Journal of Hydrology 367 (2009) 276–282

Contents lists available at ScienceDirect

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

A technique for the calibration of hydraulic models using uncertain satellite

observations of flood extent
Giuliano Di Baldassarre *, Guy Schumann, Paul D. Bates
School of Geographical Sciences, University of Bristol, University Road, Bristol BS8 1SS, UK

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

Article history: This paper presents a methodology to calibrate hydraulic models using possibility of inundation maps
Received 22 October 2008 derived from satellite imagery. The study was performed on a river reach of the Lower Dee, UK, where
Received in revised form 13 January 2009 a coarse resolution image (ENVISAT ASAR) and high resolution satellite image (ERS-2 SAR) were acquired
Accepted 17 January 2009
at the same time during the December 2006 flood event. Ten different flood extent maps were derived
from the two flood images by using five different procedures to process these data. These flood extent
This manuscript was handled by K. maps were used to perform a sensitivity analysis of a simple raster-based inundation model (LIS-
Georgakakos, Editor-in-Chief, with the FLOOD-FP). The sensitivity analysis enabled us to investigate the capability of the two different resolution
assistance of Attilio Castellarin, Associate images to calibrate the friction parameters of the flood inundation model. The analysis showed that the
Editor. optimal parameters of the model depend on the type of satellite image used to evaluate the model as well
as on the particular procedure used to derive the flood extent map. Finally, the study developed a novel
Keywords: methodology to calibrate flood inundation models by comparing the model results to a possibility of
Hydraulic models inundation map obtained by combining the ten different flood extent maps.
Remote sensing Ó 2009 Elsevier B.V. All rights reserved.
Calibration
Flood inundation maps

Introduction Advances in numerical modelling techniques, the availability of

The increased socioeconomic relevance of river flood studies
and a shift of these studies towards integrated flood risk manage-
ment concepts have triggered the development of complex meth-
odologies for the simulation of the hydraulic behaviour of river
systems. This has strengthened knowledge in the field of hydraulic
numerical modelling and has resulted in the development of a
plethora of numerical codes that have proved to be useful tools
for floodplain management, flood risk assessment and understand-
ing sediment dynamics (e.g. Horritt et al., 2007).
The availability of distributed remote sensing data for model
calibration and evaluation has led to a sudden shift from a data-
poor to a data-rich environment for flood modelling (Bates,
2004). Remote sensing techniques allowed considerable progress
in the field scale application and testing of flood inundation models
(e.g. Schumann et al., 2007). Specifically, airborne laser altimetry
data produces a wealth of topographic information for inclusion
into inundation models (e.g. Cobby et al., 2001; Castellarin et al.,
2009) and flood extent maps derived from satellite synthetic
aperture radar (SAR) sensors have been used to calibrate and eval-
uate hydraulic models (e.g. Horritt et al., 2007).
* Corresponding author. Tel.: +44 393802956454.
E-mail address: g.dibaldassarre@bristol.ac.uk (G. Di Baldassarre).
distributed remote sensing data and increased computational
resources have enabled the production of uncertain flood inunda-
tion maps conditioned on single observed inundation images from
single events (Aronica et al., 1998, 2002; Hall et al., 2005; Pappen-
berger et al., 2005; Romanowicz and Beven, 1998, 2003). Bates
et al. (2004) extend this to show how single images from multiple
events can be combined to update model-predicted flood inunda-
tion probabilities using Bayes’ theorem. As a further advance Hor-
ritt (2006) proposed a new technique for the verification of
hydraulic models based on the comparison of uncertain flood inun-
dation models with a single observed wet/dry map. However, all
these techniques treat the observed flood extent data as determin-
istic when in reality the flood extent data themselves will be sub-
ject to considerable uncertainty.
Pappenberger et al. (2007) proposed a methodology for the
estimation of inundation extent uncertainty, which implements
a fuzzy evaluation methodology and considers the uncertainty
in the observed spatially distributed information. More recently,
Schumann et al. (2009) proposed an approach to derive, from sa-
tellite images, possibility of inundation maps rather than the tra-
ditional deterministic binary maps. This type of observation
requires the development of a novel methodology for flood model
evaluation within an uncertainty framework that deals with the
presence of errors in both model and observation (Beven, 2006)
and this therefore is the primary focus of this paper. Accordingly,

