Professional Documents
Culture Documents
To cite this article: André Twele, Wenxi Cao, Simon Plank & Sandro Martinis (2016) Sentinel-1-
based flood mapping: a fully automated processing chain, International Journal of Remote
Sensing, 37:13, 2990-3004, DOI: 10.1080/01431161.2016.1192304
Article views: 25
Download by: [Library Services City University London] Date: 02 July 2016, At: 20:37
INTERNATIONAL JOURNAL OF REMOTE SENSING, 2016
VOL. 37, NO. 13, 2990–3004
http://dx.doi.org/10.1080/01431161.2016.1192304
1. Introduction
Flood extent maps derived from synthetic aperture radar (SAR) data can be a key
information source for an effective disaster management, helping humanitarian relief
organizations and decision-makers to obtain spatially explicit information about inun-
dated areas in a time- and cost-efficient manner (Voigt et al. 2007). Furthermore, such
maps can provide valuable distributed calibration and validation data for hydraulic
models of river flow processes (e.g. Schumann, di Baldassarre, and Bates 2009; Matgen
et al. 2010; Giustarini et al. 2011; Dung et al. 2011; García-Pintado et al. 2015; Mason
et al. 2015) and support the derivation of hazard maps within the scope of flood
prevention activities, insurance risk management, and spatial planning (e.g. De Moel,
van Alphen, and Aerts 2009).
Since flood situations often occur during long-lasting periods of precipitation and
persistent cloud cover, which, in many cases, hamper the utilization of optical imaging
instruments, spaceborne SAR systems are usually the preferred tool for observing flood
CONTACT André Twele andre.twele@dlr.de German Aerospace Center (DLR), German Remote Sensing Data
Center (DFD), Muenchener Str. 20, 82234 Oberpfaffenhofen, Germany
© 2016 Informa UK Limited, trading as Taylor & Francis Group
INTERNATIONAL JOURNAL OF REMOTE SENSING 2991
end users.
Recently, a TerraSAR-X-based flood mapping service has been presented by Martinis,
Kersten, and Twele (2015) and Martinis et al. (2015), which consists of a fully automated
processing chain geared towards near-real-time pixel-based flood detection based on
TerraSAR-X data. Since TerraSAR-X is acquiring data non-systematically, the flood map-
ping service needs to be activated on-demand in case of emergency situations by
programming dedicated satellite acquisitions over flood-affected regions. In most
cases, new data acquisitions are tasked when disaster management authorities request
the service of rapid mapping entities and/or mechanisms such as the Copernicus
Emergency Management Service (EMS) or the International Charter ‘Space and Major
Disasters’. Unfortunately, in a number of cases, such satellite-based emergency response
mechanisms are only activated when a flood situation has already become very severe.
Therefore, subsequently tasked satellite data acquisitions may be acquired too late to
capture the peak of the flood, thereby weakening the value of derived crisis information
for emergency management. In the case of flash floods, time constraints are even more
critical than for regular flood situations. For a successful mapping of flash floods,
Mouratidis and Sarti (2013) recommend a high observation frequency of 6–12 hours
with a high spatial resolution of 20 m.
In contrast to e.g. TerraSAR-X, Cosmo SkyMed, and Radarsat-2, the Sentinel-1 mission,
operated by the European Space Agency (ESA) in the frame of the European Union’s
Copernicus Programme, will consist of two systematically acquiring satellite sensors
(Sentinel-1A and Sentinel-1B) with a repeat cycle of 6 days for the final constellation.
Sentinel-1A and B are equipped with a C-Band SAR payload (at 5.405 GHz). Over land
masses, the interferometric wide swath (IW) mode is used by default, acquiring dual-
polarized SAR data with VV/VH polarization. The mission is based on a pre-defined
conflict-free observation scenario making optimum use of the SAR duty cycle with
respect to the technical constraints of the system (Torres et al. 2012). This is a major
advantage for the implementation of fully automated processing chains as the time-
consuming step of tasking satellite data can be omitted. Furthermore, the time delay
between data delivery and dissemination of the extracted crisis information can be
significantly reduced. Therefore, an automatic flood mapping service based on Sentinel-
1 data can significantly enhance the utility of Earth Observation for flood disaster
monitoring. However, in comparison to X-band SAR data (e.g. acquired by TerraSAR-X
or Cosmo-SkyMed), the system parameters of Sentinel-1 are more challenging for
mapping water surfaces: As the contrast between non-water and water areas decreases
