any detail of upstream discharge hydrographs A and B. Hydrograph B was
topography. greater in intensity, with a peak of 0·356 m3/s, which over-
topped the reservoir embankment (Fig. 1), whereas hydrograph
The Godunov-type scheme includes four steps:13 A was of lesser intensity (peak of 0·21 m3/s) and did not
overtop the embankment. Discharges and water depths were
(a) Compute the slope of each one of the three variables h (or provided at the upstream cross-section with a time step of 1 s
z, water level), hu and hv in every cell on the x- and y-axes during a period of 180 s, which included rise and fall of the

2 Water & Maritime Engineering 156 Issue ? Urban flood modelling Haider et al.

FIRST PROOFS TH i:/Thomas Telford/Wme/WME-12879.3d WME-12879 Page: 2 KEYWORD

 served to represent the col-
lective influence of the
houses on the flow, but the
detail regarding the individ-
ual units was lost.

(b) We were unable to repre-

sent the exact points in the
horizontal plane corre-
sponding to the vertices of
the house groups. Note that
this constraint could be over-
come by opting for a finer
mesh, so that the size of a
Fig. 1. Toce model: refined mesh and flooded area after 120 s rectangle making up the
mesh was of the order of the
smallest house, but the com-
water levels. The results detailed here below concern hydro- putation time would be greatly increased.
graph B only, but the behaviour of hydrograph A is quite
similar. 3.3. Computation and results
The initial computation was performed with a maximum space
The topographic data was supplied in two forms: step of 30 cm, a total of 8400 cells, zero diffusion, and a
Manning’s value of 0·0162 m1=3 /s (Strickler coefficient of
(a) a digital terrain model that divided the projected surface 60 m1=3 /s) recommended by the ENEL. This last value—
area of the model into a 5 cm2 mesh and gave the corresponding to the concrete from which the model was
altitude—that is, coordinate z—of each node. Finally the made—was kept constant through all the computations because
three coordinates (x, y and z) of 140 985 points were given tests from the CADAM project proved it was a convenient
(b) 67 cross-sections, containing 14 433 points. value.17 A second computation was conducted with a refined
mesh defined by a maximum space step of 15 cm comprising
In addition, topographical data concerning the dam and bridges 15 800 cells (Fig. 1). Each of the two computations considered

DRAFT
were also supplied. two cases, namely with and without houses.18

3.2. Representation of the houses in the model The inclusion of the houses led to an increase of water depth
The Toce model contained 173 houses, spread over the model only at those measuring stations that were situated close to a
area. These are concrete blocks of varying height designed to large group of houses. In fact, only three groups of houses (G1,
represent the real houses in the valley. Some blocks are G2 and G3 in Fig. 1) exercised any influence on the results at
grouped (interval between them smaller than their own the stations in their vicinity. One explanation is the strong
dimension) whereas other ones are isolated and scattered. supercritical flow in the model, which precluded any transmis-
sion of waves carrying information upstream from the
When ‘houses’ are grouped, it was decided to merge the obstacles. Fig. 2 shows that the representation of the houses by
individual houses into a large
house, of which the perimeter
is the periphery of the group.
Starting from 173 houses, 17
groups of houses were
obtained.

The coordinates of the ver-

tices of these house groups
were than matched with the
coordinates of the nodes of
the grid, and as the space
discretisation left a finite
number of points, the node
nearest to a vertex in the (x,
y) plane was selected. This
node then replaced the corre-
sponding vertex. Thus two
types of approximation arise
in the houses representation:
Fig. 2. Toce model: water level at gauge S4 (coarse mesh)
(a) The merger into groups

Water & Maritime Engineering 156 Issue ? Urban flood modelling Haider et al. 3

FIRST PROOFS TH i:/Thomas Telford/Wme/WME-12879.3d WME-12879 Page: 3 KEYWORD

 4. THE 1988 FLOOD IN
7 .35
THE CITY OF NI“MES
7 .32
4.1. General features of
7 .29 the model
On 3 October 1988 an
Water level: m

7 .26 extreme flood occurred in the

southern French city of
7.23 Nı̂mes. It was caused by a
generalised storm that poured
7 .2 down about 150 mm of rain
in 3 h and 250 mm in 6 h.
7 .17 Modelling the event in part of
Experiment
Coarse mesh: houses the city was undertaken. First,
7 .14 Coarse mesh: no houses
Refined mesh: houses a database of the zone con-
Refined mesh: no houses
cerned, comprising around 50
7 .11
0 20 40 60 80 100 120 140 160 180 200 streets, was built from varied
Time: s data (urban database and
field surveys). The zone
Fig. 3. Toce model: water levels at gauge P21 measured about 1 km along
the north–south axis, which is
also the principal flow direc-
their description as topography provides more ‘realistic’ results tion. Structurally, the zone is varied, characterised by a rather
than a representation by increased friction (Manning’s value of steep upper part followed by a middle part where the width
0·04 m1=3 s), although neither representation shows the same attains its minimum value, and a third southern zone of
variations of water level as the experimental measurements. narrow, straight streets crossing at right angles.