0022-1694/$ - see front matter Ó 2009 Elsevier B.V. All rights reserved.
doi:10.1016/j.jhydrol.2009.01.020
 G. Di Baldassarre et al. / Journal of Hydrology 367 (2009) 276–282 277

we report the development of a new methodology for the calibra-
tion of flood inundation models based on the comparison of mod-
el results and uncertain spatial observation of flood extent
derived from satellite imagery. In addition, the ability of different
image resolutions to parameterise flood inundation models is
investigated.
Test site and satellite data
The study was performed on a river network consisting of a
10 km reach of the River Dee, between the two UK Environment
Agency (EA) gauging stations at Farndon and Iron Bridge, and an
8 km reach of the River Alyn, between the EA gauging station of
Pont-y-Capel and the Dee confluence (Fig. 1). In this test site, the
River Dee consists of a main channel with a width equal to about
60 m and a wide floodplain with an overall width which varies
from 200 m to 2 km (Fig. 1). Fig. 1 shows also the airborne LiDAR
DTM, at a resolution of 2 m, obtained from the EA for this reach.
The raw LiDAR data has been processed using the EA’s in-house
techniques to remove surface artefacts such as vegetation and
buildings and has a vertical accuracy of 10 cm RMSE. In-channel
bathymetry was characterised by ground survey consisting of 36
cross-sections conducted by the EA.
In the middle of December 2006, this area experienced low
magnitude prolonged flooding. During this period a coarse resolu-
tion image (ENVISAT ASAR) and high resolution satellite image
(ERS-2 SAR) were acquired at (very nearly) the same time
(Fig. 2). Specifically, the ENVISAT ASAR image was acquired on
the 12th of December 2006 at 10:39 while the ERS-2 SAR image
was acquired on the same day at 11:07. The ENVISAT ASAR image
is characterised by a pixel size of 75 m and a ground resolution of
approximately 150 m while the ERS-2 SAR image is characterised
by a pixel size of 12.5 m and a ground resolution of approximately
25 m. The difference in the timing of the two flood images (28 min)
is negligible compared to the overall duration of the prolonged
December 2006 event (around 15 days). Also, Table 1 shows the
hydrometric data observed at the EA gauging stations (Fig. 1) at
the time of the two satellite images. By analysing Table 1 one
can observe that the boundary conditions did not change during

Fig. 1. Test site: River Dee between Farndon and Iron Bridge and River Alyn between Pont-y-Capel and the confluence (black line); Environment Agency gauging stations
(black circles); flood extent derived flood extent derived by using the ERS-2 SAR using a histogram threshold procedure (gray area); LiDAR DTM at 2 m resolution (gray scale).