2992 A. TWELE ET AL.
with increasing wavelength of the SAR system (Drake and Shuchman 1974), lower
classification accuracies can be expected for C-band Sentinel-1 data. While the on-
demand programming capability of, for example, TerraSAR-X, permits the acquisition
of HH-polarized data, which are generally considered as superior to other polarizations
in the context of flood mapping (Henry et al. 2006; Brisco et al. 2008), the Sentinel-1
processing chain is dependent on systematically acquired VV/VH-polarized IW-mode
data. In addition, most acquisition modes of X-band satellites have a higher spatial
resolution than the Sentinel-1 IW mode with 5 × 20 m. This enables to derive informa-
tion about the flood extent in more detail.
The main objective of this work is therefore to investigate if and how the processing
chain of the X-band-based TerraSAR-X flood service (Martinis, Kersten, and Twele 2015)
Downloaded by [Library Services City University London] at 20:37 02 July 2016
can be adapted to C-band Sentinel-1 data and to present modules, which enable
automatic ingestion, correction, and analysis of Sentinel-1 data. Compared to previous
versions of the methodology, the robustness of the flood processor under challenging
environmental conditions (e.g. prevalence of water-lookalikes such as snow or radar
shadow in mountainous areas) has been improved through the integration of an
exclusion mask derived from the ‘Height above nearest drainage’ (HAND) index
(Rennó et al. 2008). The complete workflow has been iteratively refined based on
comprehensive tests, which involved the processing of over 400 Sentinel-1 IW-mode
scenes. The thematic accuracy of the flood processor was quantitatively assessed for a
flood event in Greece/Turkey in March 2015. Since studies comparing the suitability of
VV and VH polarization for flood mapping are scarce, the accuracy assessment includes a
detailed analysis of the two standard polarizations (VV and VH), which are available
when using routinely acquired IW-mode Sentinel-1 data.
2. Methodology
The workflow of the Sentinel-1-based processing chain, as outlined in Figure 1, is
composed of the following main elements, which are described in Sections 2.1–2.7: (a)
automatic data ingestion through a Python-based script, which routinely queries the
ESA Sentinel Data Hub for new acquisitions matching user-defined criteria and down-
loads them, (b) geometric correction and radiometric calibration using the graph pro-
cessing tool (GPT) of the ESA Sentinel-1 toolbox (S1TBX), (c) initial classification using
automatic thresholding, (d) fuzzy-logic-based classification refinement, (e) final classifi-
cation including auxiliary data, and (f) dissemination of the results.
slope information as well as the mean and standard deviation of the elevation of all
water objects derived from the automatic thresholding approach (see Section 2.4). Both
of these parameters are used as input for the fuzzy-logic-based post-classification (see
Section 2.5).
Besides fuzzy-logic-based classification refinement, the thematic accuracy of the
processor is further enhanced through the integration of the ‘Height above nearest
drainage’ (HAND) index (Rennó et al. 2008), which helps us to reduce water-lookalikes
depending on the hydrologic–topographic setting. The HAND index has been calculated
near-globally based on elevation and drainage direction information provided by the
Hydrosheds mapping product (Lehner, Verdin, and Jarvis 2008). Based on the HAND
index, a binary exclusion mask (termed ‘HAND-EM’ in the following) has been calculated
to separate flood-prone from non-flood prone areas. Both binary classes are determined
using an appropriate threshold value. Choosing the threshold value too high may lead
to misclassifications (i.e. the inclusion of flood-lookalikes in areas much higher than the
actual flood surface and drainage network) while a threshold value set too low would
eliminate valid parts of the flood surface. The choice of an appropriate threshold is thus
critical, but could only be derived through a series of empirical tests including over 400
TerraSAR-X and Sentinel-1 scenes of different hydrological and topographical settings.
Due to the global application scope of the Sentinel-1 flood processing chain, a rather
conservative threshold value of ≥15 m was finally chosen for defining non-flood prone
areas. The HAND-EM has been further shrunk by one pixel using an 8-neighbour
function to account for potential geometric inaccuracies between the exclude layer
and SAR data.
