See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/269168809

VALIDATION OF A QUASI-2D HYDRODYNAMIC

RIVER FLOOD MODEL USING HISTORICAL AND
ERS-SAR DERIVED FLOOD MAPS
Conference Paper September 2005

CITATIONS

READS

79

3 authors:
Luis Timbe

Patrick Willems

University of Cuenca

University of Leuven

18 PUBLICATIONS 109 CITATIONS

390 PUBLICATIONS 2,882 CITATIONS

SEE PROFILE

SEE PROFILE

J. Berlamont
University of Leuven
124 PUBLICATIONS 831 CITATIONS
SEE PROFILE

All in-text references underlined in blue are linked to publications on ResearchGate,

letting you access and read them immediately.

Available from: Luis Timbe

Retrieved on: 03 May 2016

XXXI IAHR CONGRESS 137

VALIDATION OF A QUASI-2D HYDRODYNAMIC RIVER FLOOD

MODEL USING HISTORICAL AND ERS-SAR DERIVED FLOOD MAPS
LUIS TIMBE, PATRICK WILLEMS and JEAN BERLAMONT
Laboratory of Hydraulics, K.U. Leuven, Kasteelpark Arenberg 40,
B-3001 Leuven, Belgium
(Tel: +32-16-321656, Fax: +32-16-321989, e-mail: Luis.Timbe@bwk.kuleuven.be,
Patrick.Willems@bwk.kuleuven.be)
Abstract
There is an increasing interest in river flood modelling due to the severe flood events in the
last decade throughout Europe, Asia and other parts of the world. Modelling of these events
can be performed using one-dimensional (1D) or two-dimensional (2D) hydrodynamic
models. Although 2D hydrodynamic model are the forefront for river flood modelling and
prediction, they are constrained by the high requirements of data and computational time. The
aim of this study is to test a quasi-2D modelling approach for the prediction of the flood
extent along floodplains. The model used was implemented for the river Dender in Belgium,
where floods occurred frequently during the past 10 years. The historical flood events of Dec.
1993 and Jan. 1995 were simulated using the river modelling system Mike 11/Mike GIS. The
model results were validated using historical flood maps delineated by water authorities and
ERS-SAR satellite radar derived flood maps. Both types of floodmaps delivered
complementary validation information. After detection and correction of the coarse model
inaccuracies, the model became accurate in the description of the spatial extent of floods and
its temporal evolution.
Keywords: DEM; Floods; Floodplain modeling; Hydrodynamic models; ERS-SAR
1. INTRODUCTION
Many recent studies have focused on river flood modelling due to the extreme flood events
occurred the last decade over the world (Badji and Dautrebande, 1997; van der Sande et. al.,
2002; Delmeire, 1997; Nachtnebe, 2003). Been flooding one of the most costly natural hazard
in human lives and infrastructure (Nachtnebe, 2003; Brown and Damery, 2002; Wolfgang,
2002; Oberstadler et al., 1997) its prediction and prevention has become an important issue in
the policy of water and environmental authorities
Nowadays, the availability of high resolution Digital Elevation Models (DEMs) to
represent the earth surface allows coupling hydraulic models with Geographic Information
Systems (GIS) to obtain the flood extent and water levels in floodplains. Many studies on
flood mapping have been conducted using 1D or 2D hydrodynamic models (Horritt and Bates,

138 September 11~16, 2005, Seoul, Korea

2002; Sinnakaudan et al., 2002; Bates et al., 1997; Ahmad and Simonovic, 1999). As stated
by Masson et al. (2002) the 2D hydraulic models are the state of the art for river flood
modelling. Although 1D models are accurate in the main river channel they are not valid for
over bank flow (Bates et al., 1997). Nevertheless, 2D hydrodynamic models are also
constrained by the high requirements for data, hardware and software.
A compromise solution between a 1D and 2D model is the quasi-2D approach. In this
approach the floodplains are modelled as separated river branches connected to the main river
by link channels. The link channels work as weirs, allowing the water overflow to the
floodplain when the water level exceeds the river embankment or dike. The aim of this
investigation is to test the quality of a quasi-2D hydrodynamic model for flood modelling
along floodplains for some historical flood events. The generated flood maps were validated
using historical and ERS SAR derived flood maps.
2. STUDY AREA
The study was performed for the river Dender located in the south-west part of Flanders
(Belgium) with a length of 50 km, and a contributing area of 708 km2 (see Fig. 1). Along the
river there are eight hydraulic structures (weir/sluice combinations) to regulate the water level.
Time series of hydrometric data (discharge/stage) were available at the upstream and
downstream boundaries and water levels upstream/downstream of the hydraulic structures.
Along the river floodplains, a high resolution DEM based on Laseraltimetry (LiDAR) was
available; with a horizontal resolution of 4 m. This DEM was not accurate at the line elements
(e.g. river dike embankments, roads). It therefore was corrected particularly for the river dike
levels, for which the accuracy is important in the flood modelling and mapping applications.
Along the main river Dender, there was also a cross sections survey available, with crosssections approximately every 50 m. These were used to correct the DEM at the left and right
embankment tops. Also along the river bed the DEM was corrected to replace the water level
elevations by the real river bed elevations derived from the cross-sectional survey.

