Professional Documents
Culture Documents
Delineation
with Landsat TM Data
Paul Shane Frazier and Kenneth John Page
Abstract
The aim of this project was to determine the accuracy of using
simple digital image processing techniques to map riverine
water bodies with Landsat 5TM data. This paper quantifies
the classification accuracy of single band density slicing of
Landsat 5 TM data to delineate water bodies on riverine
floodplains. The results of these analyses are then compared
to a 6-band maximum likelihood classification over the same
area. The water boundaries delineated by each of these digital
classification procedures were compared to water boundaries
delineated from colour aerial photography acquired on the
same day as the TM data. These comparisons show that Landsat
TMdata can be used to map water bodies accurately. Density
slicing of the single mid-infrared band 5 proved as successful
as multispectral classification achieving an overall accuracy
of 96.9%, a producer's accuracy for water bodies of 81.7% and
a user's accuracy for water bodies of 64.5%.
Introduction
Accurate information on the extent of water bodies is important
for flood prediction, monitoring, and relief (Smith, 1997;
Tholey et al., 1997; Baumann, 1999); production of wetland
inventories (Bennett, 1987; Johnston and Barson, 1993; Blackman
et al., 1995; Shaikh et al., 1998; Phim et al., 1999); and the
evaluation of water resources (Morse et al., 1990; Manavalan et
al., 1993). Often this information is difficult to produce using
traditional survey techniques because water bodies can be fast
moving as in floods, tides, and storm surges or may be inaccessible.
Remotely sensed data provide a means of delineating
water boundaries over a large area at a given point in time. To
capture fast moving hydrological features, the data need to be
either of a high temporal resolution or in a substantial archive
to cover a range of hydrological conditions. Landsat MSS and
TM provide high spatial resolution data at 16-day intervals over
a long archival history, exceeding 25 years in most locations.
The long archive period and repetitive capture make the data
useful for mapping water bodies at a regional scale over a range
of hydrological conditions.
Since Landsat data became available in 1972, they have
been used to map water extent. Smith (1997) cites several early
studies where Landsat MSS data, in particular Band 7, were
used to distinguish water bodies from surrounding dry soil or
vegetation. Comparison with aerial photography gave error
estimates of less than 5 percent.
Bennett (1987) used density slicing of Landsat MSS band 7
to map water bodies to the west of Griffith, New South Wales,
Australia (NsW). He compared the area of MSS-derived water
around 40 percent.
(C)
Methods
In order to assess the accuracy of using Landsat TM data to accurately
locate and delineate water bodies, classified Landsat TM
data were compared with manually mapped aerial photography
of the same area. Luckily for this study, both the Landsat TM
data and the aerial photography were captured on the same date
(26 October 1990), thus ensuring identical hydrological
conditions.
Materials
The color aerial photography was acquired by the NSW Central
TABLE1.CLASSESA ND DESCRIPTIONSO F DIGITIZEDW ATERB ODIES.
Marginal trees
< 10% crown cover
Nil
1462 December 2000 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
20WX)
1-
~ ~ z ~ q ~ ~ g ~ ~ ~ ~ s~~s$>~g2>~- ~- -$
s~:a:~~
t-4
AREAS.
Band 1
Band 2 Band 7
Band 3
Key to densty sl~c~ng
Water Not water
Ground truth Image
Band 4
I
I
Key to ground truth Image
R~ver Dam Not Water
Lagoon Small
9 00 5 ; K i
Q5"m
.I 0 h
0 255 0 255
Band 4
g5;m
95q-J
.I P
rn
0 255 0255
Band 6
5 0 255
9yrl
8 0 255
-0
8K
Range ofwater training pixels
26.496.74.8
38.096.54.9
493.677.142.6
I!
596.981.764.5
794.474.445.4
Classification Data
Percentage
Band 5Error Matrix Water Not Water Total Correct
Reference Small Pool 21222844048.2%
Dam 1642919385.0%
River 2626419304586.2%
Lagoon 77717395081.8%
Not Water 2079886639074297.7%
Total 58588951295370Percentage Correct 64.5%99.1%- 96.9%
TABLE6 . SUPERVMISAEXIDM UML IKELIHOOCD LASSIFICATIEORNR ORM ATRIX.
Classification Data
Maximum-Likelihood Percentage
Classification Error Matrix Water Not Water Total Correct
Reference Small Pool
Dam
River
Lagoon
Not Water
Total
Percentage Correct
lrr~gatedP asture
II
mm
r
Small unclassi
polygons
Conclusion
From this study it is clear that Landsat TMdata can be used to
map water bodies associated with the Murrumbidgee River
and its floodplain in the Wagga Wagga region. A multispectral
maximum-likelihood classification was able to produce an
overall classification accuracy of 97.4percent. This classification
was able to locate all of the major water bodies but underestimated
the number of water pixels present on the image,
achieving a producer's accuracy of only 59.6percent.
Importantly, a simple density slice classification of the
mid-infrared band 5produced a classification accuracy (96.9
percent) similar to that of the maximum-likelihood classification.
The density slice gave a better estimate of water pixels but
tended to include more pixels that were not water bodies (user's
accuracy 64.5 percent).
The other infrared bands, band 4and band 7, were also useful
for locating water bodies but tended to include more errors
of commission. Lush crop areas were included in the band 7
density slice classification, and poor pasture, hill shadow, and
parts of the urban area were included in the band 4density
slice classification.
All of the visible bands proved to be inadequate for successful
density slice classification. The brightness of the turbid
water showed a range of pixel values similar to that of the
majority of the surrounding land cover, resulting in gross overestimation
of the water area.
The infrared wavelengths are the most useful that optical
scanners can offer for water mapping. The strong absorption of
infrared light by water bodies gives them a distinctively low
spectral response in this range. The mid-infrared bands of Landsat
TM proved marginally more successful than the near-infrared
band. The narrow nature of the water bodies in question
creates many fringing mixed pixels that often contain significant
responses from vegetation. Also, some of the water bodies
are very turbid, which increases the upper range of values in
near-infrared more than it does in the mid-inhared bands. The
combination of mixed pixels and high turbidity limits the use
of a single near-infrared band to classify water bodies. This
limitation has significant ramifications for the use of optical
remotely sensed data such as those acquired by Landsat MSS
Or SPOT.
Acknowledgments
The authors thank Rodney Rumbachs for preparation of the
aerial photography and the production of the location map and
John Louis, David Roshier, David Lamb, and the anonymous
reviewers from PE&RS for their comments on the manuscript.
References
Baumann, P., 1999. Flood Analysis, http://www.research.umbc.edu/
tbenja~/baumann/mod2.html1,0 February 1999.
Bennett, M.W.A., 1987. Rapid monitoring of wetland water status using
density slicing, Proceedings of the 4th Australasian Remote Sensing
Conference, [14-18 September], Adelaide, pp. 682-691.
Blackman, J.G., S.J. Gardiner, and M.G. Morgan, 1995. Framework for
biogeographic inventory, assessment, planning and management
prs. org
Abstract
The aim of this project was to determine the accuracy of using
simple digital image processing techniques to map riverine
water bodies with Landsat 5TM data. This paper quantifies
the classification accuracy of single band density slicing of
Landsat 5 TM data to delineate water bodies on riverine
floodplains. The results of these analyses are then compared
to a 6-band maximum likelihood classification over the same
area. The water boundaries delineated by each of these digital
classification procedures were compared to water boundaries
delineated from colour aerial photography acquired on the
same day as the TM data. These comparisons show that Landsat
Introduction
Accurate information on the extent of water bodies is important
for flood prediction, monitoring, and relief (Smith, 1997;
Tholey et al., 1997; Baumann, 1999); production of wetland
inventories (Bennett, 1987; Johnston and Barson, 1993; Blackman
et al., 1995; Shaikh et al., 1998; Phim et al., 1999); and the
evaluation of water resources (Morse et al., 1990; Manavalan et
al., 1993). Often this information is difficult to produce using
traditional survey techniques because water bodies can be fast
moving as in floods, tides, and storm surges or may be inaccessible.
Remotely sensed data provide a means of delineating
water boundaries over a large area at a given point in time. To
capture fast moving hydrological features, the data need to be
either of a high temporal resolution or in a substantial archive
to cover a range of hydrological conditions. Landsat MSS and
TM provide high spatial resolution data at 16-day intervals over
a long archival history, exceeding 25 years in most locations.
The long archive period and repetitive capture make the data
useful for mapping water bodies at a regional scale over a range
of hydrological conditions.
Since Landsat data became available in 1972, they have
been used to map water extent. Smith (1997) cites several early
studies where Landsat MSS data, in particular Band 7, were
used to distinguish water bodies from surrounding dry soil or
vegetation. Comparison with aerial photography gave error
estimates of less than 5 percent.
Bennett (1987) used density slicing of Landsat MSS band 7
to map water bodies to the west of Griffith, New South Wales,
Australia (NsW). He compared the area of MSS-derived water
bodies with that derived from digitized aerial photography and
found that the MSS data underestimated the area of water by
around 40 percent.
(C)
Methods
In order to assess the accuracy of using Landsat TM data to accurately
locate and delineate water bodies, classified Landsat TM
data were compared with manually mapped aerial photography
of the same area. Luckily for this study, both the Landsat TM
data and the aerial photography were captured on the same date
(26 October 1990), thus ensuring identical hydrological
conditions.
Materials
The color aerial photography was acquired by the NSW Central
TABLE1.CLASSESA ND DESCRIPTIONSO F DIGITIZEDW ATERB ODIES.
The aerial photographs were scanned at 300 DPI using a Micro- borhood intensity interpolation, and a pixel
resampling size of
3 m. Examination of the fit between the photographs and the TM
data indicated that there were significant registration errors.
Each photograph was then re-registered using an additional
seven GCPs on the margins of the water bodies, achieving an
average RMSerror of 23 pixels (9 m). This time overlaying the
photograph onto the TMimage indicated a good fit.
All water bodies greater than 625 m2 (lmm by lmm) in size
were then highlighted directly on the hardcopy prints of photographs
143,145, and 147 using magnified stereoscopic examination,
On the photographs, water bodies appear as uniform
regions of dark green to bright khakibrown in the low lying
channels and depressions. Each water body was then digitized
directly onto the scanned, registered aerial photograph using
the ERMapper regions tool. The polygons were then labeled as
River, Lagoon, Small Pool, or Dam (Table 1, Figure 2). The polygons
were then converted into a five-class raster image for comparison
with the Landsat TM classification results.
Landsat Image CIassHkation
1-
~ ~ z ~ q ~ ~ g ~ ~ ~ ~ s~~s$>~g2>~- ~- -$
s~:a:~~
t-4
Band 1
Band 2 Band 7
Band 3
Key to densty sl~c~ng
Water Not water
Ground truth Image
Band 4
I
I
Key to ground truth Image
R~ver Dam Not Water
Lagoon Small
9 00 5 ; K i
Q5"m
.I 0 h
0 255 0 255
Band 4
g5;m
95q-J
.I P
rn
0 255 0255
Band 6
5 0 255
9yrl
8 0 255
-0
8K
Range ofwater training pixels
26.496.74.8
38.096.54.9
493.677.142.6
596.981.764.5
794.474.445.4
I!
Classification Data
Percentage
Band 5Error Matrix Water Not Water Total Correct
Reference Small Pool 21222844048.2%
Dam 1642919385.0%
River 2626419304586.2%
Lagoon 77717395081.8%
Not Water 2079886639074297.7%
Total 58588951295370Percentage Correct 64.5%99.1%- 96.9%
TABLE6 . SUPERVMISAEXIDM UML IKELIHOOCD LASSIFICATIEORNR ORM ATRIX.
Classification Data
Maximum-Likelihood Percentage
Classification Error Matrix Water Not Water Total Correct
Reference Small Pool
Dam
River
Lagoon
Not Water
Total
Percentage Correct
lrr~gatedP asture
II
mm
r
Small unclassi
polygons
Conclusion
From this study it is clear that Landsat TMdata can be used to
map water bodies associated with the Murrumbidgee River
and its floodplain in the Wagga Wagga region. A multispectral
maximum-likelihood classification was able to produce an
overall classification accuracy of 97.4percent. This classification
was able to locate all of the major water bodies but underestimated
the number of water pixels present on the image,
achieving a producer's accuracy of only 59.6percent.
Importantly, a simple density slice classification of the
mid-infrared band 5produced a classification accuracy (96.9
percent) similar to that of the maximum-likelihood classification.
Acknowledgments
The authors thank Rodney Rumbachs for preparation of the
aerial photography and the production of the location map and
John Louis, David Roshier, David Lamb, and the anonymous
reviewers from PE&RS for their comments on the manuscript.
References
Baumann, P., 1999. Flood Analysis, http://www.research.umbc.edu/
tbenja~/baumann/mod2.html1,0 February 1999.
Bennett, M.W.A., 1987. Rapid monitoring of wetland water status using
density slicing, Proceedings of the 4th Australasian Remote Sensing
Conference, [14-18 September], Adelaide, pp. 682-691.
Blackman, J.G., S.J. Gardiner, and M.G. Morgan, 1995. Framework for
biogeographic inventory, assessment, planning and management
of wetland systems: the Queensland approach, Workshop Proceedings,
Wetland Research in the WetlDry Tbopics of Australia, (C.M.
Finlayson, editor), Jabiru, NT,Australia, Supervising Scientist
Report No. 101, Department of Environment and Heritage, Canberra,
pp. 114-122.
Brady, A., M. Shaikh, A. King, and P. Sharma, 1999. Remote sensing
and the Great Cumbung Swamp, Wetlands Australia, 7:596-606.
Jensen, J.R., 1996. Introductory Digital Image Processing: A Remote
Sensing Perspective. Prentice Hall, Upper Saddle River, New Jersey,
316 p.
Johnston, R., and M. Barson, 1993. Remote sensing of Australian wetlands:
An evaluation of Landsat TM data for inventory and classification,
Australian Journal of Marine and Freshwater Research,
44235-252.
Kingsford, R.T., R.F. Thomas, P.S. Wong, and E. Knowles, 1997. GIs
prs. org
SPECTRAL
REFLECTANCE OF
VEGETATION, SOIL
AND WATER
LWR314
Fifteenth International Water Technology Conference, IWTC-15 2011, Alexandria, Egypt
delclaux@msem.univ-montp2.fr
5 IRD/Legos 14 Av Edouard Belin, 31400, Toulouse, France, Stephane.calmant@ird.fr
ABSTRACT
In ungauged basin, space-based information is essential for the monitoring of
hydrological water cycle, in particular in regions undergoing large flood events where
satellite data may be used as input to hydrodynamic models. A method for near 3D flood
monitoring has been developed which uses synergies between radar altimetry and high
temporal resolution multi-spectral satellite. Surface Reflectance from the MODIS Terra
instrument are used to map areas of open water as well as aquatic vegetation on a weekly
basis, while water level variations in the inundated areas are provided by the radar
altimetry from the Topex / Poseidon (T/P) and Envisat satellites. We present this
synergistic approach to three different regions: Niger Inner delta and Lake Tchad in
Africa, and Ganga river delta in Asia. Based mainly on visible and Near Infra Red (NIR)
imagery is suitable to the observation of inundation extent. This method is well adapted
for arid and semi arid regions, but less for equatorial or boreal ones due to cloud
coverage.
This work emphasizes the limitations of current remote sensing techniques for full 3Ddescription
of water storage variability in ungauged basins, and provides a good
introduction to the need and the potential use of the future SWOT (Surface Water and
Ocean Topography) satellite mission.
Keywords: radar altimetry, MODIS, Flood mapping, SWOT, Arid climate
1. INTRODUCTION
The main purpose of this study is to provide space-based tool for the monitoring of
inundated areas in large wetlands and floodplains located in arid and semi arid regions.
Space and airborne technologies are increasingly found to be a key and unique source of
spatial information for wetlands conservation and management as many of the Worlds
Fifteenth International Water Technology Conference, IWTC-15 2011, Alexandria, Egypt
wetlands have insufficient on-ground data partly due to their size, number and limited
accessibility. Finding complementary and new methods to monitor inundation patterns
for large wetlands and floodplains is consequently important with expectation to
assimilate such space-based hydrological information into models (e.g.climate, land
surface, water-management and eco-hydrological models)
Comprehensive time series of inundation are required from space observations to: i)
investigate links between floods in remote areas and climate variability; ii) model the
hydrodynamic of a floodplain at high spatio-temporal resolution; iii) understand the
interaction among inter-annual and seasonal flood cycle and land use patterns in and
around area of flooding; iv) to examine the vulnerability of an ecosystem to inundation;
v) to develop alert systems for inundation in a given wetland.
In recent years, remote sensing techniques have clearly shown their capability to monitor
components of the water balance of large river basins on time scales ranging from
months to decades. Satellite altimetry, which has been developed and optimized for open
oceans, was also widely used in different fields of continental hydrology based on
coupled satellite altimetry / in-situ gauges measurements (Crtaux and Birkett [1];
Cretaux et al. [2], and Calmant et al. [3]). The global altimetry data set has now an 18
years-long lifetime and is intended to be continuously updated in the coming decade.
Moreover several authors have addressed the issue of water extent mapping over
floodplain from Remote Sensing data. Toyra et al., [4], [5], have used a combination of
Radarsat and Spot scenes to study extent of water in wetlands and produced multi-year
maps time series of flooded area in the Peace-Athabasca delta in Canada with high
spatial resolution. Frappart et al., [6], [7], have combined radar altimetry with SAR
images onboard the Japanese Earth Resources Satellites (JERS-1) or visible images of
the Vegetation instrument onboard the Spot Satellites in order to study floods over the
Rio Negro (Amazon) and Mekong Basin. The ASAR instrument is also very effective
sensor to detect flooded areas in the particular cases of cloud cover regions with high
spatial resolution (Henry et al. [8]). Bartsch et al. [9] have also used ENVISAT ASAR
data for mapping of fresh water ecosystems in Siberia, and their analysis have shown that
numerous areas previously mapped as tundra are in fact covered by water. Peng et al,
[10] have used the MODIS data to develop a method of water extent and level
monitoring, which however depends on the knowledge of topographic map of the surface
study or on a relation between surface and level of water. Over large flood regions,
medium resolution multi-spectral imagery like MODIS is well suitable as demonstrated
by Sakamoto et al. [11]. They have produced from this instrument weekly time series of
inundation maps of the Mekong basin over a multi-year period. Synthetic Aperture Radar
(SAR) Interferometry (Alsdorf and Lettenmaier, [12], Alsdorf et al. [13]) and passive
and active microwave observations (Prigent et al. [14]) also offer important information
on land surface waters, such as changing area extent of large wetlands. A common
problem in remote sensing of wetlands inundation is the detection of water under aquatic
vegetation. Leblanc et al. [15] used Meteosat thermal images to derive monthly flood
maps of Lake Chad that captured water under aquatic vegetation and adequately
reconstructed the drying of Lake Chad since the 1980s.
Current challenges remain especially in providing global mapping of inundation and
estimates of water storage changes and this has driven the decision of CNES and NASA
to propose a new concept satellite mission (Surface Water and Ocean Topography:
SWOT).
Fifteenth International Water Technology Conference, IWTC-15 2011, Alexandria, Egypt
technique, which does not allow global view of the Earth surface, hence limits
worldwide surveying as well spatial resolution in the cross-track satellite direction.
Despite those limitations the altimetry is a technique, which has a proven potential for
hydrology science: the data are freely available on all the whole earth, and for a lot of
remote areas it is the only source of information, as for the flood plain where in-situ data
(when they exist) are limited to some points near main stream of the rivers. It however
remains a significant limitation when using radar altimetry for flood monitoring. Due to
revisit interval of several days (10 for T/P or Jason1 & 2, 35 for Envisat) the chance to
collect altimetry data at the time of a flood peak is reduced, with very few measurements
for quick inundation.
The methodology developed is illustrated by the Fig. 1. It is separated into 3 steps: a
preprocessing
phase of MODIS (georeferencing and mosaicking) and altimetry data, a
processing of the MODIS data for pixel classification (to detect open water and aquatic
vegetation in particular), and a processing of the radar altimetry for the measurements
that have been classified as open water by MODIS. The MODIS data along each track of
altimetry satellite above the area of study are interpolated in space and time, then, this
information gives a criterion of selection for altimetry data. Below we present the main
results obtained from this method over few case studies around the world
Fifteenth International Water Technology Conference, IWTC-15 2011, Alexandria, Egypt
Fig. 1: Algorithm developed for the flood mapping from MODIS and water height
determination from radar altimetry. Band Unit of reflectance is internal HDF-EOS data
format specific to the Modis data and do not correspond to usual reflectance unit.
Water + dry
Water pixel (Selected)
Non Water pixel (Rejected)
Yes
MODIS PREPROCESSING
Extraction of HDF
images files from
Georeferencing and
mosaicking
Extraction of Band 1,2
and 5 over the ROI
Extraction of Envisat,
& T/P over the ROI
Pre-selection of valid data
from Modis cassification
Open Water
Band 5 < 1200
No
MODIS Data
No
No
Yes
Yes No Yes
Altimetry tracks
Pre-selection of valid data
from editing parameters
Water level calculated from altimetry data selected
ALTIMETRY
Products
Fifteenth International Water Technology Conference, IWTC-15 2011, Alexandria, Egypt
3. CASE STUDIES
3.1 Niger Inner Delta
The Inner Niger Delta (IND) is located in Central Mali in the semi-arid Sahelian zone, in
the south of the Sahara Desert. It is a shallow area of about 70,000 km2 constituted by
channels, swamps and lakes, and its water extent varies seasonally from around 1000 to
12000 km2 with large inter-annual changes (a factor of 1 to 5 has been observed
according to Mahe et al. [17]). The IND is at the junction of 2 rivers, the Niger and the
Bani which supply approximately 1490 m3/s of water to the delta (Mahe et al. [17]).The
factor, generating this variability is the precipitation rate change, in time and with
latitude. From October to May, the IND is hot and dry, and water balance is mainly
driven by evaporation. From June to September, direct precipitation on the IND and in
the upstream part of Niger and Bani Basins generate the seasonal flood. Precipitation
over the IND are weak (~300mm/yr over the northern part of the IND) while they can
between 1200 and 1500 mm/yr over the upstream parts of the Niger and Bani Basins,
mainly occurring during the four months of the wet seasons (June to September).
Interannual variability of rainfall is also marked as shown in Fig. 2. Therefore, the extent
of water in the IND is changing year to year with very complex patterns depending on
the topography of the river channels in the delta, or the presence of vegetation, and on
the total amount of water filling the IND. During flood periods, the IND land is also
subject to noticeable vegetation growth, mixed with water and dry soil. In that context,
reliable frequent information on the water extent, spatial distribution and temporal
variation of IND floodplain are vital for: i) water resources management, ii) a better
understanding of the influence of climate change and the feedback of this floodplain on
regional climate, iii) the monitoring of this important ecosystem and socio-economical
resource. Moreover, our ability to measure, monitor, and forecast supplies of fresh water
in the IND using in-situ methods is almost impossible because of limited access and the
physics of water flow across vast lowlands. This reinforce the needs for methodologies
based on space techniques, e.g., visible and radar satellite imagery to measure water
extent over the wetlands and floodplains, and radar altimetry to measure the water level
variations with time. The flood over the IND is a combination of high water in both Bani
and Niger rivers but with changes in their role from one year to another one. This is a
factor of complexity of understanding the flood over the IND as this implies to model
both river basins hydrodynamic. Modis mapping of inundation every week over interannual
period can therefore provide an invaluable source of information for both
understanding the flood regime and to serve as external validation data for everyone who
aim to develop model of the flood over the IND and more generally to model the Niger
basin hydrodynamic.
We applied the synergistic mapping with radar altimetry and MODIS to the IND for the
2000-2008 period. From January to May, the IND and surrounding regions are drying
out, and active vegetation is fully disappearing. In June, the permanent lake in the IND
starts to grow and in July aquatic vegetation initiates its seasonal cycle. Until August the
aquatic vegetation is growing regularly and it covers the entire region. At end of August
flood waters start to reach the IND and continue to rise during two months until a peak
Fifteenth International Water Technology Conference, IWTC-15 2011, Alexandria, Egypt
between mid and late of October. From November to January the water over the IND
evaporates or flows downward and the only region which still presents some vegetation
covers are the areas that just dried out, exclusively in the IND.
Precise correlations between flood sequence and precipitation patterns in the upstream
part of the Niger and Bani river watershed and in the IND have been established over the
period 2000-2010. Fig. 2a shows that open water maximum is shifted by one month and
half from the maximum of rainfall over the Bani River (the same has been observed with
Niger upstream watershed rainfall). Fig. 2b shows that a first apparition of open water is
observed in August over the IND due to direct precipitation, which is however much
weaker than the main inundation due to flow from Niger and Bani. Fig. 2c shows that
vegetation starts growing immediately after the first rainfall occurring in the IND, in
July, and also that vegetation covers over the IND presents a second smaller peak in end
of November after the main inundation.
Fifteenth International Water Technology Conference, IWTC-15 2011, Alexandria, Egypt
Fig. 2: Modis classification over the IND versus precipitation rate over the Bani River and
the IND. (TRMM data set has been used for the computation of rainfall).
Modis classification / rainfall (TRMM)
0
0,2
0,4
0,6
0,8
1
1,2
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011
normalized data
open water
Rainfall on Bani
Modis Classification / rainfall (TRMM)
0
0,2
0,4
0,6
0,8
1
1,2
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011
normalized data
open water
rainfall on IND
Modis Classification / Rainfall (TRMM)
0
0,2
0,4
0,6
0,8
1
1,2
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011
normalized data
Vegetation
rainfall on IND
The 8-days mapping of flood and vegetation dynamic over the IND has also been done
for the 10 years period, and one sees on figure 3 the high inter-annual variability of water
extent from dry to wet years.
Fig. 3: maps of IND during the maximum of inundation of four different years,
from dry years (2002, 2004) to a wet year (2001, 2005). Dark blue is representing
free water, light blue: mixed water and dry land, light green: aquatic vegetation,
dark green: vegetation on dry land, and yellow: the other type of surface.
To complete the information given by MODIS data, satellite radar altimetry
measurements from T/P (one track) and Envisat (5 tracks) mission have been used. They
have allowed calculating the water level variations over the IND, at the location where
the satellite altimetry tracks are passing over open waters pixel as determined by the
MODIS weekly. Fig. 4 shows some results obtained with Envisat satellite along a track
which crosses the IND from South to North. The low temporal resolution of Envisat
satellite is a clear limitation if one needs to monitor the exact water height variations
from the beginning to the end of each inundation, but relative water height from upper
part to lower part of the IND is measured, and it also can serve as control data of
hydrodynamic model of the flood over the IND.
Fifteenth International Water Technology Conference, IWTC-15 2011, Alexandria, Egypt
Fig. 4: level variations (m) over the IND from Envisat track n 474
In summary, the monitoring of the flood inter-annual variability also offers an interesting
feedback to the study of convective storms in the surrounding regions. It has been
demonstrated in Taylor [18] that inundation over the IND influences convection which
provokes rainfall over large regions in West Africa. They conclude that planned new
dams in Guinea (upstream to the delta) could have enormous influence, both in the
inundation in the IND and in rainfall patterns across hundreds of km2.
Niger, Nigeria, Chad and Cameroon, and the borders of Algeria, Libya, Sudan and the
Central African Republic, making it the largest endoreic basin in the world. Enclosed
between mountain ranges (Tibesti, Darfour) and high plateaus (Adamoua), this entire
drainage network converges on a central depression, in which Lake Chad is situated. The
climate of this region is related to the movement of the Inter Tropical Convergence Zone
(ITCZ) (Olivry et al. [19]). In the south, the monsoon flow brings heavy precipitation
whereas in the north, the Harmattan produces a dry climate, at the latitude of Lake Chad,
the climate is of Sudano Sahelian type because the ITCZ only reaches this zone between
June and August, the rainy season during which total precipitation barely exceeds 300
mm.
The water balance of the lake is represented in Figure 5c. During the period preceding
the 1970s drought, the Chari and Logone rivers supplied Lake Chad with, in average,
nearly 85% of the total inflows. The other significant tributary into the lake is the
intermittent Nigerian river the Komadougou Yob which contribution to the lake is about
1 km3/yr. Precipitation over the lake contributes 6 km3/yr. Concerning losses,
Fig. 5: (from up to low) Level variation of Lake Tchad from altimetry multi-mission (T/P
(up to 2002), Envisat (up to 2008) and Jason-2(up to 2010)) (a), surface variations from
Modis images (b) and water balance of the lake (c).
Multi-satellite altimetry water level variation (m)
277
277,5
278
278,5
279
279,5
280
280,5
281
1993 1995 1997 1999 2001 2003 2005 2007 2009
0
500
1000
1500
2000
2500
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
of the region if India. In pre-monsoon period, April to June is the driest season and this
time river sink to its minimum.
Like the other rivers in India, the inundation along the river Ganga is also mostly
depending upon the rain in catchments of the Ganga basin and its tributaries. The water
drained to river from catchments is at his maximum in the middle of the rainy season.
Although the region is rather cloudy in rainy season, the Modis images provide an
interesting view of phase of inundation over the Ganga delta with however a proportion
of rejected images higher than what was observed in the other regions described in this
study. One hence could estimate average water extent on the delta at inter-annual time
scale.
Modis classification over the Ganga river
0
5000
10000
15000
20000
25000
30000
35000
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011
Fig. 6: Surface of open water detected from Modis classification for the down Ganga
River.
For example the variation of water logging, Fig. 6, in lower Ganga basin shows a good
annual repetition in rainy season (maximum in August-September). There is a high peak
in 2007 (also 2008) which is because of the huge amount of water discharge in Kosi
River, due to high rain, from Nepal. Large flood area south to the Ganga River is also
Fifteenth International Water Technology Conference, IWTC-15 2011, Alexandria, Egypt
observed from MODIS in Bihar in 2000 during which, the width of the Ganga River
increases significantly, causing lateral overflow from the bank of the river. Analysis of
images from 2000 to 2010 has also showed high inter-annual amplitudes of the
inundation over the Ganga River, therefore over the delta in Bangladesh. Between the 2
dry years of 2005 and 2006, and the wet years of 2007 and 2008 there is a factor two of
the inundated surface (Fig. 6).
Indian Space Research Organisation (ISRO) will launch the Saral/Altika mission, which
will be the first altimeter operating in Ka band which will have the main advantage of a
better spatial resolution due to smaller footprint of the radar signal (150 m instead of few
km), allowing a better discrimination of water in floodplains and anastomosed rivers
with a large number of small channels. This mission will be placed on the same orbit
than Envisat and should cover the needs for continental water level monitoring, for lakes,
rivers, and floodplains. In 2013, the CNES, EUMETSAT, and NASA will continue the
Jason program, with the launch of Jason-3 radar altimeter in Ku and C bands.
However, none of those missions is dedicated exclusively to continental hydrology. The
future SWOT mission is the first satellite mission dedicated to the measurement of
continental surface water. SWOT will provide a global inventory of all terrestrial water
bodies whose surface area exceeds (250 m2) and river whose width exceeds 100 m, at
sub-monthly, seasonal and annual time scales (Biancamaria et al., [24]). The principal
instrument of SWOT will be a Ka-band Radar Interferometer (KaRIN), which will
provide heights and co-registered all weather imagery of water over 2 swaths, each 60
km wide, with an expected precision of 1cm/km for water slopes, and absolute height
level precision of 10 cm / km2 (Fig. 7). SWOT will provide an estimate of river
discharge, and map floodplain topography and channel reaches. SWOT will allow us to
better quantify the exchange between rivers and floodplains for improved prediction of
inundations (Biancamaria et al., [25]). For floodplain mapping the improvement should
be very significant since SWOT will have a global coverage with a resolution of 250 by
Fifteenth International Water Technology Conference, IWTC-15 2011, Alexandria, Egypt
ACKNOLEDGMENT
This work has been supported by the Centre National dEtudes Spatiales (CNES) in
the frame of the TOSCA program. It has also been supporter by the CEFIPRA
program. The altimetry data are downloaded from the Centre de Topographie des
Ocans et de lHydrosphre (CTOH) of Legos. The Modis data are downloaded from
the WIST data centre: Earth Observing System Data and Information System
(EOSDIS). 2009. Earth Observing System ClearingHOuse (ECHO) / Warehouse
Inventory Search Tool (WIST) Version 10.X [online application]. Greenbelt, MD:
EOSDIS, Goddard Space Flight Center (GSFC) National Aeronautics and Space
Administration (NASA). URL: https://wist.echo.nasa.gov/api/
REFERENCES
[1] Crtaux J-F and Birkett, C., Lake studies from satellite altimetry, C R Geoscience,
doi: 10.1016/J.cre.2006.08.002, 2006
[2] Crtaux J-F, Calmant, S., Abarca Del Rio, R., Kouraev, A., and Berg-Nguyen, M.,
Lakes studies from satellite altimetry, Handbook on Coastal altimetry, Springer ed.,
chap. 19, 2010
[3] Calmant, S., Seyler, F. and Cretaux, J.F., Monitoring Continental Surface Waters by
Satellite Altimetry, Survey in Geophysics, special issue on Hydrology from Space, Vol
29, Issue 4-5, pp 247-269, 2008.
[4] Tyr, J., Pietroniro, A. and Martz, L.W. Multisensor Hydrologic Assessment of a
Freshwater Wetland, Remote Sensing of Environment, 75(2), 162-173, 2001
[5] Tyr, J., and Pietroniro A., Towards operational monitoring of a northern wetland
using geomatics-based techniques, Remote Sensing of Environment, 97, 174-191, 2005
[6] Frappart F., Seyler, F., Martinez, J-M., Leon J., and Cazenave, A., Determination of
the water volume in the Negro River sub basin by combination of satellite and in situ
data, Remote Sensing of Environment, 99, 387-399, 2005.
[7] Frappart, F., Dominh, K., Lhermitte, J., Ramilllien, G., Cazenave, A. and LeToan, T.,
Water volume change in the lower Mekong Basin from satellite altimetry and imagery
data, Geophys. J. Int., 167, 570-584, 2006a.
[8] Henry, J-B., Chastanet, P., Fellah, K., and Desnos, Y-L., ENVISAT Multi-Polarised
ASAR data for flood mapping, International Journal of Remote Sensing, Vol 27, no 10,
1921-1929, 2006
[9] Bartsch A., Pathe, C., K. Scipal, K., and Wagner, W., Detection of permanent open
water surfaces in central Siberia with ENVISAT ASAR wide swath data with special
emphasis on the estimation of methane fluxes from tundra wetlands, Hydrology
Research, 39.2, 89-100, 2008
[10] Peng, D.Z., L. Xiong, S. Guo, and N. Shu, Study of Dongting Lake area variation
and its influence on water level using MODIS data, Hydrological Sciences, 50(1), pp
31-44, 2005
Fifteenth International Water Technology Conference, IWTC-15 2011, Alexandria, Egypt
[11] Sakamoto T., Nguyen, N.V., Kotera, A., Ohno, H., Ishitsuka, N., and Yokozawa,
M., Detecting temporal changes in the extent of annual flooding within the Cambodia
and the Vietnamese Mekong Delta from MODIS time series imagery, Remote Sensing of
virtual wide swath altimetry to improve Arctic river modeling. Remote Sensing of
Environment, doi:10.1016/j.rse.2010.09.008, in press, 2010b.
[26] Biancamaria, S., Boone, A., Bates, P., and Mognard, N.M., Large-scale coupled
hydrologic and hydraulic modelling of the Ob river in Siberia, J. of Hydrology, 379,
136150, 2009.
See discussions, stats, and author profiles for this publication at:
http://www.researchgate.net/publication/222427328
CITATIONS
119
READS
284
1 AUTHOR:
Brett Sanders
University of California, Irvine
88 PUBLICATIONS 1,893 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: Brett Sanders
Retrieved on: 30 November 2015
*,1
Department of Civil and Environmental Engineering, University of California, Irvine, CA, United States
Received 19 October 2006; received in revised form 6 February 2007; accepted 8 February 2007
Available online 21 February 2007
Abstract
Recent and highly accurate topographic data should be used for flood inundation modeling, but this is
not always feasible given time
and budget constraints so the utility of several on-line digital elevation models (DEMs) is examined
with a set of steady and unsteady test
problems. DEMs are used to parameterize a 2D hydrodynamic flood simulation algorithm and
predictions are compared with published
flood maps and observed flood conditions. DEMs based on airborne light detection and ranging (LiDAR)
are preferred because of horizontal
resolution, vertical accuracy (_0.1 m) and the ability to separate bare-earth from built structures and
vegetation. DEMs based on
airborne interferometric synthetic aperture radar (IfSAR) have good horizontal resolution but gridded
elevations reflect built structures
and vegetation and therefore further processing may be required to permit flood modeling. IfSAR and
shuttle radar topography mission
(SRTM) DEMs suffer from radar speckle, or noise, so flood plains may appear with non-physical relief
and predicted flood zones may
include non-physical pools. DEMs based on national elevation data (NED) are remarkably smooth in
comparison to IfSAR and SRTM
but using NED, flood predictions overestimate flood extent in comparison to all other DEMs including
LiDAR, the most accurate. This
study highlights utility in SRTM as a global source of terrain data for flood modeling.
_ 2007 Elsevier Ltd. All rights reserved.
Keywords: Flood inundation; Flood modeling; DEM; DTM; LiDAR; IfSAR; SRTM
1. Introduction
Flooding is the most common and damaging natural
hazard faced by civilization, and flooding threats are likely
to increase given current climate change predictions that
suggest higher sea levels and more intense cyclonic weather
systems and precipitation. The frequency, distribution and
causes of floods over the last thirty years has been analyzed
to the south, and the great Central Valley and the San
Francisco Bay Area to the north.
Fig. 2 shows how topography is represented by IfSAR,
NED and SRTM DEMs. NED appear smooth in comparison
to SRTM and IfSAR data, and depressions associated
with river channels and overbank flood plains can be seen
clearly. Comparing 1 s NED and SRTM, NED appear
smooth while SRTM show a subtle waviness that is a signature
of radar speckle [13]. IfSAR data reveal the most
detail, and upon close inspection it becomes clear that
the apparent roughness is linked to topographic attributes
such as vegetation and buildings. For example, large rectangular
structures of a commercial business district are visible
east of I-5, and tree tops appear as small mounds or
humps along the SCR corridor. The magnified views in
B.F. Sanders / Advances in Water Resources 30 (2007) 18311843 1833
SCR. This and the I-5 crossing are well resolved by 1/3 s
NED and visible in 1 s NED, but poorly resolved by 1
and 3 s SRTM. Fig. 4 shows what happens when such
hydraulically important features are not included in the
geometric representation of the channel. The model predicts
a relatively uniform-width flood path using SRTM,
Acknowledgments
This study was made possible by a Grant from the National
Science Foundation Multi-Disciplinary Center for
Earthquake Engineering Research (MCEER) and by research
support provided by the University of Nottingham
School of Civil Engineering, where the first author spent
his sabbatical during the 20062007 academic year. The
authors thank H. Morvan, M. Smith and N. Wright (U.
Nottingham), P. Bates (U. Bristol), J. Carter (NOAA/
CSC) and D. Finnegan (USACE/CRRE) for discussions
and assistance that contributed to this study.
References
[1] Baltavias EP. Airborne laser scanning: basic relations and formulas.
ISPRS J Photogramm 1999;54:199214.
[2] Bates PD, Marks KJ, Horritt MS. Optimal use of high-resolution
topographic data in flood inundation models. Hydrol Process
2003;17:53757.
[3] Berthier E, Arnaud Y, Vincent C, Remy F. Biases of SRTM in highmountain
areas: implications for the monitoring of glacier volume
changes. Geophys Res Lett 2006;33:L08502. doi:10.1029/
2006GL025862.
[4] Begnudelli L, Sanders BF. Unstructured grid finite volume algorithm
for shallow-water flow and transport with wetting and drying. J
Hydraul Eng 2006;132(4):37184.
[5] Begnudelli L, Sanders BF. Conservative wetting and drying methodology
for quadrilateral grid finite volume models. J Hydraul Eng
2006;133(3).
[6] Begnudelli L, Sanders BF. Simulation of the St. Francis dam-break
flood. J Eng Mech, in review.
[7] Bradford SF, Sanders BF. Finite-volume model for shallow-water
flooding of arbitrary topography. J Hydraul Eng 2002;128(3):28998.
[8] Bradford SF, Sanders BF. Performance of high-resolution, non-level
bed, shallow-water models. J Eng Mech 2005;131(10):107381.
[9] Cobby DM, Mason DC, Davenport IJ. Image processing of airborne
scanning laser altimetry data for improved river flood modeling.
ISPRS J Photogramm 2001;56:12138.
[10] Cobby DM, Mason DC, Horritt MS, Bates PD. Two-dimensional
hydraulic flood modelling using a finite-element mesh decomposed
according to vegetation and topographic features derived from
airborne scanning laser altimetry. Hydrol Process 2003;17:19792000.
[11] Dartmouth Flood Observatory http://www.dartmouth.edu/
~floods.
[12] Evans E, Hall J, Penning-Roswell E, Sayers P, Thorne C, Watkinson
A. Future flood risk management in the UK. Proc Inst Civ Eng
Water Management 2006;159(WM1):5361.
[13] Falorni G, Teles V, Vivoni ER, Bras RL, Amaratunga KS. Analysis
and characterization of the vertical accuracy of digital elevation odels
from Shuttle Radar Topography Mission. J Geophys Res
2005;110:F02005. doi:10.1029/2003JF000113.
[14] Guth PL. Geomorphometry from SRTM: comparison to NED.
Photogramm Eng Rem S 2006;72(3):26977.
[15] HEC-RAS software is described and accessed at www.hec.usace.
army.mil.
[16] Hodgson ME, Jensen JR, Schmidt L, Schill S, Davis B. An evaluation
of LIDAR- and IFSAR-derived digital elevation models in leaf-on
conditions with USGS Level 1 and Level 2 DEMs. Remote Sens
Environ 2003;84:295308.
[17] Hogg AJ, Pritchard D. The effects of hydraulic resistance on dambreak
and other shallow inertial flows. J FluidMech 2004;501:179212.
[18] Horritt MS, Bates PD. Evaluation of 1D and 2D numerical
models for predicting river flood inundation. J Hydrol 2002;268:
8799.
[19] Intermap Corp. Product handbook and quick start guide. Version
3.3. June 15; 2004.
[20] Katopodes ND, Strelkoff T. Computing two-dimensional dam-break
flood waves. J Hydraul Div ASCE 1978;104(HY9):126988.
[21] Lin B, Wicks JM, Falconer RA, Adams K. Integration of 1D and 2D
hydrodynamic models for flood simulation. Proc Inst Civ Eng
Water Management 2006;159(WM1):1926.
[22] Marks K, Bates P. Integration of high-resolution topographic
data with floodplain flow models. Hydrol Process 2000;14:
210922.
[23] Mason DC, Cobby DM, Horritt MS, Bates PD. Floodplain friction
parameterization in two-dimensional river flood models using vegetation
heights derived from airborne scanning laser altimetry. Hydrol
Process 2003;17:171132.
[24] Morris MW. CADAM concerted action on dambreak modelling.
Report SR 571, HR Wallingford; 2000.
[25] Mrse RD. Impact of resolved bed forms on model predictions of flow
resistance, M.S. Thesis, Department of Civil and Environmental
Engineering, University of California, Irvine; 2006.
[26] Outland CF. Man-made disaster: the story of St. Francis Dam. The
Arthur H. Clark Company; 1963. p. 249.
[27] PBS&J, LJA Engineering and Land Surveying. Flood flow frequency
analysis for tropical storm allison recovery project, Final Report,
See discussions, stats, and author profiles for this publication at:
http://www.researchgate.net/publication/256756488
CITATIONS
16
READS
145
4 AUTHORS, INCLUDING:
Salvatore Grimaldi
Tuscia University
A. Petroselli
Tuscia University
Fernando Nardi
Universit per Stranieri di Perugia
a,b,c,,
Dipartimento per la innovazione nei sistemi biologici, agroalimentari e forestali (DIBAF Department), University of Tuscia, Via San Camillo De
Lellis snc, 01100 Viterbo, Italy
b Honors Center of Italian Universities (H2CU), Sapienza University of Rome, Via Eudossiana 18, 00184 Roma, Italy
c Department of Mechanical and Aerospace Engineering, Polytechnic Institute of New York University, Six MetroTech Center, Brooklyn, NY
11201, United States
d Dipartimento di scienze e tecnologie per lagricoltura, le foreste, la natura e lenergia (DAFNE Department), University of Tuscia, Via San
Camillo De Lellis snc, 01100 Viterbo, Italy
e WARREDOC, Universit per Stranieri di Perugia, Piazza Fortebraccio 4, 06123 Perugia, Italy
a
articleinfo
Article history:
Received 15 August 2012
Received in revised form 8 February 2013
Accepted 14 February 2013
Available online 26 February 2013
This manuscript was handled by Andras
Bardossy, Editor-in-Chief, with the
assistance of Niko Verhoest, Associate Editor
Keywords:
Flood hazard
Flood mapping
Continuous hydrologic model
2D hydraulic model
WFIUH
summary
In this work, a fully-continuous hydrologichydraulic modeling framework for flood mapping is introduced
and tested. It is characterized by a simulation of a long rainfall time series at sub-daily resolution
that feeds a continuous rainfallrunoff model producing a discharge time series that is directly given as
an input to a bi-dimensional hydraulic model. The main advantage of the proposed approach is to avoid
the use of the design hyetograph and the design hydrograph that constitute the main source of subjective
analysis and uncertainty for standard methods. The proposed procedure is optimized for small and ungauged
watersheds where empirical models are commonly applied. Results of a simple real case study confirm
that this experimental fully-continuous framework may pave the way for the implementation of a
less subjective and potentially automated procedure for flood hazard mapping.
_ 2013 Elsevier B.V. All rights reserved.
1. Introduction
Flood hazard mapping is a fundamental non-structural measure
implemented by governments for sustainable urban and land planning,
protecting human properties, activities, lives, and preserving
the ecohydrology of river corridors (European Parliament, 2007).
While hydrogeomorphic methods, that make extensive use of Digital
Elevation Models (DEMs) and terrain analysis have been recently
adopted for preliminary floodplain characterization at the
basin scale (e.g. Dodov and Foufoula-Georgiou, 2006; Nardi et al.,
2006), detailed inundation maps are generally based on advanced
hydrologic and hydraulic modeling. In particular, the hydrologic
forcing is estimated through a rainfallrunoff model that quantifies
flood peak discharges or a flow hydrograph of given return period
(T). The hydraulic analysis is, developed, by means of a mono
dimensional (1D) surface water model in a steady state condition
and/or by means of an unsteady 1D or bidimensional (2D) flood
routing algorithm for the accurate spatially distributed evaluation
of the flow and velocity dynamics.
The more complex and detailed 2D models suffered in the past
of limited amounts of data and of the computational issue. However,
latest 2D models, overcoming these limitations, are now
increasingly used as the new standard approach for accurately simulating
spatial and temporal dynamics of the flooding process (e.g.
Horritt et al., 2007; Di Baldassarre et al., 2009). As a result, 2D simulations
represent a mandatory choice for detailed small scale urban
development projects as well as for large scale ones, especially
for complex urban settings where the 1D hypothesis is often not
applicable and the socio-economic impact of floods must also be
assessed (European Parliament, 2007).
Nevertheless, current hydrologichydraulic modeling procedure
Journal of Hydrology
that is arbitrary and far from representing real rainfall storms, and
the hypothesis that the design hyetograph and corresponding hydrograph
have the same return period.
For estimating the IDF, we selected the three-parameter formula
(Chow et al., 1988):
id; T
aT
bT d_cT 1
where i(d,T) is the gross rainfall intensity, and d the rainfall duration.
a(T), b(T) and c(T) are determined through a nonlinear leastsquares
method. The i(d,T) is estimated on the annual maxima for
the corresponding durations of 1, 3, 6, 12 and 24 h according to
the empirical cumulative distribution of the Weibull formula:
j
n11_
1
T 2
where j = 1,. . . ,n represents the jth observation of the samples arranged
in ascending order.
The selected hyetograph estimation method is the symmetric
Chicago hyetograph (Keifer and Chu, 1957), that, given as an input
to the rainfallrunoff procedure (step 4), provides the corresponding
design hydrograph.
2.2. Semi Continuous Approach (SCA)
The above mentioned four issues of the EBA correspond to four
significant theoretical and conceptual hypotheses representing
sources of uncertainty and/or user-derived subjective choices
affecting the design hydrograph estimation procedure. The SCA
aims to reduce the impact of these conceptual hypotheses by
substituting the IDF-derived critical rainfall with a simulated rainfall
time series. This calibrates a statistical model on the observed
rainfall data and provides it to the rainfallrunoff model without
any prior estimation of a single design event. In this way, the impact
of the uncertainty of the IDF, critical rainfall, and design hyetograph
hypotheses is substituted by the need for the proper
selection of a rainfall simulation model. The return period is also
estimated directly on the simulated runoff series allowing it to
omit another source of uncertainty given by the assumption that
the design discharge preserves the rainfall time period.
Fig. 1. Flow-chart of the three flood modeling approaches: event-based, semi-continuous, and fully-continuous.
S. Grimaldi et al. / Journal of Hydrology 487 (2013) 3947 41
Gx 1 _ pHx pFx 3
where x is the random variable for the flow depth for each cell, p is
the probability of wet cell (x > 0), 1 _ p is the probability of a dry
event (x = 0), G(x) is the mixed distribution function considering
both wet and dry events, H(x) is the step function: H(x) = 0, if
x = 0; H(x) = 1, if x > 0, and F(x) is the probability density function
estimated for non-zero values.
3. The Rio Torbido case study
The Rio Torbido, a 61.67 Km2 river basin located in central Italy
draining directly into the main channel of the Tiber River, is selected
for this case study. Elevations range from 85 mto 625 m, the average
slope is 21.9% and the maximum hydrologic distance of the outlet
from the watershed divide is 25.8 Km. The selected hydraulic modeling
domain is located in the proximity of the confluence of the Tiber
where the Torbido meanders within a large floodplain. This area,
shown in Fig. 2, is 1.5 Km2 large and 2 Km long wide.
The DEM, with integer precision and 20 mof spatial resolution, is
extracted from the Italian Geographic Military Institute (IGMI, 2003)
MATRIX raster dataset. The IGMI Matrix DEM, derived from the
interpolation of 1:25.000 contour lines, is characterized by a spatial
and vertical accuracy of respectively 510 mand 15 m. Land cover
is derived from the CORINE project (European Commission, 2000).