Fig. 2. Satellite imagery: (a) ENVISAT ASAR image (75 m resolution, acquired the 12th of December 2006 at 10:39) and (b) ERS-2 SAR image (12.5 m resolution, acquired 12th
of December at 11:07). The rectangular box shows the location of the test site.
278 G. Di Baldassarre et al. / Journal of Hydrology 367 (2009) 276–282
Table 1
Hydrometric data observed at the Environmental Agency (EA) gauging stations
(Fig. 1) at the time of the two satellite imagery.
EA gauging station ENVISAT ASAR 12-12- ERS-2 ASAR 12-12-
2006 10:39 2006 11:07
River Dee at Farndon (m a.s.l) 8.770 8.772
River Dee at Iron Bridge (m a.s.l.) 6.673 6.675
River Alun at Pont-y-Capel 37.941 37.940
(m a.s.l.)
this period as the observed water depths differed by less than
0.002 m. Therefore, in terms of the inundation process, the actual
flood extent at both acquisitions can be assumed to be the same
and all apparent difference in flood mapping on both images can
be attributed to differences in spatial resolution, given that all
other significant image properties (e.g. frequency, polarization
and incidence angle) are the same for both data sets (Schumann
et al., 2009).
The two satellite images were processed in order to derive flood
extent maps. The scientific literature described many different im-
age processing procedures to derive flood extent from SAR (e.g.
Schumann et al., 2009). Nevertheless, it is well known that there
is no single method that can be considered appropriate for all
images, nor are all methods equally good for a particular type of
image (Pal and Pal, 1993). Here, five different procedures were ap-
plied (see below, Appendix): visual interpretation (e.g. Oberstadler
et al., 1997), histogram threshold (e.g. Deshmukh and Shinde,
2005), active contour (e.g. Horritt et al., 2001), image texture var-
iance (Sali and Wolfson, 1992) and Euclidean distance (e.g. Kokare
et al., 2003). Hence, ten different flood extent maps were derived
from the two flood images. Fig. 1 shows, as an example, the flood
extent map derived from the ERS-2 SAR using the image histogram
threshold procedure.
Flood inundation model
Numerical code
The December 2006 flood event was simulated using the raster-
based inundation model LISFLOOD-FP (Bates and De Roo, 2000;
Horritt and Bates, 2001). The channel was discretised as a single
vector along its centreline separate from the overlying floodplain
raster grid. At each point along the vector the required channel
parameters are the width, Manning’s n value and bed elevation.
The latter data give the bed slope and also the bankfull depth when
the channel vector is combined with the floodplain DTM. Each
channel parameter can be specified at each point along the vector
and the model interpolates linearly between these. Flows along the
channel are simulated using an approximation to the 1D St. Venant
equations, and when bankfull depth is exceeded water spills out
onto adjacent floodplains. Floodplain flows are treated using a stor-
age cell approach first developed by Cunge et al. (1980) and imple-
mented for a raster grid to allow an approximation to a 2D
diffusive wave.
The LISFLOOD-FP model was applied to the test site using a grid
spatial resolution of 20 m. The model grid was constructed by areal
averaging of the 2 m LiDAR DTM (Fig. 1) to this larger scale. This
was found to erase some key topographic features such as embank-
ments. Therefore these embankments were identified manually
and their height values were retained at the model grid scale.
The Dee’s channel centreline was discretised with 110 points while
the Alyn channel centreline was discretised with 310 points. The
width and the slope were parameterised by using the channel
bathymetry data (‘‘Test site and satellite data”).
Sensitivity analysis
The December 2006 flood event was simulated by means of the