Fig. 1 Location of the Dender river basin and overlay with the Dender subbasins,
the main river network and the locations of the hydraulic structures

XXXI IAHR CONGRESS 139

3. METHODS
For the river Dender, a quasi-2D model has been set-up in previous studies (Willems et al.,
2001; Willems et al., 2002a,b) on the basis of the river modelling system Mike 11. Mike 11
solves the vertically integrated equations of Saint Venant for the conservation of momentum
and volume in a one-dimensional way:
Q A
+
=q
x t

(1)

Q2

A
h qQ Q
Q

+
+ gA + 2
=0
(2)
t
x
x C AR
where: Q is the discharge, A the flow area, q the lateral inflow, h the stage above datum, C
the Chezy resistance coefficient, R the hydraulic or resistance radius and the momentum
distribution coefficient.
The hydrodynamic module of Mike 11 uses an implicit, finite difference scheme for
computation of unsteady flows in rivers and estuaries. In addition, critical and subcritical flow
conditions can also be described (DHI, 2002).
The quasi-2D approach means that the floodplains along the river are modelled using a
network of 1D river branches (hereafter called floodbranches) and spills (overflows) in
between the main river branch and the floodbranches to represent the river dikes. Spills may
also be used in between different floodplain areas.
In the first stage the potential flood risk zones for the Dender were identified based on the maps
of recent floods ROG for Flanders which describe the maximal spatial extent of historical
floods for the last 12 years (SADL, 2000). To setup the quasi-2D hydrodynamic model for the
Dender a first set of floodbranches were implemented for the floodplains along the ROG regions
(Fig. 2). The cross sections for these floodbranches were extracted from the DEM.

Fig. 2 Overlay of the DEM with the ROG maps for the upstream part of the Dender, together
with an illustration of the implementation of floodbranches in the quasi-2D
hydrodynamic model

140 September 11~16, 2005, Seoul, Korea

The Mike GIS interface was used to perform the flood mapping (the spatial 2D-mapping of
the water level simulation results in the quasi-2D model), by linking the Mike 11 results with
the geographic information system ArcView. The flood mapping is performed using a
weighted extrapolation routine on the water levels in the calculation nodes of the
hydrodynamic model results (DHI, 2001).
The flood model is validated, comparing the spatial extent results from the flood model
with the ROG flood maps for a number of historical events. The historical flood events of
December 1993 and January-February 1995 were simulated for the Dender basin. The
hydrodynamic model results were imported in the Mike GIS interfase to perform the flood
mapping at the peak moment, and to allow comparison with the ROG maps. At this level
coarse inaccuracies were detected and corrected when possible. The model was updated
adding floodbranches for areas where the model shows flooding, but which are not present in
the ROG maps. In the second phase, the model results were also compared with the flood
maps derived from ERS-SAR satellite images (Willems et al., 2003). The comparison was
made at the specific moments of the time acquisition of the images (Table 1). It has been
shown by Willems et al. (2003) that the ERS-SAR derived flood maps have significant
underestimations of the flooded areas due to water turbulence, wind effects and shallow water
problems. In section 4, it will become clear that these flood maps are, however, still very
useful for flood validation purposes.
Table 1. ERS SAR derived flood maps acquisitions
Flood event
Event 1
Event 2

Date ERS-SAR Imagery

12/31/1993 23:00
02/02/1995 23:00

To measure the fit between the historical flood maps (observed) and the simulated flood
maps (modelled) the goodness-of-fit index presented by Bates and De Roo (2000) and Horrit
and Bates (2002) was used:
A I Amod
F = obs
(3)
Aobs U Amod
where Aobs and Amod represent the total number of observed and predicted inundated pixels
respectively.
The index F lies in the range [0-1] satisfying the general criterion to measure model
performance. The numerator represents the common pixels in both the observed and predicted
flood maps, while the denominator is the union of both flood maps. F is equal to zero when
there is no overlap between the observed and predicted flooded area, and is equal to one when
both flooded areas are exactly the same.