Rainfall measurements include 49 years of observed data at a daily
time scale and 10 years at 5 min of resolution (Serinaldi, 2010).
Fig. 3 shows the Rio Torbido basin DEM with the simulated river
network implementing the terrain analysis procedure described in
Section 2. The detailed topography, bathymetry and roughness conditions
of the 2D hydraulic modeling project also including the surveyed
river cross sections (Fig. 2) were derived by integrating the
available digital data with the results of an extensive field campaign.
The final 2D flood model resolution is 20 m that is the size
of the grid cell representing the domain topography. Upstream
and downstream boundary conditions for the hydraulic model are
respectively the inflow hydrograph for the first channel grid element
and the uniform flow assigned to the boundary grid elements.
4. Comparison test
EBA, SCA and FCA are compared by means of flood maps that are
derived using the above described procedure. Although a case
study with available observations would be better representative
and would allow a quantitative FCA validation, the aim here is to
provide a qualitative evaluation of the differences among the three
procedures. The comparison is here proposed to visualize the impact
6. Conclusions
In this paper three hydrograph estimation procedures for defining
the design hydrologic input of flood modeling and mapping are
investigated. By taking into consideration small and ungauged river
basins that usually requires subjective and empirical approaches
for rainfallrunoff modeling, the following three methodologies
have been selected and compared with specific regard to the eventual
advantages of the FCA.
The so-called event-based approach, or EBA, is based on a synthetic
representation of both the rainfall input (i.e. IDF method)
and runoff output for hydrograph characterization. The EBA has
Fig. 8. Water depth distribution in selected inundation domain related to 20-years return period: (a) EBA approach; (b) SCA approach; (c) FCA
approach.
S. Grimaldi et al. / Journal of Hydrology 487 (2013) 3947 45
been the most simple and commonly applied procedure for decades,
but it is severely impacted by conceptual hypotheses,
assumptions, related errors and uncertainties in hydrologic terms.
The more recent Semi Continuous Approach, or SCA, overcomes
most of the conceptual hypotheses of the EBA and introduces the
use of a simulated rainfall time series by which the rainfallrunoff
model is applied using a continuous implementation. This allows
the extrapolation of a design hydrograph from the runoff synthetic
time series. This SCA also removes the conceptual steps of the EBA
method in the rainfall input estimation, but it is still predominantly
synthetic in the extrapolation of the corresponding
hydrograph.
The proposed Fully Continuous Approach, or FCA, neglects any
preliminary conceptualization of the hydrologic modeling procedure
for design hydrograph estimation by implementing a method
that provides, as input to the hydraulic model, the entire simulated
runoff time series. In this way, there is no conceptualization until
the last step of flood modeling and mapping analysis. Furthermore,
it performs the last necessary analytical and statistical evaluation,
without prior simplification, of the hydraulic modeling results to
identify the flood inundation map and the corresponding design
flow depths and velocities for each cell of any given return period.
Even if the selected coarse resolution DEM is not appropriate for
a detailed quantitative evaluation, it is appositely used for demonstrating
the expected different behavior on the three approaches
while using a topographic resolution of common use for practical
flood mapping in ungauged basins. More specifically, results show
that the three methodologies provide significantly different outputs
with major differences characterizing the EBA with respect
to the SCA and FCA. Those differences confirm that the EBA should
not be used for flood modeling applications that are mainly governed
by the extra-channel physical processes and dynamics. For
these cases the hydrograph shape, duration and volume play the
major role rather than just the flood peak. This is always the case
of high return time flood mapping studies that require an accurate
and detailed characterization of the input hydrograph and the
application of a 2D hydraulic model. It is also shown that the differences
in design hydrograph estimation of the SCA with respect
to the FCA are not transferred into the corresponding flood mapping,
but this is reasonable considering that the two approaches
differ only for the hydrograph shape.
The described proof of concept suggests to prefer the FCA procedure
since it focuses on the implementation of a more realistic
representation of the rainfallrunoff process by means of a simple,
parsimonious model that could be easily implemented by professionals
for practical flood modeling and mapping applications in
small and ungauged river basins.
Future research shall validate the FCA on a small watershed
where flood maps as well as runoff observations shall be available
with the aim of investigating the potential use of such method for
gauged basins. As a result several DEMs at different resolution
should be selected to evaluate the topographic data impact on
model performances. Moreover, the validation, using a DEM with
proper resolution and precision, (<5 m of resolution and <1 m of
precision), shall be also developed. Finally, since the proposed
FCA method provides flow height and velocity time series for each
cell of the inundated domain, a specific statistically-based post
processing analysis shall be implemented for the estimating the
uncertainty of flood inundation maps with the final aim of providing
to floodplain managers and urban planners a detailed decision
making framework.
Acknowledgments
This research was partially supported by the Honors Center of
Italian Universities.
Authors are particularly grateful to Dr. Francesco Serinaldi for
his contribution in the statistical analysis of the inundation maps,
to the eponymous reviewer, Dr. Giuliano Di Baldassarre, and to the
anonymous reviewer for their useful comments.
Table 2
Summary of flood volumes and areas for six return period values estimated for the EBA, SCA and FCA.
T (years) 2 5 10 20 50 100
Event Based Approach (EBA)
V (floodplain and channel) (m3 _ 103) 69.8 115.6 142.9 182.2 269.1 341.5
V (floodplain)(m3 _ 103) 0.0 2.0 7.6 23.6 80.0 134.7
A (floodplain and channel) (m2 _ 103) 34.8 42.0 58.8 80.8 130.4 158.0
A (floodplain) (m2 _ 103) 0.0 7.2 24.0 46.0 95.6 123.2
Semi Continuous Approach (SCA)
V (floodplain and channel) (m3 _ 103) 76.0 134.5 175.5 236.2 297.7 365.7
V (floodplain) (m3 _ 103) 0.0 5.2 20.4 56.6 101.2 153.5
A (floodplain and channel) (m2 _ 103) 34.8 51.2 77.6 122.8 143.6 160.8
A (floodplain) (m2 _ 103) 0.0 16.4 42.8 88.0 108.8 126.0
Fully Continuous Approach (FCA)
V (floodplain and channel) (m3 _ 103) 77.4 136.1 180.7 249.8 310.3 382.9
V (floodplain) (m3 _ 103) 0.0 5.6 23.0 66.1 110.7 166.9
A (floodplain and channel) (m2 _ 103) 34.8 51.2 80.0 126.4 147.2 165.2
A (floodplain) (m2 _ 103) 0 16.4 45.2 91.6 112.4 130.4
Table 3
Summary of flood volume and area percentage differences for the EBA, SCA and FCA.
T (years) 2 5 10 20 50 100
Semi continuous event based
DV (%) (floodplain and
channel)
8.9 16.4 22.8 29.6 10.6 7.1
DV (%) (floodplain) 162.9 167.8 139.3 26.4 14.0
DA (%) (floodplain and
channel)
0.0 21.9 32.0 52.0 10.1 1.8
DA (%) (floodplain) 127.8 78.3 91.3 13.8 2.3
My continuous event based
DV (%) (floodplain and
channel)
10.9 17.7 26.5 37.1 15.3 12.1
DV (%) (floodplain) 183.3 201.9 179.7 38.3 23.9
DA (%) (floodplain and
channel)
0.0 21.9 36.1 56.4 12.9 4.6
DA (%) (floodplain) 127.8 88.3 99.1 17.6 5.8
Fully continuous semi continuous
DV (%) (floodplain and
channel)
1.8 1.1 3.0 5.8 4.2 4.7
DV (%) (floodplain) 7.7 12.7 16.9 9.4 8.7
DA (%) (floodplain and
channel)
0.0 0.0 3.1 2.9 2.5 2.7
DA (%) (floodplain) 0.0 5.6 4.1 3.3 3.5
46 S. Grimaldi et al. / Journal of Hydrology 487 (2013) 3947
References
Alfieri, L., Laio, F., Claps, P., 2008. A simulation experiment for optimal design
hyetograph selection. Hydrol. Process. 22, 813820.
Aronica, G., Bates, P.D., Horritt, M.S., 2002. Assessing the uncertainty in distributed
model predictions using observed binary pattern information within GLUE.
Hydrol. Process. 16, 20012016.
Bates, P.D., Horritt, M.S., Aronica, G., Beven, K., 2004. Bayesian updating of flood
inundation likelihoods conditioned on flood extent data. Hydrol. Process. 18,
33473370.
Blazkova, S., Beven, K., 2009. A limits of acceptability approach to model evaluation
and uncertainty estimation in flood frequency estimation by continuous
simulation: Skalka catchment, Czech Republic. Water Resour. Res. 45, W00B16.
Calver, A., Stewart, E., Goodsell, G., 2009. Comparative analysis of statistical and
catchment modelling approaches to river flood frequency estimation. J. Flood
Risk Manage. 2, 2431.
Chow, V.T., Maidment, D.R., Mays, L.W., 1988. Applied Hydrology. Mc-Graw-Hill,
New York.
European Commission. 2000. CORINE (Coordination of Information on
Environment) Database, a Key Database for European Integrated
Environmental Assessment. Programme of the European Commission,
properties of simulated drainage patterns using DEMs: the flat area issue.
Hydrol. Sci. J. 53 (6), 11761193.
Nishat, S., Guo, Y., Baetz, B.W., 2010. Antecedent soil moisture conditions of
different soil types in South-western Ontario, Canada. Hydrol. Process. 24,
24172424.
Nnadi, F.N., Kline, F.X., Wary, H.L., Wanielista, M.P., 1999. Comparison of design
storm concepts using continuous simulation with short duration storms. J. Am.
Water Resour. Assoc. 31 (1), 6185.
NRCS. 1997. Ponds-Planning, Design, Construction. Agriculture Handbook N. 590.
US Natural Resources Conservation Service, Washington DC.
OBrien, J.S., Julien, P.Y., Fullerton, W.T., 1993. Two-dimensional water flood and
mudflow simulation. J. Hydraulic Eng. 119 (2), 244261.
Pappenberger, F., Matgen, P., Beven, K.J., Henver, J.B., de Pfister, L., Fraipont, P., 2006.
Influence of uncertainty boundary conditions and model structure on flood
inundation predictions. Adv. Water Resour. 29, 14301449.
Pramanik, N., Panda, R.K., Sen, D., 2010. Development of design flood hydrographs
using probability density functions. Hydrol. Process. 24 (4), 415428.
Rahman, A., Weinmann, P.E., Hoang, T.M.T., Laurenson, E.M., 2002. Monte Carlo
simulation of flood frequency curves from rainfall. J. Hydrol. 256, 196210.
Sampson, C.C., Fewtrell, T.J., Duncan, A., Shaad, K., Horritt, M.S., Bates, P.D., 2012.
Use of terrestrial laser scanning data to drive decimetric resolution urban
inundation models. Advances in Water Resources, 41, 117, ISSN 0309-1708.
10.1016/j.advwatres.2012.02.010.
Santini, M., Grimaldi, S., Nardi, F., Petroselli, A., Rulli, M.C., 2009. Pre-processing
algorithms and landslide modelling on remotely sensed DEMs. Geomorphology
113 (12), 110125.
Schertzer, D., Lovejoy, S., 1987. Physical modeling and analysis of rain and clouds by
anisotropic scaling of multiplicative processes. J. Geophys. Res. 92, 96939714.
Serinaldi, F., 2009. A multisite daily rainfall generator driven by bivariate copulabased
mixed distributions. J. Geophys. Res. 114, D10103. http://dx.doi.org/
10.1029/2008JD011258.
Serinaldi, F., 2010. Multifractality, imperfect scaling and hydrological properties of
rainfall time series simulated by continuous universal multifractal and discrete
random cascade models. Nonlin. Processes Geophys. 17, 697714. http://
dx.doi.org/10.5194/npg-17-697-2010.
Serinaldi, F., Grimaldi, S., 2011. Synthetic design hydrographs based on distribution
functions with finite support. Journal of Hydrologic Engineering, 16, (5) 434
446; doi:10.1061/(ASCE)HE.1943-5584.0000339.
Soczyska, U., Nowicka, B., Somorowska, U., Kupczyk, E., Suligowski, R., 1997.
Prediction of design storms and floods. IAHS-AISH Publ. 246, 297303.
Tarboton, D.G., Ames, D.P., 2001. Advances in the Mapping of Flow Networks from
Digital Elevation Data. World Water and Environmental Resources Congress,
Orlando, Florida, May 2024, ASCE.
Tarboton, D.G., Bras, R.L., Rodriguez-Iturbe, I., 1991. On the extraction of channel
networks from digital elevation data. Hydrol. Process. 5 (1), 81100.
USDA-SCS. 1986. Urban Hydrology for Small Watersheds. Technical Release 55.
Washington, DC.
USDA-NRCS. 2010. National Engineering Handbook, Part 630 Hydrology. <http://
www.info.usda.gov/viewerFS.aspx?hid=21422>.
Verhoest, N.E.C., Vandenberghe, S., Cabus, P., Onof, C., Meca-Figueras, T.,
Jameleddine, S., 2010. Are stochastic point rainfall models able to preserve
extreme flood statistics? Hydrol. Process. 24, 34393445.
Viglione, A., Bloschl, G., 2009. On the role of storm duration in the mapping of
rainfall to flood return periods. Hydrol. Earth Syst. Sci. 13, 205216.
Viglione, A., Merz, R., Bloschl, G., 2009. On the role of the runoff coefficient in the
mapping of rainfall to flood return periods. Hydrol. Earth Syst. Sci. 13, 577593.
Yoo, C., Jung, K.S., Kim, T.W., 2005. Rainfall frequency analysis using a mixed
Gamma distribution: Evaluation of the global warming effect on daily rainfall.
Hydrol. Process. 19 (19), 38513861.
S. Grimaldi et al. / Journal of Hydrology 487 (2013) 3947 47
SpectralReflectancePropertiesof
Hydrocarbons:
RemoteSensingImplications
EDWARDA.CLOUTIS
Thespectralreflectancepropertiesofbituminoustarsandswere
examinedinthe
wavelengthrangefrom0.35to2.6micrometers.Uniqueabsorption
featuresduetoall
themajorphases,exceptquartz,appearinthespectra.The
intensitiesoftheabsorption
featurescorrelatewiththeabundancesofthevariousphases.The
resultsofthisstudy
haveapplicationstotheremotesensingofmanyterrestrialand
extraterrestrialtargets
becauseofpotentialsimilaritiesbetweentarsandandother
hydrocarbonoccurrences.
Forexample,itisfoundthathighlypolymerizedhydrocarbonsare
aplausible
constituentofthedarkmaterialononeofSaturn'ssatellites,
Iapetus.
ASYSTEMATICSTUDYOFTHESPECtral
reflectancepropertiesofbituminous
tarsandswasundertakenin
ordertodeterminewhetherthisclassof
materialshasdistinctspectralproperties(1).
Tarsandsarecomposedofamixtureof
clays,bitumen(acomplexarrayofvariously
polymerizedhydrocarbons),quartzgrains,
water,andminoraccessoryminerals(24).
Understandingthespectralpropertiesof
thismaterialisessentialforgeologicalremote
sensingofterrestrialandextraterrestrial
targets,becauseverylittleisknownabout
thespectralreflectancepropertiesofhydrocarbon
bearingmaterials.
Amongtheleadingextraterrestrialcandidates
forpossiblehydrocarbonoccurrences
areTrojanasteroids,cometarynuclei,and
thedarksideofoneofSaturn'ssatellites,
Iapetus(5,6).Tarsandsmayserveasreasonable
spectralanalogsoftheseobjectsand
otherbodieswhosesurfacesarebelievedto
becomposedofvariouscombinationsof
organicmatter,clays,andH20.Spectral
analysisofthetarsandsmayalsomake
possibletheremotesensingdetectionofterrestrial,
surficialhydrocarbonseeps,because
highlypolymerizedhydrocarbonsarecharacteristic
ofbothtarsandsandsurfaceexposures
(7).Thetermkerogenisoftenusedto
refertoextraterrestrialoccurrencesofhydrocarbons.
Kerogenandbitumendiffer
primarilyinthehydrogen/carbonratio;in
bitumentheratioishigherbecauseofgreater
maturation(7).Rapidcharacterizationof
tarsandsandheavyoilsampleswouldbe
possiblewiththedevelopmentofquantitative
spectralanalysistechniques.Thiscould
improvecommercialbitumenextractionefficiencies
andsimplifytheonerousprocedures
nowrequiredfortarsandcharacterization.
Spectralreflectancetechniquesmay
eventuallybeextendedtotheanalysisof
othernonrenewableenergyresourcessuch
asoilshalesandcoals.
AllthephasespresentintheAthabascatar
sands,exceptquartz,displaydistinctabsorption
bandsthatarepotentiallyresolvable.
Theshapes,intensities,andwavelengthpositions
ofthesebandsmayprovideinformation
onthephysicalandchemicalproperties
ofthevariousphases(811).
Manyorganiccompoundsdisplayelectronic
transitionsarisingfromexcitationsof
bondingelectronsinthewavelengthregion
from0.1to0.35,um.Asthecomplexityof
theorganicmoleculesincreases,themaximum
absorptionshiftstowardlongerwavelengths
andindividualbandsbecomeless
distinctbecauseofincreasingoverlap(12
14).Amaterialascomplexasbitumenisnot
expectedtoexhibitindividual,resolvable
absorptionbandsintheultravioletandvisible
spectralregion.Abroadoverallreflectance
decreasetowardshorterwavelengthsis
seenincoals,chars,oilshales,andtarsands
(12,1417).
Thewavelengthpositionsofthemost
intense,majororganicfundamentalbands
arelistedinTable1.Becausetheenergiesof
thefimdamentalabsorptionbandsdepend
onanumberoffactorssuchaslocalatomic
configurations(18),thewavelengthpositions
oftheovertoneandcombination
bandsaresomewhatuncertain(Table2).
Thelowoverallreflectanceoftarsandsis
expectedtosuppressallbutthemostprominent
ofthesebands.Themostpromising
regionsinwhichtosearchfororganicabsorption
bandsarenear1.7,umandbetween
2.2and2.6,um(16,19).The1.7,umregion
isdominatedbyvariousCHstretching
overtonesandcombinationbands.Thefundamental
bandsthatcontributetothisfeature
arethemostintenseabsorbersinthe
infraredspectraofbitumen(20).Theregion
from2.2to2.6,umisaffectedbynumerous
overlappingcombinationandovertone
bands.Bitumentransmissionspectrashow
moderatelyintenseabsorptionbetween
1750and900cmlwhichcancombine
inanumberofwaysintheregionfrom2.2
to2.6,um(2,12,20,21).Becauseofthe
sheernumberofpossiblebands,theregion
hasalowoverallreflectancewithonlythe
mostintenseabsorptionbandsbeingpartially
resolvable(14,15,22).
Transitionserieselements,particularlyvanadium
andnickel,complexedwithporphyrin
typemoleculesarequitecommoninthe
Table1.Positionsofthemajororganicfundamental
absorptionbandsinenergyspace.
Fundamental
BandfrequencyAssignment
(cm')
a3030Alkene,aromaticCH
stretch
b2950AsymmetricCH3
stretch
c2920AsymmetricCH2
stretch
d2875SymmetricCH3stretch
e2850SymmetricCH2stretch
f1700CarbonylcarboxylCO
stretch
g1600Aromaticcarbon
stretch
h1450AsymmetricCH2,CH3
bend
i1375SymmetricCH3bend
Table2.Predictedwavelengthpositionsofthe
mostintenseovertonesandcombinationsofthe
majorfundamentalorganicabsorptionbandslisted
inTable1.
Combinationovertone
band
2a
a+c
2b
a+e
2c
b+d
c+e
2d
2e
e+f
c+g
e+g
b+h
c+h
b+i,d+h
e+h
d+i
c+j
Wavelength
(Lm)
1.65
1.68
1.69
1.70
1.71
1.72
1.73
1.74
1.75
2.20
2.21
2.25
2.27
2.29
2.31
2.33
2.35
2.53
REPORTSI65
DepartmentofGeology,UniversityofAlberta,Edmonton,
Alberta,CanadaT6G2E3.
I4jULYI989
Downloaded from www.sciencemag.org on December 5, 2013
0.51.01.52.02.5
Wavelength(gm)
Fig.1.Reflectancespectrum(0.35to2.6,um)of
alowbitumentarsand(sample8205)(1).
asphalticfractionoftarsands.Theydisplaya
numberofabsorptionbandsintheultraviolet
andvisiblespectralregions.Themost
intenseoftheseoccursat0.4p.mandhas
beendetectedinthespectraofoilshaleand
carbonaceouschondrites(23).
ThebitumencontentintheAthabascatar
sandstypicallyrangesfrom0to15%by
weight(3).Thisrangewasarbitrarilydivided
intothree5%intervalsinordertodisplay
spectralvariationsasafunctionofchanging
phaseabundances.Thereflectancespectraof
low,medium,andhighbitumensamples
weremeasured,andarepresentativesample
ofeachgroupwasselectedformoredetailed
analysis.
Thelowbitumengroupischaracterized
byprominentabsorptionbandsduetothe
clays(Fig.1).Theabsorptionbandsat
1.4,1.9,2.2,and2.3to2.6p.mare
assignedtostructuralandadsorbedwaterin
theclaysandtocationOHvibrational
bands(8).Thewavelengthpositionofthe
1.9p.mband(1.92p.m)doesnotcoincide
withthatexpectedforfreewater[1.94p.m
(11)].Thebroadnessofthisbandandthe
1.4p.mbandisconsistentwiththepresence
ofwaterinavarietyofsitesorwithstructurally
disorderedclays(2).Thedecreasein
reflectancefrom2.3to2.6p.misalsocharacteristic
ofaclaydominatedspectrum(8).
Discretebitumenabsorptionbandsareweak
orabsentinspiteoftheamountofbitumen
present(3.2%)andthedarkappearanceof
thesample.Indirectevidenceforbitumen
comesfromthelowoverallreflectanceofthe
sample,whichisnotcharacteristicofmost
clays(8,9).
Spectraofmediumbitumensamplesare
richerindetailthanthoseofthelowbitumen
samplesanddifferinotherways.The
depthofthe1.9p.mbandisreduced,anew
bandappearsat1.7,um,thedepthofthe
bandintheregionfromthe2.3to2.6p.mis
greater(Table3),andtheoverallreflectance
slopeinthisrangeisslightlypositive(Fig.
2).Theabsorptionbandthatappearsat1.7
p.mcanbeattributedtofirstorderovertones
andcombinationsofthevariousCH
stretchingfundamentals(Tables1and2).
Theincreaseinabsorptionintheregion
I66
Table3.Wavelengthpositionsoftheobserved
absorptionbandsandtheirdepths,Db.Some
bandpositionsaregivenasrangesbecauseof
variationsbetweendifferentsamplespectra.
BandSample
minmum
(ILm)820587048619
0.407.0
1.391.414.56.03.0
1.721.751.89.918.6
1.921.9519.010.43.8
2.105.0
2.152.184.19.611.4
2.216.515.514.3
2.3114.840.254.3
2.3513.938.955.0
2.4516.935.647.9
2.5522.133.938.6
from2.3to2.6p.misalsoamanifestationof
theincreasingbitumencontent.Theslightly
positiveslopeinthisregionisnotcharacteristic
ofclays,whichshowanegativeslope.
Thissuggeststhatbitumenisoverridingthe
spectralsignatureoftheclaysinthisregion.
Spectraofsampleswithhighbitumen
contentaredominatedbybitumen(Fig.3),
thesecondmostabundantphaseafterquartz
inthesesamples(Table3).Allthemajor
absorptionbandscanbeattributedtohydrocarbons.
Theexpectedclaywaterbands
at1.4and1.9p.marevirtuallyabsent.The
1.7p.mbitumenabsorptionregionisvery
prominent,anditsnonsymmetricalshapeis
consistentwithmultiple,overlappingCH
absorptionbands.Theabsorptionisstrong
intheregionfrom2.3to2.6p.mandhasa
positiveslope,neitherofwhichischaracteristic
ofclays.Theabsorptionbandat2.3p.m
isveryintense,andthepositiveslopeinthe
regionfrom2.3to2.6p.mistheresultof
thedominanceofthisfeature.
Eachspectralgroup,dividedonthebasis
ofbitumencontent,exhibitsuniquespectral
properties.Thesevariationscanberelatedto
thephysicalandchemicalpropertiesofthe
samplesthathavebeenindependentlydetermined.
Bitumenandclayabundancesare
generallyinverselycorrelated.Thereisno
simplecorrelationbetweenspectralalbedo
andanyofthemajorphaseabundances.The
sameholdstruefortheabundancesof
quartzversusclay.Bitumenandquartz
abundancesarepositivelycorrelated.Spectra
correspondingtosampleswithhighand
lowbitumencontentsaredominatedby
hydrocarbonandclayabsorptionfeatures,
respectively.
Themostprominentabsorptionbandattributable
tothevariousformsofwateris
presentnear1.9p.mandismostintensein
themostwaterandclayrichsamples(Table
4).Thiswavelengthregionisusefulforclaywater
determinationbecause,unlikethere
0.20
=
00..115_/
;0.10
0.05
0.51.01.52.02.5
Wavelength(gm)
Fig.2.Reflectancespectrum(0.35to2.6,um)of
amediumbitumentarsand(sample8704).
gionfrom2.3to2.6,um,itisnotsignificantly
overlappedbyhydrocarbonabsorptions.
Thebroadnessandcomplexshapeof
the1.9,umbandisconsistentwiththe
presenceofwaterinanumberofsitesinthe
clay,structurallydisorderedclays,ormore
thanonespeciesofclay.Allthreeinterpretations
areborneoutbyindependentstudies
oftheAthabascaclays(2,3).Theremaybea
moderatecontributionbyfreewaterat1.94
,um,butitspresencecannotbeunambiguously
resolved.
Theabsorptionbandnear1.4,umismost
prevalentinthelowbitumencontentspectra
andisassignedtostructuralandbound
waterintheclays.Theavailablespectral
informationonvariousformsofwater
showsthatfreewaterandpolymerizedwater
absorbatdifferentwavelengths(>1.45p.m)
thanwaterinclays(1.38to1.42,um).The
broadnessofthisbandisinterpretedinthe
samewayasthatofthe1.9p.mband(24).
ClaylatticeOHabsorptionbandsare
expectedintheregionfrom2.2to2.6p.m.
Thisbandisunexpectedlyweakestinthe
spectrumofthemostclayrichsample(Fig.
1).Thehighlyabsorbingnatureofbitumen
inthisregion,anditslowabundanceinthe
clayrichsamples,explainsthisdiscrepant
behavior.Clayspectranormallyexhibithigh
overallreflectance(8,9).Thelowbitumen
spectrum(Fig.1)contains28%claysized
particlesandshowsonlyclayabsorption
bandsbuthasaverylowoverallreflectance.
Thelowreflectancemaybecausedbythe
presenceofthebitumen,ironoxidecontaminants,
ortransitionserieselementsubstitutions
(3).Areflectancespectrumofatransition
elementbearingclay,glauconite(Fig.
4),showsanumberofsimilaritiestothe
clayrichspectrumofFig.1:aconcave
reflectancerisebetween0.9and1.5p.m,a
weakorabsent1.4p.mabsorptionband,
andareflectancedecreasebeyond2.3p.m.
Theoverallreflectanceoftheglauconiteis
stillsignificantlyhigherthanthatofthe
mostclayrichsample(Fig.1).Eventhough
transitionseriesbearingclaysmaybepresent,
glauconiteitselfhasnotbeenidentified
inAthabascatarsands(2,3).Thespectrum
ofthelowbitumensampleisconsistent
SCIENCE,VOL.245
0.10h
co
40.05
0.51.01.52.02.5
Wavelength(pm)
Fig.3.Reflectancespectrum(0.35to2.6,um)of
ahighbitumentarsand(sample8619).
withacationsubstitutedclaybutrequires
thepresenceofaphasewithlowoverall
reflectance.Bitumenseemsthelikeliestcandidate
forthedarkeningmaterial.
Thehighbitumenspectrumcorresponds
toasamplewith5%claysizedparticlesand
hasavirtuallyundetectable1.9p.mclaywater
band.Thusalowerlimitof5%can
beplacedonspectraldeterminationofclays
inorganicrichmaterialsofthistype.Bitumen
servesasaneffectivesuppressorofwhat
arenormally,moderatelyintenseabsorption
bands.
Theregionsat1.7and2.3to2.6p.mare
thebestareasforthedetectionofvariousCH
absorptionbands.The1.7p.mregionis
preferablebecauseitisnotoverlappedby
clayandwaterabsorptionbandsandisthe
mostrecognizablefeatureshortwardof2.1
p.m.Thecomplexshapeofthisbandis
consistentwiththepresenceofmultiple,
partiallyoverlappingabsorptionbands,although
noattempthasyetbeenmadeto
deconvolvethisfeatureintoitsconstituent
bands.Theuseofthisabsorptionfeature
placesalowerdetectionlimitof4%on
bitumenabundancesintarsands.
Thenearabsenceofa1.9p.mclaywater
bandinFig.3stronglysuggeststhatthis
spectrumisagoodrepresentativeofpolymerized
hydrocarbonsplusspectrallyneutral
material.Themoderatestrengthabsorption
bandsthatappearinallthesample
spectrabettween2.17and2.25p.mcanbest
beassignedtobothclaylatticeOHand
organicbands,becausethereisnosimple
correlationbetweentheirbanddepthsand
anyonephaseabundance.Theoverallslope
oftheregionfrom2.3to2.6p.missensitive
tobothbitumenandclaycontents.Clays
consistentlyexhibitanegativeslopeinthis
region[see(8,9)andFig.4],whereas
organicmaterialsshowapositiveslope(15,
16,22).Thechangeoverfromnegativeto
positiveslopeoccursinthemediumbitumen
group.
Changesinbitumencontentaffectthe
ultravioletandvisiblespectralregions.The
pointofminimumreflectanceshiftstolonger
wavelengthsasthebitumencontentincreases.
Thewavelengthpositionofthere
Table4.Phaseabundancesofthesamples(in
percentagesbyweight).
Abundances(%)<400>400
pieBituWaSolmeshmesh
menterids(%(%
82053.28.188.732.167.9
87048.16.685.329.370.7
861913.22.583.96.094.0
0.60
0.40
O
aC D
x0.20
0.oo0|I, ,
. ,I..I.I.....
0.300.600.901.201.501.802.102.40
Wavelength(pm)
Fig.4.Reflectancespectrum(0.35to2.6,um)of
glauconite,whichhasanumberofsimilaritiesto
theIapetusdarkmaterial.
flectanceminimuminFig.3is0.47p.m,
whichcorrespondstoanaverageofsix
condensedaromaticringspermoleculeifwe
usethecalibrationofBadger(25).Thisisin
goodagreementwiththeconventionalanalytical
resultsoffiveringspermoleculefor
theresinfraction,themostabundantorganic
group(3,20).Theshapeofthereflectance
spectrumbetween0.5and1.5p.m
changesfromtwolinearsegmentswitha
slopebreakat0.9p.mtoasmoother
concaveslopewithincreasingbitumenand
decreasingclaycontents.Theformershape
issimilartothatoftheglauconitespectrum
(Fig.4);thelattershapeisconsistentwitha
continuumofchargetransferenergylevels
expectedincomplexhydrocarbons.
TheintenseSoretband,whichischaracteristic
ofporphyrins,shouldbepresentat
0.4p.m(23).Thelowoverallreflectance
ofthesamplespectramakesidentificationof
thisfeatureverydifficult,anditsunambiguous
presencehasnotyetbeenestablished.
Hydrocarbonsaresuspectedtobepresent
onanumberofextraterrestrialbodies(5,6)
andhavebeenfoundincarbonaceouschondrites
(26).Iapetuswasselectedformore
detailedspectralanalysis.Onehemisphereis
almostentirelycoveredwithaverydark
substanceofunknowncomposition.Earlier
investigatorsusedamixingmodeltoremove
thespectralsignatureofwatericeandisolate
thedarkmaterial(6),whichshowsagradual
riseinreflectancetowardlongerwavelengths
withonlyminorabsorptionbands.
Iftheremovalprocedureforthewaterice
signaturedidnotinadvertentlydeleteadditional
features,thereisnoevidenceforany
oftheexpectedorganicabsorptionbands.In
3
2
0
0.51.01.52.02.5
Wavelength(pm)
Fig.5.Normalizedreflectancespectrumofthe
lowbitumentarsandsamplewiththetelescopic
spectraldatafortheIapetusdarkmaterialfrom
(6)overlain(filledcircles).
particular,theregionsat1.7and2.3to2.6
p,mshownoneofthecharacteristicsof
highlypolymerizedhydrocarbons.
Thebestmatchtothedarkmaterialwas
foundtobeamixtureof90%clayand10%
coaltarrepresentingorganicmatter(6).The
lowbitumensample(Fig.1)consistsofthe
sameratioofclaytoorganicsand,likeits
coaltarcounterpart,isfreeofdetectable
organicabsorptionbands.Theclaycoaltar
spectrumprovidesagoodmatchtothe
overallchangeinreflectancewithwavelength
oftheIapetusspectrum,althoughthe
fittotheregionfrom0.8to1.4p.misnot
verygood.Thelowbitumenspectrum(normalized
tomatchtheIapetusdarkmaterial)
providesamuchbetterfittotheregionfrom
0.8to1.4,umandalsoshowsachangein
slopeat0.9p.m(Fig.5).Beyond1.4,um
thereflectanceofthetarsandisconsistently
higherthanthatoftheIapetusdarkmaterial;
theclayabsorptionbandsat1.9p.mand
2.2to2.5p.marealsomuchmoreintensein
thetarsandspectrum.
Neithertheclaycoaltarmixturenorthe
tarsandareperfectspectralmatchestothe
Iapetusdarkmaterial.Thereflectancespectrum
ofglauconite(Fig.4)exhibitsanumber
ofdesirablespectralfeaturesformatching
withtheIapetusdarkmaterial.The
differencebetweentheminimumandmaximum
reflectanceiscomparabletothatfound
forthedarkmaterial,theslopebreakat
0.9p.mispresent,andaweak1.9p.m
claywaterbandispresent.Mostotherclays
donotshowthesamereflectancedifference
betweenminimumandmaximumreflectance
(8),becausetheycontainonlyminor
amountsoftransitionserieselements.A
darkmaterialmustbeaddedtothisspectrum
tofurthersuppressthe1.9p.mband
andneutralizethereflectancedropoffbeyond
2.2p.m.
Nooneofthesamplespectraisaperfect
matchtotheIapetusdarkmaterial.Because
eachmaterialexaminedhascertaindesirable
spectralcharacteristics,acomplex,intimate
mixtureofdifferentmaterialsmayberequired.
Anironsubstitutedclayseemstobe
YREPORTSI67
.1.
14.JULY1989
anecessaryconstituent(5).Thepresenceof
someamountofahighlypolymerizedhydrocarbon,
likethebitumenpresentintar
sands,maybeneededtomodifytheclay
spectrumintheregionfrom0.8to1.4,um.
Afewpercentofbitumencanbepresentand
notexhibitdistinctabsorptionbands.Finally,
amaterialspectrallysimilartothecoaltar
extractisneeded,whichcaneffectivelymask
thenegativeslopeoftheclaysbetween2.2
and2.6,um,reducethereflectanceriseat
longerwavelengths,andsuppressclayorganic
bandsat2.2to2.6,um.
REFERENCESANDNOTES
1.Thetarsandsamplesusedinthisstudyarefromthe
AthabascadepositinnortheasternAlbertaandconsist
ofviscousorganicmatterembeddedinclastic
sedimentsoftheCretaceousMcMurrayandClearwater
formations[B.NagyandG.C.Gagnon,
Geochim.Cosmochim.Acta23,155(1961)].They
wereobtainedfromtheAlbertaOilSandsTechnology
andResearchAuthority[(AOSTRA),Edmonton,
Alberta]OilSandsSampleBank,alongwith
someanalyticaldata.Thereflectancespectrawere
acquiredattheU.S.GeologicalSurveyspectrometer
facilityinDenver,CO,andtheNationalAeronautics
andSpaceAdministrationReflectanceExperiment
Laboratory(RELAB)spectrometerfacilityat
BrownUniversityinProvidence,RI.Comprehensive
descriptionsoftheinstrumentsanddataprocessing
proceduresareavailableinthefollowing
references:R.N.Clark,Publ.Astron.Soc.Pac.92,
221(1980);C.M.Pieters,J.Geophys.Res.88,
9534(1983);T.V.V.KingandW.I.Ridley,ibid.
92,11457(1987).Thehomogeneityofthetarsand
samplescouldonlybeassessedvisually.Thesamples
weregentlychoppedandmixeduntilavisually
uniformmixtureresulted.Theywerethenplacedin
sampleholdersandlightlycompressedtoprovidea
flatsurfaceforspectralmeasurement.Allsamples
weremeasuredrelativetoHalon.Thetarsand
spectrawereacquiredwiththeuseofanintegrating
spherearrangementattheU.S.GeologicalSurvey
facility.Theglauconitewasspectrallycharacterized
inabidirectionalreflectancemodeattheNASA
RELABfacility,usinganincidenceangleof0and
anemissionangleof150.Aduplicatetarsand
spectrum,acquiredtocheckforreproducibility,was
withintheerrorlimitsofthedata.Theerrorbarsin
thespectraldatarepresent1SDofthemean.Band
depth(Db),usedextensivelyinspectralanalysis,is
heredefinedas
Db=1Rb/Rc
whereRcisthereflectanceofastraightlinecontinuum
atthewavelengthpositionofareflectance
minimumandRbistheminimumreflectanceatthe
samewavelength.Thecontinuumwasconstructed
asastraightlinetangenttothespectrumoneither
sideoftheabsorptionfeatureofinterest.Thecontinuum
fortheregionfrom2.3to2.6,umwastaken
asahorizontallinetangenttothereflectancemaximum
near2.2,um.
2.E.CzarneckaandJ.E.Gillot,ClaysClayMiner.28,
197(1980);T.M.Ignasiaketal.,Fuel62,353
(1983).
3.J.A.Bichard,OilSandCompositionandBehaviour
Research(AOSTRATechnicalPublicationSeries4,
Edmonton,Alberta,Canada,1987).
4.M.L.Seluckyetal.,Fuel56,369(1977);J.W.
Bungeretal.,ibid.58,183(1979).
5.J.GradieandJ.Veverka,Nature283,840(1980);
C.ChybaandC.Sagan,ibid.330,350(1987);E.
K.Jessberger,A.Christoforidis,J.Kissel,ibid.332,
691(1988);D.P.CruikshankandR.H.Brown,
Science238,183(1987).
6.D.P.Cruikshanketal.,Icarus53,90(1983);J.F.
Belletal.,ibid.61,192(1985).
7.J.M.Hunt,PetroleumGeochemistryandGeology
(Freeman,SanFrancisco,1979).
i68
8.G.R.HuntandJ.W.Salisbury,Mod.Geol.1,283
(1970);,C.J.Lenhoff,ibid.4,85(1973);G.
R.Hunt,Geophysics42,501(1977);R.N.Clark,
J.Geophys.Res.86,3074(1981);T.V.V.King,
thesis,UniversityofHawaii(1986).
9.J.D.LindbergandD.G.Snyder,Am.Mineral.57,
485(1972);S.C.Feldman,J.V.Taranik,D.A.
Mouat,inProceedingsoftheAirborneImagingSpectrometer
DataAnalysisWorkshop,G.VaneandA.F.H.
Goetz,Eds.(JetPropulsionLaboratory,Pasadena,
CA,1985),p.56;G.VaneandA.F.H.Goetz,
RemoteSensingEnviron.24,1(1988);S.A.Drury
andG.A.Hunt,Photogramm.Eng.RemoteSensing
54,1717(1988);C.M.PietersandJ.F.Mustard,
RemoteSensingEnviron.24,151(1988).
10.J.E.Zajic,D.G.Cooper,J.A.Marshall,D.F.
Gerson,Fuel60,619(1981);K.BuijsandG.R.
Choppin,J.Chem.Phys.39,2035(1963);H.
Yamatera,B.Fitzpatrick,G.Gordon,J.Mol.Spectrosc.
14,268(1964);M.R.Thomasetal.,J.Phys.
Chem.69,3722(1965);K.B.Whetsel,Appl.
Spectrosc.Rev.2,1(1968);J.D.WorleyandI.M.
Klotz,J.Chem.Phys.45,2868(1966).
11.J.G.Bayly,V.B.Kartha,W.H.Stevens,Infrared
Phys.3,211(1963);R.GoldsteinandS.S.Penner,
J.Quant.Spectrosc.Radiant.Transfer4,441(1964).
12.M.L.BoydandD.S.Montgomery,J.Inst.Pet.
London49,345(1963).
13.B.E.Hudson,Jr.,R.F.Robey,J.F.Nelson,in
ProceedingsoftheFifthWorldPetroleumCongress(Fifth
WorldPetroleumCongress,Inc.,NewYork,1959),
vol.V,p.1;R.M.SilversteinandG.C.Bassler,
SpectrometricIdentificationofOrganicCompounds(Wiley,
NewYork,1967).
14.L.A.Gilbert,Fuel39,393(1960).
15.0.Ito,H.Seki,M.lino,ibid.67,573(1988).
16.W.Kaye,Spectrochim.Acta6,257(1954);R.F.
Goddu,Anal.Chem.29,1790(1957);S.A.Fyshet
al.,Appl.Spectrosc.39,354(1985);N.A.McAskill,
ibid.41,313(1987).
17.T.Yokotaetal.,Fuel65,1142(1986);E.A.
Kmetko,Phys.Rev.82,456(1951).
18.G.Svehla,ComprehensiveAnalyticalChemistry,vol.
VI,AnalyticalInfraredSpectroscopy(Elsevier,Amsterdam,
1976);P.C.Painteretal.,Appl.Spectrosc.35,
475(1981);M.P.Fulleretal.,Fuel61,529(1982);
P.C.Painteretal.,ibid.66,973(1987);P.W.Yang
etal.,EnergyFuels2,26(1988).
19.R.J.Moore,R.J.Gordon,R.C.Eiffert,in
ProceedingsoftheFifthWorldPetroleumCongress(Fifth
WorldPetroleumCongress,Inc.,NewYork,1959),
vol.V,p.13.
20.M.L.BoydandD.S.Montgomery,CanadianDep.
MinesTech.Surv.R78(1961);CanadianDep.Mines
Tech.Surv.R88(1961);CanadianDep.MinesTech.
Surv.104(1962).
21.B.NagyandG.C.Gagnon,Geochim.Cosmochim.
Acta23,155(1961).
22.J.C.DoniniandK.H.Michaelian,InfraredPhys.26,
135(1986);E.K.PlylerandW.S.Benedict,J.Res.
Natl.Bur.Stand.47,202(1951).
23.J.B.F.ChamplinandH.N.Dunning,Econ.Geol.
55,797(1960);G.W.HodgsonandB.L.Baker,
Chem.Geol.2,187(1967);E.W.FunkandE.
Gomez,Anal.Chem.49,972(1977);P.N.Holden
andM.J.Gaffey,Meteoritics22,412(1987).
24.R.B.SingerandT.L.Roush,LunarPlanet.Sci.
Conf14,708(1983).
25.G.M.Badger,StructuresandReactionsoftheAromatic
Compounds(CambridgeUniv.Press,London,
1957).
26.B.Nagy,CarbonaceousMeteorites(Elsevier,Amsterdam,
1975).
27.Supportedbyascholarshipandresearchgrantfrom
AOSTRAandanAmericanAssociationofPetroleum
Geologistsgrantinaid(5821201).Ithank
D.WallaceandD.HenryoftheAOSTRAOil
SandsSampleBankforprovidingthetarsand
samples.IthankR.ClarkandG.SwayzeoftheU.S.
GeologicalSurveyinDenver,C.PietersandS.Pratt
atBrownUniversityforprovidingaccesstotheir
respectivespectrometerfacilities,andM.Gaffeyfor
providingthespectrumanalysisprogramsandencouragement
throughout.Iamalsogratefultotwo
anonymousreviewersformanyusefulsuggestions
andcorrections.
30December1988;accepted19April1989
TemperatureMeasurementsinCarbonatiteLava
LakesandFlowsfromOldoinyoLengai,
Tanzania
MAURICEKRAFFTANDJORGKELLER
Thepetrogenesisofcarbonatiteshasimportantimplicationsfor
mantleprocessesand
forthemagmaticevolutionofmantlemeltsrichincarbon
dioxide.OldoinyoLengai,
Tanzania,istheonlyactivecarbonatitevolcanoonEarth.Its
highlyalkalic,sodiumrich
lava,althoughdifferentincompositionfromthemorecommon
calciumrich
carbonatites,providestheopportunityforobservationsofthe
physicalcharacteristics
ofcarbonatitemelts.Temperaturemeasurementsonactive
carbonatiticlavaflowsand
fromcarbonatiticlavalakeswerecarriedoutduringaperiod
ofeffusiveactivityinJune
1988.Temperaturesrangedfrom491to519C.Thehighest
temperature,measured
fromacarbonatiticlavalake,was544C.Thesetemperaturesare
severalhundred
degreeslowerthanmeasurementsfromanysilicatelava.Atthe
observedtemperatures,
thecarbonatitemelthadlowerviscositiesthanthemostfluid
basalticlavas.The
unusuallylowmagmatictemperatureswereconfirmedwith1
atmospheremelting
experimentsonnaturalsamples.
ARBONATITESAREIGNEOUSROCKSOldoinyoLengai,Tanzania,theonly
richinprimarycarbonates,domiactivecarbonatitevolcanoon
Earth,isfanantly
calciteanddolomite.Almousforitsunusualalkalirichmagma,
thoughrarecomparedwithsilicaterocks,termednatrocarbonatite.
Thevolcano,a
carbonatitesprovideimportantconstraints
onpartialmeltingprocesses,volatileconM.Krafft,Centre
Vulcain,F6700Cernay,France.
tent,andchemicalcompositionintheJU.nivKeerllseirt,at,MiDn
e7ra8l0o0giFsrcehi
bPuertgr,ogFreadpehrailscRheepsubIlnistcitouftGedrer
earth'smantle.many.
SCIENCE,VOL.245
Accepted Version
www.reading.ac.uk/centaur
CentAUR
Central Archive at the University of Reading
Readings research outputs online
1
A change detection approach to flood mapping in urban areas using TerraSARX
Laura Giustarini, Renaud Hostache, Patrick Matgen, Guy Schumann, Paul D. Bates,
and David C. Mason
L. Giustarini, R. Hostache and P. Matgen are with the Centre de Recherche Public Gabriel Lippmann, Dpartement Environnement et Agro-biotechnologies, L-4422
Belvaux, Grand-Duchy of Luxembourg (email: giustari@lippmann.lu;
hostache@lippmann.lu; matgen@lippmann.lu)
G. Schumann and P. D. Bates are with the School of Geographical Sciences,
University of Bristol, BS8 1SS Bristol, U.K. (email: guy.schumann@bristol.ac.uk;
paul.bates@bristol.ac.uk)
D. C. Mason is with the Environmental System Science Centre, University of
Reading, RG6 6AL Reading, U.K. (email: dcm@mail.nerc-essc.ac.uk)
ABSTRACT
Very high-resolution Synthetic Aperture Radar sensors represent an alternative to
aerial photography for delineating floods in built-up environments where flood risk is
highest. However, even with currently available SAR image resolutions of 3 m and
higher, signal returns from man-made structures hamper the accurate mapping of
flooded areas. Enhanced image processing algorithms and a better exploitation of
image archives are required to facilitate the use of microwave remote sensing data for
monitoring flood dynamics in urban areas. In this study a hybrid methodology
combining radiometric thresholding, region growing and change detection is
2
introduced as an approach enabling the automated, objective and reliable flood extent
extraction from very high-resolution urban SAR images. The method is based on the
calibration of a statistical distribution of open water backscatter values
inferred
from SAR images of floods. SAR images acquired during dry conditions enable the
identification of areas i) that are not visible to the sensor (i.e. regions
affected by
layover and shadow) and ii) that systematically behave as specular
reflectors (e.g.
smooth tarmac, permanent water bodies). Change detection with respect to a pre- or
post flood reference image thereby reduces over-detection of inundated areas. A case
study of the July 2007 Severn River flood (UK) observed by the very high-resolution
SAR sensor on board TerraSAR-X as well as airborne photography highlights
advantages and limitations of the proposed method. We conclude that even though the
fully automated SAR-based flood mapping technique overcomes some limitations of
previous methods, further technological and methodological improvements are
necessary for SAR-based flood detection in urban areas to match the flood mapping
capability of high quality aerial photography.
I. INTRODUCTION
The support of remote sensing for mapping changes in water surface extents and
elevations has been demonstrated widely (for detailed reviews see [1]-[3]). The
success of these research studies together with recent public and political awareness
for quantifying global environmental change has led to a significant increase in the
number of satellites dedicated to flood monitoring and hydrology in the wider sense.
Importantly, flood monitoring from space has the advantage of large area coverage
and relatively fast response services (see for example the International
Charter Space
3
and Major Disasters initiated by major space agencies:
http://www.disasterscharter.org/).
The vast majority of a flooded area is rural rather than urban and accordingly most
literature on remote sensing based flood detection to date has focussed on the rural
case. However, it is perhaps more important to detect the urban flooding because of
the increased risks and costs associated with it. Flood extent can be detected in rural
floods using SARs such as ERS and ASAR, but these have too low a resolution (25
m) to detect flooded streets in urban areas. However, a number of SARs with spatial
resolutions as fine as 3 m or better have recently been launched and are potentially
capable of detecting urban flooding. They include TerraSAR-X, RADARSAT-2, and
the four COSMO-SkyMed satellites.
In an operational context, reference [4] have proposed a hybrid methodology for
SAR imagery, which combines radiometric thresholding and region growing as an
approach enabling the automatic, objective and reliable flood extent extraction from
SAR images. First results on moderate and low resolution image data indicate that the
proposed method may outperform manual approaches if no training data are available
even if the parameters associated with these methods are determined in a non-optimal
way. The results demonstrate the algorithms potential for accurately
processing data
from different SAR sensors.