In order to distinguish between permanent water bodies (i.e. related to normal water
levels) and inundated areas in the overall Sentinel-1 detected water extent, a global
2994 A. TWELE ET AL.
reference water mask is required. To achieve global coverage, the reference water mask
is a combination of two data sources: SRTM Water Body Data (SWBD) are used for areas
between 56° southern latitude and 60° northern latitude at a resolution of approximately
30 m at the equator. For all northern and southern latitudes not covered by SWBD data,
the Moderate Resolution Imaging Spectroradiometer (MODIS) 250 m land–water mask
(MOD44W) is employed.
criteria, for example, the time frame or geographical location of suitable Sentinel-1
acquisitions can be specified. Once corresponding scenes are found, they are down-
loaded to the local file system and the thematic processor is executed. After unzipping
the data, the folder structure is searched for files relevant for the further processing,
namely Sentinel-1 data in GeoTIFF-format and Extensible Markup Language (XML)
metadata used for radiometric calibration.
as a measure of the degree of variation within the data and can therefore be used as an
indicator of the probability that the tiles are characterized by spatial inhomogeneity and
contain more than one semantic class. The selected parent objects should also have a
mean individual backscatter value lower than the mean of all parent tiles on L+. This
ensures that tiles lying on the boundary between water and no water areas are selected.
In case that no tiles fulfil these criteria, the tile size on L+ and L− is halved and the
quantile for the tile selection is reduced to 90% to guarantee a successful tile selection
also in data with a relatively low extent of water surfaces or with smaller dispersed water
bodies. In comparison to previous work (Martinis, Kersten, and Twele 2015), the robust-
ness of the automatic threshold derivation has been improved by restricting the tile-
selection only to pixels situated in flood-prone regions defined by the HAND-EM. The
Downloaded by [Library Services City University London] at 20:37 02 July 2016
potential for erroneous thresholds derived from non-water areas (e.g. radar shadow and
snow-covered areas in mountainous terrain) is hence reduced. Tiles are only considered
in case less than 20% of its data pixels are excluded by HAND-EM. Out of the number of
the initially selected tiles, a limited number of N parent tiles are finally chosen for
threshold computation. This selection is accomplished by ranking the parent tiles
according to the standard deviation of the mean backscatter values of the respective
child objects. Tiles with the highest values are chosen for N. Extensive testing yielded
that N = 5 is a sufficient number of parent tiles for threshold computation. The para-
metric Kittler and Illingworth minimum error thresholding approach (Kittler and
Illingworth 1986) is then employed to derive local threshold values using a cost function
which is based on the statistical parameterization of the sub-histograms of all selected
tiles as bi-modal Gaussian mixture distributions. In order to derive a global (i.e. scene-
based) threshold, the locally derived thresholds are combined by computing their
arithmetic mean. In rare cases, only very small water bodies exist in the scene or the
water bodies do not appear as dark backscatter regions, for example, due to a wind-
induced roughening of the water surface. As a result of such circumstances, the
assumption of a bi-modal distribution of the sub-histogram cannot be fully maintained.
This can lead to local thresholds located close to the boundaries of the sub-histograms.
In such cases, a relatively conservative threshold value of −18 dB, which has been
derived empirically, is applied to the scene.
Based on the preliminary classification, the mean elevation of water objects and the
size of all individual flood objects are calculated. Together with the earlier derived land–
water threshold, these parameters calculated from the initial classification are later being
used to define fuzzy thresholds for the standard S and Z membership functions (see
Section 2.5). This means that the initial classification based on the automatic threshold-
ing procedure is mandatory to build elements for the fuzzy-logic-based refinement.
by standard S and Z membership functions (Pal and Rosenfeld 1988). Real numbers
within the interval of [0, 1] are used to define the degree of membership. A membership
degree of 0 denotes minimum and 1 maximum membership to the class water. The
average of the individual membership degrees is computed for each pixel in order to
combine all fuzzy elements into one composite fuzzy set. The membership degree of the
composite fuzzy set is fixed to zero in case one or more individual fuzzy sets have a
membership degree of zero. Subsequently, the flood mask is derived through a thresh-
old defuzzification step, which transforms each image element with a membership
degree >0.6 into a discrete thematic class. A more in-depth description of the fuzzy-
logic approach can be found in Martinis, Kersten, and Twele (2015). For increasing the
spatial homogeneity of the detected flood plain and integrating image elements at the
Downloaded by [Library Services City University London] at 20:37 02 July 2016
2.7. Dissemination
The whole processing chain from the ingestion of Sentinel-1 data to the final classifica-
tion result is implemented with an Open Geospatial Consortium (OGC) compliant Web
Processing Service (WPS) framework (PyWPS). The processing results are stored as
GeoTIFF raster files and are registered within a PostgreSQL/PostGIS database. Every
output layer (‘Flood’, ‘Non-flood’, and ‘Permanent water’) is deployed via a Geoserver
as a single web mapping service (WMS) layer set and visualized within a dedicated web
client.