XXXI IAHR CONGRESS 141

4. RESULTS AND DISCUSSION
The ROG historical flood maps were collected from different sources: photos, newspapers,
the footprint/trace left by the flood, notes taken by water authorities; therefore their
delineation is not always that accurate. When a comparison is made with topography,
discrepancies are often found in the sense that flooded areas are missed or overestimated. An
attempt to improve the accuracy of the ROG maps was made by Massari (2004) calculating
the mean elevation along the boundary of the ROG together with a confidence interval. Later
these elevations were plotted as inundation depth in Mike GIS to get new corrected ROG
maps. The corrected historical maps were used for the final validation of the quasi-2D model.
The flood mapping based on interpolation and extrapolation of water levels gives accurate
results only for areas where water levels have been simulated by the quasi-2D hydrodynamic
model. The initial simulated flood maps revealed some areas where additional floodbranches
were needed. When the water level in the river channel exceeds the embankment or dike, the
flooded area can be largely overestimated if no floodbranch is implemented along this area.
This is due to the extrapolation routine. The floodbranch therefore had to be enlarged when
the flooded area was larger than the area covered by the floodbranch.
Another problem found was the presence of gaps in the DEM. Due to the grid resolution of
4 m, the line elements such as dikes, roads, railways are not accurate (they can contain unreal
gaps). The model is very sensible to this problem and just a few cells can cause a large
overestimation. This problem was avoided along the river embankments integrating the cross
section survey with the DEM. Another problem encountered was the underestimation of the
flooded area because of line elements stopping the water level extrapolation in the DEM,
while in reality the water can flow under these elements. This mistake is common at roads and
railways, with the presence of brook culverts. This problem could be solved implementing a
culvert or another hydraulic structure in the hydrodynamic model.
4.1. FLOOD EVENT OF DECEMBER 1993
Fig. 3 presents the final flood mapping results for the river reach between chainages 10470
m and 25050 m. This region presents most of the floods along the Dender. As can be seen the
predicted flooded area (Fig. 3b) at the peak moment and the corrected ROG (Fig. 3a) show on
the average a good agreement, with a value F equal to 0.56. In general there is an
underestimation of the flooded area by the model. This can be explained by the inaccuracies
in the boundary delineation of the ROG maps. Even after correction there remains a high
uncertainty on the delineated boundary for these maps. Fig. 3a presents also an overlay with
the ERS-SAR derived flood map. The ERS-SAR acquisition was just after the peak moment,
and it shows that in some areas the model is more correct than the ROG data, see for instance
the area neighbouring the Idegem weir. At the right side of the Dender, the ERS-SAR map
confirms the flood in this floodplain while in the ROG map this zone is missing. Also at the

142 September 11~16, 2005, Seoul, Korea

downstream part (close to chainage 25050) the ERS-SAR image shows a larger flooded area
(and closer to the model results) than the ROG.
The floodplain of Denderbellebroek (see Fig. 4) is a natural flood storage area affected by
the tidal influence of the downstream Dender boundary along the river Schelde and it is
flooded quite frequently. For the flood event of Dec. 1993 there is no information for this area
in the ROG map, but according to the model results this area is completely flooded. For
comparison reasons, the predicted flooded area in Fig. 4a is plotted at the same time moment
than the derived ERS-SAR flood map. As mentioned before the ERS-SAR acquisition is not
at the peak moment; therefore we can presume an underestimation of the flood extent in the
ERS-SAR derived flood map. Based on a comparison with the topography, it was furthermore
concluded that the flooded area detected by ERS-SAR in this region was underestimated even
for the time moment given. After correction based on the floodplain topography, similar
flooded area was reached as the model results.

Fig. 3 Flood maps at the region between chainages 10470 m - 25050 m a) corrected ROG at
the peak moment and ERS-SAR derived flood map at 12/31/1993 23:00, b) Mike
11/Mike GIS at the peak moment

XXXI IAHR CONGRESS 143

Fig. 4 Flood maps for the floodplain Denderbellebroek, chainages: 43994-46300 at

12/31/1993 23:00, a) ERS-SAR derived flood map, b) Mike 11/Mike GIS
4.2. FLOOD EVENT JANUARY 1995
The simulated event at the end of January 1995 has a higher return period than the event of
December 1993. The model results for this event are similar than the previous one in terms of
agreement with the ROG map. Fig. 5 presents the results for the area between chainage 10470
m 25050 m. The F value equals 0.59. In contrast with the ERS-SAR image of Dec. 1993,
for the flood event of Jan. 1995 the ERS-SAR image acquisition was 3 days after the peak
moment. For this reason, the additional comparison with the ERS-SAR derived flood map
was less useful. Only for the Denderbellebroek floodplain, where the residence time of the
flooded water is long, the comparison with the ERS-SAR made sense.