At gauge P21, situated about 50 cm from a large group of The primary building block for the database was the street
houses (Fig. 1), various computations yielded intersecting cross-sectional profiles (about 200). As no other topographical
results. The relative proximity of this measurement point or geometrical information was available concerning the street

DRAFT
allowed us to assess the effect of various changes in the mesh intersections, they were interpolated from the street profiles. A
on the results. Fig. 3 shows that one clear tendency witnessed simplified typical street profile contained seven points (coordi-
with both types of mesh after the introduction of the houses in nates x, y, z for each point) and corresponded to two sidewalls
the model was the rise in the water level, which is under- (four points), the two gutters (two points), and the mid point of
standable as such a change leads to a reduction of the section the road section. Each sidewall was slightly inclined so that the
available for the flow. The refined mesh computation with coordinates of the top and bottom of a wall had different
houses overestimates the water level, whereas the coarse mesh coordinates in the horizontal plan. This operation is necessary
with houses agrees quite closely with the measurements. It in the case of Rubar 20 to distinguish two points located on the
seems that, whereas the introduction of the houses led to same vertical that, otherwise, will be projected as the same
improved results, the refinement contributes to no further point on the horizontal plane. This configuration divided the
improvement except the capability to simulate the slight width of the street into four computation cells if walls were
decrease of water level just after the initial quick rise. One excluded (Fig. 4).
reason might be that even the refined mesh is not sufficiently
refined to describe every house individually, and consequently From the seven-point profiles, a mesh was created by linear
the flow between the houses is not taken into account: the interpolation along the streets. The mesh at the intersections
higher water level with the refined mesh is then only a (Fig. 4) followed from the selected street profile, and repre-
consequence of the narrowing of the valley because, in that sented the intersection points of the two sets of curves
precise place, refining the mesh had the consequence of emanating from the crossing streets. The minimum space step
increasing the area of houses slightly. Further tests were in the crossroads is 0·2 m, whereas the average space step along
performed locally including cells of average size 5 cm by 5 cm, the streets is about 50 m. Such a mesh includes 9093 cells, but
which makes possible the description of the topography of only about half of them are flooded.
individual blocks;18 the conclusion was similar at gauge P21.
Thus, by using this mesh, the representation of the city is
Nevertheless, generally, the introduction of the houses provided simplified to a network of streets. By contrast, the Toce
water levels closer to the experimental measurements. Simi- example shows the case of isolated obstacles. It can be hoped
larly, the wave arrival time for certain stations increased after that, by joining the two methods, the complexity of the flow
the introduction of the house data, resulting nearer from through the city could be modelled.
measurements. Although it is difficult to generalise, the tests
show that the effect of an obstacle or a group of obstacles One of the difficulties of modelling part of the flooded area
depends on its location in the flow, and thus even a detailed only is the definition of suitable boundary conditions. In the
topographical description is not enough to model local vari- present case, upstream inputs were perfectly located because a
ations of hydraulic variables. railway embankment closes the two upstream valleys and thus

4 Water & Maritime Engineering 156 Issue ? Urban flood modelling Haider et al.

FIRST PROOFS TH i:/Thomas Telford/Wme/WME-12879.3d WME-12879 Page: 4 KEYWORD

 narrow streets. The total
length of this street is 517 m
and it has a width of 5·5 m;
the part in which the obstacle
was simulated measured
95 m. The row of cars was
simulated by raising the z
coordinate of the point
between the mid point of the
road and the bottom point of
Fig. 4. Typical seven-point street cross-section and plan view of the grid at intersection the adjacent wall by 1·25 m
(see Fig. 6) in the streets,
whereas no change was
effected in the z coordinate value at the intersections on the
60
two extremities of the street part. The objective was to model a
50 Hydrograph 1 (west)
volume similar to that occupied by parked cars, without
Discharge: m3/s