3. Experimental results
3.1. Study site and data set
The methodology was applied to a test site at the lower course of the Evros (or Meriç in
Turkish, Maritsa in Bulgarian) River, which represents the border between Greece and
INTERNATIONAL JOURNAL OF REMOTE SENSING 2997
Downloaded by [Library Services City University London] at 20:37 02 July 2016
Figure 2. Overview of validation site locations. Background: Sentinel-1 scene acquired 12 March
2015, VV polarization, © Copernicus data/ESA 2015. HAND-EM: Exclusion mask derived from the
‘Height above nearest drainage index’ (HAND).
Turkey. With a total length of 515 km, the Evros River is the second largest river of
Eastern Europe. The drainage basin covers an area of 52,900 km2, which is divided
among Bulgaria (66%), Turkey (27.5%), and Greece (6.5%). Highest discharge can be
usually observed in the period between December and April (Kanellopoulos et al. 2008).
Two subsets of a Sentinel-1 GRD VV/VH scene (ascending, relative orbit: 29) acquired on
12 March 2015 are used for validating the flood processor (see Figure 2). Each subset
comprises an area of approximately 5000 × 3800 m at River Evros, where a long-lasting
flood situation, which started in February 2015, caused widespread flooding of farmland
(Figure 2).
A reference water mask was generated by visual interpretation and manual digitaliza-
tion of a pan-sharpened WorldView-2 scene of 0.5 m spatial resolution, which was acquired
on 11 March 2015 (see Figures 3(a) and 4(a)). Due to a time difference of around 31 hours
between Sentinel-1 and WorldView-2 data sets, stable flood conditions have been
2998 A. TWELE ET AL.
Downloaded by [Library Services City University London] at 20:37 02 July 2016
Figure 3. Subsets for validation site A: (a) WorldView-2 scene acquired 11 March 2015, band
combination 5-3-2 (natural colour), © European Space Imaging/DigitalGlobe 2015. Overlay: reference
water mask. (b) Sentinel-1 scene acquired 12 March 2015 in VV polarization and (c) in VH polariza-
tion. (d) Classification results for VV-polarized and (e) for VH-polarized data. See Figure 2 for the
location of the validation site.
ascertained through visual interpretation of the flood extent of consecutive SAR and
optical satellite acquisitions. No significant change in water levels could be observed.
Figure 4. Subsets for validation site B: (a) WorldView-2 scene acquired 11 March 2015, band
combination 5-3-2 (natural colour), © European Space Imaging/DigitalGlobe 2015. Overlay: reference
water mask. (b) Sentinel-1 scene acquired 12 March 2015 in VV polarization and (c) in VH polariza-
tion. (d) Classification results for VV-polarized and (e) for VH-polarized data. See Figure 2 for the
location of the validation site.
accuracy as well as the kappa coefficient (κ) have been calculated for the classes ‘Open
water’ and ‘Other’.
In Table 1, the classification accuracies for the first validation site (‘A’) are listed. While
no major differences between the classification accuracy of VV and VH polarization can
be observed, VV-polarized data consistently reached higher user’s, producer’s, and
overall accuracies. The kappa coefficient (κ) of 0.910 for VV in contrast to 0.879 for VH
polarization confirms this observation.
With respect to overall accuracies and kappa coefficients (κ), the results obtained for
validation site B (see Table 2) are fairly consistent to validation site A with VV
3000 A. TWELE ET AL.
Table 1. Overall (OA), producer’s (PA), user’s (UA) accuracies and kappa coefficient (κ) for validation
site A.
Polarization OA (%) PA open water (%) UA open water (%) PA other (%) UA other (%) κ
VV 94.7 91.4 99.3 99.4 92.5 0.910
VH 94.0 88.3 99.2 99.3 90.0 0.879
Table 2. Overall (OA), producer’s (PA), user’s (UA) accuracies and kappa coefficient (κ) for validation
site B.