144 September 11~16, 2005, Seoul, Korea

Fig. 5 Flood maps at the region between Idegem and Pollare, a) corrected ROG at the peak
moment, b) Mike 11/Mike GIS at the peak moment

Fig. 6 Flood maps for the floodplain Denderbellebroek, chainages: 43994-46300, a) ROG and
Mike 11/Mike GIS at the peak moment, b) ERS-SAR and Mike 11/Mike GIS at
2/2/1995 23:00
The comparison between the predicted model results at the peak moment and the ROG
derived flood map is shown for the Denderbellebroeck in Fig. 6a. As can be seen the results

XXXI IAHR CONGRESS 145

are quite accurate, with F close to 1. Fig. 6b presents the overlay of Mike 11/Mike GIS model
results with the ERS-SAR derived flood map at the same time moment as the ERS-SAR
acquisition. For this floodplain once again there is a good match between the model results
and the ERS-SAR based flood map. For this event the underestimation of the ERS-SAR
derived flood map is less than for the Dec. 1993 event, possibly due to the higher water level
in the floodplain.
5. CONCLUSIONS
The results of flood mapping using a quasi-2D flood model show in general a good
agreement with available historical flood information. The flood modelling approach uses a
1D physically based hydrodynamic model, with river branches on the floodplain connected to
the main river. In comparison with a full 2D model this approach has the advantage of lower
requirements for data and computational time.
Historical flood maps were useful to identify the potential flood risk zones along the river,
where floodbranches needed to be implemented in the quasi-2D flood model. ERS-SAR
derived flood maps based on satellite radar data systematically underestimate the flooded
areas. Some problems of radar based images for detecting water bodies have been confirmed
in previous studies (Horrit and Bates, 2002; Oberstadler, 1997; Smith, 1997) because of
meteorological conditions and land cover, i.e. when the water surface in not smooth due to
wind or turbulence, or due to some land cover type. Regardless of their inaccuracies (mostly
underestimations) ERS-SAR derived flood maps have shown to be useful complementary
information for the validation and refinement of flood simulation models. They provide
information on the zones along the river where some flooding occurred and an approximation
of the spatial extent. In some cases ERS-SAR flood maps can be the only source of
information.
Comparing the simulation results of the flood model with the different flood maps the
model could be evaluated and the inaccuracies identified. The most sensitive factor is the
wrong representation of the line elements in the DEM. This problem can be avoided
correcting the DEM using measured topographical elevations along river embankments, dikes,
roads and railways (e.g. cross-sectional survey). The agreement between the predicted and
observed flooded area using the F coefficient with a value close to 0.6 for both studied
historical flood events is acceptable taking into account the quality of the historical maps. As
seen, the ERS-SAR images can help in identifying additional flooded areas not present in the
other historical flood information (the ROG flood maps in this case), and therefore to improve
the F coefficient.
ACKNOWLEDGEMENTS
The work is supported by the Selective Bilateral Agreements with Latin American
Universities, through the PhD scholarship of L. Timbe and by the Fund for Scientific