LISFLOOD-FP model. Hydrometric data observed at the EA gauging
stations (Fig. 1) were used as boundary conditions. In order to min-
imise the effects of the initial conditions, the unsteady simulation
was performed starting 6th of December 2006 at 11:00 (around
144 h before the satellite overpasses). Similar to other applications
of the same code (e.g. Horritt, 2006), the model sensitivity to flood-
plain friction was found to be small. Therefore the model is run
with a uniform floodplain Manning’s coefficient of 0.10 m1/3 s,
representative of floodplain conditions for this reach (e.g. Chow,
1959), and channel Manning’s coefficient in the range 0.02–
0.10 m1/3 s, a range that has been found in previous studies to
encompass the optimum channel friction coefficient value (e.g.
Aronica et al., 2002; Bates et al., 2006; Horritt, 2006).
The 10 flood extents maps derived from the two flood images
(‘‘Test site and satellite data”) are compared to that predicted by
the model using the following measure of fit F:
AB
F¼ ð1Þ
AþBþC
where A is the size of the wet area correctly predicted by the model,
B is the area predicted as wet that is observed dry (over-prediction)
and C is the wet area not predicted by the model (under-prediction).
F ranges from 1 to +1. This measure of fit has been found to give
good results for flood inundation modelling problems and allows
the comparison of model results on different reaches and flood
events (e.g. Horritt et al., 2007). In Eq. (1), the term B in the
numerator is used to penalise model over-prediction. This is consis-
tent with the fact that remote sensing imagery are expected to give
a larger flood extent as they are an aggregated response of all flood-
related processes (floodplain inundation as a result of excess rain-
fall, bank overtopping, backwater effects, complex 2D and 3D flows,
etc) whereas the model only reproduce few of these to inundate the
floodplain.
The results of the sensitivity analysis of the model are shown in
Fig. 3. Specifically, each panel of Fig. 3 is related to a given proce-
dure to derive flood extent and shows the model response to
changes in Manning’s coefficient, in terms of measure of fit F, eval-
uated by using the two satellite images. The sensitivity analysis en-
abled us to investigate the capability of the two different resolution
images to calibrate the friction parameters of the flood inundation
model.
By analysing each panel of Fig. 3, one can observe that,
although the absolute value of the measure of fit differs, Man-
ning’s channel values obtained by using the two different resolu-
tion images are similar when applying visual interpretation,
histogram threshold or Euclidean distance (Fig. 3a, b and e) to de-
rive the flood extent map. This result shows the useful information
content in coarser resolution SAR images, such as the ASAR WSM,
confirming that this type of data can be used successfully to delin-
eate flood extent on larger floodplains (Di Baldassarre et al., 2009).
In contrast, the optimal parameters obtained with the two differ-
ent resolution images are different when the image texture vari-
ance or active contour technique is applied (Fig. 3c and d). This
is in accordance with the findings by Schumann et al. (2009),
showing that these procedures are probably not appropriate for
this case study as they do not produce consistent results. The most
striking result is that different optimal Manning’s channel values
are identified during model calibration by using different proce-
dures to derive flood extent maps from satellite imagery (Fig. 3).
Therefore, the sensitivity analysis suggests that a more rigorous
calibration of the flood inundation model that can cope with the
presence of uncertainty in the flood extent observations needs to
be carried out.
 G. Di Baldassarre et al. / Journal of Hydrology 367 (2009) 276–282 279