40 Hydrograph 2 (east) introducing a change in friction or diffusion coefficients,

30 because such changes are difficult to define from field
observations.
20

10
The computations performed showed that the presence of the
0
obstacle caused a rise in the water level in the street section in
0 50 100 150 200 250 300 350
Time: min question but had nearly no effect elsewhere, which is consistant
with the hydraulics of the problem as the obstacle created is of
Fig. 5. Upstream discharge hydrographs quite small proportions. A curve of water depth against time at
point P1 situated in the Bons Enfants street compared the cases
before and after raising the z coordinate: it shows an increase
the flow enters the modelled area only by the two structures of peak of about 10 cm and a faster water rise (Fig. 7). Thus,
(located on Fig. 8) crossing the embankment that constitutes the because of the balancing of the water levels, even locally, the
northern border of the area; the estimated peak discharges are

DRAFT
influence is limited.
respectively 43 m3/s and 50 m3/s (Fig. 5). Definition of the
downstream conditions was not so clear, because the down-
stream adjacent areas were also flooded. It was admitted that 4.3. Introduction of a water storage area
the direction of the flow follows the bottom slope, and thus that The flooding in an urban environment is a complex phenom-
no exchange of water occurs between adjacent areas that are enon owing to the availability of storage areas for the invading
separated from the modelled area along the steeper gradients. waters. Therefore any modelling of the flood propagation
should be able to simulate this behaviour with a reasonable
A constant Strickler coefficient of 40 m1/3/s (Manning coeffi- degree of ease and accuracy. This effect was modelled for the
cient 0·025 m–1/3s) was applied to model the bottom friction, case of flooding in Nı̂mes.
and a constant diffusion coefficient of 0·01 m2/s was used to
model the exchanges between cells of various velocities. This The storage area is situated in the upper reaches of the
last value was the one calibrated from laboratory experiments modelled zone at the site of the hospital (Fig. 8). It is
that modelled the flow around an obstacle.19 The influence of surrounded on the eastern, western and northern sides by
changes in this parameter was relatively weak, possibly because streets that are bordered by walls: thus the water previously did
the highly varying space step (0·2–50 metres) creates numerical not flood this part. As several doors exist to enter the hospital,
diffusion. Similarly, the variation of bottom friction coefficient three openings were created, one for each street, in order to
changed the mean maximum water level without changing allow water to enter this area. Whereas the previous situation
main features of the flow: thus the selected value resulted from with no opening at all was unrealistic, this hypothesis certainly
a calibration based on this mean maximum water level without increases the exchanges between the streets and the storage
any local calibration, as no difference between the surfaces of area by too much.
the various streets can be noticed. With these assumptions, on
the 82 observation points, the mean maximum water level was
different by 0·03 m from measurements, although the standard
deviation reaches 0·93 m.18 This relatively high value seems to
be caused mainly by the insufficient discretisation and by the
errors or uncertainties in the bottom elevation; the possible
influence of local singularities is discussed below.

4.2. Representation of a row of cars

In order to simulate a row of cars, one part of the Bons Enfants
Fig. 6. Cross-section for the simulation of a row of cars by
street was selected. This street is aligned from west to east, and raising the level of one point
the part concerned is the centre of the southern area with

Water & Maritime Engineering 156 Issue ? Urban flood modelling Haider et al. 5

FIRST PROOFS TH i:/Thomas Telford/Wme/WME-12879.3d WME-12879 Page: 5 KEYWORD

 The slope of the area is
1. 4
adjusted to conform to the
general slope—that is, inclin-
1. 2
ing from north to south and
from west to east. The area
1. 0 Without obstacle
serves to store about
Water depth: m

With obstacle 47 000 m3 (about 10% of the

0. 8
volume of the flood), with an
average water depth of about
0. 6
1 m. Fig. 9 shows the evolu-
tion of the depth of water at
0. 4
point P2 located in the
southern part in which the
0 .2
water depth is higher: it
0
reveals the difficulty for
0 30 60 90 120 150 180 210 240 270 stored water to evacuate
Time: min because no opening was
modelled in the southern side
Fig. 7. Bons Enfants street (point P1). Effect of the simulation of a row of cars of the storage area.

In order to examine the beneficial effects of the creation of the

storage in the upstream part in terms of reduction in the flood
peak, the peak water depths are compared at the same 82 points
with and without storage. The results show a noticeable
decrease, as summarised in Table 1.