146 September 11~16, 2005, Seoul, Korea

Research-Flanders (F.W.O.-Vlaanderen), through the postdoctoral scholarship of P. Willems.
The data were made available and the model set-up supported with projects by the Flemish
Water Administration AWZ.
REFERENCES
Ahmad, S. and Simonovic, S.P. (1999). Comparison of one-dimensional and two-dimensional
hydrodynamic modelling approaches for Red river basin, Natural resources institute,
University of Manitoba, Canada.
Badji, M. and Dautrebande, S. (1997). Characterization of flood inundation areas and
delineation of poor drainage soil using ERS-1 SAR imagery, Hydrological Processes, Vol.
11, pp. 1441-1450.
Bates, P. D., Horrit, M. S., Smith, S. N. and Mason, D. (1997). Integrating remote sensing
observations of flood hydrology and hydraulic modeling, Hydrological Proccesses, Vol.
11, pp. 1777-1795.
Bates, P. D. and De Roo, A. P. J. (2000). A simple raster-based model for floodplain
inundation, Journal of Hydrology, Vol. 236, pp. 54-77.
Brown, J. D. and Damery, S. L. (2002). Managing flood risk in the UK: towards an
integration of social and technical perspectives, Transactions of the Institute of British
Geographers, Vol. 27, pp. 412-426.
Delmeire, S. (1997). Use of ERS-1 data for the extraction of flooded areas, Hydrological
Processes, Vol. 11, pp. 1393-1396.
DHI (2001). Mike 11 GIS Floodplain Mapping and Analysis, User Guide Manual, Danish
Hydraulic Institute, Hrsholm, Denmark.
DHI (2002). Mike11 A microcomputer-based modelling system for rivers and channels, User
and reference manual, Danish Hydraulic Institute, Hrsholm, Denmark.
Horritt, M. S. and Bates, P. D. (2002). Evaluation of 1D and 2D numerical models for
predicting river flood inundation, Journal of Hydrology, Vol. 268, pp. 87-99.
Mason, D.C., Cobby, D.M., Horritt, M.S. and Bates, P.D. (2002). Two-dimensional
hydraulic flood modelling using floodplain topographic and vegetation features derived
from airborne scanning laser altimetry, EGS XXVII General Assembly, Nice, book of
abstracts.
Massari, P. (2004). Flood risk assessment combining hydraulic modelling with probabilistic
approaches, Ungraduated thesis, K.U. Leuven, Belgium and Politechnico di Torino, Italy.
Nachtnebel, H. P. (2003). New strategies for flood risk management after the catastrophic
flood in 2002 in Europe, Proc. 3th DPRI-IIASA International Symposium on Integrated
Disaster Risk Management: Coping with Regional Vulnerability, Kyoto.
Oberstadler, R., Hnsch, H. and Huth D. (1997). Assesment of the mapping capabilities of
ERS-1 SAR data for flood mapping: a case study in Germany, Hydrological Processes,
Vol. 11, pp. 1415-1425.

XXXI IAHR CONGRESS 147

SADL (2000). Mapping of the natural and recent flooding areas in Flanders, Report for the
Ministry of the Flemish Community, Administration AMINAL, by K.U. Leuven, Division
SADL, Leuven (in Dutch).
Smith, L. C. (1997). Satellite remote sensing of river inundation area, stage, and discharge: a
review, Hydrological Processes, Vol. 11, pp. 1427-1439.
Sinnakaudan, S. A., Ab Ghani, A. and Kiat, C.C. (2002). Flood inundation analysis using
HEC-6 and ArcView GIS 3.2a, Proc. 5th International Conference on Hydrosciences and
Engineering, Warsaw, Poland.
van der Sande, C. J., de Jong, S. M. and de Roo, A. P. J. (2003). A segmentation and
classification of IKONOS-2 imagery for land cover mapping to assist flood risk and flood
damage assessment, International Journal of Applied Earth Observation and
Geoinformation, Vol. 4, pp. 217-229.
Willems, P., Vaes, G., Berlamont, J., Cristiaens, K., Popa, D., Timbe, L. and Feyen, J. (2001).
Methodology for river flood modelling by the quasi two-dimensional approach, Proc.
World Water and Environmental Resources Congress, Orlando.
Willems, P., Timbe, L. and Popa, D. (2002a). Mike11 model Dender, version April 2002,
Report including the results of the research project Implementation of the quasi 2D
approach for floodplain modelling for the Mike 11 model of the river Dender for the
Flemish Ministry for Waterways and Sea Affairs (AWZ - WL Borgerhout), Report by
K.U.Leuven, Hydraulics Laboratory, Leuven (in Dutch).
Willems, P., Vaes, G., Popa, D., Timbe, L. and Berlamont, J. (2002b). Quasi 2D river flood
modeling, River Flow 2002: International Congress on Fluvial Hydraulics, Louvain-laNeuve, Bousmar & Zech eds., Vol. 2, pp. 1253-1259.
Willems, P., Thompson, S. and Barbieri, M. (2003). Flood risk and damage assessment using
modelling and earth observation techniques (FAME), Final Report, ESA-DUP 2 Small
Services Project, K.U. Leuven, Spatial Applications Division Leuven, Belgium.
Wolfgang, K. (2002). Flood risk = hazard x exposure x vulnerability, Flood Defense 2002,
In: Wu et al. (eds), Science Press, New York Ltd., Vol. 1, pp. 82-97.