a b c

d e

Fig. 3. Response of LISFLOOD-FP model to changes in Manning’s n channel, in terms of measures of fit F, calculated by using ten flood extent maps derived from ENVISAT SAR
(asar_) and the ERS-2 SAR (ers_) flood imagery and applying five different procedures: (a) visual interpretation (_vi), (b) histogram threshold (_t), (c) active contour (_s), (d)
image texture variance (_va) and (e) Euclidean distance (_m).

Model calibration

Methodology

Schumann et al. (2009) produced a possibility of inundation

map by combining the ten flood extent maps derived for this site
according to a particular measure of consistency (see below,
Appendix). Fig. 4 shows this possibility of inundation map where
Pj denotes the likelihood of observing inundation at the jth cell
for the 2006 flood event (see also, Appendix). The possibility of
inundation map may be a useful tool for flood risk mapping, as it
expresses our belief in whether a particular image pixel is flooded
by an event of a given magnitude (Schumann et al., 2009).
However, no studies have looked at the use of such data to con-
strain inundation models.
Accordingly, a method was developed to calibrate flood inunda-
tion models using the possibility of inundation map derived from
satellite imagery (Fig. 4), rather than the traditional single observa-
tion of flood extent. The methodology is based on the use of reli-
ability diagrams (Horritt, 2006). In particular, for each simulation
Fig. 4. Possibility of inundation map derived from satellite imagery (gray scale,
i, the reliability diagram is constructed by classifying the possibil- from black, Pj equal to 1, to white, Pj equal to 0). Note that for overlay operations
ity of inundation map (Fig. 4) into regions of similar Pj and count- following the binary procedure, the ERS-2 SAR maps have been rescaled to the
ing the number of simulated wet cells in each region. In a perfectly coarser resolution of the ASAR, i.e. 75 m.
accurate model, the proportion of simulated wet cells (model pro-
portion, Fig. 5a) would match the possibility of inundation (for Results
example half the cells classified with Pj = 0.5 are simulated as
wet). Therefore, the reliability diagram for a perfectly accurate Fig. 5b shows the RMSE values calculated in this way against
model consists of a 1-to-1 line (Horritt, 2006), and the accuracy the Manning’s channel values. The optimum lies at n = 0.06–
can be assessed by calculating the RMSE (Horritt, 2006) of the ac- 0.07 m1/3 s (Fig. 5b). It is interesting to note that the proposed
tual line from the 1-to-1 line relationship (Fig. 5a). The RMSE re- methodology enabled us to constrain the parameter of the model
flects the discrepancy between model proportion and the whereas traditional standard methodologies lead to subjectivity
possibility of inundation map. The RMSE is weighted by the num- in identifying an optimal parameter (e.g. Fig. 3).
ber of cells in that class, with areas of Pj = 0 (i.e. cells that are sim- The results of the model can be combined to produce an uncer-
ulated dry) being discounted. tain flood inundation map, conditioned on the observed possibility
280 G. Di Baldassarre et al. / Journal of Hydrology 367 (2009) 276–282
a b
Fig. 5. (a) Reliability diagram by using the possibility of inundation map and (b) response of LISFLOOD-FP model to changes in Manning’s n channel, in terms of RMSE.
of inundation map, in a Generalized Likelihood Uncertainty Esti-
mation (GLUE, Beven and Binley, 1992) framework (e.g. Aronica
et al., 2002; Horritt, 2006). In particular, each simulation i was ac-
corded a likelihood weight Li in the range [0, 1] according to the
RMSEi values (Fig. 5):
maxðRMSEÞ  RMSEi
Li ¼
maxðRMSEÞ  minðRMSEÞ
where max(RMSE) and min(RMSE) are the maximum and minimum
measures of fit found throughout the ensemble. Then, given the
simulation results for the jth computational cell of wij equal to 1
for wet and wij equal to 0 for dry, the uncertain flood inundation
map can be produced by evaluating:
P
i Li wij
Mj ¼ P
i Li
where Mj indicates a weighted average flood state for the jth cell.
Fig. 6 shows the uncertain flood inundation map obtained by com-
bining the results of these numerical simulations. While Mj is not a
probability in the strict sense, it does range between 0 and 1, and for
a good model should bear some relationship with the likelihood of
observing inundation at the jth cell for a real flood event (Horritt,
2006). The flood uncertain inundation map (Fig. 6) was compared