Polarization OA (%) PA open water (%) UA open water (%) PA other (%) UA other (%) κ
VV 96.1 95.2 99.2 98.2 90.0 0.910
VH 95.2 98.5 94.7 87.7 96.2 0.883
Downloaded by [Library Services City University London] at 20:37 02 July 2016
polarization performing slightly better than VH polarization. However, user’s and pro-
ducer’s accuracies obtained for the individual classes (‘Open water’ and ‘Other’) show a
larger variability. While VH polarization offers the highest producer’s accuracy and VV
polarization the highest user’s accuracy for the open water class, the producer’s accuracy
for the remaining class (‘Other’) is over 10% higher for VV polarization compared to VH
polarization. In contrast, the user’s accuracy of VH polarization is over 6% higher than
that of VV polarization.
3.3. Discussion
Irrespective of validation site and polarization, the results obtained can be regarded as
very satisfying with overall accuracies between 94.0% and 96.1% and kappa coefficients
(κ) between 0.879 and 0.910. Sources of misclassifications can be best identified when
comparing the backscatter behaviour of both polarizations (Figures 3(b/c) and 4(b/c))
with the corresponding classification results (Figures 3(d/e) and 4(d/e)).
In the upper right part of Figure 3, thin lines of vegetation within open water areas
can be perceived. However, these lines are much wider in the SAR image compared to
the optical WorldView-2 scene. Due to geometric distortions such as foreshortening and
layover resulting from the side-looking geometry of the sensor, the SAR-derived flood
extent is commonly underestimated in these areas. In addition, when comparing the VV
and VH polarization images, a higher contrast between open water areas and the thin
lines of vegetation can be observed for VV polarization. This is due to double bounce
effects, i.e. strong signal return of SAR waves, scattered at the water surface towards e.g.
partially submerged tree trunks and then directly back to the SAR sensor. Within this
scattering mechanism, a great percentage of the backscattered waves show preserved
polarization (i.e. V-polarized wave transmitted returns as V-polarized signal). Therefore,
VV shows a higher signal response in double bounce conditions than VH.
The channel at the peninsula (right part of Figure 3) shows a higher rate of false alarms
in VH polarization (Figure 3(e)) compared to VV polarization (Figure 3(d)). Also, when
comparing the two corresponding SAR intensity images (Figure 3(b/c)), it is much easier
to visually separate the channel from surrounding vegetation using VV polarization instead
of VH polarization. According to Lee and Pottier (2009), VH polarization is induced by
volume scattering. Therefore, VH polarization shows a higher signal return at the tree
INTERNATIONAL JOURNAL OF REMOTE SENSING 3001
crowns, which are flipped towards the channel due to the side-looking SAR geometry. In
contrast to this, VV polarization is characterized by a higher penetration depth into the tree
crowns, i.e. VV is mainly influenced by double bouncing between tree trunks and the water
surface and not by multiple scattering inside the tree crown as VH polarization.
A further observation is the larger backscatter variability of VH-polarized data in vege-
tated areas. Within the land areas of Figure 3, forest areas are much more highlighted in
VH- compared to VV-polarized data. In contrast, agricultural farmland shows comparably
low backscatter in VH polarization. The higher sensitivity of VH polarization to volume
scattering (an indication for vegetated areas) is the reason for its higher contrast within the
land part compared to VV. The stronger contrast between forested areas and farmland in
VH polarization increases the risk of water-lookalike areas. Due to the application of the
Downloaded by [Library Services City University London] at 20:37 02 July 2016
HAND-EM (see Figure 2), lookalike areas situated in higher altitudes do not constitute a
source of misclassification in VH-polarized data. In lower altitude areas, however, certain
more sparsely vegetated land cover types can exhibit similar backscatter profiles like open
water surfaces. An example of such misclassifications is visible in Figure 4(c/e), where a
relatively high false alarm rate for open water can be observed.