Notable examples of research into automatic near real-time flood detection
algorithms using single-polarisation high resolution (greater than a few metres) SAR
imagery have been shown by references [5]-[6] on TerraSAR-X data and reference [7]
on COSMO-SkyMed data. The algorithms by references [5]-[6] search for water as
regions of low SAR backscatter using a region-growing iterated segmentation /
classification approach whereas the technique by reference [7], is based on a fuzzy
4
logic approach which integrates theoretical knowledge about the radar return from
inundated areas based on backscattering models, with simple hydraulic considerations
and contextual information. Both algorithms are very effective at detecting rural
floods, but would require substantial modification to work in urban areas containing
6
introduce an enhanced version of M2a, which we term M2b. This method addresses
some of the shortcomings of M2a that reference [4] identified in two representative
case studies. This section provides a detailed overview of all processing steps of the
flood extraction algorithm M2b, together with the associated parameters defining each
process and a list of differences with respect to the M1 and M2a algorithms
previously introduced. Besides standard pre-processing steps commonly involved
with Level 1 SAR data, the M2b algorithm consists of four processing steps (Fig. 1).
A. Statistical distribution of the open water backscatter
The flood extraction algorithm uses as input Level 1 SAR data that are geo-coded,
co-registered and calibrated. The first step is the estimation of the probability density
function (PDF) of backscattering values associated with open water. The
aim of this
processing step is the calibration of a theoretical PDF that optimally fits the empirical
distribution of backscatter values from open water inferred from the SAR
image.
According to reference [13] the backscatter variability on a homogeneous surface is
mainly due to speckle and the theoretical PDF that best describes the distribution of
backscatter originating from a homogeneous surface is of gamma type. Here, we
hypothesize open water to be a homogeneous surface, which means that a
potential
limitation of the approach and SAR mapping of inundated surfaces in general relates
to the possible roughening of open water caused by emerging vegetation, wind or
rainfall. We therefore assume that the PDF of type gamma can be used to estimate the
distribution function of open water backscatter:
7
1
0
0
1
0
0
1
0
01
1
0
0
0
1
1
/
k
e
k
fk=m
k
m
k
m
(1)
where k is the shape parameter of the gamma distribution and 0
m is the gamma
distribution mode. The parameter 0
1 is the minimum backscatter value in the SAR
image, which needs to be applied so that the gamma distribution is calculated only for
positive values.
Two parameters thus need to be optimized to identify the theoretical gamma
function f that best fits the empirical distribution of backscatter values from
open
water h (i.e. image histogram). The optimization of the two parameters k and 0
m
consists in minimizing the root mean squared error (RMSE) between the image
histogram and the gamma distribution for backscatter values lower than 0
thr , with the
parameter 0 0
thr m representing the point where the distributions f and h start
deviating. The optimization is performed with sequentially increasing values of 0
m
and 0
thr values. For each set of 0
m and 0
thr values, the parameter k is optimized
using the nonlinear fitting process of reference [14]. The RMSE between the
theoretical density function f and the empirical density distribution h is calculated for
each parameter set and over all backscatter values lower than 0
thr . Finally, the
parameter set ( 0
m , k, 0
thr ) providing the lowest RMSE is set as optimal.
B. Radiometric thresholding
The aim of the following step of the algorithm is to extract seeds of open
water
areas from the flood image, being either individual pixels or regions. The parameter
8
0
thr
represents the maximum backscatter value for which the fit between the
theoretical and empirical PDF is satisfactory. For backscattering values higher than
0
thr
the distribution functions f and h start deviating, which leads to an immediate and
together with the change detection parameter introduced in the next section.
Region growing, with the same threshold value 0
rg is also applied to dilate the
seeds of permanent smooth surfaces obtained from the reference image. The approach
provides a mask of water surface-like radar response areas that is used to limit the
rg
region growing applied on the flood image, thereby preventing the spreading of
flooded areas into permanent smooth areas.
D. Change detection
Reference [4] argued that flood maps resulting from region growing should
include all open water pixels connected to the seeds. The region growing should
thus extend into the high percentiles of the gamma distribution. However, the
resulting over-detection needs to be removed by the subsequently applied change
detection step. Change detection thus aims at removing pixels from the flood extent
map that do not correspond to flood water. To do so, only pixels that significantly
change their backscatter values with respect to their baseline backscatter values are
kept in the flood extent map, while pixels that did not decrease their backscatter
values by a minimum amount are removed. This means that the main river channel,
10
which is a permanent water body, is not any longer an integral part of the flooded
area.
The specific parameter of the change detection is 0, defined as the required
minimum change in backscatter between the reference and the flood image for a pixel
being considered as flooded. In order to determine the optimal criterion for the
required minimum change in backscatter, an iterative procedure is adopted.
As mentioned earlier, the two parameters 0
rg and 0 are optimized through a
simultaneous calibration, minimizing the RMSE computed over the whole range of
backscatter values in the flood image between the theoretical gamma distribution and
the empirical distribution of open water pixels. This means that different
threshold
values, 0
rg , which correspond to different percentiles of the theoretical gamma
distribution, are sequentially selected from an interval of plausible values, and a
corresponding minimum change detection parameter 0 is optimized for each tested
0
backscattering values associated with open water. The red box area in Fig.
2
presents the area of interest for the city of Tewkesbury: it refers to a rectangular area
of 1135x998 pixels (1.5 m pixel spacing) for a total surface of ~3 km2.
B. Validation dataset
The validation data set, consisting of very high-resolution 0.2 m aerial
photographs acquired during the flooding event in July 2007, enables a
comprehensive evaluation of the algorithms performance in terms of SARbased
15
flood delineation. An aircraft operated by the Environment Agency of England and
Wales (EA) carried out the overflights.
The flood extent was obtained through manual photo-interpretation (Fig. 3a).
Taking advantage of existing landuse maps of the area, permanent water bodies
associated with rivers and canals have been removed from the validation map. While
in general the delineation of flood boundaries from such high-resolution optical
products is relatively straightforward, it is important to note that the flooding of
densely vegetated and built-up environments can lead to some ambiguities. For
instance, in the case of bare soil fields, the accurate positioning of the separation line
between muddy flood waters and non-flooded areas is non-trivial.
In addition, it is important to bear in mind that the aerial photographs were
acquired on July 24 (at 11:30 GMT) while the TerraSAR-X image was obtained 19
hours later on July 25 (at 06:34 GMT). Although there was no significant decrease in
main river discharge between the acquisition time of aerial photographs and the
TerraSAR-X overpass [20], this time gap might be responsible for some discrepancies
between the SAR-derived and aerial photography-derived flooded areas. To illustrate
potential discrepancies, simulations with a previously calibrated hydraulic 2m
LISFLOOD-FP flood model [21], have been carried out both at the aerial photographs
acquisition time and at the TerraSAR-X overpass. These simulations show a reduction
of the flooded area of approximately 5% between the two time steps. In particular,
Fig. 3(b) shows the differences between the two simulated flood inundation maps The
most notable differences can be observed on a triangular-shaped field (see the middleright
part of the domain of interest) from which flood water was drained between the
two overpasses according to the model simulations. This location was also
problematic in terms of identifying its flooding status through photo-interpretation, as
16
explained in more details in the discussion section. These factors, all unrelated to the
processing of the SAR images, need to be taken into account during the analysis, as
all the observed differences may not be necessarily due to the inability of the
proposed algorithms for accurately extracting the flood extent from SAR imagery.
IV. RESULTS AND DISCUSSION
This section assesses the accuracy of the classification obtained with the fully
automated flood detection algorithm M2b and contrasts its performance with those of
the previously introduced M1 and M2a algorithms. The section also provides insights
into the added-value of reference images for flood delineation in urban areas.
A. Extraction of flooded areas
The flood extent has been extracted from the TerraSAR-X image using the three
methods M1, M2a and M2b.
In particular, for method M2b, Fig. 4 illustrates the optimization of the four
parameters: the mode of the open water backscatter gamma PDF, 0
m , the
radiometric threshold, 0
thr , the tolerance criterion for the region growing step, 0
rg ,
and the minimum CD value, 0. Panel (a) reports the optimization of the mode
parameter while panel (d) provides the corresponding optimized gamma PDF in red
together with the histogram of the backscatter values in the flood image. Panel (d)
displays the value of the second parameter, 0
thr (i.e. in this case study equal to 15.5
dB) as the maximum backscatter value for which there was no overlap between the
empirical histogram and the theoretical gamma PDF. The optimized 0
thr value is also
provided in Fig. 5(a), together with the backscatter histogram of pixels in the
17
reference image. This value is used to derive the reference mask of permanent water
surface-like radar response areas through thresholding of the reference image (Fig.
5(b)). From Fig. 4(d) and Fig. 5(a) it can be observed that for high backscatter values,
there is a systematic noise in the return signal of both TerraSAR-X images. This is
arguably due to the high complexity of urban topography and its considerable impacts
on the high-resolution backscattering signal. Considering the region growing step of
method M2b, it is important to mention that the parameters 0
rg and 0 are
optimized together. Therefore, the subplot in panel (b) of Fig. 4 illustrates the impact
that different 0
rg parameters (each associated with the corresponding optimal 0
value) have on the RMSE. Similarly, panel (c) provides an example of the
performance plot of parameter 0 for a given 0
rg value: Fig. 4(c) refers to the
optimal 0
rg value of the case study. The backscatter value corresponding to the
optimized 0
rg value is also displayed in panels (e) and (f). Finally, panel (f) shows the
empirical histogram of flood pixel values before and after change detection. The
histograms in panel (f) are computed only from pixels inside the SAR-derived flooded
area. The reduction of the distribution tail and the related reduction of over-detection
are indicated by the empirical histogram approaching the theoretical gamma PDF.
B. Evaluation at city level (quantitative analysis)
Three flood extent maps were obtained through the application of the three image
processing algorithms. The corresponding contingency matrices were computed using
the evidence provided by aerial photography. The binary pattern of flooded and nonflooded
pixels was compared against the reference flood map (in this case, see Fig. 3).
18
The result is a matrix (or contingency table) of four possible outcomes. With respect
to the reference flooded area, there are two ways for a remote sensing-derived flooded
area to be correct (either by correctly representing flooded or non-flooded pixels) and
two ways to be incorrect (either by erroneously under- or over-predicting the observed
inundation extent). The values of the contingency matrix for all methods are reported
in Table II (and also displayed as contingency maps in Fig. 6) for a quantitative
evaluation of the performances. Moreover, the optimized (and/or fixed) parameter
values for the region growing and the change detection are indicated.
From Table II it can be concluded that in the present case study the three
algorithms provide very similar performance levels. When the evaluation is carried
out at a regional scale (i.e. at city level), the differences seem to be marginal.
However, methods M2a and M2b still slightly outperform method M1 with respect to
all evaluation criteria listed in Table II, suggesting that change detection with respect
to a non-flood reference image does provide some advantages. The results do not
reflect the added-value that we expected from the methodological improvements of
method M2b. This result is due to the fact that, in this particular case study, the
optimized region growing threshold 0
rg equals 98%, which is very close to the predefined
99% value that reference [4] proposed for M2a. Here, the trade-off involves a
controlled growing of the seed region in order to be able to limit the over-detection of
flooded areas. While the latter can be partly removed by the subsequent change
detection, the results indicate that the reduced over-detection comes at the cost of an
increased under-detection of flooded areas. The results also indicate that method M2b,
which provides an optimal empirical distribution with respect to the targeted gamma
distribution of open water backscatter values, does not necessarily
generate a more
accurate flood inundation map than M2a. On a more positive note, it can be observed
19
that the simultaneous optimization of region growing threshold 0
rg and change
detection parameter 0, computed by minimizing the RMSE between empirical and
theoretical distribution functions, led to the maximum value of correctly detected
pixels (81.7% as reported in Table II). Moreover, from Fig. 7 it can be observed that
in this case study the optimum parameter set also yields the best performance with
respect to the validation data.
A comparison with the flood extent detected for the same test case with the semiautomatic
procedure of reference [8] cannot be carried out in a very meaningful way
due to the fact that the input data sets in both studies differ. Reference [8] took
advantage of a regional DEM so that SAR-derived water ground heights smoothly
vary along the river reach. Furthermore, reference [8] did not make use of a pre-flood
reference image. The comparison of contingency matrices reveals that the M2b
method performance of 81.7% of correctly detected pixels is rather close to the
percentage of 85.4% obtained with the flood inundation map provided by reference
[8] on a common area of interest and reference dataset. This result indicates that
topography data could be used more efficiently than pre-flood reference images for
increasing the accuracy of SAR-derived flooded areas. However, this assessment
shadow region, this result should be viewed as an error as the right answer is obtained
23
for the wrong reason. As the objective of the developed method was to generate a
mask of surfaces that produce a radar signal response similar to that of inundated
areas in order to constrain the flood extent outside the shadow areas, in the following
we focus only on the overlap between the obtained flood extent and the shadow areas
derived by reference [8]. Note that the shadow mask itself, obtained with the SAR
simulator and the LiDAR data, might contain some degree of uncertainty. However,
we noticed that the number of such pixels was not significant in comparison to the
total number of extracted flood pixels (see Table III). In general, the overlap between
the SAR-derived flood extent and the shadow mask is restricted to the border regions
of large clusters of pixels, which were correctly classified as flooded.
Furthermore,
it can be observed from the results in Table III that M2b helps in significantly
reducing the number of pixels classified as flood water in the shadow regions.
This
is due to the fact that parts of the shadow-affected areas are included in the mask of
permanent water surface-like radar response areas described earlier. By considering a
reference image acquired from the same orbital track as the target image, the method
termed M2b reduces the risk of classifying shadow areas as flooded.
C. Evaluation at street level (qualitative/thematic analysis)
The benefits of using a reference image (including the masking of permanent
smooth areas) become obvious when looking at the spatial distribution of errors (Fig.
6). The application of algorithms M2a and M2b leads to the expected reduction of
misclassified pixels in urban areas. Numerous scattered clusters of pixels that were
initially erroneously classified as flooded could be removed, thereby
significantly
reducing over-detection. In Table IV a thematic analysis with a special focus on urban
features complements the quantitative analysis presented earlier (Table II).
24
The objective is to understand the advantages and limitations of the three variants
of the SAR-based flood delineation algorithms for correctly identifying flooding in
urban areas. From the results depicted in Table IV, it can be observed that in spite of
the high-resolution SAR imagery used in this study, the detection of flooding in builtup
environments remains a very challenging task. All algorithms struggle to recognize
the flooding status of many small-scale features that might be crucial as their state of
flooding could mean significant interruptions of everyday life. However, overall, the
enhanced algorithm M2b performs best with a slightly reduced number of
misclassified areas. In particular, M2b enables the a priori delineation of areas
characterized by specular-like reflections (i.e. areas with permanent water surface-like
radar responses). This is helpful given that smooth areas (e.g. R2 & R3) tend to be
systematically classified as flooded by M1 and, to a lesser extent, by M2a. On the
other hand, more open areas, such as the main roads R12 and R13, are correctly
classified by all three methods. It is worth mentioning that, despite these somewhat
encouraging results, M2b fails to correctly delineate flooding in many densely
vegetated and built-up environments.
These errors will be analyzed in more detail in the following sections. In this
analysis we will consider ancillary data (e.g. land use map, oral communications from
local experts) to better understand the reasons that are at the origin of the remaining
misclassifications. Moreover, the mask of the region unseen by the satellite, i.e.
shadow and layover, has also been taken into account for error detection at street
level.
25
1) Analysis of errors: the problem of over detection
As it can be seen from the urban flood maps presented in Fig. 6 both over- and
under-detection are reduced as a result of applying the M2b algorithm rather than its
predecessors. This is particularly evident in the case of the large shopping mall
labelled R3. Due to the flatness of its roof and resulting specular reflection it was
erroneously classified as flooded by M1, while M2a and M2b correctly excluded it
from the flooded area. This emblematic example best illustrates the potential added
value of reference images as they enable the a priori identification of the majority of
smooth areas.
Similarly, other wide flat regions, such as parking lots and airfields, are
recognizable in the reference image. For instance, the region labelled R1 corresponds
to large parking lot composed of three parts. M2b completely removes one of them
from the flood extent map, while the two other parts are significantly reduced in size.
The sub-optimal performance of M2b is arguably due to a difference in the number
and placement of vehicles at the time of the two satellite overpasses. In very high
resolution SAR imagery the presence or not of an object like a car inevitably impacts
the radar response. This necessarily influences the capability of the M2b algorithm to
reliably identify areas of smooth tarmac and unfortunately may not be resolvable at
all, for obvious reasons.
The region labelled R2, an area both flat and made of tarmac but not used as a
parking space, shows the capability of M2b to avoid the typical misclassifications of
smooth areas as flooded.
Finally the thematic analysis confirms the algorithms ability for identifying
permanent water bodies. The permanently flooded bed of the River Avon and some
adjacent boat marinas are removed from the flood extent map when taking into
26
account the reference image: this becomes evident when looking at the areas of over
detection in the panels (b) and (c) of Fig. 6. Clearly, this result is not achievable with
a single flood image, as M1 would invariably classify permanent water bodies as
flooded (see panel (a) of Fig. 6).
Despite the previously mentioned ability of M2b to detect areas with permanent
low backscatter values, there are still some shadow-affected areas that are erroneously
classified as flooded. A typical example is a large inclined rooftop in region R9.
This is classified as flooded by all three methods due to the fact that one
side is not
visible to the SAR sensor. Other examples of this behaviour can be found
on
various inclined rooftops in the R7 region.
To summarize, some risk of over-detecting flooded areas in built-up environments
inevitably remains. Non-flooded areas that appear smooth and water-surface like at
radar wavelengths as well as areas unseen by the satellite because of the side-looking
nature of SAR systematically produce very low signal returns and are not easily
distinguishable from flooded areas. The results of this study suggest that taking into
account the baseline backscatter values from dry reference images partly
addresses
the problem. Wide, open areas of tarmac or concrete (roads, parking lots, airfields
etc.) can be identified and removed from the final flood map (or, alternatively,
categorized as areas impossible to classify), while the situation is more problematic
with shadow areas. To check the plausibility of both types of regions to be flooded,
we expect that the use of high-resolution high-precision DEM data may be helpful.
More research on the integration of additional data sources into the image-processing
algorithm is needed for this to provide significant advantages.
27
2) Analysis of errors: the problem of under detection
With respect to the problem of under-detecting the true flood extent using SAR
observations, the results depicted in Table II indicate that method M2b leads to a
decrease in performance. In fact, the percentage of under-detected flood pixels rises
from 15.6 % obtained with the initial M2a method to its M2b-related value of 16.2 %.
However, it has to be underlined that this type of error is generally to be found on the
edge of inundated fields or in the vicinity of the main riverbed with tall vegetation
surrounding the areas. While the algorithm accurately retrieves most of the flooded
areas in wide, open areas, it can be observed that it systematically fails to retrieve
flooding under the vegetation canopy. These errors are not related to the imageprocessing
algorithm; rather they are due to the fact that with X-band radar systems
volume scattering originating from the vegetation canopy causes increased signal
return (i.e. layover). Furthermore, as already mentioned in paragraph 3.2, it cannot be
ruled out that the validation flood extent itself is affected by a slight overestimation,
as it was acquired closer to peak discharge than the satellite images. This could also at
least partly explain the under detection documented in the contingency matrix. Also,
the uncertainties in the delineation of the flood validation extent form aerial
photography are expected to have some marginal effect.
For example, an important area of apparent under-detection is the triangular
shaped field labelled R15. However, a closer look at the data reveals that due to the
time difference between the acquisitions of aerial photographs satellite imagery, it is
likely that that most of the floodwater was drained from the field in the 19 hours
preceding the TerraSAR-X acquisition. This hypothesis is confirmed by hydraulic
model simulations (see Fig. 3(b)).
28
The roughening of water surfaces due to wind is another inherent and potentially
significant limitation of the algorithm proposed in this study. When there are regular
waves on the surface of the water, Bragg resonance can result in very high signal
returns [23]. The misclassification of the area labelled R18 as non-flooded represents
a typical example. From the air photos and model simulations there can be no doubt
about the flooding of the area. However, waves are clearly identifiable on the standing
water. This renders accurate flood detection extremely difficult (if not impossible), as
analysis. The TerraSAR-X flood image was received under DLR SSS project
HYD0363.
31
REFERENCES
[1] D. E. Alsdorf, E. Rodriguez, and D. P. Lettenmaier, Measuring surface water
from space, Rev. Geophys., vol. 45, no. 2, pp. 1-24, May 2007.
[2] W. A. Marcus and M. A. Fonstad, Optical remote mapping of rivers at submeter
resolutions and watershed extents, Earth Surf. Processes Landforms, vol. 33, no. 1,
pp. 4-24, Jan. 2008.
[3] G. Schumann, P. D. Bates, M. S. Horritt, P. Matgen, and F. Pappenberger,
Progress in integration of remote sensingderived flood extent and stage data and
hydraulic models, Rev. Geophys., vol. 47, Nov. 2009.
[4] P. Matgen, R. Hostache, G. Schumann, L. Pfister, L. Hoffmann, and H. H. G.
Savenije, Towards an automatic SAR-based flood monitoring system. Lessons
learned from two case studies, Physics and Chemistry of The Earth, vol. 36, no. 7-8,
pp. 241-252, 2011.
[5] S. Martinis, A. Twele, and S. Voigt, Towards operational near real-time flood
detection using a split-based automatic thresholding procedure on high resolution
TerraSAR-X data, Nat. Hazards Earth Syst. Sci., vol. 9, no. 2, pp. 303314, 2009.
[6] S. Martinis, A. Twele, and S. Voigt, Unsupervised extraction of flood-induced
backscatter changes in SAR data using Markov image modeling on irregular graphs,
IEEE Trans. Geosci. Remote Sens., vol. 49, no. 1, pp. 251-263, 2011.
[7] L. Pulvirenti, N. Pierdicca, M. Chini, and L. Guerriero, An algorithm for
operational flood mapping from Synthetic Aperture Radar (SAR) data using fuzzy
logic, Nat. Hazards Earth Syst. Sci., vol. 11, pp. 529540, Feb 2011.
[8] D. C. Mason, R. Speck, B. Devereux, G. Schumann, J. Neal, and P.D. Bates,
Flood detection in urban areas using TerraSAR-X, IEEE Trans. Geosci. Remote
Sens., vol. 48, no. 2, pp. 882-894, Feb. 2010.
32
[9] R. Speck, P. Turchi, and H. S, An end-to-end simulator for high resolution
spaceborne SAR systems, in Proc. SPIE Defense Security, 2007, vol. 6568, p. 656
80H.
[12] D. C. Mason, I. J. Davenport, J.C. Neal, G. J-P. Schumann, and P. D. Bates,
Near real-time flood detection in urban and rural areas using high resolution
synthetic aperture radar images, IEEE Trans. Geosci. Remote Sens., in press, 2012.
[13] F. Ulaby, R. Moore, and A. Fung, Microwave remote sensing, active and passive,
Vol. III: Volume scattering and emission theory, advances system and applications.
Norwood (MA), USA: Artech House, 1986.
[14] G. A. F. Seber and C. J. Wild, Nonlinear Regression. Berlin: John Wiley & Sons,
2003.
[15] R. M. Haralick and L. G. Shapiro Image segmentation techniques,
Comput.
Vision Graph. Image Process., vol. 29, no. 1, pp. 100-132, Jan. 1985.
[16] R. Hostache, P. Matgen, W. Wagner, and L. Pfister, Change detection
approaches to flood extent mapping: How to select the best pre-flood reference image
acquisition (July 24 2007, 11:30 GMT) and TerraSAR-X overpass (July 25 2007, 6:34
GMT).
Fig. 4. Optimization of the parameters of the automated algorithm for M2b method:
Gamma pdf mode, 0
m , backscatter threshold, 0
thr , region growing threshold (or
maximum water backscatter), 0
rg , CD threshold, 0; the backscatter histogram of
the flood image is also displayed in dark blue colour.
Fig. 5. (a) Backscatter histogram of the reference image with superimposed the
threshold value
0
.
Fig. 6. Contingency map deriving from method: (a) M1, (b) M2a, (c) M2b. For sake
of clearness in the representation, the displayed maps have been cleaned by
neighbourhood analysis in post-processing step.
36
Fig. 7. RMSE values computed for different region growing thresholds (M2b method)
during the optimization process and corresponding performances in terms of correctly
predicted pixels (as flooded and as non-flooded).
Fig. 8. Backscatter probability density function of water pixels from the high
resolution (HR) photographs and water pixels from the method: (a) M1, (b) M2a, (c)
M2b. The histograms refer to the algorithm output, with no post-processing cleaning
step included.
Fig. 9. (a) Main rural and urban areas overlapped on the pixels in the flood image
covered by water according to the high resolution aerial photographs; (b)
corresponding backscatter probability density functions.
Fig. 10. Mask of regions unseen by TerraSAR-X due to shadow and layover, from
reference [8].
37
TABLES
TABLE I
CHARACTERISTICS OF THE AVAILABLE IMAGES (STRIPMAP MODE) FOR
THE ANALYZED FLOOD EVENT.
thr
on study area
()
flood
July 25 2007
6:34
descending 109 1.5 3 X 9.6 H/H 24
reference
July 22 2008
6:34
descending 109 1.5 3 X 9.6 H/H 24
TABLE II
QUANTITATIVE EVALUATION OF TERRASAR-X DERIVED FLOOD
EXTENT.
0
RGthr 0 over
detection
under
detection
total
good
total
error
(dB) (%) (dB) (%) of the area of interest, red box in Figures
M1 -13.4 96 - 2.8 15.6 81.6 18.4
M2a -11.7 99 -3.1 2.6 15.6 81.8 18.2
M2b -12.5 98 -3.0 2.1 16.2 81.7 18.3
rg
TABLE III
QUANTITATIVE ANALYSIS OF WATER PIXELS (PIXEL SIZE 1.5M) IN THE
SHADOW REGIONS (MASON ET AL., 2010).
n of water
pixels
in shadow
n of
water pixels
water pixels
in shadow
(%)
M1 8648 374060 2.3
M2a 8738 371734 2.3
M2b 7304 357616 2.0
38
TABLE IV
IMPROVEMENT DERIVING FROM THE USE OF A REFERENCE IMAGE:
COMPARISON OF THE DIFFERENT METHODS FOR SOME REGIONS, WITH
A SPECIAL EMPHASIS ON URBAN FEATURES. SEE FIG. 6 FOR THE
LOCATION OF STREETS, CROSSINGS AND URBAN/RURAL AREAS (THE
CORRECTLY CLASSIFIED REGIONS OF EACH METHOD ARE IN BOLD
FONT).
ID region HR photographs method M1 method M2a method M2b
R1 parking lots non-flooded flooded partially flooded partially flooded
R2 tarmac area non-flooded flooded non-flooded non-flooded
R3
supermarket roof
(Morrisons Store)
non-flooded flooded non-flooded non-flooded
R4 road crossing non-flooded non-flooded non-flooded non-flooded
R5
Road
(Barton Road)
non-flooded flooded flooded flooded
R6 field on hillslope non-flooded flooded flooded flooded
R7 roof non-flooded flooded flooded non-flooded
R8 urban area non-flooded partially flooded less flooded less flooded
R9 roof non-flooded flooded flooded flooded
R10 road crossing non-flooded flooded partially flooded partially flooded
R11
Road
(East Street)
non-flooded non-flooded flooded non-flooded
R12
Road
(Chance Street)
non-flooded non-flooded non-flooded non-flooded
R13
road
(High Street)
non-flooded non-flooded non-flooded non-flooded
R14 parking lot partially flooded flooded partially flooded partially flooded
R15 field flooded (dubious) non-flooded non-flooded non-flooded
R16
road
(Ashchurch Road)
flooded partially flooded partially flooded partially flooded
R17 road crossing flooded non-flooded non-flooded non-flooded
R18 parking lot flooded partially flooded partially flooded partially flooded
R19 urban area flooded partially flooded partially flooded partially flooded
R20 urban area flooded non-flooded non-flooded non-flooded
R21
Road
(Knights Way)
flooded non-flooded non-flooded non-flooded
39
Laura Giustarini received the B.Sc. and M.Sc. degrees in environmental engineering
from the University of Perugia, Perugia, Italy, in 2003 and 2006, respectively. She has
worked for the National Research Council, Research Institute for Geo-Hydrologicl
Protection, Perugia, Italy. Since 2010, she has been with the Department of
Environment and Agro-Biotechnologies, Public Research Centre Gabriel Lippman,
Belvaux, Luxembourg. Her current research interests are the integration, through data
assimilation techniques, of radar remote sensing into coupled hydrologic and
hydraulic models for surface water management.
Renaud Hostache received the M.Sc. degree in mechanics of geophysical media and
environment from the Joseph Fourier University, Grenoble, France, in 2003 and the
Engineering degree in hydraulics and environmental sciences from Grenoble Institute
of Technology-National School of Hydraulics and Mechanism of Grenoble, Grenoble.
He has undertaken his Ph.D. research work with the Cemagref, Montepellier, Fgrance.
His Ph.D. investigated the fine 3D characterization of flood hazard, owing to datellite
imagery and its integration in flood inundation models. He received the Ph.D. degree
in water sciences from the National School of Agriculture, Water and Forest
Engineering, Montepellier, in 2006. Since 2007, he has been with the Department of
Environment and Agro-Biotechnologies, Public Research Centre Gabriel Lippman,
Belvaux, Luxembourg. His current research interests are focused on the integration of
remote-sensing observations in hydraulic models in the hydrologic and hydraulic
(a)
(b)
(a)
(b)
(c)
(d)
(e)
(f)
(a) (b)
(b)
(c)
(a)
(a)
(b)
(c)
(a)
(b)
int. j. remote sensing, 2002, vol. 23, no. 18, 36813696
1. Introduction
During an extreme ood event it is important to be able to determine quickly
the extent of ooding and the landuse and landcover types under water. This
information can be used in developing a comprehensive relief eVort (Corbley 1993).
During ooding events remotely sensed data can provide signi. cant mapping capabilities.
However, obtaining remotely sensed data that represents the ideal combination
of . ne spatial and temporal sampling, and the ability to see through clouds and/or
to discriminate ooding under forest cover is a diYcult task. In addition, accessibility
*e-mail: wangy@mail.ecu.edu
and describe ground observations. Next, the ood mapping eVorts are explained
using TM data for Pitt County, and TM and DEM data for a subset of Pitt County.
Flood mapping using TM and DEM data 3683
We then present the results. We also discuss the potential for applying the ood
mapping methods (using TM data alone, and combining TM and DEM data) in
coastal oodplains in general, and the limitations and cautions that should be noted
when applying these methods to mapping ood extent in areas of large spatial extent,
and areas having large topographic variation.
2. Analytical approaches
2.1. Study area and ground observation
Most of eastern North Carolina lies within the Atlantic coastal plain. Pitt County
lies in the eastern coastal plain of North Carolina, at the approximate centre of the
region. The elevation of the area drops only about 60m as it extends 120160 km
from the Piedmont region in the middle of the state towards the coast. Four large
elongated river systems drain the coastal plain in a north-westsouth-east direction.
Flat broad oodplains are usually located on the northern side of the rivers with
higher ground on the south (Gares 1999). In Pitt County the land surface has very
low relief and many parts of the region have been extensively drained, cleared and
ditched for agricultural use. The soils are primarily characterized as poorly drained
or extremely poorly drained (63.0%), with the remaining area consisting of moderately
well to well-drained soils (Gares 1999). Pitt County has a population of about
126 000 (estimated in 1998). The largest city, Greenville, is centrally located and has
a population of approximately 60 000 (estimated in 1998). The additional residents
of the county are spread throughout rural towns.
In Pitt County the majority of the 1999 ooding occurred north of the Tar River.
The Tar River has been slowly migrating southward towards the drainage divide of
the Neuse River so broad primary and secondary oodplains extend northward from
the river channel. North and immediately adjacent to the Tar River is a band
following the channel that is currently de. ned as conservation/open space landuse
in the City of Greenville (. gure 1). This conservation/open space landuse zone grades
into low and medium density residential, industrial and mixed land uses. It is clear
from the photograph, however, that there is signi. cant activity within this open
space landuse. The City of Greenville alone suVered ooding to its airport, water
treatment facility, power transmission substation and numerous residential and
industrial areas that are within or nearby current open space landuse zones. In Pitt
County, some 6000 homes were ooded. Over three-quarters of these homes were
largely uninsured. Upwards of 50 000 people were displaced. More than 6000 were
housed in emergency shelters, many for over 3 weeks.
Once the oodwaters had completely subsided, but before high water marks
faded, ground data information was gathered in the . eld. Areas both north and
south of the Tar River were examined for the extent and depth of ood waters. The
aerial photo (. gure 1) was taken on 23 September 1999 during the ood event and
is centred approximately on the City of Greenville. The ood gauge information
used for this study was taken from the gauge on the Green Street Bridge (. gure 1).
Floodwaters extend into the student housing district seen in the south-eastern section
of the photo and throughout the entire area shown north of the river. Areas of
extensive tree canopy north of the river in the primary and secondary ood plain
were completely ooded. Figure 2 shows the high water marks (reaching the middle
of windows) on a house trailer in a trailer park that is located immediately adjacent
to the north-eastern most section of the photo. These areas of tree canopy were not
3684 Y. Wang et al.
Figure 1. Aerial photo of a portion of the City of Greenville, North Carolina, taken during
the September 1999 ood. The major part of the City of Greenville is on the south
side of the Tar River.
RMSE) in this area is 1 m. Four 7.5 min USGS topographic quadrangles, Greenville
NW, NE, SE and SW were downloaded, imported, and mosaiced. The four mosaiced
quads covered an area of about 600 km2. Descriptive statistics for the four-quadrangle
DEM area included min.=0 m, mode=11.9 m, median=14.0 m, mean=14.6 m,
max.=26.2 m, and standard deviation=6.4m. This study area is primarily at,
especially on the north side of the Tar River where most of the ooding occurred
(e.g. . gures 1 and 2). We then co-registered the TM and DEM data so that the same
area of interest can be easily extracted.
Flood stage on the Tar River is measured from a point 0.7m below sea level
(based on the North American Vertical Datum, NAD88). The bottom of the river
is about 2m below sea level. The Tar River leaves its banks 4m above this measuring
point. The Tar River crested at 9.2m on 21 September. On the date the imagery was
taken, 30 September, data from the USGS showed that the mean stage level for the
Tar River at the Greenville station was 6.1 m. The non- ood stage surface height of
the water in the Tar River according to the river gauge reading on 28 July 1999 was
3686 Y. Wang et al.
1.1 m. Therefore, the elevations that represented ooded areas on 30 September
ranged from 1.1 to 6.1 m. These elevations were used as a basis for classifying the
area on the Greenville topographic quadrangles into water bodies/rivers, ooded
areas, and non- ooded areas, which will be discussed in detail in a later section (see
tables 5 and 6).
2.4. North Carolina landuse and landcover data
Between 1995 and 1997, the North Carolina Center for Geographic Information
and Analysis (NCCGIA), in cooperation with the NC Department of Transportation
and United States Environmental Protection Agency Region IV Wetlands Division,
contracted Earth Satellite Corporation (EarthSat) of Rockville, Maryland to generate
comprehensive landcover data for the entire state of North Carolina (Earth Satellite
Corporation 1997). There are 21 landuse and landcover type categories in the entire
state data layer. For Pitt County, only 17 categories exist ranging from highly
developed areas to unconsolidated sediment areas. To facilitate the presentation of
this paper and to provide the reader with a better understanding of the landcover
classes, we provide brief de. nitions of some categories (in table 3). The three landuse/
landcover categories that were aVected the most by the ood were bottomland forest/
hardwood swamps, southern yellow pine, and cultivated land. Bottomland forests/
hardwood swamps are areas where deciduous, dominant, woody vegetation is above
3m in height, as well as occurring in lowland and wet areas. Crown density is at
least 25%. Southern yellow pine are areas where stocking of trees is 75% evergreen
needleleaf or broad-leaf species, including the following forest types: longleaf pine,
loblolly-slash pine, other yellow pine, and pond pine. Cultivated lands are areas of
land that are occupied by row and root crops that are cultivated in distinguishable
rows and patterns. Two other important categories aVected by the oods were high
and low intensity developed areas, which contains the housing and infrastructure
for the majority of the human population in the area. High intensity developed areas
are covered by more than 80% synthetic (man-made) landcover. Low intensity
developed areas have between 50 and 80% coverage by synthetic landcover. (See
table 3 for the ooded areas for other landuse and landcover types.)
2.5. Flood mapping using TM images
The initial goal in ood mapping was to investigate the utility of the TM images
for identifying areas that were ooded or not ooded. There were two steps: (1)
identify water versus non-water areas on the TM images before and during the ood
event, respectively, and (2) compare the areas classi. ed as water or non-water on
both TM images to determine which areas represented ooding.
2.5.1. Identifying water areas versus non-water areas
There are many possible methods for identifying water versus non-water areas
using TM data (e.g. Jensen 1996). After unsuccessful trials of using supervised and
unsupervised classi. cation, and other methods, we used the addition of two
TM bands (TM4+TM7) of 28 July, and of 30 September, respectively. TM4
(0.760.90 mm, re ective infrared) is responsive to the amount of vegetation biomass,
and is useful in identifying land and water boundaries. However, it is possible to
confuse water and asphalt areas (road pavements and rooftops of buildings) in the
developed areas such as downtown, commercial/industrial areas, etc., as they appear
black on the TM4 image or they re ect little back to the sensor. On the TM7
Flood mapping using TM and DEM data 3687
(2.082.35 mm, mid-infrared) image, the re ectance from water, paved road surfaces,
and rooftops diVers. Thus, one can identify the water ( ooded) and non-water (non ooded) area in the developed areas, by incorporating of TM4 and TM7 into the
analysis, as detailed in table 1. This addition is done separately for the July and
September TM data. Therefore, the classi. cation rule was:
If the re ectance of pixels or areas is low in the TM4 plus TM7 image, the pixels
represented water, otherwise the pixels represented non-water or dry areas.
In the analysis, we noted that the re ectance from water, paved road surfaces,
and asphalt rooftops of buildings in the developed areas may be also distinguished
on TM5. Thus, TM5 could have been used to detect water versus non-water areas.
However, the diVerences on TM5 were slightly smaller than those on TM7 image.
Also, Banumann (1996) added two (before and during ood event) TM4 images in
his 1993 Mississippi ood analysis. He then sliced the added image into water,
ooded areas, and non- ooded areas. He further added TM7 data to the combined
TM4 image to separate some confusion between the water and industrial area.
Once the representation of the re ectance values for water and non-water features
was understood, a cut-oV value could be determined to separate the two categories.
For the July TM image, the cut-oV value was 141. If a pixels DN value was less
than 141, that pixel was assigned as a water category, otherwise it would assigned
as a non-water category. For the September TM image, the cut-oV value was 109,
i.e. if a pixels DN was less than 109, that pixel was classi. ed as water, otherwise it
was classi. ed as non-water. Even though the selection of the cut-oV values may seem
to be somewhat arbitrary, we used two diVerent methods to check the cut-oV values.
One was ground truthing, and the other was the analysis of the histograms of the
TM4 plus TM7 images. In the former, ground observations along the Tar River
were used to pick the cut-oV values. These observations were made in the . eld in
early October of 1999 (e.g. . gure 2), and through the analysis of aerial photos taken
during the ood event. In the latter, the histogram of the TM4 plus TM7 image was
examined to see whether the histogram indicated the cut-oV values. This was the
case for the histogram of the July TM4 plus TM7 image; two distinct distributions
were observed from the histogram plot. There were also two identi. able distributions
in the histogram of the September TM4 plus TM7 image. However, a distinct
separation between the two distributions, was not as easily identi. ed (i.e. any DN
value ranging from 107 to 111 may be selected).
2.5.2. Determine ooded areas during the ood event
After identifying water versus non-water areas on both images (one acquired
before the ood event and the other during the ood) using the above criteria,
determination of areas that were ooded could be made. On a pixel by pixel basis,
Table 1. Re ectance of water, asphalt pavement (road surface, root of buildings, etc.), and
other non-water dry areas on TM4 and TM7 images.
Re ectance on TM4 Re ectance on TM7 Re ectance on TM4+TM7
Water low low low
Asphalt pavement low intermediate intermediate
Other dry area high high high
Using the landuse and landcover data layer obtained from the NCCGIA, we
provide the following description for the derived ood map. The ooded areas along
the Tar River and along the tributary of the Neuse River were primarily bottomland
forests/hardwood swamps (e.g. . gure 1). The patches of ooded areas in the north,
north-east, and south-east of the image were mainly southern yellow pines and
cultivated lands. The three categories most aVected by the ood, in terms of size and
percentage of the total ooded area in Pitt County, were southern yellow pine
(91.7 km2 or 5.4% out of the total areas in the County), bottomland forest/hardwood
swamps (73.6 km2 or 4.3%), and cultivated lands (40.0km2 or 2.4%). It should be
noted that most of the high and low intensity developed areas were not ooded on
30 September 1999. (The oodwater from the Tar River had receded about 3m from
its crest on 21 September. The main developed areas are near the banks of the Tar
River.) It should be also noted that there was a slight diVerence of the sizes of the
Flood mapping using TM and DEM data 3689
Figure 3. The ood extent in Pitt County, North Carolina on 30 September 1999, derived
from a pair of Landsat TM images of 28 July and 30 September 1999. The grey
rectangle indicates the Greenville study area.
Table 2. Flood mapping of Pitt County, North Carolina, as of 30 September 1999.
Area (km2) Area (%)
Water bodies/Rivers 13.3 0.8
Flooded areas 237.9 14.0
Cloud shadows 1.2 0.1
Non- ooded areas 1444.8 85.1
Total 1697.2 100
water bodies/river channels derived from the TM image pair of 1999 and NCCGIA
landuse data layer (13.3 km2 vs 12.2 km2, cf. Tables 2 and 3). This diVerence could
be due to diVerent ways to identify water, due to the errors in our analysis and/or
3690 Y. Wang et al.
Table 3. Total areas from the landuse and landcover type data layer and ooded areas of
each landuse and landcover type in Pitt County (NC) derived from TM data.
Total areas from the
landuse data layer Flooded areas on 30
(km2) Sept. 1999 (km2) Overall (%)
High intensity developed 12.6 1.1 0.1
Low intensity developed 22.8 0.9 0.1
Cultivated 612.8 40.0 2.4
Managed herbaceous cover 40.6 3.1 0.2
Unmanaged herbaceous
coverupland 13.4 0.5 0.0
Unmanaged herbaceous
coverwetland 0.1 0.0 0.0
Evergreen shrubland 177.1 16.0 0.9
Deciduous shrubland 19.5 1.9 0.1
Mixed shrubland 26.0 2.3 0.1
Mixed upland hardwoods 0.1 0.0 0.0
Bottomland forest/hardwood
swamps 369.3 73.6 4.3
Needleaf deciduous 0.1 0.0 0.0
Southern yellow pine 350.5 91.7 5.4
Mixed hardwoods/conifers 39.1 3.4 0.2
Oak/gum/cypress 0.3 0.1 0.0
Water bodies/rivers 12.2 3.1 0.2
Unconsolidated sediment 0.7 0.1 0.0
Total 1697.2 237.8 14.0
in the landuse data layer, due simply to landuse change since the creation of the
landuse data layer in 1996, or all three.
3.3. Addition of the DEM data into the ood mapping analysis
Due to the dense or continuous canopy coverage in bottomland forest/hardwood
swamps and in some dense southern yellow pine stands, and due to the lack of
canopy penetration of the TM data, ooded areas under the canopies were not
detected by classi. cation of the TM data. This underestimation of ooded areas was
veri. ed through ground truthing and visual interpretation of low-altitude oblique
aerial photos taken during the 1999 ood. On the ood map, these undetected
ooded areas show up as patches or holes along the primary oodplain near the
riverbanks. It is important to point out this underestimation, because oods in the
coastal oodplains of North Carolina, as well as the entire East Coast and the coast
of the Gulf of Mexico often occur from the mid-summer to fall, and trees in the
oodplain are almost fully leaf-on during this period of time. Radar data (especially
radar data from a long wavelength system) can penetrate the (dense) canopies and
identify whether the areas underneath the canopies were ooded or not. However,
due to the cost of the radar (such as ERS SAR or Radarsat SAR) data we did not
incorporate them in the ood mapping analysis. One alternative is to integrate DEM
data into the analysis. There are several advantages to this integration. In the USA,
the DEM data are widely available and can be directly download from USGS/EROS
web site. (It should be noted that the availability of DEM data in other countries
might be limited.) Also, most of the bottomland forest and hardwood swamps in the
oodplain are located in places of low elevation or along the banks of rivers. By
Flood mapping using TM and DEM data 3691
using river gauge reading to inundate the DEM, one can map the ood underneath
tree canopies in the bottomland forest, and hardwood swamps, as well as in some
southern yellow pine stands in low elevation areas. Additionally, high-quality DEM
data work well for ood mapping in areas of relatively at terrain, as exists in this
study area and other coastal oodplains along the East Coast and the Gulf of
Mexico. The following discussion describes the integration of the DEM and TM
data for the ood mapping study.
We used four 7.5 min quadrangle s of USGS DEM data for the Greenville areas.
A grey box shown in . gure 3 indicates the coverage of the four quads, which includes
most of the Tar River in Pitt County. We then extracted the area (of the block,
. gure 4(a)) from the ood map of Pitt County (. gure 3) derived from the TM data.
Black areas on the image represent water bodies/rivers, grey represents ooded areas,
and white represents non- ooded areas. A detailed statistical summary of ooded
areas for each landuse and landcover type is provided in table 4. The three categories
having the largest areas and highest percentage of ooding were southern yellow
pine, bottomland forest/hardwood swamps, and cultivated land.
We then inundated the DEM based on the river gauge readings before the ood
event and on 30 September 1999. The river gauge station is near the centre of the
four quads (. gure 1). These readings were 1.1m preceding the ood event, and 6.1m
on 30 September. Using the rules found in table 5, we reclassi. ed the DEM into
water bodies/rivers, ooded areas, and non- ooded areas (. gure 4(b)). The simple
threshold used for inundating the DEM was possible due to the relatively at terrain
(no sinks) away from the river channel within our study area. The ooded areas
were located in the oodplain of the Tar River, and ooded areas did not exhibit
the canopy holes found on the TM images. By inundating the DEM data, ooded
areas under the canopies of the bottomland forest and hardwood swamps could be
identi. ed. At higher elevations and away from the river, the DEM suggested that
those areas were dry or there was no ooding. The DEM does not identify water
bodies and/or ooded areas at higher elevations. The ooded areas of each landuse
and landcover type derived from the DEM data were also tabulated (table 4), and
the largest area (22.6km2) and highest percentage (3.8%) of ooded landcover type
was bottomland forest/hardwood swamps.
The . nal ood map for the Greenville areas was derived by using the logical
OR operator to combine the ooded areas from either the TM data or the DEM
data (. gure 4(c)). Flooded areas located away from the river and at high elevation
were identi. ed by the TM data. Flooded areas near the river and its tributaries were
Table 4. Total and ooded areas of each landuse and landcover type in Greenville (NC) derived from
TM data alone, DEM data alone, and TM and
DEM data combined.
Total areas Flooded areas Flooded areas Flooded areas
from the derived from derived from derived from
landuse data TM data DEM data TM and DEM
layer (km2) (km2) Overall (%) (km2) Overall (%) data (km2) Overall (%)
High intensity developed 9.2 0.7 0.1 0.2 0.0 0.8 0.1
Low intensity developed 11.9 0.3 0.1 0.1 0.0 0.4 0.1
Cultivated 208.8 15.1 2.6 8.5 1.4 19.1 3.2
Managed herbaceous cover 18.9 2.1 0.4 1.3 0.2 2.5 0.4
Unmanaged herbaceous
coverupland 0.5 0.0 0.0 0.0 0.0 0.0 0.0
Unmanaged herbaceous
coverwetland 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Evergreen shrubland 74.3 7.0 1.2 2.9 0.5 8.0 1.3
Deciduous shrubland 7.8 1.1 0.2 0.7 0.1 1.3 0.2
Mixed shrubland 9.6 1.0 0.2 0.4 0.1 1.1 0.2
Mixed upland hardwoods 0.1 0.0 0.0 0.0 0.0 0.0 0.0
Bottomland forest/hardwood
swamps 112.7 23.2 3.9 22.6 3.8 32.7 5.5
Southern yellow pine 123.9 28.2 4.8 2.0 0.3 28.8 4.9
Mixed hardwoods/conifers 11.4 1.2 0.2 2.0 0.3 2.6 0.4
Oak/gum/cypress 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Water bodies/rivers 4.7 1.3 0.2 2.0 0.3 1.3 0.2
Unconsolidated sediment 0.1 0.0 0.0 0.0 0.0 0.0 0.0
Total 593.9 81.2 13.9 42.7 7.0 98.6 16.5
Table 5. Classi. cation rules based on the gauge readings before the ood event and on
30 September 1999. (The interval of the DEM data is 0.3m.)
Min. Max.
Water bodies/rivers 1
Flooded areas >1 6
Non- ooded areas >6
In some areas the RMSE may be higher, for example, under canopies due to the
DEM generation procedure used by the USGS. We re-ran the analysis with the
DEM dataset at 1m of the river gauge reading (with the integration of TM data) .
Recoding the elevation to represent ooding at 1m less than the river gauge reading
did not signi. cantly change the pattern of the ood (i.e. 95.3km2 vs 98.6 km2 total
ooded areas and 16.5% vs 16.1%, tables 6 and 4). The elevation data recoded at
1m above the gauge reading did expand the ood to the north of the river considerably,
as this is an area of very low relief (110.4km2 vs 98.6 km2, tables 6 and 4). Due
to the estimated accuracy of the DEM data, the total ooded areas derived from
the combination of TM and DEM data in the Greenville areas could vary from
95.3 km2 to 110.4 km2, and the ooded areas from 16.1% to 18.6% of the total study
area (table 6).
Flooding of the DEM data only works for a reasonable distance from the river
gauge from which you measure stage height. This can be a signi. cant distance in
Table 6. Flooded areas of each landuse and landcover type in Greenville derived from TM
and DEM data. The DEM data were set at 1m of the river gauge reading of Tar
River on 30 September 1999.