to the possibility of inundation map (Fig. 4). It is interesting to ob-
Fig. 6. Flood uncertain inundation map derived from model results (gray scale,
from black, Mj equal to 1, to white, Mj equal to 0). Note that for overlay operations
following the binary procedure, the model results have been rescaled to the
resolution of 75 m.
ð2Þ
ð3Þ Fig. 7. Histogram of Dj values.
serve the region around the gauging station of Iron Bridge (see also,
Fig. 1). In this region, the probability of inundation map, derived
from satellite imagery, is characterised by a much higher uncer-
tainty than the flood uncertain inundation map, derived from model
results. This result reflects the fact that close to Iron Bridge the
model results are affected by the use of the observed water level
as boundary condition.
Although the flood inundation map was derived by using the
possibility of inundation map, the comparison allows us to verify
the absence of bias. In fact, the presence of bias between the two
maps can be a useful warning about the ability of the model to
reproduce the flood inundation processes. Therefore, for each jth
cell, the quantity Dj was evaluated as Mj minus Pj. Dj values were
found to be characterised by a mean value equal to 0.04 and a
standard deviation equal to 0.32. Fig. 7 shows the histogram of Dj
values pointing out the absence of bias between the flood uncer-
tain inundation map (Fig. 6) and the possibility of inundation
map (Fig. 4).
Conclusions
This study analysed the uncertainty in observed flood extent by
using coarse resolution (ENVISAT ASAR) and high resolution (ERS-2
SAR) satellite imagery, acquired at the same time during the
December 2006 flood event for a reach of the River Dee in the UK.
A sensitivity analysis was carried out in order to investigate the
ability of the two different resolution images to parameterise a
simple raster-based flood inundation model. This analysis pointed
out that the optimal parameters of the model depend on the type
of satellite image used to evaluate the model as well as on the par-
ticular procedure used to derive the flood extent map (equifinality;
Beven, 2006).
 G. Di Baldassarre et al. / Journal of Hydrology 367 (2009) 276–282
These results pointed out the need to move from determin-
istic binary wet/dry maps to uncertain flood extent maps.
Consequently, a possibility of inundation map was obtained by
combining the different estimations of flood extent from satel-
lite imagery according to an appropriate measure of
consistency.
Then, the study investigated the possibility of calibrating
hydraulic models by using this possibility of inundation map, in-
stead of a traditional flood extent map. Specifically, a new tech-
nique to produce flood inundation maps conditioned on a
possibility of inundation map was developed. This methodology
is believed to be more reliable than the standard calibration tech-
niques that do not take into account the uncertainty in the obser-
vation of flood extent and lead to subjectivity in identifying an
optimal set of parameters.
Finally, in order to verify that the hydraulic model used to sim-
ulate the flood event is an unbiased estimator of the possibility of
inundation, the flood uncertain inundation map, derived from
model results, was compared to the possibility of inundation
map, derived from satellite imagery. This comparison pointed out
the absence of bias between the two maps. Future studies will fo-
cus on the potential of this comparison as a tool for the evaluation
and comparison of different hydraulic models.
Acknowledgements
The authors are extremely grateful to the European Space
Agency (ESA) for allowing access to the flood images used in this
study (Category 1 Project ID: 5739) and to the Environmental
Agency of England and Wales for the LiDAR data. The research re-
ported in this paper has been funded by the FREE Project of the Na-
tional Environmental Research Council (NERC, Grant number NE/
E002331/1. Guy Schumann is funded by a Great Western Research
fellowship. The authors wish to thank Tim Fewtrell and Jeff Neal
for their valuable suggestions and comments. Two Anonymous
Reviewers and the Associate Editor, Attilio Castellarin, are also
acknowledged.
Appendix
This appendix summarises the five techniques used in this pa-
per to derive flood maps from satellite imagery and the methodol-
ogy proposed by Schumann et al. (2009) to produce the possibility
of inundation map.
Prior to image processing, a Sigma Lee filter (Smith, 1996) was
applied to both SAR images to remove most speckle (i.e. random