When comparing the backscatter behaviour of VV and VH polarization, it is further
important to note that the Sentinel-1 scene used for the underlying study has been
recorded under calm wind conditions of approx. 1.5 m s–1. Therefore, open water areas
are mainly characterized by specular reflection of the SAR signal, which results in
relatively low backscatter values. The amount of specular reflection decreases when
the water surface is roughened by wind, leading to an increased backscatter level and
lower contrast to land areas. In a study to compare quad-polarized C-band SAR data for
wind speed retrieval, Zhang, Perrie, and He (2011) have shown that in contrast to co-
polarized data, the NRCS of cross-polarized data does not saturate under strong wind
conditions and is further not sensitive to incidence angles or wind directions. This
finding suggests that under high winds speeds, the elevated and non-saturated back-
scatter signature of VH-polarized data might lead to a very low land–water contrast,
potentially resulting in a higher amount of misclassifications. In the same manner, VV-
polarized data might suffer from a high backscatter variability of open water surfaces
when wind directions vary across a given satellite scene. While HH polarization is
generally considered as superior to VV or VH polarization for flood mapping purposes
(Henry et al. 2006, Brisco et al. 2008) since it yields the highest contrast between open
water and land areas, this polarization is usually not available for systematically acquired
Sentinel-1 data of land surfaces.
Currently, the processing chain has been implemented for single-polarized data to
derive information about the flood extent as fast as possible. The computational effort of
the complete workflow as shown in Figure 1 is approx. 45 min for a Sentinel-1 IW-mode
GRD scene using an Intel Xeon E5-4650 CPU (8 cores) with 2.7 GHz and 16 GB of RAM on
a Linux-based 64-bit operating system.
4. Conclusion
In this work, we presented a fully automated processing chain for inundation mapping
and monitoring based on systematically acquired Sentinel-1 data. The processing chain
is comprised of modules for automatic data ingestion, geometric correction, and
3002 A. TWELE ET AL.
the thematic processor has been enhanced through the integration of an exclusion
mask (HAND-EM) which restricts the classification only to flood-prone regions, thereby
masking out potential flood-lookalikes and misclassifications in areas with a defined
positive height difference to the drainage network. The application of the HAND-EM
further allows improving the robustness of the automatic thresholding procedure by
restricting the areas for the tile-selection only to flood-prone regions.
The performance and thematic accuracy of the processor has been assessed for two
validation sites covering a flood situation at the border between Greece and Turkey. The
result of the validation suggests that the thematic processor is able to achieve satisfying
classification results with overall accuracies between 94.0% and 96.1% and kappa
coefficients (κ) ranging from 0.879 to 0.910. The thematic accuracy of VV polarization
was found to be slightly higher than that of VH polarization. Due to a higher sensitivity
to volume scattering, VH polarization shows higher backscatter variability on land
surfaces. In sparsely vegetated areas, the backscatter level of VH polarization is compar-
ably low which leads to water-lookalikes and potential misclassifications.
Since a quantitative accuracy assessment could so far only be performed for a single
acquisition acquired under calm wind conditions, the results, particularly regarding the
performance comparison of SAR polarizations, should be interpreted with caution since
the processor’s robustness in relation to different wind speed levels as well as other
environmental factors still needs to be verified. Since Sentinel-1 data are usually
acquired in VV/VH polarization over most land masses (e.g. Europe), the best possible
performance may not be reached in practice since a number of studies have demon-
strated that superior flood mapping results can be achieved with HH-polarized data. In
the future, a systematic assessment of more Sentinel-1 scenes is planned, including a
detailed investigation of the performance of VV and VH polarization in different envir-
onments and under varying wind conditions. Since Sentinel-1 imagery is being acquired
systematically, time-series analysis might further help us to improve the robustness of
the presented flood mapping workflow.
Acknowledgments
Major parts of this work have been funded by ESA in the framework of the ASAPTERRA project
(Contract No.: 4000112375/14/I-NB). The authors are grateful to two anonymous reviewers for
their helpful comments and suggestions.
INTERNATIONAL JOURNAL OF REMOTE SENSING 3003
Disclosure statement
No potential conflict of interest was reported by the authors.
Funding
Major parts of this work have been funded by ESA in the framework of the ASAPTERRA project
[Contract No.: 4000112375/14/I-NB].
ORCID
André Twele http://orcid.org/0000-0002-8035-2625
Downloaded by [Library Services City University London] at 20:37 02 July 2016
References
Brisco, B., R. Touzi, J. J. van der Sanden, F. Charbonneau, T. J. Pultz, and M. D’Iorio. 2008. “Water
Resource Applications with RADARSAT-2 – a Preview.” International Journal of Digital Earth 1 (1):
130–147. doi:10.1080/17538940701782577.