At 1m less than the At 1m above the river
river gauge data gauge data
Flooded Flooded
areas (km2) Overall (%) areas (km2) Overall (%)
High intensity developed 0.7 0.1 1.4 0.2
Low intensity developed 0.3 0.1 1.0 0.2
Cultivated 17.7 3.0 23.6 4.0
Managed herbaceous cover 2.4 0.4 4.0 0.7
Unmanaged herbaceous cover
upland 0.0 0.0 0.0 0.0
Unmanaged herbaceous cover
wetland 0.0 0.0 0.0 0.0
Evergreen shrubland 7.6 1.3 9.3 1.6
Deciduous shrubland 1.3 0.2 1.4 0.2
Mixed shrubland 1.1 0.2 1.2 0.2
Mixed upland hardwoods 0.0 0.0 0.0 0.0
Bottomland forest/hardwood
swamps 31.8 5.4 35.1 5.9
Southern yellow pine 28.6 4.8 29.3 4.9
Mixed hardwoods/conifers 2.5 0.4 2.8 0.5
Oak/gum/cypress 0.0 0.0 0.0 0.0
Water bodies/rivers 1.3 0.2 1.3 0.2
Unconsolidated sediment 0.0 0.0 0.0 0.0
Total 95.3 16.1 110.4 18.6
in the other coastal oodplains (such as the East Coast, and the coast of the Gulf
of Mexico of the USA), using similar TM images, DEM data, and river stage data.
This method should work well for areas of large spatial extent if the ( local )
topography is relatively at, as demonstrated in this study.
The total ooded areas derived from the TM data alone, on 30 September 1999
for Pitt County, North Carolina were 237.9km2 or 14.0% of the total county area.
This number may be low due to the underestimation of the ooded areas beneath
dense vegetation canopies. The landuse/landcover categories most aVected by the
ood were the southern yellow pine (91.7 km2), bottomland forest/hardwood swamps
(73.6 km2), and cultivated land (40.0 km2). Their total ooded areas were 205.3km2
or 86.3% of the total ooded areas in the County.
Through integrating the classi. cation of ooded areas from TM imagery and
from inundating a DEM to represent ooding, the following results were obtained
for the Greenville area on 30 September 1999. The total ooded area was 98.6km 2
(out of the total study area of 593.9km2) or 16.5%. The landuse/landcover categories
most aVected by ooding were bottomland forest/hardwood swamps (32.7km2),
southern yellow pine (28.8km2), and cultivated land (19.1 km2). Their total ooded
areas were 80.6km2 or 81.7% of the total ooded area in the studied area.
Incorporating the DEMdata assisted greatly in identifying the ooding that occurred
underneath the forest canopies, especially under the canopies of bottomland forest
and hardwood swamps. However, it should be noted that there were two main
limitations regarding the integration of DEM data with TM data for ood mapping.
One was the use of river gauge readings to inundate the DEM, and the other was
the handling of error in the DEM. The authors intend to investigate methods of
error representation (e.g. Hunter and Goodchild 1995) in the future. In addition, the
US Army Corp of Engineers is currently surveying high-water marks from the
3696 Flood mapping using TM and DEM data
ooding in this area, and this data will be used to assess the accuracy of the DEM
in modelling ood extent.
Acknowledgements
This research was partially funded by a grant to East Carolina University from
the Natural Hazards Research and Applications Information Center at the University
of Colorado, Boulder, CO, and partially supported by East Carolina University.
References
Banumann, P. R., 1996, Flood analysis: 1993 Mississippi ood. URL http://umbc7.umbc.edu/
~tbenja1/baumann/baumann.html, Vol. 4 of Remote Sensing Core Curriculum.
Barnes, S. B., 1996, Northwest Flood 96: GIS in the face of disaster. Geographic Information
Systems, 2225.
Brackenridge, G. R., Tracy, B. T., and Knox, J. C., 1998, Orbital SAR remote sensing of
a river ood wave. International Journal of Remote Sensing, 19, 14391445.
Brown, D. G., and Bara, T. J., 1994, Recognition and reduction of systematic error in
elevation and derivative surfaces from 7 -minute DEMs. Photogrammetric Engineering
and Remote Sensing, 60, 189194.
Corbley, K., 1993, Remote sensing and GIS provide rapid response for ood relief. Earth
Observation Magazine, September, 2, 2830.
Correia, F. N., Rego, F. C., Saraiva, M. D. S, and Ramos, I., 1998, Coupling GIS with
hydrologic and hydraulic ood modeling. Water Resources Management, 12, 229249.
Earth Satellite Corporation, 1997, Comprehensive land cover mapping for the state of
North Carolina. Final report, March 1997, Rockville, Maryland.
Dartmouth Flood Observatory, 1999, DFO-1999-076 ooding from Hurricane Floyd.
NASA-supported Dartmouth Observatory.
Gares, P., 1999, Climatology and hydrology of Eastern North Carolina and their eVects on
creating the ood of the century. North Carolina Geographer, 7, 311.
Hess, L. L., Melack, J. M., Filoso, S., and Wang, Y., 1995, Realtime mapping of inundation
on the Amazon oodplain with the SIR-C/X-SAR synthetic aperture radar. IEEE
T ransactions on Geoscience and Remote Sensing, 33, 896904.
Hunter, G. J., and Goodchild, M. F., 1995, Dealing with error in spatial databases: a simple
Shrestha Mountain Environment and Natural Resources Information System, International Centre for Integrated Mountain
Development, GPO Box 3226, Kathmandu, Nepal Abstract Floods are one of the most common hazards in the world,
affecting peoples lives and livelihoods. Flood hazard mapping and flood shelters suitability analysis are vital elements in
appropriate land use planning for flood-prone areas. This paper describes application of Remote Sensing (RS) and
Geographical Information Systems (GIS) in identifying flood hazard zones and flood shelters and are therefore important
tools for planners and decision makers. The purpose of this article is to describe a simple and efficient methodology to
accurately delineate flood inundated areas, flood-hazard areas, and suitable areas for flood shelter to minimize flood
impacts. Possible extent of flooding and suitable location flood shelter sites were modeled and mapped for Sindh Province
in Pakistan, using the software ArcGIS model builder. The output was validated using inundation maps based on flood
events that took place in 2010 in Pakistan. These were mapped using object-based image analysis (OBIA) implemented in
eCognition software. The catastrophic flood of 2010 inundated a total area of 7579 km2, while the modeled result
indicated the hazard area to be 6216 km2 out of 46138 km2. Discrepancies in modeled and mapped results are
insignificant and acceptable considering the manual flood management interventions which are beyond the capability of
models to represent. Thus, this method is robust enough to develop flood hazard zoning maps and map shelter sites for
flood management. Keywords Remote Sensing, GIS, OBIA, Flood, Hazard, Sindh River, Pakistan
1. Introduction
More than half the world's population lives in Asia, which is approximately one-fifth of the earth's land area. As natural
disasters increase in both intensity and severity around the world, the Asian region continues to suffer a disproportionate
number of hazard events and related losses in lives, infrastructure, stability, and economic progress. In the period 1994
2004 alone, Asia accounted for one third of 1,562 flood disasters worldwide and nearly 60,000people were killed in
floods[1]. A flood is an overflow of an expanse of water that submerges the land. Flood hazards are the most common and
destructive of all natural disasters and are a constant threat to life and property. Each year, flood disasters result in
tremendous losses and social disruption worldwide. Over the past 30 years, floods have been the most catastrophic natural
disaster affecting, on average, about 80 million people per year or half of the total population affected by any natural
disaster, causing * Corresponding author: kuddin@icimod.org (Kabir Uddin) Published online at http://journal.sapub.org/ajgis Copyright 2013
Scientific & Academic Publishing. All Rights Reserved
economic damage worth over US$11 million annually around the world[2]. As increasing human activity downstream of rivers results in
greater flood damage; floods themselves are also increasing in size and frequency due to human activities in the upstream section of the
river system[3] In recent years, risk-based approaches have received increasing attention as a viable means to manage flood hazards.
Hazard risk and vulnerability assessment (HRVA) examines the hazards that may affect a community in order to determine the risk that
each hazard event poses to both the community as a whole and to vulnerable elements in the community.
In Pakistan, floods have been recognized as a major natural calamity. The country has a long history of flooding from the
Indus River and its tributaries and the floods of 1928, 1929, 1955, 1957, 1959, 1973, 1976, 1988, 1992, 1995, 1996 and
1997 attest to its destructive nature and its adverse impact on lives and property in Pakistan. More than three million
homes were destroyed and 160 persons lost their lives in the great flood of 1973. Three years later in 1976, floods
demolished over 10 million houses while 425 lives were lost with other losses amounting to a Rs. 6 billion. Towards the
end of September 1988, an unprecedented 2 Kabir Uddin et al.: Application of Remote Sensing and GIS for Flood Hazard
Management: A Case Study from Sindh Province, Pakistan
flood occurred, inflicting damage nationwide in Pakistan worth about Rs. 17 billion. The catastrophic flood of 1992
surpassed all previous records with damages estimated at Rs. 50 billion. Sindh is the south eastern province of Pakistan.
The province is associated with the longest river of Pakistan which stretches to about 1976 miles (3180 km) long.
Originating in the Tibetan Plateau in the vicinity of Lake Mansarovar in the Tibet Autonomous Region, the river runs a
course through the Ladakh district of Jammu and Kashmir, then enters the Northern Areas (Gilgit-Baltistan) flowing
through the North in a southerly direction along the entire length of the country to merge into the Arabian Sea near the port
city of Karachi in Sindh. It is the major source of irrigation for Pakistans fertile agricultural lands. The rivers estimated
annual flow stands at around 207 cubic kilometers (cu km), making it the 21st largest river in the world in terms of annual
flow. Climate is not uniform over the Indus Basin. It varies from subtropical arid and semi-arid to temperate sub-humid in
the plains of the Sindh and Punjab provinces, and alpine in the mountainous highlands to the north. Annual precipitation
ranges between 100 mm and 500 mm in the lowlands to a maximum of 2000 mm (water equivalent) on mountain slopes.
Snowfall at higher altitudes (above 2500 m) accounts for most of the river runoff. Alluvial land in Pakistan is known to be
a flood-prone area.
Earth observation techniques can contribute to finding out more accurately what causes floods. Together with flood hazard
mapping, earth observation techniques can be used to assessing damage to property, infrastructure, and agricultural crops.
The uses of remotely sensed data in identifying the trends of river channel migration and as source of input data for
determining river behaviour has been gaining popularity in recent years. One of the main characteristics of remote sensing
is its capability to generate a large amount of information frequently and spatially, becoming a powerful tool for
monitoring changing aquatic environments. Remote sensing data has been used to document water quality estimate water
depths and monitor river channel changes and aquatic habitat[4-6]. Object-based analysis of multispectral imagery was
introduced early on in remote sensing literature[7]; however, the object-based approach has largely been ignored in favour
of pixel-based methods which have been easier to implement[8]. One of the advantages of object-based image analysis is
the multitude of additional information that can be derived from image objects compared to the amount of information
available from individual pixels[9]. This study aims to determine flood hazard, flood extended areas, and proposed flood
shelters using remote sensing and geographic information systems tools. An effort was made to prepare a flood hazard
zone map and a proposed flood shelter map using GIS and remote sensing techniques and incorporating a satellite image,
topographic map. This study also attempts to exploit the capabilities of remote sensing and GIS techniques to suggest an
appropriate methodology for flood hazard mapping.
2. Methodology
2.1. Study Area
The study area is located in the south-eastern part of Nepal and seven districts of Sindh province (Figure 1). Sindh lies in a
tropical to subtropical region; it is hot in the summer and mild to warm in winter. Temperatures frequently rise above 46C
between May and August, and drops to a minimum average of 2C during December and January in the northern and
elevated regions. Annual rainfall averages about seven inches, falling mainly during July and August. The province is
mostly arid, with scant vegetation except for the irrigated Indus Valley.
Figure 1. Location Map of Study Area American Journal of Geographic Information System 2013, 2(1): 1-5 3
The next step was to define classes and insert class hierarchy. Image objects were linked to class objects and each
classification link was stored in the membership value of the image object to the linked class. Image objects were
classified using user-defined rules. Objects with an area smaller than the defined minimum mapping unit were merged
with other objects. The classified inundation map was exported to shape file format for further processing and flood hazard
processing and flood shelters and modeling. Several types of factor/parameter were required to do flood hazard
assessment. Acquired datasets of stream order, river, digital elevation model and land cover including all of the data that a
flood under such circumstances would include. One of these is slope which is an essential topographic factor that guides
the flood potential of spatial areas. Slope data layers were generated using the digital elevation model of the study area.
Vegetation poses a major restraint to flooding; vegetation reduces runoff and helps in percolation. Euclidea n allocation
calculates for each cell the nearest source based on Euclidean distance and calculates the direction in degrees to the nearest
source. Normally, all nearest source areas are flood-prone areas. Euclidean allocations of stream order were generated.
Euclidean distances of rivers were generated and short disease from river ranked flood potential. In order to combine the
datasets, they must be set to a common scale. Each dataset data is then reclassified to rank the potential hazard location.
Then reclassify date sets came to a common scale ranges from 1 to 5. Each raster was assigned ranks at the time of
reclassification: a ranking of 1 means with the most flood potential, and 5, least flood potential. Once, all the four raster
layers are created using the models developed, they are overlaid together to produce the final flood hazard map of Shind
province. The weighted layers were created in order to reduce the number of inputs in the final weighted overlay. The final
output of a flood potential index was the result of equally weighting 20% each the weighted layers. The results were
validated with the recent flood map of Sindh province.
Providing flood shelter in flood-affected areas is very important. Flood shelters provide flood-affected people with areas to
run to for shelter and keeps them safe while the flood rages Digital elevation model, land cover, settlement, road and flood
map were used to model suitable locations for flood shelter. Preparing the shelters suitability slope from the elevation
dataset, point density from the settlement dataset, Euclidean distance from the road and flood dataset were used. Each
datasets data were then reclassified to rank the suitability of locations for flood shelter. Each raster was assigned a rank at
the time of reclassification, with rank 1 as less suitable for flood shelter and 5 as most suitable. Once all four raster
layers have been created using the models developed, they were overlaid together to produce the final suitability flood
shelters location map of Shind province (Figure 3). They were then weighted according to a percentage of influence and
combined to produce a map displaying suitable locations for the proposed flood shelter. Results were selected and the
optimum site identified for the new flood shelter from alternative sites.
Figure 4. Flood hazard zoning map
3. Results
Figure 5. Flood shelter suitability map Table 1. District wise flood hazard area
Restricted
Not suitable
Most Suitable
Legend
3
4
5
6
7
Distric
t
Larkan
a
Khairp
ur
Jacoba
bad
Sukkur
Naush
ahro
Feroze
Shikar
pur
Ghotki
Very
high
High
Moder
ate
Low
Very
low
1335
3087
1706
945
439
1079
1218
4182
9316
944
2788
1840
41
941
792
905
1572
1380
666
1976
6
0
0
710
1064
770
20
355
1647
2351
2070
Received: 3 February 2006 / Accepted: 13 October 2006 / Published online: 8 November 2006
# Springer Science + Business Media B.V. 2006
Abstract Floods are a common feature in rapidly urbanizing Dhaka and its adjoining areas.
Though Greater Dhaka experiences flood almost in every year, flood management policies
are mostly based on structural options including flood walls, dykes, embankments etc.
Many shortcomings of the existing flood management systems are reported in numerous
literatures. The objective of this paper is to assess flood hazard in Greater Dhaka for the
historical flood event of 1998 using Synthetic Aperture Radar (SAR) data with GIS data.
Flood-affected frequency and flood depth calculated from the multi-date SAR imageries
were used as hydrologic parameters. Elevation heights, land cover classification,
geomorphic division and drainage network data generated from optical remote sensing
and analogue maps were used through GIS approach. Using a ranking matrix in three
dimensional multiplication mode, flood hazard was assessed. All possible combination of
flood hazard maps was prepared using land-cover, geomorphology and elevation heights for
flood-affected frequency and floodwater depth. Using two hazard maps which produced the
highest congruence for flood frequency and flood depth, a new flood hazard map was
developed by considering the interactive effect of flood-affected frequency and floodwater
depth, simultaneously. This new hazard map can provide more safety for flood countermeasures
because pixels belonging to higher hazard degrees were increased due to the
Water Resour Manage (2007) 21:16011612
DOI 10.1007/s11269-006-9116-1
A. M. Dewan (*)
Department of Geography and Enviornment, University of Dhaka,
Dhaka 1000, Bangladesh
e-mail: amd1971@univdhaka.edu
M. M. Islam
Department of Civil Engineering, International University of Business Agriculture and Technology,
4 Embankment Drive Road, Sector 10 Uttara, Dhaka 1230, Bangladesh
T. Kumamoto
Department of Earth Science, Faculty of Science, Okayama University,
3-1-1 Tsushima Naka, Okayama 700-8530, Japan
M. Nishigaki
Department of Environmental Design and Civil Engineering, Okayama University,
3-1-1 Tsushima Naka, Okayama 700-8530, Japan
consideration of higher degrees of ranks. The estimation of flood hazard areas revealed that
a major portion of Greater Dhaka comprised moderate to very high hazard zone. Only a
little portion (8.04%) was found to be the least vulnerable to potential flood hazard.
Conversely, 28.70% of Greater Dhaka was found within very high hazard zone. Based on
this study, comprehensive flood hazard management strategies for land use planning
decision were proposed for the efficient management of future flood disasters.
Key words greater Dhaka . GIS . flood hazard . SAR . flood frequency . floodwater depth
1 Introduction
During the last two decades, Remote Sensing (RS) has been played an important role in the
fields of hydrology and water resources management (Bastiaansen 1998). Geographic
Information System (GIS) on the other hand has also been used extensively to model
surface water, particularly flood and associated damage (Boyle et al. 1998; Green and
Cruise 1995; Werner 2001). Recently, the integration capabilities of satellite data with GIS
have opened up opportunities for quantitative analysis of hydrological events, such as
flood, at all geographic and spatial scales. Remote sensing application is however,
considered imperative for third world countries because it is difficult for government to
update their database frequently with the ground observation techniques because of time
and cost associated with traditional methods (Dong et al. 1997). The potential of this
facility for improving hazard evaluation and risk reduction is continually being explored
(Carrara and Guzzetti 1993).
Floods are a regular feature in Greater Dhaka, the capital of Bangladesh during the
monsoon season. Dhaka has experienced many disastrous floods in the past of which the
1998 flood is said to be the worst in memorable records (Faisal et al. 1999; Dewan et al.
2004). The 1998 flood resulted in severe damage and untold sufferings to the inhabitants of
Dhaka and adjoining areas. For example, damage on water supply and sanitation sector
alone caused massive monetary loss worth US $ 10 million (Institute of Flood Control and
Drainage Research (IFCDR) 1998). Furthermore, 262,000 houses were fully or partially
collapsed (Islam 1998) and 3.7 million people were directly affected by the 1998 flood
(Jahan 2000). Though various measures including flood walls, embankments have been
constructed, the loss of properties and causalities are on the rise (Mohit and Akhter 2000).
Available data suggests that the Bangladesh government, in association with international
agencies, have spent 141.6 million US $ on flood control measures to solve the perennial
flood problem in Dhaka, but the annual flood damage and flood affected areas increased
considerably (Mohit and Akhter 2000). For example, a total of 4.55 million people were
affected by the 1998 flood (Hye 2000) and most of the part of Greater Dhaka was under
water of various depths for more than 8 weeks (Jahan 2000). It is necessary to mention here
that the construction of embankments fosters to the encroachment of floodplains in Dhaka
(Chowdhury et al. 1998). In addition, Rasid and Mallik (1996) identified several
environmental problems linked with the embankment in Dhaka.
Even though the occurrences of floods cannot be prevented, the negative consequences
can be lessened by an integrated approach to disaster management. Disaster management
includes four elements such as mitigation, preparedness, response and recovery (Quarantelli
1991). Literature suggests that flood control works and post disaster recovery are the
primary approaches for flood management in Greater Dhaka. It has become apparent in
1602 Water Resour Manage (2007) 21:16011612
1998 that such an approach is inadequate to combat recurrent flood problems. It is also
clear that the growing flood hazards of Dhaka have not been analyzed for purposes of
disaster mitigation (Mitchell 1999). As Dhaka contributes 16.7% of GDP to the national
development (Bangladesh Bureau of Statistics (BBS) 1998), the potential loss of GDP due
to flood is very high. Experts fear that the flood vulnerability of Dhaka is likely to
exacerbate in the effect of the probable climatic change. Thus, following the 1998
devastation flood monitoring, hazard assessment and disaster relief have been identified to
be the major issues by the water experts, urban planners and key policy makers of the
country. They have called for the accurate mapping of flood prone areas (Islam 1998),
assessment of flood hazard (Nishat 1998) and the development flood risk maps for land-use
planning (Hossain 1998).
Considering the above fact, the objective of this article is to assess flood hazard in
Greater Dhaka of Bangladesh for the historical flood event of 1998 using Synthetic
Aperture Radar (SAR) data with GIS. A land-use planning decision for the mitigation of
potential flood disaster is also put forward on the basis of this study.
2 The Study Area
The study area was the Greater Dhaka of Bangladesh (Figure 1). The latitudes and
longitudes of the study area are 2368 N 9033 E and 2390 N, 9050 E, respectively.
The study area is located mainly on an alluvial terrace, known as the Modhupur terrace of
the Pleistocene period. Topographically, Dhaka is a flat land. The surface elevation of the
area ranges between 1 and 14 m (Flood Action Plan (FAP) 8A 1991). It lies in the subtropical
monsoon zone and experiences humid climatic conditions. The City experiences
about 2000 mm annual rainfall, of which more than 80% occurs during the monsoon
season. Historically, the city is endowed with rivers, numerous khals (ephemeral water
bodies) and canals that drain water from the upper reaches during the monsoon season. As
population increased, these areas have been encroached on, thus compounding flood
problems. In the Figure 1, it is depicted that the urban areas in Greater Dhaka have
increased in 1996 and 2000 over the year 1990.
3 Data Preparation
Six RADARSAT SAR (Synthetic Aperture Radar) (Narrow and Wide Beam mode)
imageries were acquired that covered the entire flood season of 1998 (JulySept). The
selected SAR data were justified by analyzing the water level records of five monitoring
stations in and around Greater Dhaka (Dewan et al. 2005). The SAR data were despeckled
using the GAMA-MAP filter with 55 window sizes. After suppressing the speckle
inherent to SAR images, geometric correction was carried out using a referenced Landsat
TM image of 1999 covered the same area until the root mean square errors resulted in less
than one pixel. A second order polynomial fit was applied and the pixel values were
resampled. A pixel of SAR data covers 50 m for narrow and 100 m for wide beam mode in
beneath. This is the ground resolution of the SAR data.
The major impediment to study flood hazard in the study area was the unavailability of
GIS data in digital format. Therefore, in this study, a number of GIS layers were generated.
A Digital elevation model (DEM) of Greater Dhaka area was obtained from the Institute of
Water Resour Manage (2007) 21:16011612 1603
Figure 1 Extent of the study area.
1604 Water Resour Manage (2007) 21:16011612
Water Modeling, Bangladesh (IWM). Using the DEM, an elevation map was constructed.
The elevation data were grouped into 14 categories, at 1 m intervals. The land-cover map
was generated in a GIS platform, using one IRS-1D PAN data which was acquired during
the dry season of 29 February 2000 (path 110 and row 055). Field data and the recent
topographic map (scale 1:10,000) were aided to develop land-cover data. A geomorphic
map was also developed by using a Landsat TM of 1999. Training areas were selected by
on the TM Image by using the available analogue maps and field investigation. An
administrative division map for Greater Dhaka was also digitized using the recent
topographic map (scale 1:10,000). A total of 34 administrative boundaries were obtained
in the study area. In addition, the drainage network of Greater Dhaka was generated from
IRS-1D PAN data. Finally, all GIS data were integrated with geometrically corrected SAR
data within a GIS approach. The Bangladesh Transverse Mercator System (BTM) was used
as the coordinate system. Each of the GIS and SAR images produced 166,560 pixels in a
common coordinate system and each pixel covered 50 m on the ground.
4 Analysis of Flood-Affected Frequency
The concept of flood-affected frequency (Islam and Sado 2000) was used to develop flood
hazard maps. A flood frequency map based on flood duration was developed using the
multi-date SAR imagery. The classified imageries obtained by the threshold algorithm
(Dewan and Nishigaki 2004) were used to generate a flood frequency map. Total six
available SAR images during the flood, which were acquired on July 7 and 31 1998;
August 10 and 25 1998; and September 10 and 17 1998, were classified into water and
non-water areas. Initially, the classified images of July 7 and 31 were combined to construct
a single classified image of July; similarly two images of August and two images of
September were combined to construct the single image for August and September,
respectively. These combinations provided an opportunity to get a common boundary
needed to develop a flood-affected frequency map. Finally, these three images were
superimposed to characterize the inundated areas. The inundated area that appeared in all
the images was considered the highly damaged area. The common inundated areas that
appeared in two and one of the three images were deemed as medium damaged and low
damaged areas, respectively. Non-inundated area in all the images was considered as
non-damaged areas. Thus, four flood frequency categories were obtained, corresponding
to damage rankings of class 4, class 3, class 2 and class 1 as high, medium, low and
non-damaged areas, respectively.
5 Analysis of Floodwater Depth
Multi-temporal SAR imageries were also used to estimate the floodwater depth. It is very
difficult to obtain flood-depth information from SAR data. However, in this study we used
the technique suggested by Islam and Sado (2000), to estimate floodwater depth from SAR
images. First, a land level map of Greater Dhaka was extracted from the DEM that consist
four height classes. This land level map was used to collect representative ground truth data
for each unit. Using a Global Positioning System (GPS), a total of 100 signatures (25 for
each category) were collected and plotted on the land level map. Thus, a ground truth map
was created and subsequently brought into GIS. These signatures were then used to extract
Water Resour Manage (2007) 21:16011612 1605
training pixels. To do so, 7 July, 25 August and 10 September 1998 images were considered
that comprised early to peak flooding of 1998. Signatures were collected by superimposing
the ground truth map and a digital elevation model on each SAR image. In addition, visual
interpretation and grey level differences of digital number (DN) values were also helpful to
determine the selection of training pixels. A minimum distances algorithm of supervised
classification was then applied to classify SAR data into four flood depth categories, i.e.,
deep, medium, shallow and no-flooding. If deep floodwater depth observed in a single
image then it was regarded as deep depth, and if medium depth appeared in any single
image for the area but that was represented as shallow depth area by other two images, then
it was considered as medium depth. After the classification of deep and medium, the remaining
inundated pixels were considered as shallow depth. Non-inundated areas were considered as
noflooding
areas. Thus, four flood depth categories were obtained, corresponding to depth rankings
of class 1, class 2, class 3 and class 4 as no-flooding, shallow, medium and deep depth,
respectively. It is necessary to mention here that the DEM was not considered for the estimation
of flood depth because of its coarse spatial resolution (300 m). However, to derive flood depth,
we checked and used benchmark points to get training areas from the DEMand field data, which
was eventually used to develop flood depth categories from SAR data.
In order to evaluate the classified results of floodwater depth, derived depth map was
further authenticated by a flood depth map which was developed in Institute of Water
Modeling (IWM) (2000) using a hydrodynamic model. A comparison was made to
understand the agreement between two floodwater depth maps. It is found that deep flooded
category, for instance, covered 20.95% area in the depth map produced by IWM, whilst the
same category covered 21.21% area in the floodwater depth map developed in this study.
6 Flood Hazard Assessment
Even though satellite remote sensing has shown great potentiality for hazard assessment
(McKean
et al. 1991), the big challenge is the lack of generally accepted methods for producing hazard
maps or even on the scope of producing such maps (Rhoads 1986). Moreover, many well
developed methods usually applied in developed nations may not be accessible to developing
country particularly in Bangladesh, due to the lack of digital data or restriction to access to data
sources. For example, DEM based detail flood mapping is not possible for the entire Bangladesh
since high resolution digital elevation data of the whole country or part of it, is not available.
Therefore, a simple procedure is adopted in this paper for flood hazard assessment.
To assess flood hazard, a model was constructed in Model Maker utility of ERDAS
IMAGINE (v. 8.6) software. In order to assess the flood hazard for each category of land cover,
geomorphology and elevation units, a weighted score was estimated and hazard ranks were
decided using the technique suggested by Rahman et al. (1991). Following steps were involved
in the development of flood hazard maps. Step 1: Flood-affected frequency and floodwater
depth maps were superimposed on thematic layers to estimate the percentage of area occupied
by each category of flood-affected frequency and floodwater depth. We used land cover data
with elevation height and geomorphic data to signify the different types of land covers. Because
of the importance of the land use, value and the significance uses of different types of land cover
are not same. Hydrological parameters like flood frequency and flood depth maps obtained from
remote sensing data were superimposed onto the land cover, elevation height and
geomorphologic
maps to view the interactive effect between hydrological parameters and thematic
information.
1606 Water Resour Manage (2007) 21:16011612
Step 2: A weighted score for the acquired area percentage for each category of land
cover, elevation, and geomorphic unit was estimated using the following equation:
Weighted Score Class 1 _ 1:0 Class 2 _ 3:0 Class 3 _ 5:0 Class 4 _ 7:0 1
Where, class 1 to class 4 represents the area percentage occupied by the categories of each
hydraulic component i.e., flood-affected frequency and floodwater depth, for each ID
category of the respective GIS component. The coefficients 1.0, 3.0, 5.0 and 7.0 in Eq. 1
were used to describe the weight for the flood loss. Calculated weighted score and their
corresponding hazard ranks for each geomorphic category for flood-affected frequency is
presented in Table I. Step 3: After calculating the weighted score, points for each category
of land cover, elevation and geomorphology was estimated on the basis of a linear
interpolation between 0 and 10, where 0 represents the lowest score and 10 represents the
highest score. This represents the degree of flood hazard. Step 4: On the basis of the
estimated points, hazard ranks were determined. In order to quantify flood hazard, three
hazard ranks were established. For example, hazard ranks were fixed by assigning 13.5 for
hazard rank 1, 3.56.5 for hazard rank 2 and 6.510.0 as hazard rank 3. The higher the
ranking value, the more susceptible that particular class is considered to be to the
occurrence of flood. Step 5: Using three dimensional multiplication mode suggested by
Islam and Sado (2000), assessment of flood hazard was performed for flood-affected
frequency and floodwater depth. The resulting flood hazard maps were further reclassified
according to the intensity of hazards. Thus, five distinct hazard zones corresponded to least,
less, moderate, high and very high were obtained. To illustrate flood hazard on the basis of
mapping unit, three dimensional multiplication mode applied for three thematic maps.
Two hazard maps were developed for flood-affected frequency and floodwater depth by
considering three dimensional multiplication mode. Although the two flood hazard maps
showed only 34% congruence, these maps provide hazard information for flood frequency
and floodwater depth, independently. They cannot represent higher hazard ranks for two
hydraulic parameters at the same time. Moreover, these maps may not be perceptible to
many people since they represent two hydrologic parameters. Therefore, a new hazard map
was constructed using the interactive effect of flood-affected frequency and floodwater
depth simultaneously, which can provide more safety for flood countermeasures because
Table I Flood hazard ranks for geomorphological division by flood-affected frequency (ID: geomorphic
identification number; HR: hazard rank)
ID Class 1 Class 2 Class 3 Class 4 Score Point HR 13
1 38.94 23.86 13.18 24.02 344.58 5.01 2
2 83.08 10.01 3.26 3.66 154.99 2.25 1
3 28.63 17.86 15.01 38.50 426.76 6.20 2
4 0.44 0.80 2.93 95.82 688.27 10.00 3
5 56.58 13.67 8.74 21.01 288.37 4.19 2
6 63.35 17.26 9.52 9.88 231.85 3.37 1
7 47.83 15.69 11.03 25.46 328.24 4.77 2
8 0.55 0.95 3.18 95.32 686.55 9.98 3
9 23.18 16.45 12.73 47.63 469.64 6.82 3
10 42.79 23.41 12.53 21.27 324.55 4.72 2
11 5.91 5.40 6.20 82.49 630.55 9.16 3
12 9.66 7.43 7.59 75.32 597.13 8.68 3
13 55.36 7.14 5.36 32.14 328.57 4.77 2
Water Resour Manage (2007) 21:16011612 1607
Figure 2 Flood hazard map developed using land-cover, elevation and geomorphic categories for floodaffected
frequency and floodwater depth simultaneously.
1608 Water Resour Manage (2007) 21:16011612
the pixels belonging to higher ranks were increased due to the consideration of higher
hazard degrees. If one hazard map shows the higher rank than the other for a pixel, then the
higher rank was assigned for that pixel. The final hazard map is shown in Figure 2.
The hazard area estimation using the final flood hazard map developed by the interactive
effect of flood frequency and floodwater depth concurrently revealed that only 8.04% areas
were within the least hazard zone (Table II) that corresponded to the non-flooded areas in
Greater Dhaka. Highly elevated lands (>13 m) of the study area represented the least hazard
zone. Moderately high lands were within the group of less hazard zone and constituted
21.65%. These are the most extensively developed area in Greater Dhaka. Urban peripheral
zones of Greater Dhaka were characterized by moderate hazard zone (15.05%). High
hazard (26.56%) and very high hazards (28.70%) were assigned to cells where higher
degrees of ranks observed. Extreme lowlands and areas adjacent to river banks are
designated as very high hazard zone.
7 Disaster Management Policy for Land-Use Planning
The developed flood hazard map further can be incorporated to land-use planning
decisions. This may help to reduce potential flood damage in Greater Dhaka. Micro scale
flood management plan can be of little help in the effort of flood damage abatement. A
comprehensive plan addressing flood hazard management is therefore, necessary. This plan
should combine land use strategies for each zone with the careful consideration of certain
structural controls. This can be achieved by the minimal disruption of natural environments.
Table III presents an example of general management strategies based on this study and
rigorous field investigation. These strategies could serve as basic components in a
comprehensive flood management plan for Greater Dhaka.
Infilling of natural channels and lowlands is the primary cause for increasing flood
hazard in Greater Dhaka. Therefore, preservation of natural channels and infiltration
processes should be the top priority when formulating policy. It is postulated that during
urban developments small channels are often viewed as being insignificant and are
destroyed, but in reality they have very vital role in easing flood peaks (Dunne and Leopold
1978 cited in Rhoads 1986). Greater Dhaka is endowed with natural channels and canals
that used to drain floodwater. Due to the increasing population pressure, those canals have
lost their ability to drain floodwater during the monsoon. Therefore, further infilling of
those channels must be prohibited and every effort should be made to prevent the
destruction of any channel regardless of size. Human settlements should not be allowed in
those areas comprised of high and very hazard zones. If necessary, structures must be flood
Table II Areas in Greater Dhaka with various degree of flood hazard
Hazard zone (s) Area (ha) Percentage (%)
Least 3348.50 8.04
Less 9013.75 21.65
Moderate 6266.75 15.05
High 11061.00 26.56
Very high 11950.00 28.70
Total 41640.00 100.00
Water Resour Manage (2007) 21:16011612 1609
proofed and must be put in elevated places. Proper maintenance of the existing storm water
facilities in the least and less hazard zones should be another priority.
8 Conclusions
In this paper, flood hazard assessment for the historical flood event of 1998 in Greater
Dhaka of Bangladesh was examined using an integrated approach of GIS and remote
sensing. The study demonstrated a simple and cost effective way to use geographical
information system for creating flood hazard maps from the available dataset. Although it
was very time consuming to derive the GIS database because of the size of the study area,
GIS was invaluable in reducing the complexity associated with hazard assessment.
Flood hazard maps by considering the interactive effect of land cover, elevation, and
geomorphic units for flood affected frequency and floodwater depth were constructed. The
hazard maps showed that a major portion of Dhaka were within moderate to very high
hazard zone, especially fringe areas. It is projected that these areas will be fully urbanized
by the year 2010, therefore higher priorities must be paid for the development of apposite
flood countermeasures for areas having higher hazard potential. Comprehensive land-use
policies for flood disaster management have been proposed on the basis of this study. It is
anticipated that proposed land-use policies can contribute to effective flood forecasting,
relief and emergency management for future flood event.
Table III General flood hazard management strategies Greater Dhaka of Bangladesh
General management strategies
Least hazard zone
(i) High development densities allowable provided that stormwater retention basins are adequately
constructed. It also requires of their proper maintenance.
Less hazard zone
(i) Moderate development densities can be maintained.
(ii) Require structures should be put on higher places.
(iii) Should have separate drainages facilities wherever possible.
(iv) Existing natural channels should be kept for flood flow.
(v) Require proper maintenance and cleaning of storm water facilities on timely basis.
Moderate and high hazard zone
(i) Detail hydrologic analysis of flood characteristics must be performed prior to allocate for the new
development.
(ii) Low development densities may be allowed on condition.
(iii) Structures must be flood-proofed.
(iv) Adequate drainage facilities of floodwaters must be secured.
(v) Require maintenance of drainage facilities, natural channels, canals and khals.
(vi) Detail flood zoning should be conducted and require to introduce proper land development rules.
(vii) Should have emergency evacuation facilities and adequate flood shelters.
(viii) Community based flood management program should be introduced.
Very high hazard zone
(i) Must be kept open for flood conveyance as permanent natural open space.
(ii) Any kind of developments must be prohibited.
(iii) Structures must not be allowed.
(iv) Require preservation of natural drainage systems, existing floodplains.
1610 Water Resour Manage (2007) 21:16011612
the vacant land of the City by filling up of natural channels and floodplains. Consequently,
flood risk is increasing. In order to ameliorate flood induced damage, the developed flood
hazard map would be invaluable. Government can use hazard maps for ensuring the proper
development planning of the very high and high hazard zones which is suppose to be
urbanized by the year 2010. Urban planners can use this information to make
environmentally sound land-use decisions. One more advantage is that the hazard maps
can be quickly updated if modification, such as changes in land use, occurs subsequent to
the original study. Furthermore, this type of study will provide information about flood
protection measure such as construction and development of infrastructure and preparedness
of aid and relief operation for high hazard areas for future flood event.
References
Bangladesh Bureau of Statistics (BBS) (1998) Statistical yearbook sof Bangladesh. Ministry of Planning,
Dhaka, p 224
Bastiaansen WGM (1998) Remote sensing in water resources management: the state of the art, International
Water Management Institute (IWMI), Colombo
Boyle SJ, Tsanis IK, Kanaroglou PS (1998) Developing geographic information system for land use impact
assessment in flooding condition. J Water Resour Plan Manage 124:8998
Carrara A, Guzzetti F (eds) (1993) GIS in assessing natural hazards. Kluwer, Italy
Chowdhury JU, Rahman R, Bala SK, Islam AKM Saiful (1998) Impact of 1998 flood on Dhaka City and
performance of flood control works. Institute of Flood Control and Drainage Research, Dhaka, p 76
Dewan AM, Nishigaki M (2004) Using Synthetic Aperture Radar with GIS for flood damage estimation, in
Proceedings of the 6th Annual Meeting of the Japan Society for Disaster Information Studies, Japan
Society for Disaster Information Studies, Tokyo
Dewan AM, Nishigaki M, Komatsu M (2004) Use of multi-temporal RADARSAT-1 ScanSAR data to
differentiate between flooded and non-flooded areas, in Annual Meeting of the Chugoku Branch of the
Japan Society of Civil Engineers, Tottori, Japan
Dewan AM, Yeboah KK, Nishigaki M (2005) Flood mapping and damage evaluation in Greater Dhaka,
Bangladesh with remote sensing. Asian Profile 33(5):495512
Dong Y, Forster B, Ticehurst C (1997) Radar backscatter analysis for urban environments. Int J Remote Sens
18:13511364
Faisal IM, Kabir MR, Nishat A (1999) Non-structural flood mitigation measures for Dhaka City. Urban
Water 1:145153
Flood Action Plan (FAP) 8A (1991) Master plan study for Greater Dhaka protection project. Japan
International Cooperation Agency, Dhaka
Green RG, Cruise JF (1995) Urban watershed modeling using geographic information system. J Water
Resour Plan Manage 121:318325
Hossain MM (1998) Bangladesh floods and its management initiatives. Steps Towards Development 3:1219
Hye A (2000) Effects on livelihood. In: Nishat A, Reazuddin M, Amin R, Khan AR (eds) The 1998 flood:
impact on environment of Dhaka City. Department of Environment and IUCN, Dhaka
Institute of Flood Control and Drainage Research (IFCDR) (1998) Impact of 1998 flood on Dhaka City and
performance of flood control works, Dhaka
Institute of Water Modeling (IWM) (2000) Baseline information study of the Dhaka combined flood control
cum eastern bypass road project, Final Report, Dhaka. p 86
Islam N (1998) Flood 98 and the future of urban settlements in Bangladesh. National Seminar on Flood 98
and Management of Floods in Future, 8th December 1998, Dhaka
Islam MM, Sado S (2000) Development of flood hazard maps of Bangladesh using NOAA-AVHRR images
with GIS. Hydrol Sci J 45:337355
Jahan S (2000) Coping with flood: the experience of the people of Dhaka during the 1998 flood disaster.
Aust J Emerg Manag 15(3):1620 (Spring)
McKean J, Buechel S, Gaydo L (1991) Remote sensing and landslide hazard assessment. Photo Eng Rem
Sen 57(9):11851193
Mitchell JK (1999) Megacities and natural disasters: a comparative analysis. GeoJournal 49:137142
Water Resour Manage (2007) 21:16011612 1611
Mohit MA, Akhter S (2000) Delineation of flood damaged zones of Dhaka City based on the 1998 flood by
using GIS. In: Ali MA, Seraj SM, Ahmed S (eds) Engineering concerns of flood. Bangladesh University
of Engineering and Technology, Dhaka, pp303318
Nishat A (1998) A discussion on flood management in Bangladesh. The 1998 Deluge-Developing Coping
Capacities, Special Workshop on the 1998 flood, Nagorik Durjog-Mokabela Uddogh, Dhaka,
Bangladesh
Quarantelli EL (1991) Some aspects of disaster planning in developing countries, in disaster management
and regional development planning with Peoples Participation, Vol. II, United Nations Center for
Regional Development, Dhaka
Rahman NM, Ochi S, Murai S, Hashimoto T, Kakiuchi H (1991) Flood risk mapping in Bangladesh-flood
disaster management using remote sensing and GIS, in application of remote sensing in Asia and
Oceania-environmental change monitoring, Asian Association of Remote Sensing, Tokyo
Rasid H, Mallik A (1996) Living on the edge of stagnant water: an assessment of environmental impacts of
construction-phase drainage congestion along Dhaka City Flood Control Embankment, Bangladesh.
http://ppg.sagepub.com/
Abstract: In the event of a natural disaster, remote sensing is a valuable source of spatial information
and its utility has been proven on many occasions around the world. However, there are many
different types of hazards experienced worldwide on an annual basis and their remote sensing
solutions are equally varied. This paper addresses a number of data types and image processing
techniques used to map and monitor earthquakes, faulting, volcanic activity, landslides, fl ooding, and
wildfi re, and the damages associated with each. Remote sensing is currently used operationally for
some monitoring programs, though there are also diffi culties associated with the rapid acquisition
of data and provision of a robust product to emergency services as an end-user. The current
status of remote sensing as a rapid-response data source is discussed, and some perspectives given
on emerging airborne and satellite technologies.
Key words: image processing, natural hazards, optical, remote sensing, SAR, thermal.
_
*Author for correspondence. Email: k.joyce@gns.cri.nz
The Author(s), 2009. Reprints and permissions:
http://www.sagepub.co.uk/journalsPermissions.nav
I Natural hazards and disasters
The use of remote sensing within the
domain of natural hazards and disasters has
become increasingly common, due in part to
increased awareness of environmental issues
such as climate change, but also to the increase
in geospatial technologies and the
ability to provide up-to-date imagery to the
public through the media and internet. As
technology is enhanced, demand and expectations
increase for near-real-time monitoring
and visual images to be relayed to emergency
services and the public in the event of a
natural disaster. Recent improvements to
earth monitoring satellites are paving the way
to supply the demand. Techniques needed
to exploit the available data effectively and
rapidly must be developed concurrently,
to ensure the best possible intelligence is
reaching emergency services and decisionmakers
in a timely manner.
A comprehensive review of remote sensing
for some natural hazards was completed by
Tralli et al. (2005), and Gillespie et al. (2007)
Downloaded from ppg.sagepub.com by Giyanto Giyanto on October 11, 2010
Karen E. Joyce et al.: A review of satellite remote sensing and image processing techniques
185
Satellite Sensor Swath (km) Nadir spatial
resolution (m)
Revisit capability
SPOT-5 Panchromatic 6080 5 11 times every 26 days
Multispectral 6080 10
Kompsat Panchromatic 15 1 23 days
Multispectral 15 1
Landsat-5 TM Multispectral 185 30 Every 16 days
TM Thermal 185 120
Landsat-7* ETM+ Panchromatic 185 15 Every 16 days
ETM+ Multispectral 185 30
ETM+ Thermal 185 60
NOAA AVHRR 2399 1100 Several times per day
Envisat MERIS 575 300 23 days
Radarsat-2 Ultra-fi ne 20 3 Every few days
Radarsat-1/-2 Fine 50 8
Radarsat-2 Quad-pol fi ne 25 8
Radarsat-1/-2 Standard 100 25
Radarsat-2 Quad-pol standard 25 25
Radarsat-1 Wide 150 30
Radarsat-1/-2 ScanSAR narrow 300 50
Radarsat-1/-2 ScanSAR wide 500 100
Radarsat-1/-2 Extended high 75 25
Radarsat-1 Extended low 170 35
ERS-2 100 30 35-day repeat cycle
Envisat ASAR standard 100 30 36-day repeat cycle
ASAR ScanSAR 405 1000
TerraSAR-X Spotlight 10 1 11-day repeat cycle;
Stripmap 30 3 2.5-day revisit capability
ScanSAR 100 18
Cosmo-Skymed^ Spotlight 10 <1 ~37 hours
Stripmap 40 315
ScanSAR 100-200 30100
*Landsat-7 nearing the end of its useful life; problems with scan line corrector resulting in data gaps
^Figures quoted for one satellite in constellation
Table 1 Continued
and infrastructure.
1 Optical detection of earthquakes and faulting
The technique of choice in the use of remote
sensing for fault mapping with optical data
is manual interpretation, regardless of the
data source (Fu et al., 2004; 2005; Walker,
2006; Walker et al., 2007). Frequently, the
effects of earthquake activity and faulting are
not manifested in spectral variations within
image data, but in topographical changes.
Image interpretation therefore relies on the
expertise of the analyst, rather than spectral
classifi ers. It is possible that this fi eld could
benefit from the use of filters specifically
designed to detect linear features. Note that
fault detection is more of an exercise in preparedness
than rapid response.
A number of different techniques have
been reported in the literature to map the
extent of earthquake damage, particularly in
urban areas. Image differencing of multidate
spectral ratios demonstrated better results
than synthetic aperture radar (SAR) in Turkey
and Iran, though a combination of optical and
SAR coherence was reported to give the
most accurate result (Stramondo et al.,
2006). Sertel et al. (2007) used semivariogram
analysis of SPOT panchromatic imagery
obtained both before and after the Izmit
earthquake in Turkey. This technique demonstrated
the possibility of mapping earthquake
severity based on changes in the shape
of semivariograms, although further research
was suggested before the relationship to
a quantifi able amount of damage could be
determined. It may also be of use where the
spatial resolution of the image data is sufficient
to detect textural changes, though
insuffi cient to detect specifi c damages. See
also section IV for details of landslide mapping
as a result of earthquake damage.
2 Thermal and microwave detection of
earthquakes and faulting
As an alternative to mapping earthquake
damage, several studies have sought to characterize
short-term temperature increases
immediately prior to earthquakes. While the
detection of thermal anomalies has thus far
been conducted retrospectively, refi nement
of this technique and routine investigation
may hold information key to earthquake
prediction and warnings. A normal temperature
for a region can be calculated using
a time series of image data and an image of
interest compared with this to determine
Karen E. Joyce et al.: A review of satellite remote sensing and image processing techniques
187
Table 2 Remotely sensed data types and image processing techniques for
information extraction about natural hazards
Data type Sensor examples Technique Application Advantages Disadvantages
Multispectral
high to
moderate
resolution
Ikonos,
Quickbird,
SPOT, ASTER,
ALOS
Manual
interpretation
Infrastructure and
property damage
due to fl ooding,
earthquakes,
landslides, etc
Benefi ts from analysts knowledge
of the area in addition to other
interpretation cues such as context,
site, association, shape, size;
immediate vector output fi le
Can be subjective, time-consuming
for widespread events, and nonrepeatable
Spectral
classifi cation
Location and
extent of fl ooding,
landslides, volcanic
debris, fi re scars
Relatively rapid to apply over a large
area
Non-unique spectral response
values, may require additional
manual editing, appropriate
algorithm must be selected for
optimal result
Semivariogram
analysis and
other textural
classifi ers
Damage due to
earthquakes; location
of landslides
May be useful when spatial resolution
is lower than desired
Only returns relative damage
estimates
Image
thresholding
(including band
ratios)
Location and
extent of fl ooding,
landslides, volcanic
debris, fi re scars
Simple and rapid to apply, band ratios
reduce illumination variability, can be
applied with panchromatic data
Determination of threshold values
may be subjective
Image
differencing
Location and
extent of fl ooding,
landslides, volcanic
debris, fi re scars
Can be conducted on panchromatic
data, band ratios or SAR backscatter
imagery
Requires before and after imagery
that is accurately co-registered
and radiometrically balanced, only
takes the spectral information from
a single band (though this may be
a ratio combination), all changes
will be identifi ed regardless of their
relevance to the particular natural
hazard (eg, crop rotations); still
need to determine a threshold of
change
(Continued)
Downloaded from ppg.sagepub.com by Giyanto Giyanto on October 11, 2010
Karen E. Joyce et al.: A review of satellite remote sensing and image processing techniques
189
SAR JERS-1, ERS1/2, ENVISAT,
ALOS
PALSAR,
TerraSAR-X,
Radarsat-1/2,
Cosmo-SkyMed
Coherence Change detection
due to landsliding,
fl ooding, fi re, etc
Provides quantitative estimation of
ground changes
Does not work well in densely
vegetated regions, affected by
seasonal changes, accuracy
decreases with time
Backscatter
intensity
Change detection
due to landsliding,
fl ooding, fi re, etc
Can be used in cloudy conditions,
side-looking acquisition geometry is
benefi cial for certain applications
Quantitative analysis is complicated
and varies signifi cantly for different
regions, may be diffi cult to interpret
for non-experienced end-users
Interferometry/
DEM
generation
DEM is used as
supplementary
information in variety
of studies
Independent of weather conditions Accuracy depends on acquisition
geometry, wave-length and
coherence, side-looking acquisition
geometry creates distortion and
shadowed areas
Differential
interferometry
Surface deformation
due to volcanic or
tectonic activity;
velocity and extent
of slow moving
landslides
High precision, high resolution of
some new sensors
Dependent on spatial baseline and
DEM accuracy; cannot determine
difference between vertical and
horizontal components, high
Karen E. Joyce et al.: A review of satellite remote sensing and image processing techniques
191
this technique of hazard mapping and
monitoring is not appropriate for providing
information in a rapid-response emergency
situation due to the time it takes to acquire
and process the data to a point where it can
be manually interpreted.