The mean maximum water level is decreased by 0·32 m

resulting in a level that is 0·29 m too low, which means that, if
the input discharges are not underestimated, this storage is by
far too important (which is the impression from the observa-

DRAFT
tions although they did not result in measurements as the water
depths were low). However, it seems that the reduction of the
flood downstream of this storage area improved the results in
some places, particularly, as it resulted in a larger proportion of
the water going to the western part of the southern flooded
area. This point is well observed by the standard deviation of
maximum water depths, which reduced from 0·93 m to 0·79 m.
So, although the global result of the introduction of the storage
area is a higher deviation, it proves that consideration of some
local or complementary flow may improve the results, or at
Fig. 8. Peak water depths with storage area least improve our knowledge of the uncertainty about these
results.

25 5. CONCLUSIONS
Two situations of floods in
urbanised milieu have been
20
analysed: one with low-den-
sity housing, the other one
Water depth: m

with dense housing. In both

15
cases the velocities are very
high (often higher than 2 m/s)
10
owing to longitudinal slopes
above 1%: consequently, flow
becomes supercritical in some
05 areas.

Using Rubar 20, which solves

0 2D shallow water equations
0 50 100 150 200 250 300
by an explicit second-order
Time: min
finite volume scheme, pro-
Fig. 9. Water depth in the hospital storage area (point P2) vides water levels close to
measurements. The tests pre-

6 Water & Maritime Engineering 156 Issue ? Urban flood modelling Haider et al.

FIRST PROOFS TH i:/Thomas Telford/Wme/WME-12879.3d WME-12879 Page: 6 KEYWORD

 Total number Increase No Decrease <10% Decrease Decrease Decrease Decrease Decrease
of points change of peak value 10^20% 20^30% 30^40% 40^50% >50%

82 4 11 7 31 6 16 4 3

Table 1. Change in maximum water depths when adding the storage area

sented in this paper proved the capability of 2D codes to area. Proceedings of Hydroinformatics 2000, Iowa City,
simulate flow around or through various obstacles simply 2000, 8.
represented by their shape. One limit is the size of the cells, as 5. BARKDOLL B. B., HAGEN B. L. and ODGAARD A. J. Experimental
too fine a discretisation may lead to such a high number of comparison of dividing open-channel with duct flow in
cells that it becomes practically impossible to compute (too T-junction. Journal of Hydraulic Engineering, 1998, 124,
long a computing time) and difficult to define all the details. No. 1, 92–95.
One other limit is the interaction between obstacles close 6. BEST J. L. and REID I. Separation zone at open-channel
together that creates waves and, more generally, essentially 3D junctions. Journal of Hydraulic Engineering, 1984, 110, No.
processes that cannot be directly represented by a 2D model. 11, 1588–1594.
7. GURRAM S. K. and KARKI K. S. Subcritical junction flow.
In the Toce case, unless an appropriate or detailed enough way Journal of Hydraulic Engineering, 1997, 123, No. 5, 447–
of representing houses is selected, a finer mesh will not 455.
necessarily provide average water levels closer to measurements 8. WEBER L. J., SCHUMATE E. D. and MAWER N. Experiments on
anywhere, although the details of the hydrographs are flow at a 90 open-channel junction. Journal of Hydraulic
generally modelled in a more accurate way. Engineering, 2001, 127, No. 5, 340–350.
9. SHETTAR A. S. and MURTHY K. K. A numerical study of
division of flow in open channels. Journal of Hydraulic
In the Nı̂mes case, although the complexity of the processes
Research, 1996, 34, No. 5, 651–675.
and the uncertainty of the data hinder a precise comparison
10. KHAN A., CADAVID R. and WANG S. S. Y. Simulation of
between observations and calculation results, it seems that the
channel confluence and bifurcation using the CCHE2D
representation of specific features (such as cars and water
model. Proceedings of the Institution of Civil Engineers—