De Moel, H., J. van Alphen, and J. C. J. H. Aerts. 2009. “Flood Maps in Europe – Methods, Availability
and Use.” Natural Hazards and Earth System Sciences 9: 289–301. doi:10.5194/nhess-9-289-2009.
Drake, B., and R. A. Shuchman. 1974. “Feasibility of Using Multiplexed SLAR Imagery for Water
Resources Management and Mapping Vegetation Communities.” Proceedings of the 9th
International Symposium on Remote Sensing of Environment, Environmental Research
Institute of Michigan, Ann Arbor, MI, April 15–19, 219–249. doi:10.1080/01431161.2015.1060647.
Dung, N. V., B. Merz, A. Bárdossy, T. D. Thang, and H. Apel. 2011. “Multi-Objective Automatic
Calibration of Hydrodynamic Models Utilizing Inundation Maps and Gauge Data.” Hydrology and
Earth System Sciences 15 (4): 1339–1354. doi:10.5194/hess-15-1339-2011.
García-Pintado, J., D. C. Mason, S. L. Dance, H. L. Cloke, J. C. Neal, J. Freer, and P. D. Bates. 2015.
“Satellite-Supported Flood Forecasting in River Networks: A Real Case Study.” Journal of
Hydrology 523: 706–724. doi:10.1016/j.jhydrol.2015.01.084.
Giustarini, L., P. Matgen, R. Hostache, M. Montanari, D. Plaza, V. R. N. Pauwels, G. J. M. De Lannoy,
et al. 2011. “Assimilating SAR-Derived Water Level Data into a Hydraulic Model: A Case Study.”
Hydrology and Earth System Sciences 15 (7): 2349–2365. doi:10.5194/hess-15-2349-2011.
Henry, J.-B., P. Chastanet, K. Fellah, and Y.-L. Desnos. 2006. “Envisat Multi-Polarized ASAR Data for
Flood Mapping.” International Journal of Remote Sensing 27 (10): 1921–1929. doi:10.1080/
01431160500486724.
Kanellopoulos, T. D., V. Kapsimalis, S. E. Poulos, M. O. Angelidis, A. P. Karageorgis, and K.
Pavlopoulos. 2008. “The Influence of the Evros River on the Recent Sedimentation of the
Inner Shelf of the NE Aegean Sea.” Environmental Geology 53 (7): 1455–1464. doi:10.1007/
s00254-007-0754-2.
Kittler, J., and J. Illingworth. 1986. “Minimum Error Thresholding.” Pattern Recognition 19 (1): 41–47.
doi:10.1016/0031-3203(86)90030-0.
Lee, J.-S., and E. Pottier. 2009. Polarimetric Radar Imaging - From Basics to Applications. Boca Raton,
FL: CRC Press. doi:10.1201/9781420054989.
Lehner, B., K. Verdin, and A. Jarvis. 2008. “New Global Hydrography Derived from Spaceborne
Elevation Data.” Eos, Transactions American Geophysical Union 89 (10): 93–94. doi:10.1029/
2008EO100001.
3004 A. TWELE ET AL.
Lu, J., L. Giustarini, B. Xiong, L. Zhao, Y. Jiang, and G. Kuang. 2014. “Automated Flood Detection
with Improved Robustness and Efficiency Using Multi-Temporal SAR Data.” Remote Sensing
Letters 5 (3): 240–248. doi:10.1080/2150704X.2014.898190.
Martinis, S., J. Kersten, and A. Twele. 2015. “A Fully Automated TerraSAR-X Based Flood Service.”
ISPRS Journal of Photogrammetry and Remote Sensing 104: 203–212. doi:10.1016/j.
isprsjprs.2014.07.014.
Martinis, S., C. Kuenzer, A. Wendleder, J. Huth, A. Twele, A. Roth, and S. Dech. 2015. “Comparing
Four Operational SAR-Based Water and Flood Detection Approaches.” International Journal of
Remote Sensing 36 (13): 3519–3543. doi:10.1080/01431161.2015.1060647.
Martinis, S., A. Twele, and S. Voigt. 2009. “Towards Operational near Real-Time Flood Detection
Using a Split-Based Automatic Thresholding Procedure on High Resolution TerraSAR-X data.”
Natural Hazards and Earth System Sciences 9: 303–314. doi:10.5194/nhess-9-303-2009.