III Volcanic activity
Detection and monitoring of volcanic activity
spans a number of different data types and
processing methods. Thermal anomalies are
commonly detected by comparing a location
with its background or average temperature;
volcanic deposits are best detected with
optical data, often using the normalized
difference vegetation index (NDVI) for
spectral enhancement; the split-window
method is used for detecting ash composition
within clouds; and InSAR is best for
volcanic deformations.
1 Optical detection of volcanic activity
A variety of sensors are available to provide
data suitable for debris mapping, with a
preference noted towards the higher spatial
Karen E. Joyce et al.: A review of satellite remote sensing and image processing techniques
193
landslide mapping (Singhroy, 1995; Singhroy
et al., 1998; Ostir et al., 2003; Nichol and
Wong, 2005a; Kumar et al., 2006; Voigt et al.,
2007). Orthophotography in particular has
demonstrated its utility for mapping landslides
in detail and IKONOS with pansharpening
has been suggested to be of
equivalent if not superior use (Nichol and
Wong, 2005b). Manual techniques benefi t
from the analysts knowledge of the area,
though they cannot be automated, and are
impracticably time-consuming when mapping
widespread numerous landslides. Some
emerging studies are attempting to use more
automated extraction techniques, utilizing
band ratios (Cheng et al., 2004; Rau et al.,
2007) and unsupervised (Dymond et al.,
2006) and supervised classifi cation (Nichol
and Wong, 2005b; Joyce et al., 2008a; 2008b)
to reduce the level of manual interpretation,
while still providing reasonably high accuracy
levels (up to 80%). In a study to test the most
accurate method for mapping landslides with
SPOT-5 imagery, it was determined that
the spectral angle mapper (SAM) supervised
classifi cation and NDVI thresholding were
the most accurate semi-automated techniques
compared with the results achieved
with parallelepiped, minimum distance to
means, principal components, and multitemporal
image differencing (Joyce et al.,
2008a; 2008b). This study also noted that
manual digitizing produced a higher accuracy
than any of the aforementioned techniques,
but was considerably more time-consuming
for a widespread event.
Multitemporal image analysis is a valuable
technique that can be used if imagery is
available both before and after a landsliding
event. This is perhaps the most promising
option for rapid response. The process
applied then relies on digital change detection
a number of methods of which are available
(Singh, 1989; Jensen, 1996). An overall accuracy
of approximately 70% was achieved
using the postclassification comparison
technique for landslide detection in Hong
Kong (Nichol and Wong, 2005b). While
apparently effective, and commonly used in
other areas of interest, this method is not
documented frequently in landslide detection
literature. An alternative technique requires
Karen E. Joyce et al.: A review of satellite remote sensing and image processing techniques
195
visible under a closed vegetation canopy. The
utility of high temporal resolution sensors
such as AVHRR is realized by Sandholt
et al. (2003), who state that although the
spatial resolution is coarser than many other
satellite sensors the frequent revisit time
offers a greater probability of obtaining cloudfree
imagery. They used linear spectral unmixing
with thermal imagery to determine
inundated areas, but were faced with the
difficulty of selecting pure endmembers.
Alternatively, they also tested supervised
maximum likelihood and ISODATA clustering
classifi cations with the higher spatial
resolution Landsat ETM+, concluding that
no technique is necessarily better than the
other, rather that each has its advantages and
disadvantages depending on the flooding
extent, cloud cover and temporal variability.
Manual analysis of MISR imagery was
completed to determine quantitative characteristics
of the 2004 tsunami development
along the eastern coast of India (Garay and
Diner, 2007). This provided information about
wave propagation and behaviour, but was
Karen E. Joyce et al.: A review of satellite remote sensing and image processing techniques
197
similar to those used for detection of thermalbased
volcanic activity. Thermal differences
have also been noticed between burnt and
Karen E. Joyce et al.: A review of satellite remote sensing and image processing techniques
199
of acquiring early imagery since there was
an air traffi c ban in force over the USA that
was not lifted until 13 September. From
then on, aircraft-mounted sensors were
able to provide daily multispectral imagery,
thermal imagery, photographs, hyperspectral
imagery, and LiDAR. Some time later,
some aerial photographs were converted to
three-dimensional information by use of proprietary
software from Pictometery, which
can make all angles of a structure visible and
measurable. Such images would have been
useful for assessing hazards, eg, overhanging
debris, during the clean-up (Huyck and
Adams, 2002).
There are still many countries that do
not have access to direct reception stations
within their territory for medium to high
spatial resolution imagery. In an emergency,
commercial satellite services can be tasked
to collect data. Depending upon the position
of the particular satellite within its orbit, the
time interval between an urgent data programming
request and the fi rst acquisition
attempt could be as short as 24 hours or
as long as several days. As a general rule,
satellite services more commonly used for
emergencies are better at rapid response.
A prime example of such a satellite service
is Radarsat, which can typically schedule a
data acquisition at very short notice and then
supply the data to the data requester within
hours of a successful acquisition. Other
Table 3 Utility of various data types for providing information about natural hazards
Spectral Visible NIR SWIR Hyperspectral Thermal SAR LiDAR
Spatial (very high = <5 m;
high = 520 m; medium =
20250 m; coarse = >250 m)
Very high High Medium Coarse High Medium Coarse Very high Medium Coarse Medium Coarse Single
polarization
Polarimetric DEM
Sensor example Quickbird,
Ikonos
ASTER,
SPOT,
ALOS
Landsat MODIS,
AVHRR
ASTER,
SPOT
Landsat MODIS,
AVHRR
CASI,
Hymap
Hyperion MODIS ASTER,
Landsat
MODIS,
AVHRR
Radarsat-1,
ERS-1/2
Radarsat-2,
TerraSAR-X
Airborne
Sensor
Volcano Thermal
anomaly
< 100C
EEEEEEEEEEAAEEE
Thermal
anomaly
>100C
EEEEBBBBBBAAEEE
Thermal
anomaly
>1000C
BBBBBBBBBBAAEEE
Lahar B A A E E E E B B E B E B D B
Ash clouds
detection B B B B B B A E * B B B B E E E
Ash clouds
quantifi cationE E E E B B A E * B B B B E E E
Gas clouds
detection B B B B B B B E * B B B A E E E
Gas clouds
quantifi cationE E E E B B B E * B B B A E E E
Deformation E E E E E E E E E E E E A D A
Debris B B B C E E E B B C E E C D B
Lava fl ow B B B C B B C B B C B B C D B
Pyroclastic
fl ow B B B C B B C B B C B B C D B
Downloaded from ppg.sagepub.com by Giyanto Giyanto on October 11, 2010
Karen E. Joyce et al.: A review of satellite remote sensing and image processing techniques
201
Earthquakes
and faulting
(see also
fi res, fl ooding
and
landslides)
Fault location B C C E C C E B C E E E B E A
Deformation E E E E E E E E E E E E A C A
Aftermath
building
and property
damage
BBBCEEEBBCEECCB
Landslides Scar + debris
fl ow B A B C E E E B B E E E C D B
Isolate scar
from debris
fl ow
CCEEEEECCEEECDB
Flooding Inundated areaA A B C B B C B B C B C A A B
Aftermath
building
and property
damage
ABBCCCEBBEEECDC
Wildfi re Fire front B B B B B B A B B B B A E E E
Aftermath
building
and property
damage
BBBCBBCBBCEECDB
Landscape
scars B A A C E E E B B C E E C A B
A: Clearly demonstrated to work using standard image processing systems and is openly available in the literature
B: Shown to work with experimental image data sets or over limited areas with very small pixels or over global scales with large
pixels
C: If extent is bigger than several pixels
D: Not widely available in literature but theoretically should be a potential use
E: Not feasible
* Listed as not feasible because of aircraft restrictions on fl ying over volcanic ash/gas clouds, rather than sensor inability
Downloaded from ppg.sagepub.com by Giyanto Giyanto on October 11, 2010
References
Ambrosia, V.G. and Hinkley, E. 2008: NASA
science serving society: improving capabilities for fi re
characterization to effect reduction in disaster losses.
IEEE International Geoscience and Remote Sensing
Symposium, Boston, MA, 4, IV-628-631.
Begg, J.G. and Mouslpoulou, V. 2007: Rangitaiki
Plains: another veil is lifted. New LiDAR data from
Bay of Plenty, New Zealand. Geological Society of
New Zealand Miscellaneous Publication 123A, 9.
Belliss, S.E., Pairman, D. and McNeill, S.J. 1998:
Use of Radarsat data to map Landslide erosion in
Downloaded from ppg.sagepub.com by Giyanto Giyanto on October 11, 2010
Karen E. Joyce et al.: A review of satellite remote sensing and image processing techniques
203
steep landforms. Application Development and
Research Opportunity (ADRO) Final Symposium,
1315 October, Montreal, Quebec, Canada,
Electronic Proceedings.
Berardino, P., Fornaro, G., Lanari, R. and Sansoti,
E. 2002: A new algorithm for surface deformation
monitoring based on small baseline differential SAR
interferometry. IEEE Transactions on Geoscience and
Remote Sensing 41, 237583.
Bonn, F. and Dixon, R. 2005: Monitoring fl ood extent
and forecasting excess runoff risk with Radarsat-1
data. Natural Hazards 35, 37793.
Bourgeau-Chavez, L. and Kasischke, E. 2002:
Mapping fire scars in global boreal forest using
imaging radar data. International Journal of Remote
Sensing 23, 421134.
Calomarde, R. 1998: An application of remote sensing
and geographic information system (GIS): a case
study in Sto. Tomas-Marela River, Mt Pinatubo,
Zambales, Philippines. Asian Conference on Remote
Sensing, Manila.
Carn, S.A. and Oppenheimer, C. 2000: Remote
monitoring of Indonesian volcanoes using satellite
data from the Internet. International Journal of
Remote Sensing 21, 873910.
Cartwright, S. 2005: National Civil Defence Emergency
Management Plan Order 2005. Published
under the authority of the New Zealand Government,
Wellington, New Zealand, 68 pp.
Casson, B., Delacourt, C. and Allemand, P. 2005:
Contribution of multi-temporal remote sensing
images to characterize landslide slip surface application
to the La Clapiere landslide (France). Natural
Hazards and Earth System Sciences 5, 42537.
Castro, O.T. and Carranza, E.J.M. 2005: Remote
sensing of temporal variations in spatial distributions
of lahar and pyroclastic-fl ow deposits, West Mount
Pinatubo, Philippines. In Erasmi, S., Cyffka, B. and
Kappas, M., editors, Remote sensing and GIS for
environmental studies, volume 113, Gottingen:
Gottinger Geographische Abhandlungen, 22329.
Centre for Remote Sensing and Spatial Information
Science (CRSSIS) 2006: Coastal remote
sensing mapping capability matrix. Retrieved 17 May
2009 from http://www.gpem.uq.edu.au/CRSSIS/
tools/rstoolkit/pdf/Mapping%20capability%20mat
rix.pdf
Chen, R.F., Chang, K.J., Angelier, J., Chan, Y.C.,
Deffontaines, B., Lee, C.T. and Lin, M.L. 2006:
Topographical changes revealed by high-resolution
airborne LiDAR data: the 1999 Tsaoling landslide
induced by the Chi-Chi earthquake. Engineering
Geology 88, 16072.
Karen E. Joyce et al.: A review of satellite remote sensing and image processing techniques
205
Kaneko, T. and Wooster, M.J. 1999: Landsat infrared
analysis of fumarole activity at Unzen Volcano:
time-series comparison with gas and magma fl uxes.
Journal of Volcanology and Geothermal Research 89,
5764.
2005: Satellite thermal analysis of the 1986 IzuOshima lava flows. Journal of Volcanology and
Geothermal Research 148, 35571.
Kerle, N., Froger, J.L., Oppenheimer, C. and
de Vries, B.V. 2003: Remote sensing of the 1998
mudfl ow at Casita volcano, Nicaragua. International
Journal of Remote Sensing 24, 4791816.
Kumar, K.V., Martha, T.R. and Roy, P.S. 2006:
Mapping damage in the Jammu and Kashmir caused
by 8 October 2005 M-w 7.3 earthquake from the
Cartosat-1 and Resourcesat-1 imagery. International
Journal of Remote Sensing 27, 444959.
Kwoun, O.I., Lu, Z., Neal, K. and Wicks, C. 2006:
Quiescent deformation of the Aniakchak Caldera,
Alaska, mapped by InSAR. Geology 34, 58.
Lewis, A.J., Henderson, F.M. and Holcomb,
D.W. 1998: Radar fundamentals: the geoscience
perspective. In Henderson, F.M. and Lewis, A.J.,
Karen E. Joyce et al.: A review of satellite remote sensing and image processing techniques
207
Waite, W.P. and MacDonald, H.C. 1971: Vegetation
penetration with K-band radars. IEEE Transactions
on Geoscience and Remote Sensing GE-9, 14755.
Walker, R.T. 2006: A remote sensing study of active
folding and faulting in southern Kerman province,
SE Iran. Journal of Structural Geology 28, 65468.
Walker, R.T., Nissen, E., Molor, E. and Bayasgalan,
A. 2007: Reinterpretation of the active faulting in
central Mongolia. Geology 35, 75962.
Whitworth, M.C.Z., Giles, D.P. and Murphy, W.
2005: Airborne remote sensing for landslide hazard
assessment: a case study on the Jurassic escarpment
slopes of Worcestershire, UK. Quarterly Journal of
Engineering Geology and Hydrogeology 38, 285300.
Wright, R. and Flynn, L.P. 2003: On the retrieval of
lava-fl ow surface temperatures from infrared satellite
data. Geology 31, 89396.
Wright, R., Carn, S. and Flynn, L.P. 2005: A
satellite chronology of the MayJune 2003 eruption
of Anatahan volcano. Journal of Volcanology and
Geothermal Research 146, 10216.
Wright, R., Flynn, L.P., Garbeil, H., Harris,
A.J.L. and Pilger, E. 2004: MODVOLC: nearrealtime thermal monitoring of global volcanism.
Journal of Volcanology and Geothermal Research 135,
2949.
Downloaded from ppg.sagepub.com by Giyanto Giyanto on October 11, 2010
Wilson &Wilson's
COMPREHENSIVE ANALYTICAL CHEMISTRY
VOLUME XLII
NON-DESTRUCTIVE MICROANALYSIS OF
CULTURAL HERITAGE MATERIALS
Edited by
K. J'ANSSENS
R. VAN GRIEKEN
University ofAntwerp
Department of Chemistry
Universiteitsplein, I
B-2610 Antwerp
Belgium
~~
a -c tir_ t
2004
ELSEVIER
AMSTERDAM -- BOSTON - HEIDELBERG -- LONDON - NEW YORK - OXFORD - PARIS
SAN DIEGO - SAN FRANCISCO - SINGAPORE - SYDNEY- TOKYO
Contents
Contributors to Volume XLII . . . . . . . . . . . . . . . . . . . . . vi
Volumes in the Series . . . . . . . . . . . . . . . . . . . . . . . . . xi
Series Editor's Preface . . . . . . . . . . . , , . . . . . . . . . . . xxv
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxvii
Chapter 1. Introduction and overview . . . . . . . . . . . . . . . . . 1
K. Janssens and R. Van Grieken
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Overview of the analytical reference section . . . . . . . . 5
1.3 Overview of the case studies section . . . . . . . . . . . . 7
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
PART I. ANALYTICAL REFERENCE SECTION
K. Janssens
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . 129
4.2 Basic principles . . . . . . . . . . . . . . . . . . . . . . . 131
4.2.1 X-ray wavelength and energy scales . . . . . . . . 131
4.2.2 Interaction of X-rays with matter . . . . . . . . . 131
4.2.3 The photoelectric effect; X-ray
fluorescence . . . . . . . . . . . . . . . . . . . . . 133
4.2.4 Scattering and diffraction . . . . . . . . . . . . . 137
4.2.5 X-ray absorption fine structure and
spectroscopy . . . . . . . . . . . . . . . . . . . . 143
4.3 Instrumentation for X-ray investigations . . . . . . . . . 147
4.3.1. X-ray sources . . . . . . . . . . . . . . . . . . . . 148
4.3 .2 X-ray detectors . . . . . . . . . . . . . . . . . . . 155
4.3 .3 X-ray fluorescence instrumentation . . . . . . . . 164
4.3 .4 XRD instrumentation . . . . . . . . . . . . . . . 186
4.3 .5 XAS instrumentation at SR beamlines . . . . . . . 193
4.4 A survey of applications of X-ray methods in the cultural
heritage sector . . . . . . . . . . . . . . . . . . . . . . . 194
4.4.1 Compositional analysis of historic glass . . . . . . 195
4.4.2 Pigments . . . . . . . . . . . . . . . . . . . . . . 195
4.4.3 Lustre ware . . . . . . . . . . . . . . . . . . . . 200
Teresa E. Jeffries
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . 313
7.2 The inductively coupled plasma mass spectrometer . . . . 314
7.2.1 Historical account . . . . . . . . . . . . . . . . . 314
7.2.2 Operational rationale . . . . . . . . . . . . . . . . 315
7.2.3 The inductively coupled plasma . . . . . . . . . . 317
7.2.4 The plasma sampling interface . . . . . . . . . . 321
7.2.5 Ion focusing . . . . . . . . . . . . . . . . . . . . 323
7.2 .6 Quadrupole mass analyser . . . . . . . . . . . . . 323
7.2 .7 The vacuum system . . . . . . . . . . . . . . . . 327
7.2 .8 Ion detection and signal handling . . . . . . . . . 329
7 .3 Laser ablation : essential components . . . . . . . . . . . 329
7.3 .1 Development of the laser . . . . . . . . . . . . . . 329
7.3.2 The association of lasers with ICP-MS . . . . . . . 331
7.3.3 Stimulated emission . . . . . . . . . . . . . . . . 333
7.3.4 Nd:YAG laser (resonator) cavity . . . . . . . . . . 334
7.3.5 Harmonic generation . . . . . . . . . . . . . . . . 337
7.3.6 Harmonic separation . . . . . . . . . . . . . . . . 338
7.3.7 Energy attenuation. and control . . . . . . . . . . . 340
7.3 .8 Beam delivery and viewing optics . . . . . . . . . 341
7.3 .9 Ablation cell and sample transport , . . . . . . . 342
7.4 Analytical concepts and factors affecting analysis . . . . . 343
7.4.1 Why use the technique? . . . . . . . . . . . . . . 343
xvu
Contents
7 .4.2 Sample preparation and mounting . . . . . . . . . 344
7 .4.3 Analysis of transient signals . . . . . . . . . . . . 345
7.4.4 Factors affecting analysis . . . . . . . . . . . . . 347
7.4 .5 Optimization and calibration . . . . . . . . . . . . 350
7.4.6 Figures of merit and analytical performance
targets . . . . . . . . . . . . . . . . . . . . . . . 352
7.5 Continuing developments and final remarks . . . . . . . . 355
7.5.1. Continuing developments . . . . . . . . . . . . . 355
7.5.2 Final remarks . . . . . . . . . . . . . . . . . . . 356
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357
Chapter 8. Infrared, Raman microscopy and fibre-optic
Raman spectroscopy (FORS) . . . . . . . . . . . . . . . . . . . 359
Howell G.M. Edwards and Dalva L.A. de Faria
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . 359
8.2 Comparison of the potential use of IR and Raman
spectroscopies for the non-destructive analysis
of art works . . . . . . . . . . . . . . . . . . . . . . . . 360
8.3 Some theoretical aspects of IR and Raman
spectroscopies . . . . . . . . . . . . . . . . . . . . . . . 367
8.4 Instrumentation . . . . . . . . . . . . . . . . . . . . . . 373
8.5 Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . 387.
8 .6 Resonance Raman . . . . . . . . . . . . . . . . . . . . . 387
8 .7 SERS . . . . . . . . . . . . . . . . . . . . . . . . . . . . 390
8 .8 Intensity measurements in Raman scattering . . . . . . . 390
8 .9 Raman spectroscopy with fibre optics . . . . . . . . . . . 392
8.9.1 Sampling considerations . . . . . . . . . . . . . . 392
8.9.2 Probe design . . . . . . . . . . . . . . . . . . . . 392
8.9.3 Probe background . . . . . . . . . . . . . . . . . 393
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393
Chapter 9. Secondary ion mass spectrometry. Application to
archaeology and art objects . . . . . . . . . . . . . . . . . . . . 397
Evelyne Darque-Ceretti and Marc Aucouturier
9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . 397
9.2 Principles and equipment . . . . . . . . . . . . . . . . . 399
9.2.1 Principles . . . . . . . . . . . . . . . . . . . . . . 399
9.2.2 Equipment and choice of analytical parameters . . 47.0
9.3 Analysis procedures . . . . . . . . . . . . . . . . . . . . 420
9.3.1 Elemental identification, sensitivity , . . . . . . . 420
9.3.2 Quantitative analysis . . . . . . . . . . . . . . . 425
9.3.3 In-depth analysis and depth resolution . . . . . . 429
9.3.4 Surface analysis . .
-'---. . . 435
9.3.5 Imaging, lateral resolution . . . . . . . . . . . . . 438
9.3 .6 Chemical compound analysis and distribution . . . 439
9.4 Examples of applications for cultural heritage . . . . . . . 440
9.4.1 Dating and/or provenance studies based
on isotopic analysis . . . . . . . . . . . . . . . . 440
9.4.2 Dating (not based on isotopic analysis) . . . . . . 444
9.4.3 Provenance studies not based on isotopic
analysis . . . . . . . . . . . . . . . . . . . . . . 447
9 .4.4 Surface layer analysis on artefacts . . . . . . . . 448
9 .4.5 Interface studies on coated layers . . . . . . . . . 453
Contents
11.2.2 Illustration of the analytical performances of
non-vacuum PIXE for gold artefacts . . . . . . . . 496
11.3 The soldering of gold . . . . . . . . . . . . . . . . . . . . 498
11 .3 .1 Ancient recipes for gold soldering . . . . . . . . . 498
11.3.2 Iranian goldsmithery from. the 4th century BC . . . 499
11.3.3 Tartesic gold artefacts . . . . . . . . . . . . . . . 502
11.3 .4 Later Iranian goldsmithery . . . . . . . . . . . . . 506
11.3 .5 Preparations of low-melting brazing alloys . . . . . 509
11.3 .6 A new reading of Elder Pliny's Natural History . . 517
11.3 .7 Italian jewellery . . . . . . . . . . . . . . . . . . 520
11.3 .8 Gold artefacts from Slovenia . . . . . . . . . . . . 521
x1.3 .9 The Guarrazar treasure . . . . . . . . . . . . . . 525
11.3 .10 Merovingian and late Byzantine jewellery . . . . . 528
11.4 Pre-Hispanic gold artefacts of Mesoamerica . . . . . . . . 530
11.4.1 Archaeological context . . . . . . . . . . . . . . . 530
11.4.2 A selection of typical artefacts . . . . . . . . . . . 534
11.4.3 Differential PIXE . . . . . . . . . . . . . . . . . . 536
11.4.4 Application to the measurement of the gold
enhancement at the surface of tumbaga . . . . . . 538
11.5 Characterization of complex items . . . . . . . . . . . . . 544
11.5.1 XRF induced by a -y-ray source . . . . . . . . . . . 544
11.5.2 Gamma-ray transmission measurements . . . . . 544
11 .5.3 Study of a composite gold jewellery artefact . . . . 545
See discussions, stats, and author profiles for this publication at: http://www.researchgate.net/publication/41806088
CITATIONS
49
READS
510
4 AUTHORS:
Rodorico Giorgi
University of Florence
69 PUBLICATIONS 921 CITATIONS
SEE PROFILE
Michele Baglioni
University of Florence
10 PUBLICATIONS 114 CITATIONS
SEE PROFILE
Debora Berti
University of Florence
106 PUBLICATIONS 1,505 CITATIONS
SEE PROFILE
Piero Baglioni
University of Florence
437 PUBLICATIONS 7,586 CITATIONS
SEE PROFILE
Available from: Piero Baglioni
Retrieved on: 11 December 2015
CON SPECTUS
of works of art. Accordingly, the past has witnessed a number of actively detrimental treatments, such
as the
application of acrylic and vinyl resins to wall paintings, which can irreversibly jeopardize the
appearance (or even the
continued existence) of irreplaceable works of art. Current research activity in conservation science is
largely based
on the paradigm that compatibility of materials is the most important prerequisite for obtaining
excellent and durable
results.
The most advanced current methodologies are (i) the use of water-based micelles and microemulsions
(neat or combined
with gels) for the removal of accidental contaminants and polymers used in past restorations and (ii)
the application
of calcium hydroxide nanoparticles for the consolidation of works of art. In this Account, we highlight
how
conservation science can benefit from the conceptual and the methodological background derived
from both soft (microemulsions
and micelles for cleaning) and hard (nanoparticles for consolidation) nanoscience.
A combination of different nanotechnologies allows todays conservators to provide, in each restoration
step, interventions
respectful of the physicochemical characteristics of the materials used by artists. The palette of
methods
provided by nanoscience is continuously enriching the field, and the development of novel
nanomaterials and the study
of nanoscale physicochemical phenomena will further improve the performance of restoration
formulations and our
comprehension of degradation mechanisms.
Published on the Web 03/05/2010 www.pubs.acs.org/acr Vol. 43, No. 6 June 2010 695-704 ACCOUNTS OF CHEMICAL RESEARCH 695
10.1021/ar900193h 2010 American Chemical Society
Introduction
lost (see Figure 1). This was due to the combination of the
mowilith film with extensive salt crystallization processes that
quickly destroyed the plaster beneath the paint layer.
In addition to the mechanical degradation, the natural aging
of polymer resins generates discoloration and loss of solubility,
due to the increase of the polymer molecular weight. At the end
of the degradation process, these polymers are very difficult or
impossible to remove with conventional neat solvents. Considering
the severe side effects generated by the use of polymers
and the necessity of their removal, we proposed the use of nanostructured
fluids; some examples are highlighted in the following.
FIGURE 6. Mural painting sample treated, on the right side, with mowilith DM5 under glazing light (center), and
SEM pictures of the
uncoated (left) and coated (right) surface.
FIGURE 7. Details of the painting surface after removal of mowilith DM5 coating by using micellar solution
(Mayapan system).
FIGURE 8. Linen canvas glued with mowilith DMC2 for a relining
treatment. In the red box, the appearance of the treated area after
micellar solution application.
Nanoscience for the Conservation of Cultural Heritage Giorgi et al.
Vol. 43, No. 6 June 2010 695-704 ACCOUNTS OF CHEMICAL RESEARCH 701
FIGURE 9. Wall paintings from Mayan Classic period in Calakmul (Mexico). Calcium hydroxide nanoparticles were
used for consolidation.
Nanoscience for the Conservation of Cultural Heritage Giorgi et al.
702 ACCOUNTS OF CHEMICAL RESEARCH 695-704 June 2010 Vol. 43, No. 6
2 Ferroni, E.; Baglioni, P. Experiments About the Method for the Restoration of
Sulfated Frescoes. In Proceedings of the Symposium Scientific Methodologies
Applied to works of Art, Florence, May 2-5, 1984; Parrini, P. L., Ed.; Montedison
Progetto Cultura: Milan, 1986, pp 108-109.
3 Baglioni, P.; Giorgi, R. Soft and Hard Nanomaterials for Restoration and Conservation
of Cultural Heritage. Soft Matter 2006, 2, 293303.
4 Baglioni, P.; Giorgi, R.; Dei, L. Soft Condensed Matter for the Conservation of
Cultural Heritage. C. R. Chim. 2008, 12, 6169.
5 Ambrosi, M.; Dei, L.; Giorgi, R.; Neto, C.; Baglioni, P. Colloidal Particles of Ca(OH) 2:
Properties and Application to Restoration of Frescoes. Langmuir 2001, 17, 4251
4255.
6 Giorgi, R.; Dei, L.; Ceccato, M.; Schettino, C. V.; Baglioni, P. Nanotechnologies for
Conservation of Cultural Heritage: Paper and Canvas Deacidification. Langmuir
2002, 18, 81988203.
7 Giorgi, R.; Bozzi, C.; Dei, L.; Gabbiani, C.; Ninham, B. W.; Baglioni, P. Nanoparticles
of Mg(OH)2: Synthesis and Application to Paper Conservation. Langmuir 2005, 21,
84958501.
8 Giorgi, R.; Chelazzi, D.; Baglioni, P. Nanoparticles of Calcium Hydroxide for Wood
Conservation. The Deacidification of the Vasa Warship. Langmuir 2005, 21, 10743
10748.
9 Carretti, E.; Dei, L.; Baglioni, P. Solubilization of Acrylic and Vinyl Polymers in
Nanocontainer Solutions. Application of Microemulsions and Micelles to Cultural
Heritage. Langmuir 2003, 19, 78677872.
10 Carretti, E.; Giorgi, R.; Berti, D.; Baglioni, P. Oil-in-Water Nanocontainers as Low
Environmental Impact Cleaning Tools for Works of Art: Two Case Studies. Langmuir
2007, 23, 63966403.
11 Bonini, M.; Lenz, S.; Giorgi, R.; Baglioni, P. Nanomagnetic Sponges for the Cleaning
of Works of Art. Langmuir 2007, 23, 86818685.
12 Carretti, E.; Grassi, S.; Cossalter, M.; Natali, I.; Caminati, G.; Weiss, R. C.; Baglioni,
P.; Dei, L. Poly(vinyl alcohol)-Borate Hydro/Cosolvent Gels: Viscoelastic Properties,
Solubilizing Power, and Application to Art Conservation. Langmuir 2009, 25, 8656
8662.
13 Mora, P.; Mora, L.; Philipot, P. Conservation of Wall Painting; Butterworths: London,
1984.
14 Neher, H. T. Acrylic Resins. Ind. Eng. Chem. 1936, 28, 267271.
15 Resins in Conservation, Proceedings of the Symposium, Edinburgh, 21 and 22 May,
1982; Tate, J. O., Tennent, N. H., Eds.; Scottish Society for Conservation and
Restoration: Edinburgh, 1983.
16 Horie, C. V. Materials for Conservation; Butterworths: London, 1987.
17 Carretti, E.; Dei, L.; Baglioni, P. Aqueous Polyacrylic Acid Based Gels:
Physicochemical Properties and Applications in Cultural Heritage Conservation. Prog.
Org. Coat. 2004, 49, 282289.
18 Lazzari, M.; Chiantore, O. Thermal-Ageing of Paraloid Acrylic Protective Polymers.
Polymer 2000, 41, 64476455.
19 Chiantore, O.; Lazzari, M. Photo-Oxidative Stability of Paraloid Acrylic Protective
Polymers. Polymer 2001, 42, 1727.
20 Favaro, M.; Mendichi, R.; Ossola, F.; Rosso, U.; Simon, S.; Tomasin, P.; Vigato,
P. A. Evaluation of Polymers for Conservation Treatments of Outdoor Exposed Stone
Monuments. Part I: Photo-Oxidative Weathering. Polym. Degrad. Stab. 2006, 91,
30833096.
21 Orea, H.; Magar, V. A Brief Review of the Conservation of Wall Paintings in Mexico.
In Preprints of the 13th Triennial Meeting ICOM Committee for Conservation, ICOMCC,
Rio de Janeiro, 22-27 September 2002; Vontobel, R., Ed.; James and James
(Science Publishers): London, 2002; p 176.
22 Borgioli, L.; Caminati, G.; Gabrielli, G.; Baglioni, P. Removal of Hydrophobic
Impurities from Pictorial Surfaces by Means of Heterogeneous Systems. Sci. Tech.
Cult. Herit. 1995, 4, 6774.
23 De Gennes, P. G.; Taupin, C. Microemulsions and the Flexibility of Oil/Water
Interfaces. J. Phys. Chem. 1982, 86, 22942304.
24 Ferroni, E.; Gabrielli, G.; Caminati, G. Asportazione di Materiali Idrofobi da Superfici
Pittoriche Murali Mediante Solubilizzazione in Sistemi Dispersi. In La Cappella
Brancacci, la scienza per Masaccio, Masolino e Filippino Lippi, Quaderni del
restauro; Olivetti: Milan, 1992; pp 162-171.
25 Friberg, S. E.; Flaim, T. Surfactant Association Structures. In Inorganic Reactions In
Organized Media; Holt, S. L., Ed.; ACS Symposium Series 177; American Chemical
Society: Washington, DC, 1982; pp 1-17.
26 Fontell, K.; Ceglie, A.; Lindman, B.; Ninham, B. W. Some Observations on Phase
Diagrams and Structure in Binary and Ternary Systems of
Didodecyldimethylammonium Bromide. Acta Chem. Scand. 1986, A40, 247256.
27 The ammonium counterion was chosen because the painting, after the cleaning
from the wax, was consolidated using the Ferroni-Dini method, which uses the
ammonium carbonate in the first step of the consolidation process.
28 Palazzo, G.; Fiorentino, D.; Colafemmina, G.; Ceglie, A.; Carretti, E.; Dei, L.; Baglioni,
P. Nanostructured Fluids Based on Propylene Carbonate/Water Mixtures. Langmuir
2005, 21, 67176725.
29 Colafemmina, G.; Fiorentino, D.; Ceglie, A.; Carretti, E.; Fratini, E.; Dei, L.; Baglioni,
P.; Palazzo, G. Structure of SDS Micelles with Propylene Carbonate as Cosolvent: A
PGSE-NMR and SAXS Study. J. Phys. Chem. B 2007, 111, 71847193.
30 Carretti, E.; Fratini, E.; Berti, D.; Dei, L.; Baglioni, P. Nanoscience for Art
Conservation: Oil-in-Water Microemulsions Embedded in a Polymeric Network for
the Cleaning of Works of Art. Angew. Chem., Int. Ed. 2009, 48, 89668969.
31 Baglioni, P.; Dei, L.; Carretti, E.; Giorgi, R. Gels for the Conservation of Cultural
Heritage. Langmuir 2009, 25, 83738374, and references therein.
32 Carretti, E.; Dei, L.; Weiss, R. G. Soft Matter and Art Conservation: Rheoreversible
Gels and Beyond. Soft Matter 2005, 1, 1722.
33 Peng, F.; Li, G.; Liu, X.; Wu, S.; Tong, Z. Redox-Responsive Gel-Sol/Sol-Gel
Transition in Poly(acrylic acid) Aqueous Solution Containing Fe(III) Ions Switched by
Light. J. Am. Chem. Soc. 2008, 130, 1616616167.
34 Liu, Q.; Wang, Y.; Li, W.; Wu, L. Structural Characterization and Chemical Response
of a Ag-Coordinated Supramolecular Gel. Langmuir 2007, 23, 82178223.
35 Carretti, E.; Dei, L.; Baglioni, P.; Weiss, R. G. Synthesis and Characterization of Gels
from Polyallylamine and Carbon Dioxide as Gellant. J. Am. Chem. Soc. 2003, 125,
51215129.
36 Wolbers, R. C. Cleaning Painted Surfaces: Aqueous Methods; Archetype
Publications: London, 2000.
37 Stulik, S.; Miller, D.; Khanjian, H.; Khandekar, N.; Wolbers, R.; Carlson, J.; Petersen,
W. C. In Solvent Gels for the Cleaning of Works of Art: The Residue Question; Dorge,
V., Ed.; The Getty Conservation Institute: Los Angeles, 2004; pp 18-83 and
references therein.
38 Bonini, M.; Lenz, S.; Falletta, E.; Ridi, F.; Carretti, E.; Fratini, E.; Wiedenmann, A.;
Baglioni, P. Acrylamide-Based Magnetic Nanosponges: A New Smart
Nanocomposite Material. Langmuir 2008, 24, 1264412650.
39 Chevalier, A.; Chelazzi, D.; Baglioni, P.; Giorgi, R.; Carretti, E.; Stuke, M.; Menu, M.;
Duchamp, R. Extraction dAdhe sifs de Rentoilage en Peinture de Chevalet: Nouvelle
Approche. In Proceedings of the 15th Triennal Conference ICOM Committee for
Conservation, New Delhi, September 22-26; Bridgland, J., Ed.; Allied Publishers
Pvt. Ltd: New Delhi, 2008; Vol. II, pp 581-589.
40 Riederer, J. The Restoration of Archaeological Monuments in the Tropical Climate. In
Proceedings of 7th Triennial ICOM Meeting, Copenhagen, 10-14 September 1984;
De Froment, D., Ed.; ICOM Committee for Conservation: Paris, 1984; pp 21-22.
41 Espinosa, A. Conservation and Restoration of the Murals of the Temple of the
Paintings in Bonampak. In Proceedings of in situ Archaeological Conservation, 613 April 1986, Mexico City; Hodges, H. W. M., Ed.; INAH: Mexico City, 1987; pp
84-89.
42 Odom, T. W.; Pileni, M.-P. Nanoscience (guest editorial). Acc. Chem. Res. 2008,
41, 1565 and articles collected in this special issue.
43 Giorgi, R.; Chelazzi D.; Carrasco, R.; Colon, M.; Desprat, A.; Baglioni, P. The Maya
Site of Calakmul: In situ Preservation of Wall Paintings and Limestone by Using
Nanotechnologies. Proceedings of the IIC Congress 2006, Munich - The Object in
Context: Crossing Conservation Boundaries; Saunders, D., Townsend, J. H.,
Woodcock, S., Eds.; James and James: London, 2006; pp 162-169.
44 Giorgi, R.; Dei, L.; Baglioni, P. A New Method for Consolidating Wall Paintings Based
on Dispersions of Lime in Alcohol. Stud. Conserv. 2000, 45, 154161.
Nanoscience for the Conservation of Cultural Heritage Giorgi et al.
704 ACCOUNTS OF CHEMICAL RESEARCH 695-704 June 2010 Vol. 43, No. 6
See discussions, stats, and author profiles for this publication at: http://www.researchgate.net/publication/41806088
Nanoparticles
ARTICLE in ACCOUNTS OF CHEMICAL RESEARCH MARCH 2010
Impact Factor: 22.32 DOI: 10.1021/ar900193h Source: PubMed
CITATIONS
49
READS
510
4 AUTHORS:
Rodorico Giorgi
University of Florence
69 PUBLICATIONS 921 CITATIONS
SEE PROFILE
Michele Baglioni
University of Florence
10 PUBLICATIONS 114 CITATIONS
SEE PROFILE
Debora Berti
University of Florence
106 PUBLICATIONS 1,505 CITATIONS
SEE PROFILE
Piero Baglioni
University of Florence
437 PUBLICATIONS 7,586 CITATIONS
SEE PROFILE
Available from: Piero Baglioni
Retrieved on: 11 December 2015
CON SPECTUS
on the paradigm that compatibility of materials is the most important prerequisite for obtaining
excellent and durable
results.
The most advanced current methodologies are (i) the use of water-based micelles and microemulsions
(neat or combined
with gels) for the removal of accidental contaminants and polymers used in past restorations and (ii)
the application
of calcium hydroxide nanoparticles for the consolidation of works of art. In this Account, we highlight
how
conservation science can benefit from the conceptual and the methodological background derived
from both soft (microemulsions
and micelles for cleaning) and hard (nanoparticles for consolidation) nanoscience.
A combination of different nanotechnologies allows todays conservators to provide, in each restoration
step, interventions
respectful of the physicochemical characteristics of the materials used by artists. The palette of
methods
provided by nanoscience is continuously enriching the field, and the development of novel
nanomaterials and the study
of nanoscale physicochemical phenomena will further improve the performance of restoration
formulations and our
comprehension of degradation mechanisms.
Published on the Web 03/05/2010 www.pubs.acs.org/acr Vol. 43, No. 6 June 2010 695-704 ACCOUNTS OF CHEMICAL RESEARCH 695
10.1021/ar900193h 2010 American Chemical Society
Introduction
696 ACCOUNTS OF CHEMICAL RESEARCH 695-704 June 2010 Vol. 43, No. 6
FIGURE 1. Mural paintings in Templo de los Nichos Pintados in Mayapan (Yucatan). The pictures show the
degradation that occurred to the
paintings from the last restoration in 1999 due to the mowilith DM5 coating.
Nanoscience for the Conservation of Cultural Heritage Giorgi et al.
Vol. 43, No. 6 June 2010 695-704 ACCOUNTS OF CHEMICAL RESEARCH 697
FIGURE 6. Mural painting sample treated, on the right side, with mowilith DM5 under glazing light (center), and
SEM pictures of the
uncoated (left) and coated (right) surface.
FIGURE 7. Details of the painting surface after removal of mowilith DM5 coating by using micellar solution
(Mayapan system).
FIGURE 8. Linen canvas glued with mowilith DMC2 for a relining
treatment. In the red box, the appearance of the treated area after
micellar solution application.
Nanoscience for the Conservation of Cultural Heritage Giorgi et al.
Vol. 43, No. 6 June 2010 695-704 ACCOUNTS OF CHEMICAL RESEARCH 701
FOOTNOTES
*To whom correspondence should be addressed. E-mail: piero.baglioni@unifi.it. Phone:
+39 0555253033. Fax: +39 0555253032.
No kinship exists among the authors.
REFERENCES
1 Ferroni, E.; Malaguzzi, V.; Rovida, G. Experimental Study by Diffraction of
Heterogeneous Systems as a Preliminary to the Proposal of a Technique for the
Restoration of Gypsum Polluted Murals. ICOM-CC Plenary Meeting - The
International Council of Museums-Committee for Conservation, Amsterdam,
Netherlands, 1969 [Paper read at the Amsterdam ICOM meeting in 1969; copy in
ICCROM library, Rome].
Nanoscience for the Conservation of Cultural Heritage Giorgi et al.
Vol. 43, No. 6 June 2010 695-704 ACCOUNTS OF CHEMICAL RESEARCH 703
2 Ferroni, E.; Baglioni, P. Experiments About the Method for the Restoration of
from the wax, was consolidated using the Ferroni-Dini method, which uses the
ammonium carbonate in the first step of the consolidation process.
28 Palazzo, G.; Fiorentino, D.; Colafemmina, G.; Ceglie, A.; Carretti, E.; Dei, L.; Baglioni,
P. Nanostructured Fluids Based on Propylene Carbonate/Water Mixtures. Langmuir
2005, 21, 67176725.
29 Colafemmina, G.; Fiorentino, D.; Ceglie, A.; Carretti, E.; Fratini, E.; Dei, L.; Baglioni,
P.; Palazzo, G. Structure of SDS Micelles with Propylene Carbonate as Cosolvent: A
PGSE-NMR and SAXS Study. J. Phys. Chem. B 2007, 111, 71847193.
30 Carretti, E.; Fratini, E.; Berti, D.; Dei, L.; Baglioni, P. Nanoscience for Art
Conservation: Oil-in-Water Microemulsions Embedded in a Polymeric Network for
the Cleaning of Works of Art. Angew. Chem., Int. Ed. 2009, 48, 89668969.
31 Baglioni, P.; Dei, L.; Carretti, E.; Giorgi, R. Gels for the Conservation of Cultural
Heritage. Langmuir 2009, 25, 83738374, and references therein.
32 Carretti, E.; Dei, L.; Weiss, R. G. Soft Matter and Art Conservation: Rheoreversible
Gels and Beyond. Soft Matter 2005, 1, 1722.
33 Peng, F.; Li, G.; Liu, X.; Wu, S.; Tong, Z. Redox-Responsive Gel-Sol/Sol-Gel
Transition in Poly(acrylic acid) Aqueous Solution Containing Fe(III) Ions Switched by
Light. J. Am. Chem. Soc. 2008, 130, 1616616167.
34 Liu, Q.; Wang, Y.; Li, W.; Wu, L. Structural Characterization and Chemical Response
of a Ag-Coordinated Supramolecular Gel. Langmuir 2007, 23, 82178223.
35 Carretti, E.; Dei, L.; Baglioni, P.; Weiss, R. G. Synthesis and Characterization of Gels
from Polyallylamine and Carbon Dioxide as Gellant. J. Am. Chem. Soc. 2003, 125,
51215129.
36 Wolbers, R. C. Cleaning Painted Surfaces: Aqueous Methods; Archetype
Publications: London, 2000.
37 Stulik, S.; Miller, D.; Khanjian, H.; Khandekar, N.; Wolbers, R.; Carlson, J.; Petersen,
W. C. In Solvent Gels for the Cleaning of Works of Art: The Residue Question; Dorge,
V., Ed.; The Getty Conservation Institute: Los Angeles, 2004; pp 18-83 and
references therein.
38 Bonini, M.; Lenz, S.; Falletta, E.; Ridi, F.; Carretti, E.; Fratini, E.; Wiedenmann, A.;
Baglioni, P. Acrylamide-Based Magnetic Nanosponges: A New Smart
Nanocomposite Material. Langmuir 2008, 24, 1264412650.
39 Chevalier, A.; Chelazzi, D.; Baglioni, P.; Giorgi, R.; Carretti, E.; Stuke, M.; Menu, M.;
Duchamp, R. Extraction dAdhe sifs de Rentoilage en Peinture de Chevalet: Nouvelle
Approche. In Proceedings of the 15th Triennal Conference ICOM Committee for
Conservation, New Delhi, September 22-26; Bridgland, J., Ed.; Allied Publishers
Pvt. Ltd: New Delhi, 2008; Vol. II, pp 581-589.
40 Riederer, J. The Restoration of Archaeological Monuments in the Tropical Climate. In
Proceedings of 7th Triennial ICOM Meeting, Copenhagen, 10-14 September 1984;
De Froment, D., Ed.; ICOM Committee for Conservation: Paris, 1984; pp 21-22.
41 Espinosa, A. Conservation and Restoration of the Murals of the Temple of the
Paintings in Bonampak. In Proceedings of in situ Archaeological Conservation, 613 April 1986, Mexico City; Hodges, H. W. M., Ed.; INAH: Mexico City, 1987; pp
84-89.
42 Odom, T. W.; Pileni, M.-P. Nanoscience (guest editorial). Acc. Chem. Res. 2008,
41, 1565 and articles collected in this special issue.
43 Giorgi, R.; Chelazzi D.; Carrasco, R.; Colon, M.; Desprat, A.; Baglioni, P. The Maya
Site of Calakmul: In situ Preservation of Wall Paintings and Limestone by Using
Nanotechnologies. Proceedings of the IIC Congress 2006, Munich - The Object in
Context: Crossing Conservation Boundaries; Saunders, D., Townsend, J. H.,
Woodcock, S., Eds.; James and James: London, 2006; pp 162-169.
44 Giorgi, R.; Dei, L.; Baglioni, P. A New Method for Consolidating Wall Paintings Based
on Dispersions of Lime in Alcohol. Stud. Conserv. 2000, 45, 154161.
Nanoscience for the Conservation of Cultural Heritage Giorgi et al.
704 ACCOUNTS OF CHEMICAL RESEARCH 695-704 June 2010 Vol. 43, No. 6
Remote Sensing Department, IRRS, National Research Council, Via Bassini 15,
20133-Milano, Italy
(Received 8 April 1999; in . nal form 11 July 2000)
Abstract. This paper describes a synergetic use of satellite radar images and
ancillary information to detect ooded areas at their peak and evaluates its
potential with mapping. The procedure was tested on the catastrophic ood that
1. Introduction
In densely populated areas, both in tropical and temperate regions, oods are
one of the most important natural disasters as they inundate large areas causing
damage to agricultural crops and property, road and rail links, and the loss of
human lives. Appropriate land use planning, which requires accurate knowledge of
ood extent for locating ood prone areas, is a key tool to improve ood management
and to mitigate its catastrophic eVects. The knowledge of the spatial extent of
inundated areas is essential both during the emergency phase, when it is necessary
to have an overall view of the phenomenon in order to plan relief eVorts, and after
the event for the assessment of damaged areas by authorities and insurance
companies.
Traditional methods of ood mapping are based on ground surveys and aerial
observations, but when the phenomenon is widespread, such methods are time
consuming and expensive; furthermore timely aerial observations can be impossible
e-mail: brivio@irrs.mi.cnr.it
2000) or single-band (Sharma et al. 1996) and multi-frequency (Ormsby et al. 1985,
Hess et al. 1995) synthetic aperture radar (SAR) data. The disadvantages of the use
of radar sensors lie in the diYcult classi. cation of the acquired signal because of the
in uence of complex ground and system variables. This is particularly true when
single-band and single-polarization data are available, as is the case in the ERS-1/2
satellites. To improve this situation, multi-temporal techniques based on detection
of changes between radar images acquired before and after the inundation event are
usually recommended (Badji et al. 1994, Wang et al. 1995, Profeti and Macintosh
1997, Takeuchi et al. 1999). Moreover if a high temporal resolution dataset is
available, change detection analysis can be used to produce a ood evolution map
(Noyelle et al. 1995, Oberstadler et al. 1995, Laugier et al. 1997).
If radar images are acquired some days after the event, when only a few areas
are still submerged by water, it is advisable to use a multi-sensor approach. In this
case ooded areas as derived from radar data are complemented with information
extracted from optical images, such as the areal extent of . ne material left during
the ood (ImhoV et al. 1987, Bonansea 1995). Another possibility is the integration
in a Geographical Information System (GIS) of radar imagery acquired during the
ood event with information derived from digital topography (Giacomelli and
Mancini 1996, Brakenridge et al. 1998).
This paper proposes a new methodology which, integrating results obtained from
ERS SAR images and topography in an appropriate GIS methodology, allows a
satisfactory mapping of inundated areas, even when satellite data are acquired some
days after the ooding event. The catastrophic ood event that occurred in Regione
Piemonte (northern Italy) on November 1994 is used to illustrate the methodology.
Integration of SAR data and GIS for ood mapping 431
2. Mapping ooded areas through an integrated RS and GIS approach
The magnitude of the ood and the extent of ood-aVected areas depend on the
intensity of rainfall, its duration, the watershed topography and its conditions at the
time of the heavy rainfall. Observation from a RS satellite provides a timeinstantaneous
portrait of the ood stage over a wide area. However, the period of
submergence can vary from hours to weeks and the time delay between the ood
peak phase and the satellite observation may severely limit the capabilities of mapping
the widest spread of ooding. Figure 1 illustrates the relationships between ooded
area extent, ood peak phase and SAR image acquisition.