image noise obstructing features of interest). Thereafter, binary
thematic maps showing flooded/non flooded pixels were derived
and which could be fused later to generate a weighted possibility
of inundation map reflecting uncertainty associated with observed
flood area. Many different image processing techniques may be ap-
plied to a single satellite image. In this study, we applied five
widely used algorithms.
Visual interpretation
There are a number of studies that illustrate the potential of vi-
sual interpretation to derive flooded area from satellite imagery
(e.g. Oberstadler et al., 1997). Flood area was mapped on both
images by visually digitising the flood boundaries. A skilful delin-
eation of flood shorelines by visual interpretation requires event
as well as expert flood knowledge. The obvious breaks of slope be-
tween the floodplain and adjacent hillslopes clearly visible in the
LiDAR topography were used to constrain the delineated flood ex-
tent to the valley floor area.
281
Histogram threshold method
Thresholding is a simple but widely used and efficient method
to generate a binary image (Deshmukh and Shinde, 2005). An opti-
mal gray level threshold is found using Otsus method (Otsu, 1979).
The method applies a criterion measure to evaluate the between-
class variance (i.e. separability) of a threshold at a given level com-
puted from a normalized image histogram of gray levels.
Histogram texture measures
Image texture may be modelled as a gray level function using
simple statistical methods on the image histogram. Widely used
algorithms rely on statistical properties of a neighbourhood of pix-
els which are computed for each pixel using a moving window. In
this paper, we selected the variance and the mean Euclidean dis-
tance (Irons and Petersen, 1981) both of which can be found in
most commercial remote sensing software packages.
Active contour model
A dynamic curvilinear contour searches the edge image space
until it settles upon image region boundaries. This is achieved by
an energy function attracted to edge points. The contour is usually
represented as a series of nodes linked by straight line segments
(Horritt et al., 2001). The statistical snake is formulated as an en-
ergy minimisation (Ivins and Porrill, 1994). The total energy is
minimised if the contour encloses a large area of good pixels, and
in this respect the model behaves as a region growing algorithm
(Horritt et al., 2001).
Possibility of inundation map
Fig. 4 shows the possibility of inundation map derived by com-
bining the 10 flood extent maps derived from the two satellite
images applying the five techniques. The 10 maps were combined
assigning a weight to each single flood extent map. Given that no
field data were available, a measure of consistency (i.e. relative
performance measure) was used to set the weights. In particular,
the weights reflect the relative agreement of one given processing
algorithm between both images (Schumann et al., 2009). Roughly
speaking, the idea behind this approach is that, in theory, a reliable
procedure to derive flood extent from the two satellite images
should give exactly the same flood extent (as expected given that
the actual flood extent at both acquisitions is the same, see for
more details Schumann et al., 2009).
References
Aronica, G., Hankin, B.G., Beven, K.J., 1998. Uncertainty and equifinality in
calibrating distributed roughness coefficients in a flood propagation model
with limited data. Advances in Water Resources 22 (4), 349–365.
Aronica, G., Bates, P.D., Horritt, M.S., 2002. Assessing the uncertainty in distributed
model predictions using observed binary pattern information within GLUE.
Hydrological Processes 16 (10), 2001–2016.
Bates, P.D., 2004. Remote sensing and flood inundation modelling. Hydrological
Processes 18, 2593–2597.
Bates, P.D., De Roo, A.P.J., 2000. A simple raster based model for flood inundation
simulation. Journal of Hydrology 236, 54–77.
Bates, P.D., Horritt, M.S., Aronica, G., Beven, K., 2004. Bayesian updating of flood
inundation likelihoods conditioned on flood extent data. Hydrological Processes
18, 3347–3370.
Bates, P.D., Wilson, M.D., Horritt, M.S., Mason, D.C., Holden, N., Currie, A., 2006.
Reach scale floodplain inundation dynamics observed using airborne synthetic
aperture radar imagery: data analysis and modelling. Journal of Hydrology 328
(1–2), 306–318. 30.
Beven, K., 2006. A manifesto for the equifinality thesis. Journal of Hydrology 320,
18–36.