DRAFT
storage areas) may be used to create local changes in water
Water and Maritime Engineering, 2000, 142, No. 2, 97–
levels and thus to simulate such observed or possible
102.
discrepancies.
11. VAYSSIÈRE C. Expérimentations sur modèle physique et
modélisation 2D d’une inondation en milieu urbain: cas
6. ACKNOWLEDGEMENTS d’un croisement à quatre branches (Experiments on
This work was performed in the framework of the CADAM physical model and 2D modelling of urban milieu: case of
project supported by the European Commission (Fifth Research a crossing with four branches). Student report, Génie
and Development Framework programme) and the programme Mécanique Développement, INSA, Lyon, France, 2001.
RIO (Flood risks) supported by the French Ministry of the 12. TAYFUR G., KAVVAS M. L., GOVINDARAJU R. S. and STORM D. E.
Environment. The authors thank all the partners of the Applicability of St Venant equations for two-dimensional
concerned projects overland flows over rough infiltrating surfaces. Journal of
Hydraulic Engineering, 1993, 119, No. 1, 51–63.
REFERENCES 13. PAQUIER A. 1D and 2D models for simulating dam-break
1. GALLATI M. and BRASCHI G. Simulation of the inundation of waves and natural floods. In Concerted Action on Dam-
large areas of complex topography caused by heavy floods. Break Modelling, Proceedings of the CADAM meeting,
in: HYDROSOFT 90, Proceedings of the International Wallingford, UK (MORRIS M., GALLAND J.-C. and BALABANIS
Conference on Hydraulic Engineering Software Applica- P. (eds)). European Commission, Science Research Devel-
tions (BLAIN E. and OUAZAR D. (eds)). Computational opment, Hydrological and Hydrogeological risks., L2985,
Mechanics Publications, Lowell, MA, 1990, pp. 117–126. Luxembourg, 1998, pp. 127–140,
2. HERVOUET J. M., SAMIE R. and MOREAU B. Modelling urban 14. VANLEER B. Towards the ultimate conservative difference
areas in dam-break flood wave numerical simulations. scheme. V. A second-order sequel to Godunov’s method.
Proceedings of a RESCDAM Workshop, Seinäjoki, Finland. Journal of Computational Physics, 1979, 32, 101–136.
Finnish Environment Institute, 2000, p. 12 15. ROE P. L. Approximate Riemann solvers, parameter vectors
3. HINGRAY B., BOUVIER C., CAPPELAERE B. and DESBORDES M. and difference schemes. Journal of Computational Physics,
Inondations urbaines dans les PED: un indicateur géomé- 1981, 43, 357–372.
trique caractéristique du comportement hydrodynamique 16. PAQUIER A. Modélisation et simulation de la propagation de
du bâti (Urban flooding: a geometric indicator to describe l’onde de rupture de barrage (Modelling and simulating the
the hydraulic behaviour of built-up areas). Revue des propagation of dam-break wave). PhD thesis, Université
Sciences de l’Eau (Journal of Water Science), 2000, 13, 85– Jean Monnet, Saint Etienne, France, 1995.
100. 17. SOARES FRAZAO S. and TESTA G. Numerical model analysis:
4. LAI J. S., HUANG L. H., HSIEH P. C. and LIN M. Y. Integration the Toce River test case. In: Concerted Action on Dam-
of models for simulating flood inundation in the urban Break Modelling, Proceedings of the CADAM meeting,

Water & Maritime Engineering 156 Issue ? Urban flood modelling Haider et al. 7

FIRST PROOFS TH i:/Thomas Telford/Wme/WME-12879.3d WME-12879 Page: 7 KEYWORD

 Milan, Italy (TESTA G., MORRIS M. and FABBRI K. (eds)). of a flood in an urbanised zone: 2D approach by de Saint
European Commission, Research Directorate D.I.1, Energy, Venant equations). PhD thesis, Institut National des
environment and sustainable development, L2985, Luxem- Sciences Appliquées, Lyon, France, 2001.
bourg, 2000, pp. 1–10. 19. PAQUIER A., CETINA M., KRZYK M., PROUST S. and RIVIÈRE N.
18. HAIDER S. Contribution à la modélisation d’une inondation Comparison of Slovenian and French 2D codes on river
en zone urbanisée. Approche bidimensionnelle par les flow situations. In Lyon Fleuves 2001, CD-ROM, Agence de
équations de Saint Venant (Contribution to the modelling l’Eau Rhone Méditerranée Corse, Lyon, France, 2001.

Please email, fax or post your discussion contributions to the secretary by ?? ???????? 2003: email: [email protected];
fax: +44 (0)20 7799 1325; or post to Emma Holder, Journals Department, Institution of Civil Engineers, 1^7 Great George Street,
London SW1P 3AA.

DRAFT

8 Water & Maritime Engineering 156 Issue ? Urban flood modelling Haider et al.

FIRST PROOFS TH i:/Thomas Telford/Wme/WME-12879.3d WME-12879 Page: 8 KEYWORD