2.1. Flood mapping on SAR imagery
The fundamental characteristic recorded on a radar image is the backscattering
coeYcient, which may vary from surface to surface. The strength of the returned
signal from the surface is in uenced by combined system and ground parameters,
including the average surface roughness and soil dielectric properties. Horizontal
smooth surfaces, such as water bodies, re ect nearly all incident radiation away from
the sensor and the weak return signal is represented by dark tonality on radar
images with the result that standing water areas are easily recognizable. This specular
re ection can be decreased by bad weather conditions and /or the presence of vegetation,
roughening the surface and making the detection of ooded areas more di Ycult
(Laugier et al. 1997). Moreover, SAR data are subject to speckle, a multiplicative
random noise that considerably reduces the interpretability of the images and limits
classi. cation techniques, and SAR images have to be . ltered in order to increase
the signal-to-noise ratio. The enhanced Lee adaptive . lter is an eYcient tool for
reducing SAR image speckle by removing high-frequency noise while preserving
edges or sharp features in the image (Zhengao and Fung 1994). Shadows at the
Figure 1. Temporal relationships between ood event, ooded area and SAR observation
highlighting the diVerence between ooded area at the peak (T p) and ooded area
after the event (T p+DT ) seen on an SAR image.
inundated areas since they also appear in a dark grey tone. The determination of
paved roads and river streams can be established with the help of auxiliary
information derived from digitized topographic maps.
DiVerent kinds of approaches can be adopted to map the extent of ooded areas.
The . rst one is based on segmentation techniques applied to SAR imagery acquired
after the ood. A second approach consists of comparing two SAR images, one
taken before and the other after the ood, through visual interpretation or change
detection analysis. A third approach, which is not considered in this study, involves
the use of coherence information deduced from an interferometric pair of SAR images
(Nico et al. 2000).
When a segmentation approach is adopted to detect ooded areas on the after ood SAR image, optimal threshold de. nition is required. The threshold value can
be determined on a frequency histogram of the areas corresponding to dark grey
tones in the SAR image and to ooded areas in the aerial photographs. A possible
alternative is to use an adaptive thresholding algorithm. Ober et al. (1998) developed
an iterative process which, assuming a Gaussian distribution of the data, calculates
the mean value (m) and the standard deviation (s) and assigns to the Digital Number
(DN) values out of the interval [m s; m s], the extreme value of the same interval.
This iterative process continues until all pixels are one of only two values (resulting
in a binary image ( ooded and non- ooded areas)) or when all the image pixels
have a value belonging to the interval [m s; m s]; in a such case m is the searched
threshold value.
When using a multi-temporal approach, principal component analysis (PCA) of
SAR images, taken at diVerent times, oVers great opportunities for ooded area
detection (Calabresi 1995, Henebry 1997). In this approach a ood map is drawn
up directly on a computer screen by means of visual interpretation of the false colour
composite represented in RGB by images taken after and before the ood event.
These approaches allow satisfactory determination of ooded zones where water
is still remaining at the time of space observation. However, this situation does not
always represent the widest spread of the ooding. Often, ash ooding occurs and
then the water level decreases very quickly; in this case observation from remote
sensing satellites cannot meet the peak ooding phase.
2.2. An integrated RS and GIS approach
In order to overcome these limitations, a new procedure was developed in a GIS
environment with the purpose of estimating the ooded area at the peak using
standing water area observed after the ood event seen on SAR imagery and ancillary
information extracted from the digital elevation topography. The main steps of the
overall methodology are presented in . gure 2. The basic assumption of this approach
is that water has to ow out from the main river channel up to reach at least the
areas recognized as ooded in the SAR data. The technique, based on the computation
of least accumulative costdistance surface (Douglas 1994), . lls up the ooded
place by tracing backwards from the main river stream towards still remaining water
observed on SAR image.
From the digital topography of the study area a least accumulative costdistance
matrix is derived. In such a raster image, each cell is characterized by a value equal
to the least accumulative cost that water has to spend to get to the cell departing
from main stream of the river. Cost corresponds to the work necessary for the water
Integration of SAR data and GIS for ood mapping 433
Figure 2. Schematic diagram of the combined methodology: 1, detect ooded areas from
SAR data; 2, generate least accumulative costdistancematrix from digital topography
of the study area, where the origin is the river; 3, a cell in the study area is ooded if
its cost is lower than the cost of the cell detected from SAR data.
the saturation of both the drainage capacity of the slopes and the impoundment of
the hydrographic network. Rainfall produced exceptional discharge of the rivers and
hydrometric levels never recorded before (unit discharge 0.67.3m3 s 1km 2).
Floods caused widespread damage to 38 urban areas, with the creation of 44 victims
and 2000 homeless (Luino 1999). A comparable case in the Tanaro basin has not
been recorded since 1879.
3.1. Study area and data used
The study area is a reach of the Tanaro basin included between the towns of
Asti and Felizzano. This part of the catchment is an almost at area with an average
slope of 1% and is mainly devoted to agricultural activities. The land cover as
Integration of SAR data and GIS for ood mapping 435
derived from supervised classi. cation of Landsat 5 TM data taken in July 1994 is
as follows: 79% cropland (wheat, sugar beet, grass), 8% poplar plantations, 6%
urban areas and 7% water bodies.
The satellite data used were two ERS-1 SAR images, the . rst one acquired on 4
October 1994, one month before the ood and the second one acquired on
9 November 1994, three days after the event (. gure 3). The SAR mounted onboard
ERS-1 is a C-band system (5.6cm wavelength) with a nominal incidence angle of
23 at mid-swath, which transmits and receives vertically polarized microwave
electromagnetic
energy. ERS-1 Precision Image (PRI) products used are three-look
digital images, corrected for antenna elevation gain pattern and range spreading
loss, with pixel dimensions of 12.5m12.5 m.
In addition to the satellite images, ancillary data were used. Contour lines were
extracted from a digital topographic map in raster format (1:10 000) and used to
derive the digital elevation model (DEM); the main streams of the river network
and paved roads were extracted from the same map. Colour aerial photographs
taken on 9 and 10 November 1994 over some areas were also available and used to
de. ne ground truth areas. Moreover, maps (1:10 000 scale) of the actual inundated
areas, acquired from ground surveys by the Settore Prevenzione del Rischio
Geologico, Meteorologico e Sismico of Regione Piemonte, were used as reference
data for accuracy evaluation. The areal extension of the actual inundated areas was
estimated to be 54km2 from this reference map.
Both radar images and ancillary data were pre-processed and arranged in a GIS
environment. A set of ground control points (GCPs) taken from the 1:10 000 topographic
map was used to rectify the 9 November ERS-1 image to the Universal
Transverse Mercator (UTM) map projection, producing a root means square error
(RMSE) of 11.6 m. The pre- ood image of 4 October was then registered to the
after- ood scene (RMSE 8.75m) through a . rst-degree polynomial transformation,
to form a perfectly overlapping multi-temporal dataset. A nearest-neighbour resampling
technique was adopted for both images in order to retain radiometric integrity.
Figure 3. Tanaro river basin study area where a catastrophic ood occurred in November
1994. ERS-1 SAR image acquired on 9 November, three days after the ood event.
allowed only a limited assessment of the ooded areas. The two thresholding procedures
and visual interpretation were able to map only a small fraction of the total
inundated area, amounting to 54km2.
Visual interpretation of a false colour composite obtained from multi-temporal
SAR images and Second Principal Component (. gure 4) gave larger extension of
ooded areas (10.23km2) compared to maps determined by thresholding approaches
applied to a single-date SAR imagery. An iterative threshold algorithm applied to a
single-date SAR image allowed us to map an area of 7.08km2; using a simple
threshold value derived from the cumulative frequency curve mapped only 3.96km2
as ooded zones.
Figure 5 shows a map of ooded areas derived from visual interpretation of the
false colour composite of the multi-temporal dataset. In the same . gure, the study
area extracted from digital topography of the Tanaro watershed is also presented.
Areas classi. ed as inundated are patches spread around the main river stream. The
water surfaces in this map cover only 18.9%of the ooded area recorded by Regione
Piemonte. These large diVerences between the ooded zone detected on SAR imagery
Figure 4. False colour composite of multi-temporal ERS-1 radar images (RGB: 9 November
1994; 4 October 1994; Second Principal Component). Areas still submerged by water
three days after the ood event are easily recognizable due to their intense blue colour.
Figure 5. Flood map derived from visual interpretation of false colour composite of multitemporal
ERS-1 radar images (RGB: 9 November 1994; 4 October 1994; Second
Principal Component), compared with the reference ood map compiled by the Settore
Prevenzione del Rischio Geologico, Meteorologico e Sismico of Regione Piemonte.
and reference data can be explained by considering the time delay between the ERS-1
passage and the ood peak.
On 9 November, when the satellite image was acquired, most of the water had
already own back into the river or in. ltrated into the soil. Aerial photographs
(. gure 6) taken on 9 and 10 November 1994 over some sites con. rmed that, at the
time of the ERS-1 overpass, most of the water was not actually still standing on
land, and areas estimated as inundated corresponded to few zones still under water
three days after the ood event. Moreover, the presence of a layer of uvial wet . ne
sediments made the recognition of areas aVected by the ood event diYcult. In fact,
in radar images, wet sediments produce intermediate backscattering values between
low-intensity signals of water and the higher intensity signals of agricultural areas.
In some cases poplar plantations close to the Tanaro river were also a constraint to
ERS-1 in monitoring standing water, because of the rather low penetration capacity
of the C microwave band at which it operates.
In contrast to the previous results, which re ect the situation three days after the
ood peak, the ood map of . gure 7 obtained by application of the combined
methodology describes the maximum ooded area (60km2). The ooded zones are
seen to extend both sides of the river and form a continuous belt around the
main stream.
This map was obtained using the ooded area map given by the simple threshold
approach as a base layer to be integrated with the least accumulative costdistance
matrix. The Tanaro river basin was subdivided into diVerent sections re ecting
438 P. A. Brivio et al.
Figure 6. Enlargement of ood map of . gure 5 and oblique aerial photograph taken on
9 November 1994 over the same area (Courtesy of Fabio Luino, CNR-IRPI Torino).
The results obtained appear very promising, especially because the methodology
allows mapping of inundated areas accurately over a large area and without the
need for a complex simulations approach. Even though hydraulic models may
provide a more correct description of the ooding process, they are also characterized
by a more critical parameterization.
Integration of SAR data and GIS for ood mapping 439
Figure 7. Map of the maximum extent of ooded area estimated by the new approach based
on integration of the ood map derived from ERS-1 post- ood image and the least
accumulate costdistance matrix.
4. Conclusions
Techniques based on visual interpretation of multi-temporal satellite data and
thresholding algorithms applied to single-date satellite data show serious limitations
in determination of inundated areas at the peak of the ood. This is because of the
inappropriate time of acquisition of satellite radar data with respect to the ood
peak event. In fact, in the case study presented here, most of the water had already
own back into the river or had been in. ltrated into the soil at the time of the
satellite overpass.
To overcome these limitations, a new approach was proposed which integrates
results obtained from SAR data with ancillary information using a least accumulative
costdistance matrix. The catastrophic ood that occurred in northern Italy in
November 1994 was used to test the obtained results through detailed comparison
with a reference map of the ooded area compiled by Regione Piemonte. When
compared with the reference map showing the area actually ooded, the ooded
area map obtained from the new integrated RS and GIS approach shows an accuracy
of 96%. This result indicates that images collected by ERS-1 SAR some days after
an event can be useful in generating ood maps of the peak if used in conjunction
with an appropriate methodology, whereas without such an approach SAR data do
not allow such an accurate delimitation of ooded areas.
Acknowledgments
This study was supported by the National Group for Prevention from
Hydrogeological Hazards (GNDCI) of the National Council of Research (CNR) of
440 P. A. Brivio et al.
Italy. We are also grateful to Giovanna Ober for software implementation of the
iterative threshold technique. Suggestions of two anonymous referees greatly
improved the quality of the manuscript. The 4 October 1994 and the 9 November
1994 ERS-1 PRI images were kindly provided by ESA-ESRIN, Frascati, Italy.
References
Badji, M., Dautrebande, S., Mokaden, A. I., and Dewez, A., 1994, ERS-1 SAR imagery
applied to rural basins hydrological studies: the ood inundation mapping and monitoring.
Proceedings of the First Workshop on ERS-1 Pilot Projects, T oledo, Spain,
2224 June 1994, SP-365 (Noordwijk, The Netherlands: ESA-ESTEC), pp. 117124.
Barton, I. J., and Bathols, J. M., 1989, Monitoring oods with AVHRR. Remote Sensing of
Environment, 30, 8994.
Berg, C. P., Matson, M., and Wiesnet, D. R., 1981, Assessing the Red River of the North
1978 ooding from NOAA satellite data. In Satellite Hydrology, edited byM. Deutsch,
D. R. Wiesnet and A. Rango (Minneapolis, MN: American Water Resources
Association), pp. 309315.
Blasco, F., Bellan, M. F., and Chaudhury, M. U., 1992, Estimating the extent of oods in
Bangladesh using SPOT data. Remote Sensing of Environment, 39, 167178.
Bonansea, E., 1995,Mapping of 1994 oods in Piemonte Region, Italy: an example of remote
sensing and GIS application. Proceedings of First ERS T hematic Working Group
Meeting on FloodMonitoring, ESA-ESRIN, Frascati, Italy, 2627 June 1995 (Frascati,
Italy: ESA-ESRIN).
Brakenbridge, G. R., Tracy, B. T., and Knox, J. C., 1998, Orbital SAR remote sensing of
a river ood wave. International Journal of Remote Sensing, 19, 14391445.
Brivio, P. A., Gregoire, J. M., and Zilioli, E., 1984, The detection of hydrological indicators
in the study of Niger River regime by means of Landsat imageries. ITC Journal,
3, 191199.
Calabresi, G., 1995, The use of ERS data for ood monitoring: an overall assessment.
Proceedings of the 2nd ERS Applications Workshop, L ondon, UK, 68 December 1995,
SP-383 (Noordwijk, The Netherlands: ESA-ESTEC), pp. 237241.
Douglas, D. H., 1994, Least-cost-path in GIS using an accumulated cost surface and slope
lines. Cartographica, 31, 3751.
ESRI, Environmental System Research Institute Inc., 1992, ARC/INFO Users Guide
6.1GRID Command References. Environmental System Research Institute Inc.,
Redlands, California, USA.
Giacomelli, A., and Mancini, M., 1996, Integrazione di immagini satellitari ERS-1 PRI e
modelli digitali del terreno per lindividuazione di aree esondate. Proceedings of XXV
Convegno di Idraulica e Costruzioni Idrauliche, T orino, Italy, 1618 September 1996,
Vol. I (Torino, Italy: MAF Servizi), pp. 182192.
Henebry, G. M., 1997, Advantages of principal components analysis for land cover segmentation
from SAR images series. Proceedings of T hirs ERS Symposium on Space at the
Service of our Environment, Florence, Italy, 1421March 1997, SP-414, Vol. I
(Noordwijk, The Netherlands: ESA-ESTEC), pp. 175178.
Hess, L. L., Melack, J. M., Filoso, S., and Wang, Y., 1995, Delineation of inundated areas
and vegetation along the Amazon oodplain with the SIR-C SyntheticAperture Radar.
IEEE T ransactions on Geoscience and Remote Sensing, 33, 896904.
Hutchinson, M. F., 1989, A new procedure for gridding elevation and stream line data with
automatic removal of spurious pits. Journal of Hydrology, 106, 211232.
Imhoff, M. L., Vermillion, C., Story, M. H., Choudhury, A. M., Gafoor, A., and Polcyn,
F., 1987, Monsoon ood boundary delineation and damage assessment using space
borne imaging radar and Landsat data. Photogrammetric Engineering and Remote
Sensing, 53, 405413.
Laugier, O., Fellah, K., Tholey, N., Meyer, C., and De Fraipont, P., 1997, High temporal
detection and monitoring of ood zone dynamic using ERS data around catastrophic
natural events: the 1993 and 1994 Camargue ood events. Proceedings of T hird ERS
Symposium on Space at the Service of our Environment, Florence, Italy, 1421March
1997, SP-414, Vol. I (Noordwijk, The Netherlands: ESA-ESTEC), pp. 559564.
Lougeay, R., Baumann, P., and Nellis, M. D., 1994, Two digital approaches for calculating
the area of regions aVected by the great American ood of 1993. Geocarto International,
9, 5359.
Luino, F., 1999, The ood and landslide event of November 46 1994 in Piedmont Region
(Northwestern Italy): causes and related eVects in Tanaro Valley. Proceedings of XXII
General AssemblyEuropean Geophysical Society, V ienna, 2125 April 1997, Vol. 24,
N. 2 (Amsterdam: Elsevier), pp. 123129.
Michener, W. K., and Houhoulis, P. F., 1997, Detection of vegetation changes associated
with extensive ooding in a forested ecosystem. Photogrammetric Engineering and
Remote Sensing, 63, 13631374.
Nico, G., Pappalepore, M., Pasquariello, G., Refice, A., and Samarelli, S., 2000,
Comparison of SAR amplitude vs. coherence ood detection methodsa GIS
application. International Journal of Remote Sensing, 21, 16191631.
Noyelle, J., Demeire, S., and Marinelli, L., 1995, Identi. cation of ooded areas in the Rhone
and Var river, France, using ERS-1 data. Proceedings of First ERS T hematic Working
Group Meeting on Flood Monitoring, ESA-ESRIN, Frascati, Italy, 2627 June 1995
(Frascati, Italy: ESA-ESRIN), pp. 7080.
Ober, G., Cella, F., Meroni, F., Tomasoni, R., and Zonno, G., 1998, Characterisation of
urban areas through texture analysis. Proceedings SPIE Earth Surface Remote Sensing
II, Vol. 3496, edited by G. Cecchi and E. Zilioli (Bellingham, Washington: SPIE),
pp. 94101.
Oberstadler, R., Honsch, H., and Hutch, D., 1995, Assessment of the mapping capabilities
of ERS-1 SAR data for ood mapping: a case study in Germany. Proceedings of the
2ndERS ApplicationsWorkshop, L ondon,UK, 68 December 1995, SP-383 (Noordwijk,
The Netherlands: ESA-ESTEC), pp. 247255.
Ormsby, J. P, Blanchard, B. J., and Blanchard, A. J., 1985, Detection of lowland ooding
using active microwave systems. Photogrammetric Engineering and Remote Sensing,
51, 317328.
Profeti, G., and Macintosh, H., 1997, Flood management through Landsat-TM and ERS
CITATIONS
43
READS
163
3 AUTHORS:
Valeria Daniele
Universit degli Studi dell'Aquila
11 PUBLICATIONS 101 CITATIONS
SEE PROFILE
Giuliana Taglieri
Universit degli Studi dell'Aquila
17 PUBLICATIONS 217 CITATIONS
SEE PROFILE
Raimondo Quaresima
Universit degli Studi dell'Aquila
20 PUBLICATIONS 133 CITATIONS
SEE PROFILE
Original article
Abstract
Water and milk of lime are usually adopted for conservative surfaces treatments, thanks to the
conversion of lime into calcium carbonate.
Calcium carbonate is, as a matter of fact, very compatible with many carbonatic lithotypes and
architectonic surfaces, because its characteristics
are very similar to those of the materials to be restored. But there are some limiting aspects to
treatments effectiveness: the reduced penetration
depth, the binder concentration and the incompleteness carbonatation process. In order to improve
lime treatments, Ca(OH)2 particles with
submicrometric dimensions (nanolimes) are recently introduced in Cultural Heritage conservation.
Lime nanoparticles are typically produced
by a chemical precipitation process in supersaturated aqueous solutions of the reactants (calcium
chloride and sodium hydroxide). The aim
of the present work is to analyse the nanolime carbonatation process in relation to some parameters,
like time and the relative humidity
conditions. For this scope, lime nanoparticles are therefore synthesised and characterised by X-ray
diffraction (XRD), scanning and transmission
electron microscopy (SEMeTEM), electron diffraction measurements (ED) and dark field images (DFI).
The possibility to improve the nanolime
carbonatation process is investigated using an alcoholic suspension and by adding a baking soda
solution in order to disaggregate particles
and to increase CO2 content in the suspension respectively. The efficiency of the nanolime
carbonatation process is reported too.
After that the lime nanoparticles are applied on natural lithotypes (Estoril and Pietra Serena) and
some tests are performed in order to
estimate the superficial consolidating and protective effect of the treatment: Scotch Tape Test,
capillarity and imbibition tests. SEM analyses
are performed to evaluate penetration depth and surface adhesion of nanolime treatments.
_ 2008 Elsevier Masson SAS. All rights reserved.
Keywords: Calcium hydroxide; Consolidation; Lime; Nanoparticles; Protection
1. Introduction
The use of lime (Ca(OH)2), in building industry as in Cultural
Heritage conservation, is based on the well-known carbonation
reaction and on the characteristics of calcium carbonate
(CaCO3) obtained. The low solubility and the compatibility
between the latter compound and material substrates offer a favourable
use in many lime-based conservative treatments (preconsolidation
[1], cleaning [2], consolidation and protection [3]).
The applicationsemploy lime solutions (lime milk or limewater).
Lime water consolidation is generally obtained by spraying
the lime solution on the cleaned surface. To reach a good
penetration, the treatment is repeated several times until the surface
is able to absorb limewater [4]; some authors indicate that it
could be necessary to repeat the application for 30e40 times [5].
Lime milk is used on the same basis as lime water [6]. Compared
with limewater, lime milk treatments involve greater amounts of
lime with the same water volume; this represents an advantage
due to a reduced water percentage brought to the stone.
Lime water and lime milk treatments are characterised by
some limitations due to:
- the incomplete conversion of lime into calcium carbonate
that leaves free particles on the surfaces [6];
- the binder concentration due to the low water solubility of
lime, giving chromatic alteration to stone surfaces;
- a reduced penetration depth [4].
* Corresponding author. Tel.: 39 0862 434212; fax: 39 0862 434203.
E-mail address: taglieri@ing.univaq.it (G. Taglieri).
1296-2074/$ - see front matter _ 2008 Elsevier Masson SAS. All rights reserved.
doi:10.1016/j.culher.2007.10.007
2. Experimental
2.1. Materials
Calcium chloride (CaCl2), sodium hydroxide (NaOH), sodium
bicarbonate (NaHCO3) and 2-propanol pro analysis
products, supplied by Merck, are used without further purification.
Water is purified by a Millipore Organex system
(R _ 18 MU cm).
2.2. Synthesis of the particles
To obtain about 20 g of Ca(OH)2 nanoparticles, two different
aqueous solutions of 900 ml, containing 0.3 mol/l of CaCl 2
and 0.6 mol/l of NaOH, respectively, are prepared. The NaOH
alkaline solution (used as precipitator) is added dropwise into
the CaCl2 solution (speedz4 ml/min, temperature of 90 _C).
2-theta
Intensity (a.u.)
Ca(OH)2
* CaCO3
*
***
Fig. 5. XRD pattern of the aqueous nanolime suspension (sample WS, concentration
of 15 mg/ml) after 30 min of air exposition time.
296 V. Daniele et al. / Journal of Cultural Heritage 9 (2008) 294e301
25 30 35 40 45
2-theta
Ca(OH)2
* CaCO3
Intensity (a.u.)
Fig. 6. XRD pattern of the alcoholic nanolime suspension (sample ALS).
Table 1
Relative humidity influence on the carbonatation process
Sample RH (%) Yield (%)
ALS 40 65
ALS70 70 80
ALS90 90 93
*
***
Intensity (a.u.)
25 30 35 40 45
2-theta
Ca(OH)2
* CaCO3
Fig. 7. XRD pattern of the alcoholic nanolime suspension mixed with the baking
soda solution (sample ALBKS).
Fig. 8. Porosimetric distribution of: (a) Estoril; (b) Pietra Serena.
V. Daniele et al. / Journal of Cultural Heritage 9 (2008) 294e301 297
1. Introduction
Floods are the major disaster affecting many countries in the world year after year.
It is an inevitable natural phenomenon occurring from time to time in all rivers
and natural drainage systems that not only damages the lives, natural resources
and
environment, but also causes the loss of economy and health. In India also, floods
are
most frequent mainly in the eastern part (Orissa, Bengal, Andhra Pradesh, Bihar).
The annual precipitation in India, including snowfall is estimated at 4000 billion
m3 (BCM). Out of this, the seasonal rainfall in monsoon is of the order of 3000
BCM. The rainfall in India shows great temporal and spatial variations, unequal
seasonal distribution and geographical distribution and frequent departures from
the normal (Mohapatra and Singh, 2003).
Accurate information on the extent of water bodies is important for flood prediction,
monitoring, and relief (Smith, 1997; Baumann, 1999). Often, this information
is difficult to produce using traditional survey techniques because water bodies can
be fast moving as in floods, tides, and storm surges or may be inaccessible. The
synoptic, repetitive nature of satellite remotely sensed data have allowed
monitoring
of water bodies over large regions of land. In the optical range (visible-infrared)
water has a distinctively low spectral response. In many studies, Landsat data (TM
334 S. K. JAIN ET AL.
and MSS), IRS or SPOT have been used to determine the extent of water bodies
using simple classification procedures, usually with an infrared band. These studies
have relied on the water bodies having a unique spectral response in this range
of electromagnetic radiation when compared to the surrounding landscape (Wang,
2002; Frazier and Page, 2000). Many studies using density slicing of Landsat MSS
band 7 (Bennett, 1987), TM bands 4 and 7 (Wang, 2002; Frazier and Page, 2000)
have been reported in literature. Manavalan et al. (1993) used Landsat TM band 4
to map the extent of the Bhadra reservoir, India. Overton, 1997 used density slicing
of Landsat TM band 5 and a high flood spatial mask to map water bodies on the
Murray river between Blanchetown and Wentworth, South Australia. Shaikh et al.
(1997) used band 4 of Landsat TM to map flood extent of the Mississippi river
but experienced problems separating water from certain urban features without the
inclusion of an additional band.
The river Ganges as it flows down in the flatter plain of Bihar and further to
Bengal, its flood plain extended much beyond the defined riverbanks. In both banks
of the Ganges exists such land zones, which behaves not only as flood plain but
also as detention basin for up-land flows. Kahalgaon Super Thermal Power Project
(KhSTPP) of NTPC is located near Kahalgaon town, District Bhagalpur of Bihar
State, India. The plant area and its adjoining areas were subjected to severe flood
in August 1993, September 1995 and September 1999. The flood was brought by
Koa river coupled with back water effect due to high flood level of Ganges river
and inundated a vast areas upstream and adjoining the power station (situated in
Koa river basin) and also affected the drainage inside the plant area. The lower
part of Koa basin out falling to Ganges where the Kahalgaon Power plant and
other structures are situated is also in flood plain area. Topographically, the plant
is situated at the mouth of a catchment (Koa nala) near the confluence of Ganga.
Insufficient natural drainage aggravates the flood situation in the catchment of Koa
Nala. In this paper, IRS-1C LISS III and Landsat TM data for mapping of flooded
areas have been used. A digital elevation model (DEM) has also been used.
2. The Study Area and Data Used
The river Koa emerges from the hills at an elevation of about 400 m. The river
in general has very narrow width and traverses about 48 km and then joins river
Ganga. The general slope of the river is very mild except in some upper reaches
of the catchment. The catchment area of Koa river up to Eastern Railway Bridge
is 677.0 km2 and up to Marry Go Round (MGR), 584 km2. The catchment is
bounded between latitude 2500_00__N to 2515_10__N and Longitude 8713_51__E
to 8738_28__E. The catchment area has a shape of key, i.e. wide in the beginning
and
narrow at the mouth. The annual normal rainfall is about 122 cm. On an average, it
rains for about 5060 days in a year. Most of the rain occurs (about 7580%) during
the monsoon season (JuneSeptember). The PublicWorks Department road bridge
is the most downstream bridge in Kao river and is just upstream of confluence of
DELINEATION OF FLOOD PRONE AREAS 335
Koa with Ganges river. The deck level is of the order of elevation about 34.5 m.
The eastern railway bridge is approximately 100.0 upstream of the Public Works
Department road bridge and the clear water way is 100 m. The elevation of railway
track is about 35.50 (NHPC, 1995).
The satellite data used were as follows:
Satellite/sensor Path/row Date of pass
Landsat TM 13943 27 May 1995
18 October 1995
IRS-1C LISS III 10654 March 1999
December 1999
Topographical maps at a scale of 1:50,000 have been used. The area is covered
in topographical maps Nos. 72 O/4, 7, 8 and 12.
3. Methodology
3.1. PREPARATION OF DATABASE
In the present study for preparation of drainage map, contour map, etc., Survey
of India toposheets has been used. The Koa catchment is covered in two
topographical
maps at a scale of 1:50,000. The drainage and contour maps of the Koa
catchment were digitized directly from topographical maps. The drainage map of
the Koa catchment along with MGR and the plant is shown in Figure 1. The data
Figure 1. The study showing MGR and road.
The satellite data of the years 1995 and 1999 received in digital format
from the National Remote Sensing Agency (NRSA), Hyderabad, India, was loaded
in ERDAS.
3.2. PROCESSING OF SATELLITE DATA
To study the extent of flood inundation, the analysis of Landsat TM scenes and
IRS-1C LISS III digital data have been performed. For flood mapping, two sets of
the remotely sensed data are required; one set consisting of data acquired during
the
occurrence of the flood. In reality, data availability may cause some compromise,
because it is difficult to obtain cloud free data during floods. Therefore, in this
study, efforts were made to take satellite data just before and after the flood event.
Selection of the dates of the satellite scenes is made based on observed flood
events.
Satellite data as mentioned earlier were geocoded with topographical maps using
ground control points. The Polyconic projection system is used for geocoding the
digital data. The geocoded scenes are masked by boundaries forKoa catchment
digitized
from topographical maps and imported in ERDAS IMAGINE. Some clearly
identifiable features like crossing of rivers, canals, sharp turns in the rivers, roads,
bridges, etc., were located on both map and image and were selected as control
points. About 10 such points were selected for geo-referencing in all the cases.
Now looking at the statistics, some points that generated big errors were deleted
and replaced by other points so as to obtain the better geo-referencing. After
completing
this process, the image was displayed over the map and the superimposition
was compared. The geo-referencing was found to be quite good. Then the
catchment
area was masked. The False Colour Composite (FCC) of Landsat TM for two
seasons, i.e. May and October 1995 are shown in Figures 2 and 3.
DELINEATION OF FLOOD PRONE AREAS 337
Figure 2. False Colour Composite (FCC) of Landsat TM for May 1995.
Figure 3. False Colour Composite (FCC) of Landsat TM for May 1995.
satellite data. Therefore, the density-slicing approach using different band have
been attempted. Then two approaches, one based on the Tasseled Cap
Transformation
(wetness) and the other based on water index approaches were applied. The
results of the all the methods have been analysed and compared. These methods
are discussed in the following sections.
3.4. DENSITY SLICING OF SINGLE BAND
In many studies, Landsat data (TM and MSS), IRS or SPOT have been used to
determine the extent of water bodies using density-slicing approach, usually with
an infrared band. These studies have relied on the water bodies having a unique
spectral response in this range of electromagnetic radiation when compared to the
surrounding landscape. Baumann (1999) carried out a study using all the bands
of Landsat TM. He concluded that band 4 (NIR) represented the best spectral
band that identifies the flood-inundated area well. Wang et al. (2002) carried out
a study for identifying the water-affected area and he observed that bands 4 and 7
(2.082.35 m) are well suited for the purpose. In this study, initially, six bands of
Landsat TM (except thermal band) and three bands of IRS LISS III (except MIR
band) have been analysed. In Figure 4, six Landsat TM individual band histograms
for the October 1995 scene are presented. It is seen that in bands 13, there is
considerable spectral overlap between the digital values found on the water bodies
and those in the surrounding urban/vegetation areas. The infrared band gives a
much
better representation in the water-related features than the visible bands did. Each
of the visible bands displays a uni-model histogram with no indication of a separate
group of data for water pixel. The infrared bands show a bi-model histogram. The
density slicing ranges for infrared band starts at near zero, meaning thatwater
pixels
are the darkest in the image. In this study, the analysis of bands 4 and 7 of TM data
DELINEATION OF FLOOD PRONE AREAS 339
Figure 4. Individual Landsat TM band histograms October 1995.
and band 3 of LISS III have been made. It was found that the results obtained
using band 7 are not representing water-related features very well when compared
with raw data; therefore, only band 4 has been taken. Initially, near infrared band
(band 4 in TM data and band 3 in IRS LISS III) has been taken for mapping of
flood. As the reflectance of water in near infrared (NIR) band is minimum, the area
340 S. K. JAIN ET AL.
of low reflectance has been taken using this approach. Once the representation of
the reflectance values for water and non-water features was clear, a cut-off value of
digital numbers could be determined to separate the two categories. For the
October
TM image the cut-off value was 53, means if the digital number was less than 53,
that pixelwas assigned aswater category. For the May image ofTMdata, the cut-off
value was 60, i.e. if digital number was less than 60, that pixel was classified as
water, otherwise as non-water. In case of IRS LISS III data, the cut-off values for
December 1999 scene was 54 and for March 1999 scene it was 71. NIR band is
able to identify most of the major water features, but there is still some mixing with
the hill shadows, vegetation area, and therefore another method based on Tasseled
Cap Transformation (wetness) has been applied.
3.5. TASSELED CAP TRANSFORMATION
There are numerous methods available for enhancing spectral information content
of
satellite data. The Tasseled Cap Transformation compressed the total information
into three bands: greenness, brightness and wetness. Besides expressing a large
amount of image variability within three bands, tasseled cap bands could be directly
related to physical scene characteristics. A Tasseled Cap Transformation to data
from the Landsat TM has been given (Crist and Cicone, 1984). The coefficients for
wetness functions are:
1234567
0.1509 0.1793 0.3299 0.3406 0.306 0.7112 0.4572
One of the main reasons for supporting the use of the Tasseled Cap Transformation
method against, for example, the principal component technique is that the
coefficients of the transformation are defined a priori. This method was applied on
the Landsat TM data only. The coefficients for IRS LISS III data were not available.
The earlier wetness function was applied and threshold values of this index
52.0 to 34.0 for October and 35.0 to 19.0 for May have been chosen for water
features. This method was good for identification of water-related features when
compared with original satellite data. But it could not eliminate the shadow effects
caused mainly due to forest, etc. To overcome the problem of shadows, a band ratio
technique has been applied.
3.6. NDWI APPROACH
There are numerous vegetation indices developed to estimate vegetation cover with
the remotely sensed imagery. A vegetation index is a number that is generated by
some combination of remote sensing bands. The most common spectral index used
to evaluate vegetation cover is the Normalized DifferenceVegetation Index (NDVI).
McFeeters (1996) developed an index similar to the NDVI, which is called the
Normalized DifferenceWater Index (NDWI). Any sensor having a green band and a
DELINEATION OF FLOOD PRONE AREAS 341
NIR band can provide for this index. TheNDWIwas derived using principles similar
to those that were used to derive the NDVI. The NDWI is calculated as follows:
NDWI = (GREEN NIR)
(GREEN + NIR)
(1)
where GREEN is a band that encompasses reflected green light and NIR represents
reflected near-infrared radiation. The selection of these wavelengths was done to:
(1) maximise the typical reflectance of water features by using green light
wavelengths;
(2) minimise the low reflectance of NIR by water features; and (3) take
advantage of the high reflectance of NIR by terrestrial vegetation and soil features.
When Equation (1) is used to process a multi-spectral satellite image that contains
a reflected visible green band and NIR band, water features have positive values;
while soil and terrestrial vegetation features have zero or negative values, owing to
their typically higher reflectance of NIR than green light. Image processing software
can easily be configured to delete negative values. This effectively eliminates the
terrestrial vegetation and soil information and retains the openwater information for
analysis. The range of NDWI is then from zero to one. Multiplying Equation (1) by
a scale factor (e.g. 255) enhances the resultant image for visual interpretation. Now
in the output obtained water-related features were identified. The cut-off values of
NDWI for Landsat TM were chosen as 21 and above for May, 1995 and 39 and
above for October 1995. For IRS LISS III, the cut-off values for the March scene
was 55 and above while for the December scene it was chosen as 60 and above.
3.7. FLOOD INUNDATION USING DEM
The flood inundation map using DEM based on the river gauge readings before
the flood event was prepared. The river gauge station is near the meeting of Kao
river with Ganga River. The simple threshold used for this purpose was possible
due to the relatively flat terrain (no sink). In the present study, the area inundated
corresponding to gauge level of 33 m were obtained.
4. Results and Discusion
To quantify the flood-affected areas the whole catchment was subdivided in two
areas; one downstream of MGR experiencing maximum flood and other upstream
MGR where flood but most of the areas, which seems flood affected is actually
forest covered/shadow area, affect some area. In Figures 5 and 6 the results of
density slicing for TM band 4 are shown. This band was able to identify major
water bodies, but there is still some mixing with the urban areas, hill shadows,
etc. In fact, upstream of MGR a very little area is affected by flood; therefore,
the flood-affected areas downstream of MGR is only shown in all the figures. The
density slicing of the other infrared bands (5 and 7) of the TM data were also
tried but they did not produce good results. The density slicing of the visible bands
342 S. K. JAIN ET AL.
Figure 5. Flood-affected area using density-slicing approach, May 1995.
Figure 6. Flood-affected area using density-slicing approach, October 1995.
343
substantially overestimate the area ofwater on the image, the explanationwas given
earlier. The combination of mixed pixels and high turbidity limits the use of a single
NIR band to classify water bodies. The second approach applied was based on the
Tasseled Cap Transformation. In this method, the wetness index based on certain
coefficient has been computed. Using this approach also, mixing of land use and
overestimation of water area has been observed. It was observed that this method
is
good during post-monsoon season (after floods) when enough water is available but
it is not effective during pre-monsoon season. Also this method could not eliminate
shadow areas. The flood-inundated areas for May 1995 scene is shown in Figure 7
and for October 1995 shown in Figure 8.
In the NDWI approach, the results was better and the mixing of pixels have been
eliminated up to certain extent. Using this approach also, flood-affected areas could
not be delineated clearly because this area is not like water body. In case of IRS
LISSIII data, the water and sediment-laden water both are represented by positive
values of NDWI. While in case of TM data, the values of NDWI are positive for
clear water but the values for sediment-laden water is negative also. For TM data,
the flood-affected areas are shown in Figures 9 and 10 for May 1995 and October
1995, respectively.
Even applying different approaches of image processing, scattered water features
were observed. In case of river gauge/DEM method area submerged withwater
can be obtained clearly, but there are other limitations of the method. First, the
contour
interval available for the study area was at a interval of 20 m which is not quite
DELINEATION OF FLOOD PRONE AREAS
345
good; therefore, the DEM prepared was not of the high quality. Second, the
gauge data available were not of the dates corresponding to the flood event.
After identifying water versus non-water areas on all the images using the
aforementioned
approaches, determination of flood-affected areas were made. The area
346 S. K. JAIN ET AL.
affected by flood in these two regions, i.e., upstream of MGR and downstream
of MGR is given in Table I. In the last column, the ratio of flood-affected area
in downstream of MGR with total flood-affected area is given. From this table, it
is observed that the flood-affected area downstream of MGR is maximum using
NDWI approach. Also in percentage it represents the maximum area (more than
50%) for post-monsoon season. It means that using NDWI approach, water-related
feature is best identified. The ratio in two cases, i.e. single band and Tasseled Cap
Transformation for pre-monsoon image is very lowi.e. about 10%. From the figures
obtained using these two methods, it is observed that the area falls in the upper
most
part of the catchment where shadow effect is prominent. In case of wetness-based
approach, the ratio is very low (about 10%) during pre-monsoon season, indicating
poor response.
5. Conclusions
In this study, methods for mapping flood extent through the use of IRS LISS III and
Landsat TM as well as DEM data have been applied. The density-slicing method
was based on a comparison of the reflectance features of the water versus nonwater
targets on a pair of satellite images (one acquired before and the other during the
flood event). All the visible bands proved to be inadequate for successful density
slicing for identification of water pixels. The combination of mixed pixels and high
turbidity limits the use of single band (NIR) to classify water. The other methods
that have been applied include wetness index from Tasseled Cap Transformation
and NDWI approaches.
NDWI-based approach is produces best results for mapping of flood-inundated
areas when verified with original satellite data. Also, this compares well with results
derived from DEM. In this study, there are certain limitations such as the data
availability during flood season due to cloud cover and quality of DEM.
References
Baumann, P., 1999, Flood analysis: 1993 Mississippi Flood, URL: http://www.research.umbc.edu/
tbenjal/baumann/mod2.html.
Bennett, M.W. A., 1987, Rapid monitoring of wetlandwater status using density slicing, in Proceedings
of the 4th Australian Remote Sensing Conference, 1418 September, Adelaide, pp. 682691.
Crist, E. P. and Ciconel, R. C., 1984, A physically based transformation of Thematic Mapper
dataThe TM Tasseled Cap, IEEE Trans. Geosci. Remote Sens. GE-22, 256263.
Frazier, P. and Page, K., 2000, Water body detection and delineation with Landsat TM data,
Photogrametric
Eng. Remote Sens. 66, 12.
McFeeters, S. K., 1996, The use of Normalised Difference Water Index (NDWI) in the delineation
of open water features, Int. J. Remote Sens. 17(7), 14251432.
Mohapatra, P. K. and Singh, R. D., 2003, Flood management in India, Nat. Hazards 28, 131143.
NHPC, 1995, Project Report on Kahalgaon Super Thermal Power Project, a Technical Report
submitted to National Thermal Power Corporation (NTPC) by Naitonal Hydro Power Corporation
(NHPC).
DELINEATION OF FLOOD PRONE AREAS
347
Kundu, B. S. and MohiKumar,K. E., 1995, Mapping and management of flood affected areas through
remote sensing. A case study of Sirsa district, Harayana, J. Indian Soc. Remote Sens. 23(3).
Smith, L. C., 1997. Satellite remote sensing of river inundation area, stage, and discharge: A review,
Hydrol. Process. 11, 14271439.
Wang, Y., Colby, J. D. and Mulcahy, K. A., 2002, An efficient method for mapping flood extent in a
coastal flood plain using Landsat TM and DEM data, Int. J. Remote Sens. 23(18), 36813696.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 49, NO. 1, JANUARY 2011 85
Sadiq I. Khan, Yang Hong, Jiahu Wang, Koray K. Yilmaz, Jonathan J. Gourley, Robert F.
Adler,
G. Robert Brakenridge, Fritz Policelli, Shahid Habib, and Daniel Irwin
AbstractFloods are among the most catastrophic natural disasters
around the globe impacting human lives and infrastructure.
Implementation of a flood prediction system can potentially
help mitigate flood-induced hazards. Such a system typically requires
implementation and calibration of a hydrologicmodel using
in situ observations (i.e., rain and stream gauges). Recently, satellite
remote sensing data have emerged as a viable alternative or
supplement to in situ observations due to their availability over
vast ungauged regions. The focus of this study is to integrate the
best available satellite products within a distributed hydrologic
model to characterize the spatial extent of flooding and associated
hazards over sparsely gauged or ungauged basins. We present
a methodology based entirely on satellite remote sensing data
to set up and calibrate a hydrologic model, simulate the spatial
extent of flooding, and evaluate the probability of detecting inundated
I. INTRODUCTION
Fig. 1. Map showing the Nzoia river basin in the Lake Victoria region, East Africa.
TABLE I
SELECTED FLOOD EVENTS, LOCATION, FLOODED AREAS/RIVER (VERIFIED WITH THE DFO FLOOD INVENTORY).
NUMBERS IN PARENTHESES ARE THE JULIAN DAYS OF THE CORRESPONDING YEAR
KHAN et al.: SATELLITE REMOTE SENSING AND HYDROLOGIC MODELING FOR FLOOD INUNDATION MAPPING 89
Fig. 2. Schematic of the satellite remote-sensing- and hydrologic-modeling-based flood monitoring system.
Fig. 3. (In blue on top abscissa) Observed precipitation, (in black) observed
runoff, and (in red) simulated runoff for the Nzoia basin during the calibration
period (19851998).
forcing CREST with the TMPA 3B42RT rainfall, the data were
accumulated into daily amounts and linearly resampled onto the
30-arc-second resolution model grid.
2) Model Calibration and Validation: CREST was calibrated
using the available daily discharge observations for the
period between 1998 and 2004. A one-year period (1998) was
used for warming up the model states. The model utilizes a
global optimization approach to capture the parameter interactions.
An autocalibration technique based on the Adaptive
Random Search (ARS) method by Brooks [40] was used to
calibrate the CREST model. The ARS method is considered
adaptive in the sense that it uses information gathered during
previous iterations to guide the parameter search in the current
step. The two most commonly used indicators to assess model
skill in matching the model-simulated streamflow with observations
are the NashSutcliffe Coefficient of Efficiency (NSCE)
[41] and the relative bias ratio (Bias). These two criteria are
used as objective functions for the automatic calibration. The
best skill occurs with NSCE 1 and Bias 0%
NSCE =1 _(Qi,o Qi,c)2
_(Qi,o Qo)2
(1)
Bias = _Qi,o _Qi,c
_Qi,o
100% (2)
where Qi,o is the observed discharge at the ith time step, Qi,c
is the simulated discharge at the ith time step, and Qo is the
90 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 49, NO. 1, JANUARY 2011
TABLE II
2 2 CONTINGENCY TABLE FOR CREST-SIMULATED AND
SATELLITE-BASED FLOOD EXTENT COMPARISONS
KHAN et al.: SATELLITE REMOTE SENSING AND HYDROLOGIC MODELING FOR FLOOD INUNDATION MAPPING 91
Fig. 4. (A1A4) MODIS-based flood inundation maps for December 4, 2006 and August 15, 22, and 24, 2007. (A5)
ASTER inundation map for November 12,
2008. (B1B5) MODIS true-color composite of bands 1, 3, and 4. (C1C5) MODIS false-color composite of bands 7, 2,
and 1.
Fig. 5. Comparison of satellite-based and CREST-simulated flood inundation extents. First legend entry is the year
and the Julian day of the flood event, followed
by the event identification number (refer to Table I).
V. CONCLUSION AND FUTURE WORK
KHAN et al.: SATELLITE REMOTE SENSING AND HYDROLOGIC MODELING FOR FLOOD INUNDATION MAPPING 93
Fig. 6. Evaluation of hydrologic model-based spatial flood extents using inundated areas derived from (A1A4)
MODIS and (A5) ASTER. Evaluation is
performed as a function of search radius.
See discussions, stats, and author profiles for this publication at:
http://www.researchgate.net/publication/258028252
2
READS
67
3 AUTHORS:
Igor Ogashawara
Indiana University-Purdue University India
38 PUBLICATIONS 45 CITATIONS
SEE PROFILE
Marcelo Curtarelli
National Institute for Space Research, Brazil
43 PUBLICATIONS 37 CITATIONS
SEE PROFILE
Celso M. Ferreira
George Mason University
35 PUBLICATIONS 30 CITATIONS
SEE PROFILE
All in-text references underlined in blue are linked to publications on ResearchGate,
ABSTRACT
Flood maps are a crucial tool to support emergency management, disaster recovery and risk reduction planning.
Traditional flood mapping methods are time-consuming, labor intensive, and costly. Our goal in this paper is to
introduce a novel technique to aggregate knowledge and information to map coastal flooded areas. We proposed
a Difference of Normalized Difference Water Indices (DNDWI) derived from two LANDSAT-5/TM surface
reflectance product acquired before and after the passage of Hurricane Ike, for Upper Texas in September of
2008. The reference flooded area was delineated interpolating the maximum surge in each location using a spline
with barriers method with high tension and a 30 meter Digital Elevation Model (DEM). It was noticed that
NDWI values decreased after the hurricane landfall on average from 0.226 to 0.122 for flooded area. However
for the non-flooded areas it increased from 0.292 to 0.300. Results from the Monte Carlo simulation showed that
mapping flooded areas with DNDWI got an accuracy of 85.68% while the non-flooded areas got an accuracy of
92.13%. Thus, DNDWI is promising tool for mapping flooded areas since it is a cheaper and simple technique
which can be applied rapidly for several areas of the planet.
Keywords - Flood mapping, Hurricane, Monte Carlo Simulation, NDWI, Optical Remote Sensing
I. INTRODUCTION
Hurricanes are one of the most costly natural
disasters in the United States [1] and recent storms
such as Hurricane Sandy (2012), Ike (2008) and
Katrina (2005) have caused major infrastructure
damage and losses of lives along the East and Gulf
Coasts [2]. Tropical cyclones are also a major cause of
flooding and damage in several regions worldwide
including recent events in the Bay of Bengal, Typhoon
Phailin (2013), and historically deadly storms such as
Tropical cyclone Thelma (1991), in the North Pacific
Ocean, and Cyclone Zoe (2002), in the South Pacific
Ocean, among others. Coastal flooding is one of the
major hazards to accompany a tropical cyclone
landfall [3] and can be aggravated by the combination
of the storm tidal surge and rainfall-runoff from the
heavy precipitations.
Traditionally, coastal flooding due to tropical
cyclones has been estimated by measured water levels
on buoys and coastal gages (e.g., [4-5]). Although
these monitoring networks provide good historical
data for coastal flooding, it lacks spatial information
due to the limited number of stations over large areas.
Recent developments of physics based numerical
models using High Performance Computing (HPC)
(e.g., [6]) in addition to an increasingly large volume
of high resolution data (e.g., topo/bathymetry, land
use, wind fields) has led to an unprecedented
improvement in accuracy of tropical cyclones flood
prediction and mapping. A combination of numerical
III. DATASET
The dataset used in this study include in situ
measurements and remote sensing images. The in situ
data were collected by the USGS mobile storm surge
network [8], which provide atmospheric pressure and
water level data at each 6 minutes. The data were
measured using a pressure transducer (HOBO
Onset) and stored in the USGS database [8].
Topography was obtained from a Digital Elevation
Model (DEM) dataset extracted from the National
Elevation Dataset (NED) [13] representing the entire
region topography at a resolution of 1 arc-second.