Mason, D. C., J. García-Pintado, H. L. Cloke, and S. L. Dance. 2015. “The Potential of Flood
Downloaded by [Library Services City University London] at 20:37 02 July 2016
Forecasting Using a Variable-Resolution Global Digital Terrain Model and Flood Extents from
Synthetic Aperture Radar Images.” Frontiers in Earth Science 3. doi:10.3389/feart.2015.00043.
Matgen, P., R. Hostache, G. Schumann, L. Pfister, L. Hoffman, and H. H. G. Savenije. 2011. “Towards
an Automated SAR-Based Flood Monitoring System: Lessons Learned from Two Case Studies.”
Physics and Chemistry of the Earth Parts A/B/C 36 (7–8): 241–252. doi:10.1016/j.pce.2010.12.009.
Matgen, P., M. Montanari, R. Hostache, L. Pfister, L. Hoffmann, D. Plaza, V. R. N. Pauwels, G. J. M. De
Lannoy, R. De Keyser, and H. H. G. Savenije. 2010. “Towards the Sequential Assimilation of SAR-
Derived Water Stages into Hydraulic Models Using the Particle Filter: Proof of Concept.”
Hydrology and Earth System Sciences 14 (9): 1773–1785. doi:10.5194/hess-14-1773-2010.
Mouratidis, A., and F. Sarti. 2013. “Flash-Flood Monitoring and Damage Assessment with SAR Data:
Issues and Future Challenges for Earth Observation from Space Sustained by Case Studies from
the Balkans and Eastern Europe.” In Earth Observation of Global Changes (EOGC), edited by J. M.
Krisp, L. Meng, R. Pail, and U. Stilla, 125–136. Berlin: Springer. doi:10.1007/978-3-642-32714-8_8.
Pal, S. K., and A. Rosenfeld. 1988. “Image Enhancement and Thresholding by Optimization of Fuzzy
Compactness.” Pattern Recognition Letters 7: 77–86. doi:10.1016/0167-8655(88)90122-5.
Pulvirenti, L., N. Pierdicca, M. Chini, and L. Guerriero. 2011. “An Algorithm for Operational Flood
Mapping from Synthetic Aperture Radar (SAR) Data Using Fuzzy Logic.” Natural Hazards and
Earth System Sciences 11: 529–540. doi:10.5194/nhess-11-529-2011.
Rennó, C. D., A. D. Nobre, L. A. Cuartas, J. V. Soares, M. G. Hodnett, J. Tomasella, and M. J. Waterloo.
2008. “HAND, a New Terrain Descriptor Using SRTM-DEM: Mapping Terra-Firme Rainforest
Environments in Amazonia.” Remote Sensing of Environment 112 (9): 3469–3481. doi:10.1016/j.
rse.2008.03.018.
Schreier, G. 1993. SAR Geocoding: Data and Systems. Karlsruhe: Wichmann.
Schumann, G., G. di Baldassarre, and P. D. Bates. 2009. “The Utility of Spaceborne Radar to Render
Flood Inundation Maps Based on Multialgorithm Ensembles.” IEEE Transactions on Geoscience
and Remote Sensing 47 (8): 2801–2807. doi:10.1109/TGRS.2009.2017937.
Torres, R., P. Snoeij, D. Geudtner, D. Bibby, M. Davidson, E. Attema, P. Potin, et al. 2012. “GMES
Sentinel-1 Mission.” Remote Sensing of Environment 120: 9–24. doi:10.1016/j.rse.2011.05.028.
Voigt, S., T. Kemper, T. Riedlinger, R. Kiefl, K. Scholte, and H. Mehl. 2007. “Satellite Image Analysis
for Disaster and Crisis-Management Support.” IEEE Transactions on Geoscience and Remote
Sensing 45 (6): 1520–1528. doi:10.1109/TGRS.2007.895830.
Westerhoff, R. S., M. P. H. Kleuskens, H. C. Winsemius, H. J. Huizinga, G. R. Brakenridge, and C.
Bishop. 2013. “Automated Global Water Mapping Based on Wide-Swath Orbital Synthetic-
Aperture Radar.” Hydrology and Earth System Sciences 17 (2): 651–663. doi:10.5194/hess-17-
651-2013.
Zhang, B., W. Perrie, and Y. He. 2011. “Wind Speed Retrieval from RADARSAT-2 Quad-Polarization
Images Using a New Polarization Ratio Model.” Journal of Geophysical Research 116. doi:10.1029/
2010JC006522.