Beven, K.J., Binley, A.M., 1992. The future of distributed models: model calibration
and uncertainty prediction. Hydrological Processes 6, 279–298.
282 G. Di Baldassarre et al. / Journal of Hydrology 367 (2009) 276–282
Castellarin, A., Di Baldassarre, G., Bates, P.D., Brath, A., 2009. Optimal cross-section
spacing in Preissmann scheme 1D hydrodynamic models. ASCE Journal of
Hydraulic Engineering 135 (2), 96–105.
Chow, V.T., 1959. Open-Channel Hydraulics. McGraw-Hill, New York.
Cobby, D.M., Mason, D.C., Davenport, I.J., 2001. Image processing of airborne
scanning laser altimetry data for improved river flood modeling. ISPRS Journal
of Photogrammetry and Remote Sensing 56 (2), 121–138.
Cunge, J.A., Holly Jr., F.M., Verwey, A., 1980. Practical Aspects of Computational
River Hydraulics. Pitman, London.
Deshmukh, K.S., Shinde, G.N., 2005. An adaptive color image segmentation.
Electronic Letters on Computer Vision and Image Analysis 5 (4), 12–23.
Di Baldassarre, G., Schumann, G., Bates, P.D., 2009. Near real time satellite imagery
to support and verify timely flood modelling. Hydrological Processes,
doi:10.1002/hyp.7229.
Hall, J.W., Tarantolo, S., Bates, P., Horritt, M., 2005. Distributed sensitivity analysis of
flood inundation model calibration. ASCE Journal of Hydraulic Engineering 131
(2), 117–126.
Horritt, M.S., 2006. A methodology for the validation of uncertain flood inundation
models. Journal of Hydrology 326, 153–165.
Horritt, M.S., Bates, P.D., 2001. Effects of spatial resolution on a raster based model
of flood flow. Journal of Hydrology 253, 239–249.
Horritt, M.S., Mason, D., Luckman, A.J., 2001. Flood boundary delineation from
synthetic aperture radar imagery using a statistical active contour model.
International Journal of Remote Sensing 22, 2489–2507.
Horritt, M.S., Di Baldassarre, G., Bates, P.D., Brath, A., 2007. Comparing the
performance of 2-D finite element and finite volume models of floodplain
inundation using airborne SAR imagery. Hydrological Processes 21, 2745–2759
(ISSN: 0885-6087).
Irons, J.R., Petersen, G.W., 1981. Texture transforms of remote sensing data. Remote
Sensing of Environment 11, 359–370.
Ivins, J., Porrill, J., 1994. Statistical snakes: active region models. In: Proceedings of
the 5th British Machine Vision Conference, September 1994, vol. 2. University of
York, BMVA Press, York, UK, pp. 377–386.
Kokare, M., Chatterji, B.N., Biswas, P.K., 2003. Comparison of similarity metrics for
texture image retrieval. In: Proceedings of the IEEE 10th Conference on
Convergent Technologies for Asia-Pacific Region, October 2003, vol. 2. IEEE,
Bangalore, India, pp. 571–575.
Oberstadler, R., Hnsch, H., Huth, D., 1997. Assessment of the mapping capabilities of
ERS-1 SAR data for flood mapping: a case study in Germany. Hydrological
Processes 10, 1415–1425.
Otsu, N., 1979. A threshold selection method from gray-level histograms. IEEE
Transactions on Systems, Man, and Cybernetics 9, 62–66.
Pal, N.R., Pal, S.K., 1993. A review on image segmentation techniques. Pattern
Recognition 26, 1277–1294.
Pappenberger, F., Beven, K., Horritt, M., Blazkova, S., 2005. Uncertainty in the
calibration of effective roughness parameters in HEC-RAS using inundation and
downstream level observations. Journal of Hydrology 302, 46–69.
Pappenberger, F., Frodsham, K., Beven, K.J., Romanovicz, R., Matgen, P., 2007. Fuzzy
set approach to calibrating distributed flood inundation models using remote
sensing observations. Hydrology and Earth System Sciences 11 (2), 739–752.
Romanowicz, R., Beven, K., 1998. Dynamic real-time prediction of flood inundation
probabilities. Hydrological Sciences Journal 43 (2), 181–196.
Romanowicz, R., Beven, K., 2003. Estimation of flood inundation probabilities as
conditioned on event inundation maps. Water Resources Research 39 (3), 1073–
1085.
Sali, E., Wolfson, H., 1992. Texture classification in aerial photographs and satellite
data. International Journal of Remote Sensing 13, 3395–3408.
Schumann, G., Hostache, R., Puech, C., Hoffmann, L., Matgen, P., Pappenberger, F.,
Pfister, L., 2007. High-resolution 3D flood information from radar imagery for
flood hazard management. IEEE Transactions on Geoscience and Remote
Sensing 45 (6), 1715–1725.
Schumann, G., Di Baldassarre, G., Bates, P.D., 2009. The utility of space-borne radar
to render maps of observed possibility of inundation. IEEE Transactions on
Geoscience and Remote Sensing, in press.
Smith, D.M., 1996. Speckle reduction and segmentation of synthetic aperture radar
images. International Journal of Remote Sensing 17, 2043–2057.