The remote sensing data comprise images
collected by the TM sensor, onboard LANDSAT-5
satellite. This images are provided with 7 spectral
bands (from the visible to thermal spectral regions),
quantized in 8 bits and with 30 meters spatial
resolution, except the thermal band which has 120 m
spatial resolution. The images are acquired at each 16
days. In this study we used the surface reflectance
product from the LANDSAT CDR. This product is
generated from specialized software called Landsat
Ecosystem Disturbance Adaptive Processing System
(LEDAPS) which provide images converted for
reflectance values and corrected for the atmospheric
effects. More information about this product is
available at [14]
IV. METHODS
mm
mm
NDWI
0.86 1.24
0.86 1.24
4 5
( 4) ( 5)
TMB TMB
TMB TMB
NDWI
VI. CONCLUSION
A methodology for mapping spatial
variations of flood inundation caused by hurricanes
events using optical remotely sensed image series was
developed in this study. Since flood prevention,
management and emergency response is an important
issue for policy makers, flood detection using remote
sensing can improve the number of monitored areas in
remotely accessed places or in places without any
monitoring program. As optical orbital sensors, like
Landsat TM, ETM+ and OLI family, are collecting
non-stop data from the entire planet, the use of an
optical sensor could enhance the knowledge of
flooding spatial mapping in areas with lack of data.
Our results showed that the DNDWI and a threshold
analysis method for distinguish flooded areas could be
a potential tool to enhance the knowledge of tropical
cyclones flooding mapping. It was observed, from a
MCM technique, that 85.68% of 10000 mean values
from an interaction of 20 to 1000 DNDWI values were
accurately classified as flooded areas. For the nonflooded
area an accuracy of 92.13% was found.
However, these results were based on only
one event (Hurricane Ike) and should be extended to
other study areas. We also observed that additional
spectral behavior studies are needed to explain the
relationship between water content and vegetation
spectral response in flooded areas. Nevertheless, we
proposed a methodology which could be an useful tool
for countries without any flooding monitoring
program and RADAR imagery cover, since the
mapping of flooded areas is an important issue for the
economy and the rebuilding of the affected region.
Although the use of Synthetic Aperture Radar has
been previously applied, the cost for buying the
images is high as well as it is labor intensity. Using
optical remote sensing is not only cheaper but also
much easier to manage the data. Through the advance
of orbital hyperspectral sensors like the Hyperspectral
Imager for the Coastal Ocean (HICO), it will be
possible to identify the key spectral ranges to identify
the elevated degree of water content in vegetation.
Thus the hyperspectral studies will enhance the use of
optical remote sensing to map flooded areas.
VII. ACKNOWLEDGEMENTS
The first author is grateful to the Brazilian
CITATIONS
55
READS
248
1 AUTHOR:
Ian Overton
The Commonwealth Scientific and Indus
38 PUBLICATIONS 209 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: Ian Overton
Retrieved on: 10 December 2015
RIVER RESEARCH AND APPLICATIONS
ABSTRACT
The lower River Murray in South Australia is highly regulated through weirs and water extraction for
irrigation. Management of
the river for environmental purposes requires an understanding of the extent of floodplain inundation
from various flows and
weir manipulations. This study aimed to produce a floodplain inundation model for the 600 km long
and 15 km wide portion of
the River Murray in South Australia from the New SouthWales border to Lake Alexandrina. The model
was developed using a
Geographical Information System (GIS), remote sensing and hydrological modelling. Flood inundation
extents were monitored
from Landsat satellite imagery for a range of flows, interpolated to model flood growth patterns and
linked to a hydrological
model of the river. The resulting model can be analysed for flows ranging from minimum flow to a 1-in13-year flood event for
any month and weir configuration and has been independently tested using aerial photography to an
accuracy of approximately
15% underestimate. The results have proven the approach for determining flood inundation over a
large area at approximately
one-tenth of the cost of detailed elevation and hydrodynamic modelling. The GIS model allows
prediction of impacts on infrastructure,
wetlands and floodplain vegetation, allowing quantitative analysis of flood extent to be used as an
input into the
management decision process. Copyright # 2005 John Wiley & Sons, Ltd.
key words: flood inundation; modelling; hydrology; GIS; remote sensing; River Murray
INTRODUCTION
The River Murray is Australias largest river and plays a vital role in Australias economic and
environmental
resources. The MurrayDarling Basin catchment occupies one-seventh of Australia and
contributes 70% of
Australias irrigated crops and pastures. The River Murray provides 100% of Adelaides water
supply in dry
months and is the major water supply source to the towns in the basin. The study region
receives almost all of
the catchment run-off, minus the water extracted due to irrigation and evaporation, as all
tributaries including
the Darling River enter the River Murray further upstream.
Management of the River Murray in southeastern Australia has been implemented to
mitigate large floods and to
protect infrastructure, while maintaining storages for regular water supply to irrigators.
Concerns over river health
have increased the attention on environmental flow strategies over the past decade. The
focus is on releasing and
managing flows to provide environmental benefits to the floodplain, wetlands and in-stream
water quality. No
quantitative tool for measuring the extent and volume of flood inundation existed for the
lower part of the River
Murray prior to the original work (Overton et al., 1999). This project was commissioned to
create such a model to
provide input into the management decision process. Previous studies on modelling river
flow and inundation have
been undertaken for wetlands on the Darling River in New South Wales (Shaikh et al., 2001)
and floodplains on
Roanoke River in North Carolina (Townsend and Walsh, 1998) and the River Murray in
Australia (Overton et al.,
1999; Frazier et al., 2003).
Spatial decision support is one of the main roles of Geographical Information Systems (GIS)
and provides an
excellent framework for the integration of multi-criterion evaluation results (Taylor et al.,
1999; Jankowski et al.,
2001). The flood inundation model was developed using a spatial information system
framework that allowed the
integration of non-spatial river flow models with the extent of flood inundation. Riverine
ecosystems benefit from
spatial analysis studies because they encompass three important temporally dynamic
spatial dimensions (along the
Received 12 October 2004
Revised 13 December 2004
Copyright # 2005 John Wiley & Sons, Ltd. Accepted 19 January 2005
*Correspondence to: I. C. Overton, CSIRO Land and Water, Urrbrae, SA, 5064, Australia; School of Earth and
Environmental Sciences,
University of Adelaide, SA, 5005, Australia. E-mail: ian.overton@csiro.au
channel, river height and floodplain width) and the physical and ecological processes taking
place are complex.
Much work has been done on the longitudinal (upstream and downstream) dimension
examining the ecological
impacts of river regulation on native flora, fauna and the physical changes occurring in the
littoral zone (Stanford
et al., 1996). However, long-term modification of flow rates, changes to the frequency of
flooding events and alteration
to the timing of flows have now been identified as causing degradation beyond the littoral
zone into both the
lateral and vertical dimensions (Young, 2001). This is particularly prevalent in the River
Murray ecosystem where
rising saline groundwater has contributed to land degradation throughout the region and
increased accession of
saline groundwater to the river (Jolly, 1996). Furthermore the semi-arid nature of the Murray
floodplain means
that the variability in the size and timing of flows is extreme and unpredictable (Walker and
Thoms, 1993). Consequently,
river regulation in the Murray has severely impacted the balance of physical and ecological
processes
that maintain this unique riverine environment. Managing the river system to balance social,
economic and environmental
requirements, when resources are scarce, requires quantitative analytical tools (Young et al.,
2000). This
paper describes the development of a predictive model to assess the effect of flow
management options on floodplain
inundation.
Mapping flood inundation
The degradation of the River Murray has increased interest and the need for information and
predictive modelling
of river health and the impacts from management and land use options. An Environmental
Flows Decision
Support System (Young et al., 2000) has been developed to look at a number of river
environmental factors such as
area of wetland and aquatic and terrestrial fauna. This model has been advanced into the
Murray Flow Assessment
Tool (Young et al., 2003). Neither model, however, looks at the extent of varying flood sizes
and considers floodplains
as one unit with a commence-to-fill flow and a rate of fill.
Remote sensing is particularly useful for monitoring flood extents because it provides basic
data more cheaply
and efficiently than ground-based methods (Whitehouse, 1989). Previous studies to
determine the aerial extent
of flooding have commonly involved optical satellite image analysis (Usachev, 1985; Walker
et al., 1986;
Townsend and Walsh, 1998; Overton et al., 1999; Shaikh et al., 2001; Sheng et al., 2001;
Frazier et al., 2003),
radar remote sensing (Townsend and Walsh, 1998) or an integration of remote sensing and
GIS (Brivio et al.,
2002). These methods have proved to be very useful and economical for large-area flood
analysis. More detailed
studies have used digital elevation models to create a floodplain surface that can be
inundated at certain river
heights (Townsend and Walsh, 1998). Elevation methods are particularly useful for predictive
studies of changing
flow paths through the floodplain by manipulating flow barriers. However, surface modelling
may not give the best
representation of flood inundation, as there are numerous impediments and small channels
across a predominantly
flat floodplain. The modelling of flood inundation from surface elevation also requires
detailed information on
stage heights, backwater curves, flow impedances and roughness coefficients as simple
height levels are insufficient
in this dynamic environment.
Due to the high level of control over flows in this region, the river exists as a series of pools
under most conditions.
Floods rise and fall in days or weeks and last for several weeks to months, meaning that the
extent of flood
inundation for a given flow has usually extended to its full potential extent before the flood
peak passes. This type
of regulated river is therefore modelled sufficiently from images of various events. Upstream
in the unregulated
reaches, stream flooding is very rapid and sporadic and two flood peaks of equal magnitude
are unlikely to create
the same extent of inundation (Frazier et al., 2003).
Study area
The study area was chosen to capture the floodplain environments of the River Murray in
South Australia from
the New SouthWales border toWellington (Figure 1). This is generally referred to as the lower
River Murray and is
highly regulated by six weirs (which are situated alongside locks for river navigationthe
total structure is usually
referred to as a lock) and the barrages at the mouth of the river. It has large extents of
floodplain and is a discharge
area for the regional saline groundwater system of the MurrayDarling Basin. The ecological
significance of the
Chowilla floodplain, the largest floodplain in this region, is highlighted by its listing under the
UNESCO Ramsar
Convention as a Wetland of International Importance (National Environmental Consultancy,
1988) and its
992 I. C. OVERTON
Copyright # 2005 John Wiley & Sons, Ltd. River Res. Applic. 21: 9911001 (2005)
inclusion in the list of Significant Ecological Assets of the River Murray under the Murray
Darling Basin
Ministerial Councils The Living Murray Initiative (Murray Darling Basin Commission, 2003).
The regulation
of this section of the river allows flows to be manipulated for environmental management to
mitigate increasing
salinization on the floodplain and enhance wetland environments.
Figure 1. Study area and detail showing locks and localities
MODELLING FLOODPLAIN INUNDATION ON A REGULATED RIVER 993
Copyright # 2005 John Wiley & Sons, Ltd. River Res. Applic. 21: 9911001 (2005)
The floodplains along this stretch of the river range from gorge sections below Overland
Corner (Figure 1),
23 km wide and 3040m deep, to valleys 510 km wide flanked by broad floodplains
(Walker and Thoms,
1993). Figure 2 shows a schematic cross-section of the floodplain. GIS data are available for
the study area including
the floodplain boundary and the permanent water in the river channel and wetlands;
however, no data were
available for the flood extent under different flows other than the extent of the 1956 flood
(the largest on record)
at approximately 290 000 Ml/day which covered what we now define as the floodplain
boundary.
METHODOLOGY
Satellite imagery was obtained for this region at a range of magnitudes of flood events and
therefore correct interpretation
of these data provides information on the spatial extents of flood inundation. The process of
developing
the flood inundation model is described in detail below. Figure 3 shows a schematic flow
diagram of the process
involved.
Figure 2. Cross-section of the local geomorphology
Figure 3. Flow diagram of the process of building the Flood Inundation Model
994
I. C. OVERTON
Copyright # 2005 John Wiley & Sons, Ltd. River Res. Applic. 21: 9911001 (2005)
boundary. Rainfall events were not taken into consideration in choosing the image dates as
water not connected
to the river was able in most cases to be filtered out at a later stage.
Figure 4. Histogram of a 102 Gl/day flow with the central line indicating the point at which the reflectance value is
cut (density slice) to
indicate water or not water. Two low-reflectance-value peaks represent Lake Victoria (with deep water) and the
River Murray (with shallow
turbid water)
MODELLING FLOODPLAIN INUNDATION ON A REGULATED RIVER 995
Copyright # 2005 John Wiley & Sons, Ltd. River Res. Applic. 21: 9911001 (2005)
events to provide a more continuous predictive model. Interpolation between the discrete
flow intervals was performed
to produce finer intervals of flood extent.
The extents of the flood masks of each flow from the satellite imagery provided a boundary
line of equal flood
extent. These pixels were interpolated to obtain the flow at all other pixels in the image.
There are many ways to
Table I. Image path and dates for the 21 satellite Landsat TM images used in the study. All images were
from Row 84
Flow Path 97/Row 8485 Path 96/Row 84 Path 95/Row 84
(Gl/day) Lakes to Overland Corner Overland Corner to Lake Victoria Lake Victoria to Wentworth
Date (Gl/day) Date (Gl/day) Date (Gl/day)
<20 9-Nov-94 6.45 30-Sep-94 3.5 9-Nov-94 6.1
2030 20-May-89 26.7 20-Nov-89 25.8 8-May-90 27.6
3040 10-Aug-96 37.8 11-Oct-92 38.2 12-Aug-96 38
4045 25-Jul-90 40.6 27-Oct-86 43.8
4550 3-Aug-90 46.7 21-Sep-93 47.4
5060 20-Sep-96 57 2-Sep-95 55.1
6070 23-Nov-96 68.3 23-Nov-96 68.3 7-Oct-93 66.4
7075 31-Jul-89 70.5
7580 16-Aug-89 78.1 16-Aug-89 77.4
80100 20-Sep-90 93.4 3-Oct-89 82.4 26-Sep-89 80.2
>100 22-Nov-93 109.9 26-Oct-90 101.6 29-Sep-90 101.9
996
I. C. OVERTON
Copyright # 2005 John Wiley & Sons, Ltd. River Res. Applic. 21: 9911001 (2005)
perform this kind of interpolation. The true situation is defined by the local topography of the
area which is
unknown in this case. Therefore interpolation of the flow level is similar to the problem of
interpolating the landscape
height at each point. Kriging is the most common surface interpolation method (Burrough,
1986) but is influenced
by areas outside the adjacent known boundaries being the two closest satellite masks. It
was decided to use
an image morphological process called a marker-based watershed segmentation algorithm
rather than traditional
interpolation methods to ensure that the information from each satellite image was
preserved. The watershed algorithm
is commonly used in mathematical morphological problems and is often used in relation to
topographic
analysis of digital elevation models (Vincent and Soille, 1991). The advantage of this method
over other
approaches to contour interpolation is that it can be applied to very noisy data with broken
contours. This was
the case here since the data are quantized so coarsely.
Flood interpolation was undertaken by dividing the combined satellite mask image into
regions of constant
minimum flow. Using this method, interpolated values were ensured of lying within the
minimum and maximum
flow bounds by interpolating each region independently. In each region, contours of equal
flow are interpolated
from the boundary points, at which the minimum flow is known. This contour interpolation is
repeated iteratively,
each time based on the contours that have already been estimated. Each contour is
interpolated using a flooding
simulation extending from the next higher and lower boundaries. Regions having the
minimum flow value or representing
land that did not flood at the largest flow were not included and were used as the lower and
upper limit of
interpolation accordingly.
The difficulty in applying the watershed algorithm to flood interpolation is in choosing the
source and sink
points. The sink points are points adjacent to the region that have the next highest
quantized flow level. The source
points are those adjacent points that have the highest flow that is lower than the sink points
for this region. The
locations of these sink and source points were chosen to model the behaviour of flood
growth across the floodplain
but also to represent the filling of wetlands from a single inflow channel. A set of colour
aerial photographs of a
70 000 Ml/day flood were used to validate the growth behaviour of floods, and a small
number of edits were made
to the final flood inundation grid.
GIS development
The result of the interpolation stage was a raster grid of cell values that represent the
commence-to-flow based
on the flows on the day of the satellite images. This grid was then filtered to remove noise
and converted to a
polygon coverage.
Filtering is a technique used to enhance the quality of digital imagery by changing the
values of cells in raster
images, and can be used to ease the computation burden of raster to vector conversion by
removing isolated pixels
(Trotter, 1991). Filtering uses neighbouring cells to determine the value of the cell in
question and can sharpen or
smooth images to emphasize features or reduce the effects of erroneous cells (Wilkinson,
1996). To remove
anomalies in the interpolated flood masks, such as higher or lower value pixels in the middle
of lakes, a majority
filter was used. This filter replaced the value of each pixel with the value of the eight nearest
neighbours if there is a
majority of neighbouring cells with a different value from the cell.
At this stage of the processing the imagery was still in raster format (pixel-based data). The
raster data were
converted to vector data (area- or shape-based data) to reduce the data redundancy and
allow for easy retrieval,
updating and generalization of graphics and attributes that are especially important if
equations are to be used to
generate models.
Hydrological modelling
Output from the hydrological models was provided as input data to the GIS. The GIS model
then used table
look-ups to query the required hydrological parameters that then link to the spatial layers in
the GIS. This type
of loose coupling allows existing hydrological models to be run without reprogramming
within the GIS (Karami
and Houston, 1996; Sui and Maggio, 1999). As the hydrological parameters do not change,
this proved an effective
and economical approach.
River flow and height model. To create a model that would predict flood inundation from a
regulated river, the
river height at every kilometre is required. Actual recorded data of river height at certain
kilometres from the
MODELLING FLOODPLAIN INUNDATION ON A REGULATED RIVER 997
Copyright # 2005 John Wiley & Sons, Ltd. River Res. Applic. 21: 9911001 (2005)
mouth of the river, and flow at the gauging station during the flooding events, were obtained
from the South
Australian Water Authority, the organization which manages the river in this region. The
limited number of
gauging stations meant that a modelled series of backwater curves were also used. These
were a series of water
surface elevation curves between weirs for a steady uniform flow computed by the Murray
River Level Module
(MURLEV), part of the River Murray Flow and Salt Transport (RMFST) computer model (Water
Studies, 1992).
Figure 5 shows the flow curves derived from these methods for a section of the river from
the border to weir 2. The
flow rate of 3000 Ml/day equates to a river height referred to as weir pool level.
The river is held higher than pre-regulation by the weirs, which causes the water levels to
have a backwater curve
rather than a straight flow down hill. River heights near the mouth of the river are much
lower than near the border
due to downfall flow of the river and attenuation of the flow. Once flow levels have reached
60 000 Ml/day the river
is above the maximum height of the weirs and backwater curves become straight.
Flood units. As river heights can be manipulated by the six weirs, predominantly
independent of each other, the
floodplain needed to be divided up into regions that could be related to the local conditions
in the river. Areas where
the floodplain inundation responded to the same point in the river were identified using flood
behaviour and river
morphology as a guide. Regions identified that respond to a particular point in the river were
termed Flood Units.
Flood Units were assigned the closest kilometre marker from the hydrological model (trigger)
or the trigger that
was assumed would most likely cause that area to flood. As a height difference of several
metres can occur between
the upper and lower pool levels of the weirs, behaviour of the flow path around the weir is
complex. Aweir manipulation
trial at Lock 5 demonstrated that the area around the weir responded to changes in the river
height below the
weir. There is very little difference between the river height within 2-km intervals in the
modelled backwater
curves, which implies that there is probably an error margin of approximately two triggers
up- or downstream
in the assignment of a Flood Unit to a trigger. Each polygon for each particular reach now
contained attributes
for Reach, Flood Unit, Flow and Trigger.
Once the data had been converted to polygons in the GIS it was clear that each Flood Unit
contained small areas
of floodplain that were inundated at the same particular flow. These areas were termed
unique Ecological Units as
they will have similar ecology due to their similar elevation and the fact that distribution of
floodplain vegetation
communities is strongly related to flooding frequency (Van Der Sommen, 1987).
Hydrological inputs to the model. The same flow rate at the border gauging station can
produce different river
elevations along the river depending on the weir elevations, the antecedent conditions and
the time of year. Local
river height predominantly determines whether the river will break the banks and flood an
adjacent Flood Unit.
River height at kilometre markers from state border to lock 2
0.00
5.00
10.00
15.00
20.00
25.00
362.1
368
378
388
403
409
418
429
439
454
480
489
500
515
520
531
542
555
566
575
584
594
603
623
633
642
River km markers
River height (m)
3
26
47
55
82
102
Flow rate at border
(Gl/day)
Figure 5. River levels for kilometres of the river from weir 2 to the SANSW border for different flows
998 I. C. OVERTON
Copyright # 2005 John Wiley & Sons, Ltd. River Res. Applic. 21: 9911001 (2005)
Plate 1. Example of seasonal attenuation curves
Backwater Profile for Lock Pool 3,
at 5cm above normal pool level
9.2
9.4
9.6
9.8
10.0
10.2
10.4
10.6
10.8
11.0
11.2
11.4
11.6
11.8
12.0
12.2
12.4
12.6
12.8
13.0
13.2
13.4
13.6
13.8
14.0
430 440 450 460 470 480 490 500 510 520
AMTD (km)
AHD (m)
56295
50000
40000
30000
20000
15000
10000
5000
pool level
Flow (ML/day)
Other factors such as antecedent conditions were not included in the model. In the hot dry
summer period a flow
rate at the border will be further reduced downstream than in wetter months as water is
evaporated and extracted
from the river and less is replaced by rainfall. Curves were generated which show the
attenuation of flow at each
weir for a given month. Plate 1 shows an example of the change in flow rate during the year
in each lock reach for a
flow of 60 000 Ml/day at the border.
The RMFST model was used to generate backwater curves for a range of weir manipulations
and flow values.
The model estimates the stagedischarge matrices for a nominated water level and
discharge at the weir, which
defines the upstream water levels to the next weir. In this way relationships can be derived
between the water levels
at the downstream and upstream ends of the river segment and discharge. The river level
and discharge relationships
for each weir reach were established using a linear relationship. Plate 2 shows the backwater
curve for the
weir pool of Lock 3 at a level of 5 cm above normal pool level. The precision of the model is
5 cm in the river
height predictions.
Decision support system development
The height of the river at triggers had to be related to the management areas, the Ecological
Units derived from
the original flow values and the interpolation process. With area inundated extremely
sensitive to slight alterations
in river height, it was decided that a relationship between river height and area of flood
inundation should be used
in each Flood Unit. This relationship has the potential to be improved with further image
acquisition and statistical
analysis. The model will always be limited to the spatial resolution at which the images are
acquired, which for
Landsat TM is 30m, regardless of the number of images used. If a change in the height of
the river results in less
than a 30m change at the flood boundary, the area of inundation will not increase. Other
satellite systems may be
useful in reducing the pixel size and therefore identifying finer changes in flood inundation.
In the region downstream of Lock 3, the height of the river continues to increase without
breaking its banks until
it reaches approximately 35 000 Ml/day. As the water flows onto the floodplain there is a
gradual increase in area
with increasing flow until the river level rises to reach the edges of the trench (Figure 2). In
the region upstream of
Lock 3 the river height increases until it breaks its banks and continues to increase as it fills
up wetlands and then
gradually tapers out as the water spreads over the floodplain. The relationships at each
floodplain unit were used to
code the flood masks with a river height which would cause the Ecological Unit to be
inundated. The height of the
river was assigned to the Ecological Unit for all Flood Units using the Flood Unit trigger as
the point that would
cause this Ecological Unit to flood. A dialogue screen was then designed with input fields to
ask the user the values
for the flow at the border, the month of the year and the height of each lock. These values
are then used to perform a
query which selects all the river height codings in the map layer where the local river height
is less than that predicted
for the given flow (attenuated for the time of the year).
The flood inundation model is a steady-state model that predicts the extent of flooding from
a given flow on the
first day of the flood. It does not consider the effect of antecedent conditions or the effect of
flood duration. Further
research on the wetting and drying behaviour of the floodplain and its wetlands needs to be
incorporated into the
model to be able to predict time sequences for management scenarios. The model is limited
by its ability to predict
areas of inundation given the same floodplain topography. Changes in elevations due to
construction of levee banks
and regulators will cause changes in the area of inundation not modelled by the Flood
Inundation Model. The
model can provide wetting and drying cycles for wetlands, days since last flooding, average
return period, volume
of water, extent of inundation and other hydrological indicators that could be linked to
floodplain ecological processes.
Better understanding is needed to determine, or at least demonstrate in most cases, these
linkages between
the hydrological indicators and the benefits to the environment.
RESULTS
The Flood Inundation Model produced has been used as the basis for an environmental flow
strategy in South
Australia (SKM and Mapping and Beyond, 2000) and a number of other studies on the
hydrology and ecology
of the floodplain. The Flood Inundation Model was used to predict the area of inundation
expected to occur as
a result of the raising of Lock 5 near Renmark in 2000 (DWLBC, 2002). Aerial photography
was obtained during
MODELLING FLOODPLAIN INUNDATION ON A REGULATED RIVER 999
Copyright # 2005 John Wiley & Sons, Ltd. River Res. Applic. 21: 9911001 (2005)
the raised weir and was interpreted for area of inundation. The spatial accuracy of the model
in predicting the area
inundated was estimated to be 15% underestimate of the actual inundated area from the
photography. A database of
the surveyed heights of 126 bores on the floodplain was used to independently test the
accuracy of the height data
derived by the model. The mean difference between the surveyed height and modelled
height was 0.07m with a
standard deviation of 0.88 m. Plate 3 shows an example of the GIS Flood Inundation Model.
The initial stages of the project provided us with Ecological Units and the height of the river
at a particular
trigger that would cause these areas to flood. The hydrological modelling allows the monthly
simulation of a flow
from the border under different weir configurations, with the GIS providing the spatial view
of the area inundated.
The inputs to the decision support system are the flow at the border, the weir configurations
of all six weirs and the
month. The outputs of the model will be the river heights at each trigger kilometre and the
area of inundation. The
flood inundation model created is in a GIS framework and has incorporated layers such as
riparian vegetation,
transport and water infrastructure and major wetlands. The science of incorporating spatial
patterns in decision
support systems through the integration of GIS and process models is suggested as a critical
tool for the future
of information integration in the environmental sciences (Taylor et al., 1999; Hendriks and
Dirk, 2000).
The total area of inundation of the floodplain in South Australia, including the whole of the
Chowilla floodplain,
which is partly in New South Wales, down to Wellington, is 117 827 hectares (downstream of
the gauging station
on Figure 1). The floodplain is defined as the extent of the 1956 flood which was
approximately 258 000 Ml/day.
Plate 4 shows the increasing area of inundation of the River Murray floodplain from
5000Ml/day to 100 000Ml/
day. The area of permanent water is approximately 24 000 hectares and is made up of the
river channel, wetlands
and anabranch creeks.
CONCLUSION
Satellite imagery of known flood events has been interpolated to create a model of flood
growth. This flood growth
has been linked to a hydrological model of the river, making predictions of the extent of
flood inundation from
given river flows and manipulation possible. The GIS framework could be utilized to
determine flow patterns and
losses across the floodplain, incorporating wetlands as sources and sinks to create a more
dynamic model of flooding
(Costelloe et al., 2003).
The integration of the hydrological model with the GIS has enabled simulation for both
scientific research and
policy management. The visualization, quantitative analysis and spatial correlation of
environmental and infrastructure
data of the GIS has improved the usefulness of the hydrological modelling.
The model has been used successfully to predict flood inundation in a weir manipulation trial
(DWLBC, 2002)
and has been used as the basis for the development of a flow management strategy for the
River Murray in South
Australia (SKM and Mapping and Beyond, 2002). The model has also provided a useful
surface elevation model in
further floodplain modelling work. Current work on modelling floodplain vegetation health
affected by soil salinization
processes will be linked to the flood inundation model to improve its usefulness as a decision
support
system. Further research on the behaviour of floods on the floodplain and in wetlands is
required before it can
be used as a temporal model.
ACKNOWLEDGEMENTS
The author wishes to acknowledge funding through the Murray Darling Basin Commission,
the River Murray
Catchment Management Board and the South Australian Water Corporation (MDBC Natural
Resource Management
Strategy R6045 project). Assistance on the initial project design was provided by Brenton
Erdmann and Bob
Newman of the South Australian Water Corporation. Jamie Sherrah performed the
mathematical image interpolation,
while Suzanne Slegers and Jane Lawley, of Mapping and Beyond Pty Ltd, assisted with the
data manipulation.
REFERENCES
Brivio PA, Colombo R, Maggi M, Tomasoni R. 2002. Integration of remote sensing data and GIS for accurate mapping
of flooded areas. International
Journal of Remote Sensing 23(3): 429441.
Burrough AP. 1986. Principles of Geographical Information Systems for Land Resources Assessment. Oxford
University Press.
1000 I. C. OVERTON
Copyright # 2005 John Wiley & Sons, Ltd. River Res. Applic. 21: 9911001 (2005)
Costelloe JF, Grayson RB, Argent RM, McMahon TA. 2003. Modelling the flow regime of an arid zone floodplain river,
Diamantina River,
Australia. Environmental Modelling and Software 18(89): 693703.
DWLBC. 2002. Lock 5 Flow Enhancement Trial. Murray-Darling Division, Department of Water, Land and Biodiversity
Conservation, South
Australia. Report No. 1.
Frazier PS, Page KJ. 2000. Water body detection and delineation with Landsat TM data. Photogrammetric
Engineering and Remote Sensing
66(12): 14611467.
Frazier PS, Page KJ, Louis J, Briggs S, Robertson A. 2003. Relating wetland inundation to river flow using Landsat TM
data. International
Journal of Remote Sensing 24(19): 37553770.
Hendriks P, Dirk V. 2000. From geographical information systems to spatial group decision support systems: a
complex itinerary. Geographical
and Environmental Modelling 4: 1.
Jankowski P, Andrienko N, Andrienko G. 2001. Map-centred exploratory approach to multiple criteria spatial decision
making. International
Journal of Geographical Information Science 15(2): 101127.
Jolly ID. 1996. The effects of river management on the hydrology and hydroecology of arid and semi-arid
floodplains. In Floodplain Processes,
Anderson MG, Walling DE, Bates PD (eds). John Wiley and Sons: New York; 577609.
Karami HA, Houston BH. 1996. Evaluating strategies for integrating environmental models with GIS: current trends
and future needs. Computers,
Environment and Urban Systems 20(6): 413425.
Munyati C. 2000. Wetland change detection on the Kafue Flats, Zambia, by classification of a multitemporal remote
sensing image dataset.
International Journal of Remote Sensing 21(9): 17871806.
Murray Darling Basin Commission. 2003. The Murray Darling Basin Commission website. http://www.mdbc.gov.au
[November 2003].
National Environmental Consultancy. 1988. Chowilla Salinity Mitigation SchemeDraft Environmental Impact
Statement. Report prepared by
National Environmental Consultancy for the Engineering and Water Supply Department of South Australia: Adelaide.
Overton IC, Newman B, Erdmann B, Sykora N, Slegers S. 1999. Modelling floodplain inundation under natural and
regulated flows in the Lower
River Murray. Second Australia Stream Management Conference Proceedings, 811 February 1999, Adelaide,
Australia: 493498.
Pressey RL. 1986. Wetlands of the River Murray below Lake Hume. Environmental Report 86/1. Murray-Darling Basin
Commission: Canberra.
Shaikh M, Green D, Cross H. 2001. A remote sensing approach to determine the environmental flows for wetlands of
the Lower Darling River,
New South Wales, Australia. International Journal of Remote Sensing 22(9): 17371751.
Sheng Y, Gong P, Xiao Q. 2001. Quantitative dynamic flood monitoring with NOAA AVHRR. International Journal of
Remote Sensing 22(9):
17091724.
SKM and Mapping and Beyond. 2002. Decision Support System for Managing Flows in the River Murray, South
Australia. Report prepared for
the River Murray Catchment Management Board, the South Australian Department of Water, Land and Biodiversity
Conservation, and the
Natural Heritage Trust.
Stanford JA,Ward JV, Liss WJ, Frissell CA,Williams RN, Lichatowich JA, Countant CC. 1996. A general protocol for
restoration of regulated
rivers. Regulated Rivers: Research and Management 12: 391413.
Sui DZ, Maggio RC. 1999. Integrating GIS with hydrological modelling: practices, problems and prospects.
Computers, Environment and
Urban Systems 23: 3351.
Taylor K, Walker G, Abel D. 1999. A framework for model integration in spatial decision support. International Journal
of Geographical
Information Science 13(6): 533555.
Townsend PA,Walsh SJ. 1998. Modelling floodplain inundation using an integrated GIS with radar and optical remote
sensing. Geomorphology
21: 295312.
Trotter C. 1991. Remotely-sensed data as an information source for geographical information systems in natural
resource management: a
review. International Journal of Geographical Information Systems 5(2): 225239.
Usachev VF. 1985. Evaluation of flood plain inundations by remote sensing methods. Hydrological applications of
remote sensing and remote
data transmission. Proceedings of the Hamburg Symposium, August 1983, Goodison BE (ed.). Publication No. 145.
International Association
Hydrological Sciences: Washington; 235243.
Van Der Sommen FJ. 1987. A Review of Components and Processes Influencing the Ecological Status of the
Vegetation Component of the
Riparian Ecosystems of the River Murray Floodplain in South Australia. South Australian Department of Environment
and Planning Report.
Vincent L, Soille P. 1991. Watersheds in digital spaces: an efficient algorithm based on immersion simulations. IEEE
Transactions on Pattern
Analysis and Machine Intelligence 13(6): 583598.
Walker KF, Thoms MC. 1993. Environmental effects of Flow Regulation on the Lower River Murray, Australia.
Regulated Rivers: Research and
Management 8: 103119.
Walker RD, Aubrey MC, Fraser SJ. 1986. Remote Sensing in Australia. Canbec Press: Melbourne.
Water Studies. 1992. River Murray Flow Model, Vol. 3. User Manual. Department of Engineering and Water Supply.
Whitehouse G. 1989. Flood monitoring and floodplain studies. In Remote Sensing in Hydrological and
Agrometeorological Applications, de
Vries D (ed.). COSSA Publications: Canberra.
Wilkinson GG. 1996. A review of the current issues in the integration of GIS and Remote Sensing data. International
Journal of Geographic
Information Systems 10(1): 85101.
Young WJ (ed.). 2001. Rivers as Ecological Systems: The Murray-Darling Basin. Murray Darling Basin Commission:
Canberra, Australia.
Young WJ, Lam DC, Ressel V, Wong IW. 2000. Development of an Environmental Flows Decision Support System.
Environmental Modelling
and Software 15: 257265.
Young WJ, Scott AC, Cuddy SM, Rennie BA. 2003. Murray Flow Assessment Tool: A Technical Description. CSIRO
Client Report for the CRC
for Freshwater Ecology.
MODELLING FLOODPLAIN INUNDATION ON A REGULATED RIVER 1001
Copyright # 2005 John Wiley & Sons, Ltd. River Res. Applic. 21: 9911001 (2005)
HYDROLOGICAL PROCESSES
Hydrol. Process. 19, 36993716 (2005)
Published online 1 August 2005 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/hyp.5852
Abstract:
Flooding due to excessive rainfall in a short period of time is a frequent hazard in the flood plains of
monsoon Asia. In
late September 2000, a devastating flood stuck Gangetic West Bengal, India. This particular event has
been selected for
this study. Instead of following the conventional approach of flooded area delineation and overall
damage estimation,
this paper seeks to identify the rural settlements that are vulnerable to floods of a given magnitude.
Vulnerability
of a rural settlement is perceived as a function of two factors: the presence of deep flood water in and
around the
settlement and its proximity to an elevated area for temporary shelter during an extreme hydrological
event. Landsat
ETMC images acquired on 30 September 2000 have been used to identify the non-flooded areas within
the flooded
zone. Particular effort has been made to differentiate land from water under cloud shadow. ASTER
digital elevation
data have been used to assess accuracy and rectify the classified image. The presence of large
numbers of trees around
rural settlements made it particularly difficult to extract the flooded areas from their spectral
signatures in the visible
and infrared bands. ERS-1 synthetic aperture radar data are found particularly useful for extracting the
settlement
areas surrounded by trees. Finally, all information extracted from satellite imageries are imported into
ArcGIS, and
spatial analysis is carried out to identify the settlements vulnerable to river inundation. Copyright
2005 John Wiley
& Sons, Ltd.
KEY WORDS flood; remote sensing; settlement vulnerability; GIS
INTRODUCTION
Flood is a perpetual natural hazard in the flood plains of monsoon Asia, where over 80% of
annual precipitation
is received in the four wet months from June to September. The problem of river flooding is
of great concern
in the Indian state of West Bengal. The Irrigation and Waterways Department of West Bengal
Government
reports that, since its independence in 1947, there has only been 5 years in which it has
been spared from the
effects of monsoon inundation. One of the most devastating of its kind was experienced in
SeptemberOctober
2000. A total of 23 756 km2 land area was inundated and 221 million people were affected
(Rudra, 2001).
This particular flood event caught everybody unaware. The unprepared condition of the
administration to
cope with this kind of natural calamity was thoroughly exposed. In recent years, a number of
studies have
recognized the importance of estimating peoples vulnerability to natural hazards, rather
than retaining a
narrow focus on the physical processes of the hazard itself (Hewitt, 1997; Varley, 1994;
Mitchell, 1999).
Cannon (2000) argued that natural disaster is a function of both natural hazard and
vulnerable people. He
emphasized the need to understand the interaction between hazard and peoples
vulnerability.
Although most of the developed countries are well equipped with detailed flood hazard
maps, up-to-date
flood insurance maps and post-disaster hazard mitigation technical support (FEMA, 2003),
there is hardly any
* Correspondence to: Joy Sanyal, Department of Geography, 1 Arts Link, National University of Singapore, Singapore
117570, Singapore.
E-mail: g0202381@nus.edu.sg
detailed spatial database for flood prevention and mitigation in the developing countries. In
recent years, efforts
have been made to use remote sensing and geographic information systems (GISs) for
creating national-level
flood hazard maps for Bangladesh (Islam and Sado, 2000a). Population density and other
socio-economic
data have been integrated with hydrologic information to identify priority zones for
implementing anti-flood
measures (Islam and Sado, 2002). These studies were undertaken on a regional scale using
coarse-resolution
AVHRR imageries from NOAA satellites. The results of such investigations would only be
useful for nationallevel
macro planning.
The focus of this paper is on individual settlements. Its purpose is to analyse how the
location of an
individual settlement vis-`a-vis the flood-prone zone and its socio-economic characteristics
make it vulnerable
to monsoon floods. The theoretical framework of this study is based upon the hypothesis
that settlements
are vulnerable to flood from three aspects: (1) whether the people have access to relatively
higher ground to
take shelter during an extreme hydrological event; (2) whether a settlement falls in a zone
that is expected
to experience a high flood discharge causing extraordinary damage of life and property; (3)
whether the
population density of the area is high enough to result in huge loss of property even in a
moderate flood.
The example used for this study is Gangetic West Bengal, which is a natural flood-prone area
whose fluvial
characteristics have made it very suitable for rice cultivation. The population density in this
region is one of
the highest in the world. Although human settlement is abundant in this region, the local
people traditionally
settled only on relatively higher ground, known locally as danga. Rapidly increasing
population density, due
to both natural growth and the influx of millions of refugees in the post-independence era
from the then East
Pakistan, left very little choice for the people to settle on higher ground selectively. Severe
shortage of land
has forced people to settle indiscriminately over the highly flood-prone zone.
Landsat ETMC and ERS synthetic aperture radar (SAR) imageries are used in this study to
classify nonflooded
areas and flood depth within flooded zones, and to delineate human settlements at village
level. The
high spatial resolution of satellite imageries enables us to obtain detailed classification
results that are suitable
for formulating planning measures on a small scale. An added advantage is that the high
resolution hydrologic
information can be conveniently integrated with demographic data collected from smaller
administrative units.
This would greatly enhance the capability of the spatial database to estimate vulnerability of
individual
settlements to an extreme flood event.
STUDY AREA
The study area extends over three major river basins of southern West Bengal, namely
Bhagirathi-Hoogly,
Jalangi and Churni. All these three rivers are distributaries of the main branch of the River
Ganga. Although
we have tried to cover the natural region of the three river basins, the extent of the
investigation is marginally
compromised due to limited availability of digital terrain data. The eastern parts of the
districts of Bardhaman,
Murshidabad, and most of Nadia form the administrative entity for the area (Figure 1). The
three river
basins are overwhelmingly rural, with agriculture as the main source of livelihood. According
to Bagchis
(1945) subregional classification of the Bengal Delta, the study area is identified as a
moribund delta. In this
section of the delta, the rivers are decaying and the land-building process has entirely
ceased. Owing to its
comparatively higher elevation and high levees, this area is traditionally less flood prone
than the area that
lies further south. The area falling between the Rivers Bhagirathi and Jalangi is an elongated
depression,
and the Churni basin area is almost entirely low lying in comparison with rest of Gangetic
West Bengal.
Therefore, this zone is liable to flooding. The study states that, in the Nadia and Hoogly
districts, this belt is
bounded by 10 m contour lines. Interfluves of the numerous distributaries are ill drained
(Spate, 1965) and
very often cause waterlogging during the monsoon season. This situation ultimately led to
stagnation of water
and development of cut-off channels known as bills. An abundance of oxbow lakes and misfit
river channels
also characterize this part of Gangetic West Bengal. There is a marked distinction in the
channel pattern of
the streams lying east and west of the River Hoogly. Sinuosity indices of the rivers on the
eastern side of the
Copyright 2005 John Wiley & Sons, Ltd. Hydrol. Process. 19, 36993716 (2005)
REMOTE SENSING AND GIS-BASED FLOOD VULNERABILITY ASSESSMENT 3701
Bhatar
Chapra
Pandua
Kaliganj
Kandi
Mangolkot
Nakshipara
Domkal
Sagardighi
Chakdaha
Noada
Jamalpur
Lalgola
Berhampur
Balagar
Nabagram Jalangi
Manteshwar
Tehatta-I
Bardhaman-I Santipur
Memari-I
Hanskhali
Kalna-I
Ranaghat-II
Hariharpara
Katwa-I
Memari-II
Karimpur-II
Kalna-II
Ketugram-I
Ranaghat-I
Krishnanagar-I
Beldanga-II
Karimpur-I
Tehatta-II
Katwa-II
Bharatpur-I
Beldanga-I
Raninagar-I
Purbasthali-II
Krishaganj
Raninagar-II
Ketugram-II
Bharatpur-II
Bhagawangola-I
Nabadwip
Murshidabad-Jiaganj
Purbasthali-I
Bhagawangola-II
Krishnanagar-II
Bardhaman-II
Legend
Rivers
Development Blocks
Study Area
Study Area
Location within India
50 25 0 50 Kilometers
N
River Hoogly are very high compared with the western side (Goswami, 1983). The overall
geomorphology
of the study area depicts a degenerating fluvial system.
Owing to its geographical location, i.e. at the tail end of the extensive Ganga basin, West
Bengal has a
very limited capacity to control extreme hydrological events ensuing from the upper
catchment of the River
Ganga and its tributaries. Very high precipitation over a short period of time is cited as the
most important
factor responsible for triggering devastating floods in Gangetic West Bengal (The Times of
India, 2000). After
the Independence of India, 1956, 1959, 1978, 1995, 1999 and 2000 are identified as years
that received
abnormally high precipitation and, hence, severe floods (Basu, 2001). The 2000 flood in
SeptemberOctober
was the worst in terms of its scale and damage caused. The West Bengal Government
estimated that a total of
171 blocks of the state _23 756 km2_ area was affected. The total loss was estimated to be
56 600 million rupees
(US$1132 million) (Ganashakti, 2000). Abnormally high rainfall for 4 days in the upper
catchment areas of
the western tributaries of the River Bhagirathi were responsible for this natural calamity. The
severity of the
event was so high that many low-lying areas of the Nadia district remained waterlogged for
over 3 weeks,
with the depth of water estimated as high as 3 m (Rudra, 2001).
Copyright 2005 John Wiley & Sons, Ltd. Hydrol. Process. 19, 36993716 (2005)
3702 J. SANYAL AND X. X. LU
extracted from satellite imageries to identify effectively the settlements that are highly
vulnerable to flood
hazard. In this section, each component of the total work flow has been dealt with
separately. Special attention
has been given to illustrating how each of these components contributes to the ultimate
objective of this study.
Delineating non-flooded areas from flooded areas
The current study is concerned more with dry/land area than flooded area. Delineation of the
non-flooded
area is particularly important because these areas can serve as a temporary shelter for the
nearby settlements.
This information is necessary for identifying the settlements that are highly vulnerable to
flooding. Settlements
having no immediate access to dry areas would be considered highly vulnerable to flooding.
From the early era of passive remote sensing, special attention has been given to
distinguishing water
from dry surface. MSS band 7 _0811 m_ has been found to be particularly suitable for
distinguishing
water or moist soil from dry surface due to its strong absorption of water in the near-infrared
(NIR) range of
the spectrum (Smith, 1997). MSS data were used to deal with the flood-affected areas in
Iowa (Rango and
Solomonson, 1974), Arizona (Morrison and Cooley, 1973), and the River Mississippi basin
(Deutsch et al.,
1973; Deutsch and Ruggles, 1974; Rango and Anderson, 1974; McGinnis and Rango, 1975).
From the early 1980s, Landsat TM data with an improved spatial resolution of 30 m have
become one of
the major sources of remotely sensed data for flood management research. Landsat TM
band 4 is spectrally
a near equivalent of MSS band 7. Water yields very low reflectance in the NIR region of the
spectrum and,
therefore, can be effectively used to discriminate water from land surface. This property of
Landsat band 4 has
been extensively used to delineate flooded areas in West Africa (Berg and Gregiore, 1983),
India (Bhavsar,
1984) and Thailand (Raungsiri et al., 1984).
For the current research, two Landsat ETMC scenes of the study area acquired on 30
September 2000 have
been obtained. The imageries were acquired nearly at the peak of the flood. The scenes
were geometrically
and radiometrically corrected (level 1G product from US Geological Survey (USGS)). The two
scenes have
been accurately georeferenced using GPS control points collected during a field visit to the
study area. All
bands of the two scenes have been mosaicked. TM bands 4, 3 and 2 have been projected in
RGB to generate
a standard false colour composition (FCC) of the study area. Although TM band 4 is useful in
delineating
land and water boundaries, asphalt road surfaces and rooftops also yield very low
reflectance in this band.
Reflectance from water varies sufficiently from roads and dark rooftops in Landsat band 7
(228235 m,
mid infrared). Therefore, water and non-water can be effectively discriminated by adding TM
band 4 and
band 7 (Wang et al., 2002).
The predominance of cloud cover during the flooding season is one of the major obstacles to
using optical
remote sensing in flood management. Although SAR can penetrate cloud cover, scientific
communities in
the developing countries still prefer optical data over radar because of two reasons.
Purchasing radar data
is beyond the means of most of the government agencies in the developing countries
because of its high
price. Radar satellite needs prior programming to capture data for a particular area. Owing
to its very high
rate of data capture it requires a real-time data transfer to a receiving station near the
coverage area. Lack of
ground receiving stations and proper coordination has resulted in limited coverage of SAR
data for developing
countries. Major satellites that operate in the optical portion of the spectrum maintain a
year-round archive
of global coverage and, therefore, are still more acceptable in the field of remote sensing.
Copyright 2005 John Wiley & Sons, Ltd. Hydrol. Process. 19, 36993716 (2005)
REMOTE SENSING AND GIS-BASED FLOOD VULNERABILITY ASSESSMENT 3703
After adding TM bands 4 and 7, an effort was made to mask out the cloud-contaminated
pixels, as their
existence can interfere with any classification effort. There are a number of algorithms for
screening clouds
in optical imageries (England and Hunt, 1985; Saunders, 1986). Cloud appears very bright in
a standard FCC
image. After a detailed comparison of the band _4 C 7_ image with the FCC, it has been
found empirically
that any pixel having a digital number (DN) of over 124 in band _4 C 7_ is cloud. The pixels
with a (DN)
range >124 have been masked. It has been observed that it is very easy to extract cloudcovered pixels over a
dark background (e.g. flooded area in band _4 C 7_), but over a bright background, like
settlements or healthy
vegetation, pixels at the periphery of a cloud cluster have a very similar reflectance to the
background.
Therefore, the above-mentioned threshold is not able to remove small numbers of cloudcovered pixels
over bright and dry surfaces. Since this study focuses on flooded areas, the existence of
small numbers of
cloud-contaminated pixels in the predominantly dry areas is not of great concern.
Band _4 C 7_ is quite efficient in discriminating flood water from dry land. Major boundaries
of flood water
can be delineated without much effort. However, there are still a few water pixels that are
not classified
as flooded within the rural settlements (Figure 2). These pixels can be readily identified in a
standard FCC.
The reflectance of these pixels is very close to the nearby wet soil surface. During floods, the
albedo from
the water body increases significantly because of a high concentration of debris and silt in
the water. Thus,
the reflectance peak moves toward the red band. On the other hand, increasing soil
moisture decreases soil
albedo, making reflectance from some non-flooded pixels very similar to flood pixels (Sheng
et al., 1998).
From a close comparison of band _4 C 7_ with the FCC, a binary classification has been done
as follows for
band _4 C 7_:
Pixel value > 78 D dry land
Pixel value _ 78 D water
This classification effectively extracts water pixels within the settlement area, but the main
disadvantage of
this classification is that it cannot distinguish between dry surface and water under cloud
shadow.
Areas under cloud shadow receive only scattered sunlight, but the low illumination results in
a suppressed
reflectance from all land-cover categories. Therefore, it is very difficult to discriminate
between land and
water for the areas under cloud shadow (Sheng et al., 1998). Water has a significantly low
reflectance in
the NIR region of the electromagnetic spectrum compared with dry land surface. Owing to
low reflectance
of non-flooded areas under cloud shadow, the above-mentioned threshold spuriously
classifies it as water
(Figure 3). As a consequence, the classified image appears as an underestimation of nonflooded area and an
overestimation of water. Actually, the difference in the reflectance between flooded and nonflooded regions
becomes so low that it is not possible to separate them by means of a threshold value. Land
surface reflects
Flood affected area within
a rural settlement (water
pixel)
Figure 2. Landsat ETMC FCC (zoomed eight times from optimum resolution) showing flooded area within a
settlement
Copyright 2005 John Wiley & Sons, Ltd. Hydrol. Process. 19, 36993716 (2005)
3704 J. SANYAL AND X. X. LU
Land under cloud shadow
Water under cloud shadow
Masked out cloud
Figure 3. FCC showing flooded and non-flooded areas under cloud shadow
higher energy in the red band (Landsat band 3) compared with water. On the other hand,
reflectance from
water is significantly lower in the NIR band compared with land. Therefore, the ratio of NIR
and red (band
4/band 3) increases the difference in the reflectance of flooded and non-flooded pixels. This
difference in the
ratio image would be sufficient to distinguish water from land under cloud shadow. After
comparison with
the FCC it has been found that, although the ratio image is effective in differentiating water
and land under
cloud shadow, it is not as sensitive as band _4 C 7_ to water surface. The problem arises
from situations
where the water pixels mixed with non-flooded pixels have a higher value than the open
water pixels in the
ratio image. Sometimes, values of such pixels overlap with the non-flooded pixels
surrounded by water under
cloud shadow. Hence, the ratio image is not able to identify a cluster of water pixels
surrounded by land.
In this paper, an attempt has been made to synthesize the advantage of band _4 C 7_ and
the ratio image
for extracting the non-flooded area from the flood scene. After thorough observation of the
DN of the ratio
image and Landsat FCC, it has been found empirically that pixel values ranging from 052
to 1 in the ratio
image represent non-flooded areas under cloud shadow. Pixels that fall within the specified
range in the
ratio image have been selected and merged with non-flooded pixels derived from band _4 C
7_. It should be
pointed out that these two regions derived from different manipulation of Landsat bands
have not been found
mutually exclusive in their spatial coverage. Overlapped areas are dissolved to create one
classified layer
depicting the non-flooded area. In the synthetic image, the advantages of band _4 C 7_ and
band (4/3) have
been incorporated to extract the non-flooded area in the flood scene accurately.
There is no doubt that the above classification scheme enhances the accuracy. However, it
still suffers from
limitations. The roofs of submerged houses appear as dry surface, and in areas of compact
settlement this
error becomes significant. The tree canopy also creates confusion in delineating the
boundary between water
and land. Flooded areas under canopy appear as non-flooded. Although forest constitutes an
insignificant
land-cover category in Gangetic West Bengal, rural settlements are often surrounded by
trees. This factor
constitutes a major source of error in the classification scheme.
Site visits to some selected settlements were made during the fieldwork in 2003 and local
people were
interviewed to locate specific sites for validation. However, after incorporating that
information it was found
that it is not scientifically viable to do an accuracy assessment based on peoples
recollection of a flood
event that occurred 2 years ago. In the absence of any aerial photograph during the flood
we embarked
upon using a digital elevation model (DEM) to assess the accuracy of our classification
indirectly. To assess
the accuracy of the classified image, pixels classified as non-flooded have been
superimposed on a DEM
derived from the advanced space-bome thermal emission and reflection radiometer (ASTER)
images of the
study area. The ASTER DEM has a spatial resolution of 30 m with a relative accuracy of more
than 10 m.
These digital data are suitable to meet the 1 : 50 000 to 1 : 250 000 map accuracy standard
(USGS, 2003).
The elevation distribution of the non-flooded pixels is shown in Figure 4. The graph shows a
sudden upward
Copyright 2005 John Wiley & Sons, Ltd. Hydrol. Process. 19, 36993716 (2005)
REMOTE SENSING AND GIS-BASED FLOOD VULNERABILITY ASSESSMENT 3705
2.5105
2.0105
1.5105
Frequency
1.0105
5.0104
8m
-10 0
0
10
Elevation (m)
20 30
Figure 4. Elevation distribution of the non-flooded area extracted from the ASTER DEM
the non-flooded layer. Table I summarizes the extent of area eliminated at different levels of
correction of the
non-flooded area.
Anything other than the non-flooded area depicts the actual water area. This classified
water layer includes
permanent water bodies of the area. For calculating actual flood-affected area, permanent
water bodies must be
eliminated from the water layer (Yang et al., 1999). Two Landsat ETMC scenes of 16 April
2003, depicting
normal hydrological conditions, have been used to extract the permanent water bodies.
These pixels were
subtracted from the water layer to obtain the actual flooded area. The classified image is
shown in Figure 5.
Delineating the high flood-depth zone
Flood depth is considered as the most important indicator of flood hazard (Wadge et al.,
1993; Townsend
and Walsh, 1998; Islam and Sado, 2002). A higher depth of flood is associated with a high
discharge, which
is a determining factor in flood-induced destruction of life and property. During the 2000
September flood
the depth of water was as high as 25 to 3 m in some areas. The roofs of the majority of the
houses in rural
West Bengal are not concrete; they are commonly built with hay or thatch and corrugated
sheets of iron. This
fact excludes the local people from climbing over their roofs to escape inundation. Flood
depth determination
from remotely sensed imagery is very difficult, but an indirect method exists to classify a
flooded area into
different flood-depth zones. The amount of radiant energy reflected by water in the visible
light portion of
the electromagnetic spectrum is essentially determined by the colour of the water and its
turbidity. Except
for the blue band, all optical bands have very high correlation with the turbidity and
sediment concentration
of the water. Deeper water has more turbidity than shallower waters because of its high
velocity (Islam and
Sado, 2000b).
Interband correlation is a major impediment in analysing multispectral data. Principal
component (PC)
transformation is performed to overcome this problem. This image-processing technique
makes the bands less
correlated and reduces the dimensionality of the original dataset (Lillesand and Kiefer,
2000). To enhance
contrast and facilitate classification, a PC transformation has been applied over bands 2, 3,
4, 5 and 7 of the
Landsat ETMC data acquired during the flood. The first three components, explaining about
9965% of the
total variation, were selected for further analysis. The other components were excluded from
further analysis
as their noise-to-signal ratio is expected to be very high. Kunte and Wagle (2003) attempted
to classify depth
of water in the Gulf of Kutch and reported that the PC2 band is particularly sensitive to the
concentration
of suspended sediments and, therefore, can be effectively used for broad classification of
water depth. To
enhance the amount of information, different combinations of the three PC bands into RGB
were tried to
create an FCC, and it was found that PC2, PC1 and PC3 (RGB) generate the best FCC. Figure
6 shows the
turbidity/sediment concentration in the flooded zone.
Figure 6 clearly represents at least two turbidity zones in the shades of yellow and violet.
The general trend
of tonal variation reveals that highly turbid water exists at the core of the flooded zone and
the sediment
concentration of the water gradually decreases towards the boundary of water and land.
After consulting Islam
and Sados (2000b), we have assumed that highly turbid water is associated with high
velocity. During a
flood, high water flows at high velocity through the highly inundated zone. Highly
turbid/deep water appears
in yellow and shallow water with less turbidity is presumed to appear in violet tint. The major
rivers of
the region, namely the Bhagirathi and Jalangi, fall in the yellow zone, which further confirms
our visual
interpretation of flood depth.
Table I. Correction of non-flooded area under different levels of processing
Processing level Total area _km2_ Area reduced _km2_
Digital image processing of Landsat flood scene 351298
After eliminating area under 8 m 332498 741
After eliminating polygons less than nine Landsat pixels 328708 3790
Copyright 2005 John Wiley & Sons, Ltd. Hydrol. Process. 19, 36993716 (2005)
REMOTE SENSING AND GIS-BASED FLOOD VULNERABILITY ASSESSMENT 3707
Classification
category
Area
(Km2)
Flood water
Land
Permanent water body
Cloud cover
2745.79
3287.08
331.39
1327.53
30 15 0 30 Kilometers
Legend
Flood water
Land
Permanent Water Body
Cloud Cover
It is interesting to note that along the River Bhagirathi the majority of the flooded area falls
in the high
depth zone and there is little evidence of a shallow depth zone along the margin of the
flooded area. This
phenomenon is attributed to the presence of extensive embankments along the river. Land
use is very intensive
in the immediate flood plain of the River Bhagirathi. To protect the land, the West Bengal
Government has built
hundreds of kilometres of embankment. These embankments, along with other minor floodcontrol measures,
put an abrupt stop to advancing flood water during the flood of SeptemberOctober 2000.
The low-lying active
flood plain, located between the embankment and the river, records very high water
discharge during floods
and, consequently, the area suffers from high water depth. Owing to the existence of the
embankments, there
Copyright 2005 John Wiley & Sons, Ltd. Hydrol. Process. 19, 36993716 (2005)
3708 J. SANYAL AND X. X. LU
Masked out cloud
cover
Non-flooded area
Shallow flood depth
Deep flood water
Figure 6. Different flood depth/turbidity zones identified over an FCC (PC2, PC1 and PC3 as RGB)
was hardly any zone of transition from deep to shallow flood water along the bank. The River
Bhagirathi
represents the most dominant distributary of the River Ganga that flows through the Indian
state of West
Bengal. The discharge and carrying capacity of this river is much higher than the other
smaller rivers in this
region. Owing to a higher velocity and discharge, the sediment concentration and
consequent turbidity in the
River Bhagirathi is much higher than the other smaller rivers, like the Jalangi. This factor also
contributes to
deep inundation along the Bhagirathi river bank.
After comparing the PC2 band with the FCC, a threshold value has been selected empirically
to extract
highly turbid/deep water from the PC2 band. Although river flooding is not only a function of
terrain
configuration, the topography of the flood plain does play an important role in the course of
an advancing flood.
Therefore, the ASTER DEM has been used to verify the accuracy of the high flood-depth
zone. Elevation
distribution of the high flood-depth zone has been plotted in Figure 7. This plot ratifies that
classification of
high flood depth is, on average, accurate and realistic. The area having an elevation of more
than 16 m was
not likely to experience high flood depth. The fact that the wave drops sharply after an
elevation of more
than 16 m attests to the high accuracy of the classified high flood-depth zone.
Delineating human settlements
As mentioned in the Introduction, the main focus of this study revolves around identifying
the floodvulnerable
settlements. Human settlement is the main input layer of the present study, and all other
layers of
information are used to estimate its vulnerability to floods. Land-use maps of 1 : 250 000
have been used as
the basic source of information for delineating human settlements in this area. These maps
were published
in 1991 and, therefore, are considered relatively outdated. Satellite imageries have been
used to update these
maps. Application of remotely sensed data for identifying human settlements is not new.
Radar imageries
Copyright 2005 John Wiley & Sons, Ltd. Hydrol. Process. 19, 36993716 (2005)
REMOTE SENSING AND GIS-BASED FLOOD VULNERABILITY ASSESSMENT 3709
1.5105
1.0105
Frequency
5.0104
0
-10 -5 0 5 10 15
Elevation (m)
20 25
Figure 7. Elevation distribution of the area affected by high flood depth
are more widely used than optical data because settlements are particularly accentuated in
radar imageries
owing to their geometric shapes and dielectric constants. These parameters of SAR data are
different from
optical sensors (Henderson and Xia, 1997). Studies in Germany (Henderson, 1995), China
(Lo, 1986), and
the Ganges Plain (Imhoff et al., 1987) reported that settlements having a population of more
than 1000 are
generally recognizable from SAR imageries.
The difficulty associated with identifying human settlement from radar data is that most SAR
sensors obtain
data at a single wavelength with fixed polarization. The best band combination for any
classification should
cover major portions of the electromagnetic spectrum (visible, infrared and microwave) that
are normally
used for remote sensing (Haack et al., 2000). Haack and Slonecker (1994) reported that
neither Landsat TM
nor SAR data can independently locate villages in Sudan. For the present study, Landsat
ETMC data of 13
April 2003 and ERS-1 SAR data of 9 October 1995 have been used to interpret the land cover
of the study
area visually and to update the settlement layer previously digitized from the land-use map.
The Landsat ETMC data obtained are geometrically corrected (level-1G) from the USGS. It has
been
further registered to the 1 : 250 000 land-use maps. The radar data have been obtained as a
precision image
(PRI). ERS SAR PRI products are projected to ground range and resampled in 125 m pixel
size. The PRIs
have been coregistered with the land-use map with a root-mean-square error (RMSE) of
106 pixels. After
georeferencing, the SAR imageries have been mosaicked. A low-pass filter of a 5 5 pixel
window has been
applied to it to reduce speckle and improve visual interpretability in identifying settlements.
After stacking
TM bands 4, 3 and 2 in RGB, a coloured image was generated by fusing it with the processed
radar image.
HSV sharpening tools have been used to perform this operation. This function transforms an
RGB image to
HSV colour space, replaces the value band with the high-resolution radar image and
automatically resamples
the hue and saturation bands to the high-resolution pixel size (125 m) of the SAR PRI using
the nearest
Copyright 2005 John Wiley & Sons, Ltd. Hydrol. Process. 19, 36993716 (2005)
3710 J. SANYAL AND X. X. LU
neighbour method. Finally, it transforms the image back to the RGB colour space. The output
was a coloured
image of 125 m spatial resolution. It is necessary to point out that a coregistration RMSE of
more than one
pixel is acceptable for SAR scenes because the 125 m 125 m pixels of radar imageries
are fused with
30 m 30 m ETMC pixels.
Figure 8 represents part of the study area. The small rural settlements are easily
distinguishable by their
bright appearance over a reddish background of vast cropland. The vectorized settlement
layer, already
extracted from the land-use maps, has been superimposed on this image to update the
boundaries of settlements
manually. Automated classification methods have not been considered, as a majority of
studies reported low
accuracy for smaller settlements (Dowman and Morris, 1982; Lo, 1984; Liu et al., 1986). It
has also been
found that the reflectance of fallow land and bare soil are quite similar to the rural
settlements in the TM
bands. Such substantial error in classifying human settlements would jeopardize the whole
gamut of results.
This factor induced us to rule out the option of automated classification.
Processing different data layers in a GIS environment
In order to reach the objective of obtaining a flood vulnerability of settlements, it is
necessary to make the
other information layers compatible with the settlement layer to allow for the analysis of
their vulnerability
to river inundation. Apart from the hydrological information extracted from the Landsat ETM C
flood scenes,
some socio-economic data, like population density, have also been incorporated in the
current framework to
facilitate vulnerability analysis of the settlements. Considering the scale of this investigation,
development
blocks have been selected as the appropriate administrative unit for reporting population
density. Development
blocks are the smallest administrative unit in West Bengal, as well as India, for collecting
majority of the
data. Boundaries of the development blocks over the study area are shown in Figure 1.
Rural
settlements
Figure 8. Landsat ETMC bands 4, 3 and 2 merged with ERS SAR image to identify the rural settlements in Gangetic
West Bengal visually
Copyright 2005 John Wiley & Sons, Ltd. Hydrol. Process. 19, 36993716 (2005)
REMOTE SENSING AND GIS-BASED FLOOD VULNERABILITY ASSESSMENT 3711
Administrative boundaries of the development blocks have been digitized from a 1 : 500 000
map.
Demographic data, collected from the 2001 Indian Census, have been integrated in the
attribute table of the
development blocks. It has been observed over the study area that human settlements,
especially rural hamlets,
develop across administrative boundaries of the development blocks. The frequency of flood
occurrence in
each development block for the decade 1991 to 2000 has been considered to identify those
areas that have
been suffering from chronic flooding (Sanyal and Lu, 2005). This information was obtained
from the Annual
Flood Reports of the Irrigation and Waterways Department, Government of West Bengal,
India.
To facilitate spatial analysis, the shape of the file containing settlements has been clipped
by the
administrative boundaries of the development blocks. In the output, individual settlements
have been
subdivided where an administrative boundary cuts across them. This overlay operation
proved very useful in
subsequent analysis, as in the output layer each settlement contains certain information
about the development
block it is located in. This information includes name, population density and frequency of
flood occurrence
over the last decade. It should be noted that the above-mentioned attributes are assumed to
be constant for all
individual settlements falling within one development block. It is evident that the accuracy
of this information
decreases with increasing size of the development blocks. Owing to the monotonous flat
terrain of the study
area, there is not much scope for large variation of topography within a development block.
RESULTS AND DISCUSSION
The main purpose of this investigation is to analyse the interaction between different flood
hazard indicators
that contribute to vulnerability of human settlements in the study area. The previous section
illustrated a
systematic approach to extracting the spatial extent of these hazard factors. A GIS
environment is used to
evaluate the interaction of these factors in a spatial dimension and locate the settlements
that should be given
priority in implementing remedial measures. It should be pointed out that this methodology
is likely to yield
better perception of flood hazard if a very high magnitude flood is taken into consideration.
Using a moderate
magnitude event would increase the risk of leaving some vulnerable settlements
unidentified that have chances
of be affected should a very high magnitude event hit the area. We found that an
overwhelming majority of
the settlements contain some pixels classified as non-flooded. The reason behind this
phenomenon is that the
local people traditionally build their houses over relatively higher ground, leaving the floodprone lowlands
for paddy cultivation. This settlement pattern evolved as an outcome of local inhabitants
adaptation to living
in a flood-prone area. In Gangetic West Bengal, owing to the vast expanse of cropland, trees
are only found in
and around rural settlements and along roads. Misclassification of some tree canopy as nonflooded pixels also
contributes to overestimation of non-flooded areas. It has been decided that any individual
settlement having
very little intersection with the non-flooded pixels should be practically considered as highly
vulnerable to
flood. For computational ease we have calculated the hazard indicator as
HI1 D _A/B_ 100 _1_
where A is the area of intersection between non-flooded pixels and individual settlements
and B is the total
area of an individual settlement.
The intersection function in ArcGIS is used to calculate the common area between the
settlement and nonflooded
layer. Part of the attribute table of the output shape file is shown in Table II. This table shows
that,
very often, each settlement polygon intersects with more than one polygon representing
non-flooded areas.
Therefore, the area of intersection is distributed in more than one row, which makes it
difficult to calculate HI1.
It is noted in Table II that settlement ID 23 is distributed in 13 rows. In other words,
settlement 23 contains 13
patches of non-flooded land area. Table II is not the complete table. It only shows a typical
portion of it that
exemplifies the difficulty in calculating HI1. The attribute table of the output shape file has
been summarized
by dissolving the polygons on the basis of their settlement ID number. Part of the processed
table is illustrated
in Table III. After calculating HI1 it has been found that out of 921 settlement polygons in the
study area 390
Copyright 2005 John Wiley & Sons, Ltd. Hydrol. Process. 19, 36993716 (2005)
3712 J. SANYAL AND X. X. LU
Table II. Part of the attribute table illustrating how the intersection of a non-flooded layer with
individual settlements is distributed in different polygons. Note that 13 polygons represent the
area of intersection between settlement 23 and the non-flooded layer
Settlement
polygon ID
Intersection
area _m2_
Block
name
District Settlement
ID
56796 21 1919 Balagar Hugly 13
56791 35 1324 Balagar Hugly 13
56720 1 100 000 0000 Balagar Hugly 20
56720 1 100 000 0000 Pandua Hugly 22
56720 1 100 000 0000 Pandua Hugly 22
56793 27 8300 Chakdaha Nadia 23
56789 25 0259 Chakdaha Nadia 23
56791 35 1324 Chakdaha Nadia 23
56782 93430 Chakdaha Nadia 23
56780 29 3860 Chakdaha Nadia 23
56788 84 3854 Chakdaha Nadia 23
56778 10 2580 Chakdaha Nadia 23
56781 46 8217 Chakdaha Nadia 23
56769 99209 Chakdaha Nadia 23
56758 81 5494 Chakdaha Nadia 23
56738 37 8673 Chakdaha Nadia 23
56707 10 2459 Chakdaha Nadia 23
56762 310 0000 Chakdaha Nadia 23
56720 1 100 000 0000 Pandua Hugly 24
56525 49 3324 Pandua Hugly 26
56525 49 3324 Pandua Hugly 26
56720 1 100 000 0000 Pandua Hugly 26
56720 1 100 000 0000 Pandua Hugly 28
56720 1 100 000 0000 Pandua Hugly 28
56720 1 100 000 0000 Pandua Hugly 28
Table III. Area of intersection between settlement layers and non-flooded
area summarized on the basis of individual settlements
Development
block
District Settlement ID Intersection
area _m2_
Balagar Hugly 13 45 79000
Balagar Hugly 20 813684
Chakdaha Nadia 21 000
Pandua Hugly 22 580 05279
Chakdaha Nadia 23 685 68778
Pandua Hugly 24 1 200 00000
Chakdaha Nadia 25 000
Pandua Hugly 26 160 07156
Balagar Hugly 27 000
Pandua Hugly 28 126 63152
settlements have less than 50% of their area classified as non-flooded. In other words, more
than 50% of the
area of these settlements was affected by inundation. For more than 75% inundation, the
number is still high
at 206 (2236%). There are 124 settlements having more than 90% of their area submerged
under flood water
(Figure 9). These settlements are located in a high flood-hazard zone and, therefore, are
extremely vulnerable
Copyright 2005 John Wiley & Sons, Ltd. Hydrol. Process. 19, 36993716 (2005)
REMOTE SENSING AND GIS-BASED FLOOD VULNERABILITY ASSESSMENT 3713
Location of vulnerable settlements N
Legand
Vulnerable Settlements
Development Blocks
Studyarea
River
30 15 0 30 Kilometers
Figure 9. Location of the settlement that does not have access to higher ground as shelter during the flood on 30
September 2000
to inundation. In addition, a buffer operation reveals that 13 of these 124 settlements have
no non-flooded
areas within a buffer zone of 500 m. For the sake of simplicity, a straight-line distance has
been used to
calculate the buffer. Since Gangetic West Bengal is remarkably flat there is no conspicuous
physical barrier
such as, mountain to travel for short distance. It is been also realized that 500 m is the
maximum distance
that people can travel even at a moderate flood depth. Thus, these settlements have no
immediate access to a
potential flood shelter. Out of these 13 settlements, five are located in the Ranaghat-II block,
four are in the
Ranaghat-I block, two are in the Balagar block, one is in the Chakdah block and one is in the
Chapra block
(see Figure 1 for block names).
The location of a particular settlement with respect to the high flood-depth zone forms the
basis for
computing the second flood hazard indicator: any settlement having the majority of its area
under deep flood
water has been considered as vulnerable. The hazard indicator has been calculated as
HI2 D _C/D_ 100 _2_
where C is the area of intersection between high flood depth and an individual settlements
and D is the total
area of an individual settlement.
The vector layer of the high depth zone, extracted from Landsat imageries, has been
overlaid with the
settlement layer and the intersection between the two layers was calculated using ArcGIS.
Attribute tables
are compiled in a similar way to HI1. It is evident from previous discussion that a high HI 2
value for any
Copyright 2005 John Wiley & Sons, Ltd. Hydrol. Process. 19, 36993716 (2005)
3714 J. SANYAL AND X. X. LU
settlement indicates its vulnerability to flood-induced disaster. More than 50% of the area of
19 settlements
was affected by high flood depth. Ten of these 19 settlements are situated in Nadia District
and nine are in
the Bardhaman District of West Bengal.
Vulnerability analysis for each settlement can be done more meaningfully by synthesizing
HI1 and HI2
with population density. It was mentioned before that each settlement polygon already
contains information
regarding average population density as the number of people per square kilometre. An SQL
query was built
to identify those settlements that have a high population density and which are vulnerable
with respect to
both hazard factors.
A total of 18 settlements meet the criteria where HI 1 < 25, HI2 > 50 and Pop density >750.
These
settlements have been classified as extremely vulnerable. The distribution of these
settlements in different
development blocks is shown in Table IV. Table V shows the location coordinates of the
centroids of
the polygons. The coordinates will help local planners and administrators easily locate highly
vulnerable
settlements over the administrative boundaries of revenue villages. During any highmagnitude floods, the
local administration should place high priority on providing relief to the population of these
settlements.
Table IV. Development block-wise distribution of extremely flood-vulnerable
settlements
Development
block
District Flood occurrences
(19912000)
Vulnerable
settlements
Chapra Nadia 3 2
Kalna-I Bardhaman 2 3
Krishnanagar Nadia 3 1
Nabadwip Nadia 3 2
Purbasthali-I Bardhaman 4 3
Purbasthali-II Bardhaman 3 2
Ranaghat-II Nadia 2 1
Shantipur Nadia 2 4
Table V. Precise location of centroids of the settlements that are highly vulnerable
to flooding
Settlement ID Block Name Latitude (N) Longitude (E)
1 Kalna-I 23296 88361
2 Kalna-I 23317 88291
3 Purbasthali-I 23336 88348
4 Purbasthali-I 23356 88293
5 Kalna-I 23354 88290
6 Purbasthali-I 23356 88284
7 Purbasthali-II 23500 88340
8 Purbasthali-II 23509 88283
9 Santipur 23218 88412
10 Santipur 23296 88364
11 Ranaghat-II 23292 88650
12 Santipur 23310 88369
13 Nabadwip 23329 88380
14 Santipur 23332 88349
15 Nabadwip 23335 88349
16 Krishnanagar-I 23456 88545
17 Chapra 23501 88615
18 Chapra 23530 88606
Copyright 2005 John Wiley & Sons, Ltd. Hydrol. Process. 19, 36993716 (2005)
REMOTE SENSING AND GIS-BASED FLOOD VULNERABILITY ASSESSMENT 3715
Table V shows that these settlements are not only vulnerable to extreme flood events like in
2000, but most
of them are also located in very frequently flooded development blocks because of a rapidly
increasing
population density and hence rapid expansion of settlements to highly flood-prone zones.
This research has been funded by the National University of Singapore (NUS) research grant
R-109-000-049112. NUSs support of this research is gratefully acknowledged by the authors.
REFERENCES
Bagchi K. 1945. The Ganges Delta. University of Calcutta: Kolkata, India.
Basu PK. 2001. Flood 2000: a comprehensive study with major floods in the state Sechpatra 3(1): 612.
Berg A, Gregiore JM. 1983. Use of remote sensing techniques for rice production forecasting in West Africa (Mali and
Guinea: NigerBani
project). In ESA Satellite Remote Sensing for Developing Countries, Ispra, Italy; 161168.
Bhavsar PD. 1984. Review of remote sensing applications in hydrology and water resource management in India.
Advances in Space Research
4(11): 193200.
Cannon T. 2000. Vulnerability analysis and disasters. In Floods Vol I, Parker DJ (ed.). Routledge: New York; 4555.
Deutsch M, Ruggles F. 1974. Optical data processing and projected application of the ERTS1 imagery covering the
1973 Mississippi river
valley floods. Water Resource Bulletin 10(5): 10231039.
Deutsch M, Ruggles F, Guss P, Yost E. 1973. Mapping the 1973 Mississippi floods from the Earth Resource
Technology satellites. In
Proceedings of International Symposium on Remote Sensing and Water Resource Management. American Water
Resources Association,
No. 17. AWRA: Burlington, Ontario; 3955.
Dowman IJ, Morris AH. 1982. The use of synthetic aperture radar for mapping. Photogrammetric Record 10(60):
687696.
Copyright 2005 John Wiley & Sons, Ltd. Hydrol. Process. 19, 36993716 (2005)
3716 J. SANYAL AND X. X. LU
England CF, Hunt GE. 1985. A bispectral method for the automatic determination of parameters for use in imaging
satellites cloud retrievals.
International Journal of Remote Sensing 6(8): 15451553.
FEMA. 2003. Federal Emergency Management Agency. http://www.fema.gov/nfip [20 September 2000] (in Bengali).
Ganashakti. 2000. November 3 (In Bengali).
Goswami AK. 1983. Studies on the nature of floods in the depressed belt of southern West Bengal. Geographical
Review of India 41(2):
189201.
Haack BN, Slonecker ET. 1994. Merged spaceborne radar and thematic mapper digital data for locating villages in
Sudan. Photogrammetric
Engineering and Remote Sensing 60(10): 12531257.
Haack BN, Herold ND, Bechdol MA. 2000. Photogrammetric Engineering and Remote Sensing 66(6): 709716.
Henderson FM. 1995. An analysis of settlement characterization in central Europe using SIR-B radar imagery.
Remote Sensing of Environment
54: 6170.
Henderson FM, Xia ZG. 1997. SAR applications in human settlement detection, population estimation and urban
land use pattern analysis:
a status report. IEEE Transactions on Geoscience and Remote Sensing 35(1): 7985.
Hewitt K. 1997. Regions of risk: A Geographical Introduction to Disasters. Addison-Wesley Longman: Harlow, UK.
Imhoff ML, Vermillion C, Story MH, Choudhury AM, Gafoor A, Polcyn P. 1987. Monsoon flood boundary delineation
and damage
assessment using space borne imaging radar and Landsat data. Photogrammetric Engineering and Remote Sensing
53(4): 405413.
Islam MM, Sado K. 2000a. Flood hazard assessment in Bangladesh using NOAA AVHRR data with geographical
information system.
Hydrological Processes 14(3): 605620.
Islam MM, Sado K. 2000b. 1997. Development of flood hazard maps of Bangladesh using NOAA-AVHRR images with
GIS. Hydrological
Sciences Journal 45(3): 337355.
Islam MM, Sado K. 2002. Development of priority map flood countermeasures by remote sensing data with
geographic information system.
Journal of Hydrologic Engineering 7(5): 346355.
Kunte PD, Wagle BG. 2003. Sediment transportation and depth variation study of the Gulf of Kutch using remote
sensing. International
Journal of Remote Sensing 24(11): 22532263.
Lillesand TM, Kiefer RW. 2000. Remote Sensing and Image Interpretation, 4th edn. John Wiley: Chichester.
Liu J, Teng X, Xiao J. 1986. Application of Shuttle imaging radar data for land use investigation. Remote Sensing of
Environment 19:
291301.
Liu Y, Islam MN, Gao J. 2003. Quantification of shallow water quality parameters by means of remote sensing.
Progress in Physical
Geography 27(1): 2443.
Lo CP. 1986. Settlement, population and land use analysis of the North China Plain using Shuttle imaging radar-A
data. Professional
Geographer 38(2): 141149.
McGinnis DF, Rango A. 1975. Earth Resource Satellite System for flood monitoring. Geophysical Research Letters
2(4): 132135.
Mitchell JK (ed.). 1999. Crucibles of Hazard: Mega-Cities and Disaster in Transition. United Nations Publication: New
York.
Morrison RB, Cooley ME. 1973. Assessment of flood damage in Arizona by means of ERTS-1 imagery. In Proceedings
of Symposium on
Significant Result Obtained from the Earth Resource Satellite 1, New Carrollton, Maryland, Vol. 1; 755760.
Rango A, Anderson AT. 1974. Flood hazard studies in the Mississippi river basin using remote sensing. Water
Resources Bulletin 10(5):
10601081.
Rango A, Solomonson VV. 1974. Regional flood mapping from space. Water Resources Research 10(3): 473484.
Ruangsiri P, Sripumin R, Polongam S, Kanjanasuntorn P, Wongparn S. 1984. State of flooding in the Mun-Chi river
basin area, N. E.
Thailand by digital Landsat data analysis. Report: Remote Sensing Division, National Resource Council of Thailand,
Bangkok, Thailand.
Rudra K. 2001. The Flood in West Bengal: September 2000 . Jayasree Press: Kolkata, India.
Sanyal J, Lu XX. 2004. Application of remote sensing in flood management with special reference to monsoon Asia:
a review. Natural
Hazards 33: 283301.
Sanyal J, Lu XX. 2005b. GIS based flood hazard mapping in Gangetic West Bengal. Singapore Journal of Tropical
Geography (in press).
Saunders RW. 1986. An automated scheme for the removal of cloud contamination from AVHRR radiance over
western Europe. International
Journal of Remote Sensing 7(7): 867888.
Sheng Y, Su Y, Xiao Q. 1998. Challenging the cloud-contamination problem in flood monitoring with NOAA/AVHRR
imagery. Photogrammetric
Engineering and Remote Sensing 64(3): 191198.
Spate OHK, Learmonth ATA, Learmonth AM. 1965. India and Pakistan: A General and Regional Geography. Methuen:
Bungay, Suffolk.
Smith LC. 1997. Satellite remote sensing of river inundation area, stage and discharge: a review. Hydrological
Processes 11: 14271439.
The Times of India. 2000. Rain reality mired in recording controversy. 18 November.
Townsend PA, Walsh SJ. 1998. Modelling flood plain inundation using integrated GIS with radar and optical remote
sensing. Geomorphology
21(98): 295312.
USGS. 2003. http://edcdaac.usgs.gov/aster/ast14dem.html [6 December 2003].
Varley P (ed.). 1994. Disaster, Development and the Environment. Wiley: Chichester, UK.
Wadge G, Wisloki AP, Pearson J, Whittow JB. 1993. Mapping natural hazard with spatial modelling system. In
Geographic Information
Handling Research and Application, Mather PM (ed.). Wiley: New York; 312324.
Wang Y, Colby JD, Mulcahy KA. 2002. An efficient method for mapping flood extent in a coastal flood plain using
Landsat TM and DEM
data. International Journal of Remote Sensing 23(18): 36813696.
Yang C, Zhou C, Wan Q. 1999. Deciding the flood extent with Radarsat SAR data and image fusion. In Proceedings of
20th Asian Conference
of Remote Sensing, Hong Kong, 2225 November.
Copyright 2005 John Wiley & Sons, Ltd. Hydrol. Process. 19, 36993716 (2005)
See discussions, stats, and author profiles for this publication at:
http://www.researchgate.net/publication/226259443
CITATIONS
130
READS
748
2 AUTHORS:
Joy Sanyal
Presidency University
"X.X." Lu
National University of Singapore
146 PUBLICATIONS 2,504 CITATIONS
SEE PROFILE
Available from: "X.X." Lu
Retrieved on: 22 December 2015
283
Department of Geography, Faculty of Arts and Social Science, National University of Singapore, 1
Arts Link, Singapore 117570
(Received: 20 June 2003; accepted: 4 November 2003)
Abstract. The conventional means to record hydrological parameters of a flood often fail to record
an extreme event. Remote sensing technology along with geographic information system (GIS) has
become the key tool for flood monitoring in recent years. Development in this field has evolved
from optical to radar remote sensing, which has provided all weather capability compared to the
optical sensors for the purpose of flood mapping. The central focus in this field revolves around
delineation of flood zones and preparation of flood hazard maps for the vulnerable areas. In this
exercise flood depth is considered crucial for flood hazard mapping and a digital elevation model
(DEM) is considered to be the most effective means to estimate flood depth from remotely sensed or
hydrological data. In a flat terrain accuracy of flood estimation depends primarily on the resolution
of the DEM. River flooding in the developing countries of monsoon Asia is very acute because of
their heavy dependence on agriculture but any flood estimation or hazard mapping attempt in this
region is handicapped by poor availability of high resolution DEMs. This paper presents a review of
application of remote sensing and GIS in flood management with particular focus on the developing
countries of Asia.
Key words: flood, remote sensing, geographic information system, digital elevation model, hazard
mapping, monsoon Asia
1. Introduction
Among all kinds of natural hazards of the world flood is probably most devastating,
wide spread and frequent. In the humid tropics and subtropical climates, especially
in the realms of monsoon, river flooding is a recurrent natural phenomenon. Floods
resulting from excessive rainfall within a short duration of time and consequent
high river discharge damage crops and infrastructures. They also result in siltation
of the reservoirs and hence limit the capacity of existing dams to control floods.
For formulating any flood management strategy the first step is to identify the
area most vulnerable to flooding. This step is even more critical for the developing
countries in the monsoon Asia as the funding available for developmental activities
is very limited. Thus these funds need to be utilized optimally for the areas that
_
But in some cases a simple density slice or supervised classification is not enough
to identify the inundated area accurately.
During later stages SPOT multi spectral imageries, were also used for flood
delineation with the similar assumption that water has very low reflectance in
the near infrared portion of the spectra. SPOT imageries, for example, were used
along with a DEMfor delineation of monsoon flood in Bangladesh (Brouder, 1994;
Oberstadler et al., 1997; Profeiti et al., 1997; Sado et al., 1997).
Apart from these medium resolution imageries, coarse resolution imageries like
Advanced Very High Resolution Radiometer Radiometer (AVHRR) data have been
also found useful for floods of a regional dimension ( Wiesnet et al., 1974; Huh et
al., 1985ac; Ali et al., 1987; Islam et al., 2000ac, 2001, 2002). Although AVHRR
imageries are coarse in resolution and frequently contaminated by cloud cover their
merit lies in their high temporal resolution. This advantage enables us to monitor
the progress of a flood in near real-time.
To use the capability of the near infra red band more effectively to detect water
Normalized Difference Vegetation Index (NDVI) can be used to monitor river
inundation
from AVHRR images. It is well known that water has a unique spectral
signature in the near infrared which is very different from other surface features.
Therefore, when a surface feature is inundated its NDVI value changes considerably
from the normal situation. Wang et al. (2002) observed that in the lower
reaches of the Yangtze River, the NDVI value for inundated surface features remains
negative while the value for non inundated surface is commonly greater than
0. But choice of this threshold is critical because natural condition of river flooding
varies greatly from place to place. The main difficulties of selecting an appropriate
threshold arise from two facts. Firstly the albedo of water bodies increases signi286
JOY SANYAL AND X. X. LU
ficantly due to high concentration of sediment in the flooded water and secondly,
albedo of bare soil decreases considerably due to its high moisture content during
the monsoon season. These two factors collectively reduce the difference in NDVI
value between inundated and dry surface. In some studies, NDVI values of flood
water were found to be significantly positive (Barton et al., 1989) Thus, a straight
forward approach of using simple NDVI values might not be universally effective
in delineation of inundated area. Moreover, many other factors such as atmospheric
condition, cloud cover and satellite viewing angle also influence NDVI values and
attempts should be made to minimize these effects before calculating the NDVI.
2.2. APPLICATION OF MICROWAVE REMOTE SENSING
The existence of cloud cover appears as the single most important impediment to
capture the progress of floods in bad weather condition (Rango et al., 1977; Lowry
et al., 1981; Imhoff et al., 1987; Rashid et al., 1993; Melack et al., 1994). The
development of microwave remote sensing, particularly radar imageries, solve the
problem because the radar pulse can penetrate cloud cover. Currently the most
common approach to flood management is to use synthetic aperture radar (SAR)
imagery and optical remote sensing imagery simultaneously in one project (Honda
et al., 1997; Liu et al., 1999; Chen et al., 1999). Apart from its all weather capability
the most important advantage of using SAR imagery lies in its ability to sharply
distinguish between land and water.
Thresholding is one of the most frequently used techniques in active remote
sensing to segregate flooded areas from non flooded areas in a radar image (Liu
et al., 1999; Townsend et al., 1998; Brivio et al., 2002). Commonly, a threshold
value of radar back scatter is set in decibel (dB) and a binary algorithm is followed
to determine whether a given raster cell is flooded or not. Radar back scatter is
computed as a function of the incidence angle of the sensor and digital number
(DN) (Chen et al., 1999). The threshold values are determined by a number of
processes depending on the study area and overall spectral signature of the
imagery.
Change detection can be used as a powerful tool to detect flooded area in SAR
imagery. It is generally performed by acquiring two imageries taken before and
after the flood. Coherence and amplitude change detection techniques are widely
applied in SAR domain. In the amplitude approach, areas are estimated as flooded
where the radar back scatter is observed to be in considerable decline from before
flood to after flood imagery. In the coherence approach areas are generally
identified as flooded where the coherence or correlation of radar backscatters from
before and after flood imagery are very low (Nico et al., 2000). Multi-date SAR
scenes for the same area can be projected to red, green and blue channels to create
a colour composite. Long et al. (2001) used three ERS SAR scenes to produce
this kind of composite image. The composite image effectively depicts progress of
a flood during a specific time period. This methodology is simple to execute and
APPLICATION OF REMOTE SENSING IN FLOOD MANAGEMENT 287
provides an opportunity to readily identify the area that remains water logged for a
maximum period of time.
The existing studies pointed out some common problems encountered in accurately
extracting the flood affected area from SAR imageries. A major problem is
associated with the relation between radar wave length and roughness of the
terrain
and water body. Normally pure and calm water acts as a specular reflector to the
radar signals. Thus the radar antennae receive no backscatter and the water
appears
in dark tone in the SAR imageries. Rough water surface appears in brighter tone
in the SAR imageries than the calm water (Yang et al., 1999). During floods, bad
and windy condition usually prevails over the affected area. Wind induced ripples
in the water surface frequently creates problems for the interpreter to determine
the
threshold value to delineate the flooded area.
Forest cover also poses an obstacle to accurately identify inundated areas from a
SAR image (Hess et al., 1990). The key to identify the inundated areas under forest
cover lies in the fact that flooded forests produces a bright radar back scatter in
contrast to non flooded forests due to a double bounce effect (Kundus et al., 2001),
whereas the flooded areas without a forest canopy appear dark in SAR imageries.
The flooded forests reflect 2dB as radar back scatter in the L-band (Pope et al.,
1997). This particular property has been utilized to separate the inundated areas
from the non inundated ones by applying the thresholding technique (Townsend
et al., 1998; Ali et al., 2001; Rosenqvist et al., 2002). A correct separation of
the flooded and non-flooded settlements is also problematic. Normally the high
back scatter of the buildings overlay the back scatter of flood water within the
settlements. Rural settlements, especially in monsoon Asia are often surrounded
by trees. Therefore, due to the effect of trees, inundation within the settlements is
very difficult to detect (Oberstadler et al., 1997).
Establishing a universal threshold value for detecting flood is not justified. A
particular algorithm of deriving the threshold for differentiating inundated areas
from the dry land may not work with the same efficiency in varied natural settings.
Jin (1999) used DMPS SSM/I (Specific Sensor Microwave/Imager) data for this
purpose and came to the conclusion that a particular algorithm developed on the
basis of a wet land environment produces spurious results for inundated areas
under
a forest canopy. He emphasized the importance of knowledge in the regional
geography of the area under investigation in setting any threshold value.
Combination of wave length, incidence angle and polarization play a major role
in influencing the interpreters ability to segregate flood areas from the non-flooded
ones under a forest canopy. Given the same wave length and incidence angle the
ratio of back scatter from flooded forests to the non-flooded ones is higher in
horizontal
polarization than at vertical polarization. It can be further demonstrated
that while wave length and polarization parameters are held constant, the aforesaid
ratio is larger under a small incidence angle than large (Wang et al., 1995). Radar
back scatter also depends on the orientation of the rough surface (e.g., furrows
of a ploughed field). As radar signals are directional in nature, the same surface
288 JOY SANYAL AND X. X. LU
may produce different tonal signatures depending on the relative orientation of the
rough surface to the radar antenna.
Radar incidence angle and the consequent variation in back scatter also pose
difficulties in the delineation of inundated areas from SAR imageries. Generally,
it is found that dealing with forested area imageries taken at a lower incidence
angle prove more useful compared to imageries taken at a higher incidence angle.
Townsend et al. (1998) recommended the use of JERS SAR imageries to ERS for
identification of flooded areas under forest canopy due to the smaller incidence
angle of the former as compared to the latter. This satellite has some advantages
over other earth observation satellite systems to detect flooding under forest as the
L-Band signal of JERS1 is particularly sensitive to standing water below the forest
canopy (Rosenqvist et al., 2002). Radarsat SAR imageries is also preferable to ERS
for flood mapping as the former can rotate its sensor disseminating the radar signal
at different incidence angles. This manoeuvring capability proves very useful in
locating the flood affected area in different types of terrain and land cover (Andre
et al., 2002).
2.3. A COMBINED APPROACH
In recent years, flood mapping efforts synthesize the advantages of both optical
and microwave remote sensing technologies for better results. In some occasions
this approach also leads to the formulation of better flood management strategy. In
mountains, slopes positioned perpendicular to the radar beam only appear bright
and all other areas appear as dark or shaded. This poses an obstacle to effectively
identify the flooded areas in the mountains. Due to its shaded appearance it is very
common to erroneously identify the mountainous areas as inundated. Yang et al.
(1999) came out with a solution to this problem by fusing Landsat TM imageries
with SAR imageries. After extracting the mountain regions from a Landsat TM
data it was overlaid on the SAR imagery. To get rid of the spuriously demarcated
flooded area in the SAR imagery, the portion identified as mountain in the
TM image was eliminated from it. This process led to accurate delineation of the
inundated area. However, the authors realized that adoption of this method may
result in removal of some existing water bodies from the actual hydrological layer
of the mountainous area. To rectify this error, the TM data was consulted and the
water bodies in the mountain region have been restored in the final output. This
paper illustrates that the method of using data captured by different sensors is not
straight forward and often requires iterative experiments to arrive at an optimal
result.
In this context, the utility of generating land cover maps is realized as it provides
information about permanent water bodies in normal hydrological conditions and
rescues the interpreter from possible confusion of including permanent water bodies
as inundated area (Tholey et al., 1997). But while dealing with the monsoon
flood one should consult the imagery of a wet season to carve out the natural
APPLICATION OF REMOTE SENSING IN FLOOD MANAGEMENT 289
drainage. Use of an imagery captured in dry season would lead to underestimation
of the natural drainage, which in turn would lead to an overestimation of the flooded
area (Islam et al., 2002).
The efforts towards delineation of inundated area are ultimately aimed at assessing
the impact of flood on our economy and livelihood. Land use maps obtained
primarily from optical remote sensing are overlaid on the flood maps for assessing
the degree of damage on different kind of land uses. Varieties of land use
classification
strategies have been adopted by researchers to estimate flood risk. An
analysis of most of the papers show predominance of classifying land use pattern of
the flood prone areas into conventional categories like cropland, urban area, barren
land etc (Islam et al., 2001). However, these conventional land cover classifications
do not serve the purpose optimally. Since damage inflicted over areas of high value
land use is of greater concern a more detailed classification of the economically
intensive land use is recommended. It can be done meaningfully by clubbing low
value land use areas like barren land, forest, marshy land into one category and
applying a rigorous classification of arable land and urban areas (Honda et al.,
1997). It is also realized that a data archive for selected satellites should be built for
the developing countries. This kind of archive would depict the background land
cover condition for the study area over different seasons and facilitate the process
of change detection. Only with such a data set can remote sensing of flooded areas
become a useful tool in flood management and mitigation.
3. Flood Hazard & Risk Mapping with GIS and Remote Sensing
Another primary issue for flood management is to identify the area having higher
hazard potential. Hazard can be defined as some threat, natural, technological, or
civil to people, property and environment and risk is viewed as the probability that
a hazard will occur during a particular time period. Flood is a natural hazard and
flood risk is defined in terms of hundred-year flood (Godschalk, 1991). The issue
of preparing a reliable hazard map is one of the latest concerns within the subject
of flood management.
Different approaches have been taken to map the potential hazard. Rejesk
(1993) introduces three different methods for hazard zoning. His first method
describes
a binary model which evaluates whether the hazard is present or not in a
particular raster cell. The second method involves ranking different locations of an
area depending upon the intensity of the hazard present. In the last approach some
hazard values have been assigned to each of the raster cells based on the results
of a multivariate model which were built up on a host of variables related to river
flooding and associated hazards. But the quantification of the severity of the hazard
in some units or numbers as proposed by Rejesk (1993) in his last approach was
in the density of vegetation cover of the ground (Hodgson et al., 2003). Although
LIDAR data is even more expensive than the SAR imageries, sometimes it provides
the only appropriate option to do flood mapping in the extremely flat flood plains.
Fowler (2002) maintained that the resolution of the LIDAR data depends upon the
intensity of the laser pulses and any attempt to make the survey more intensive by
increasing the laser pulses would increase the cost of the data exponentially.
4.2. AGRICULTURAL DAMAGE ASSESSMENT
Boyle et al. (1998) classified flood damages into two categories; tangible and
intangible.
Tangible damage occurs due to direct contact with the flood water whereas
intangible damage is exemplified by the loss of historical monuments, heritage sites
etc. A disease assuming a form of an epidemic due to flood is also categorized as
an intangible damage. In the agricultural landscape of monsoon Asia, estimation
of agricultural damages requires special attention. Asias population is
predominantly
rural and its economy is heavily reliant on agriculture. Monsoon flood often
creates havoc on the economy by damaging the standing paddy. This problem is
unique for monsoon Asia because in the developed world damage to urban area
and infrastructure facilities is of concern to government authorities. Hence, in this
section we shall emphasize studies dealing with application of remote sensing in
assessment of agricultural damage.
Erosion of top soil due to a flash flood and deposition of flood borne coarse
sand reduce the fertility of soil very severely and thus have a negative impact on
agricultural economy. The process of change detection is found useful to monitor
this kind of damage to agricultural land. The most widely used procedure is to
monitor the change in brightness value (VB) at a particular wave length or different
bands to identify the erosion caused by a flood. Several change detection
techniques
like Spectral Image Differencing (SID), Tasseled Cap Brightness Image Differencing
(TCBID), Principal Component Analysis (PCA), and Spectral Change Vector
Analysis (SCVA) are employed for the purpose of detecting the erosions due to
flooding, but for Landsat TM data, SCVA is found to yield most accurate results
(Dhakal et al., 2002).
294 JOY SANYAL AND X. X. LU
In the flood plains of monsoon Asia rice is the dominant crop. Landsat TM4 band (Near Infrared) has been extensively used to estimate the damage to rice
crops due to flood (Shibayama et al., 1989; Okamoto et al., 1996) while the use of
band 3 is also considered to be very effective in assessment of paddy field damage
(Yamagata et al., 1988) due to its very high reflectance in turbid or muddy water
typically associated with flooded paddy field in the riverine plains of monsoon Asia
(Miller et al., 1983; Patel et al., 1985). The studies in this direction even attempted
to estimate the quantum of crop damage from satellite data. For example, a DN
value of 84 (in TM-4) indicates a yield of 3.0 metric tons/ha, whereas a value
around 55 corresponds to no crop in North Korea (Okamoto et al., 1998). The
main shortcoming of formulating a universal range of DN number is that it might
fail to extract accurate information for different varieties of rice crops.
Traditional supervised classification by maximum likelihood method generally
perform well in flood affected agricultural area where the instantaneous field of
view (IFOV) is less than the inundated segment of land and the land use is
dominated
by crop lands (Jensen et al., 1995; Hugenin et al., 1997). The mean value
of Normalized Difference Vegetation Index (NDVI) derived from JERS-1 OPS
data was used to differentiate damaged crops from undamaged one in North Korea
(Choen et al., 1998) because vegetation condition or biomass over a portion of land
is highly correlated with NDVI (Singh, 1989; Tappen et al., 1992; Gamon et al.,
1995; Michener et al., 1997; Lyon et al., 1998).
4.3. PROBLEM OF TEMPORAL RESOLUTION IN FLOOD MANAGEMENT
Devastating floods are generally low frequency, high magnitude natural
phenomena.
Flash floods occur within a very short interval of time and the peak stage
remains only for a couple of hours, but the most extensive and severe damage
takes
place during that time. With the current Radarsat resources it is very difficult to
capture the spatial extent of a flood at its peak. Thus, attempts have been made to
extrapolate the extent of inundation at the peak of a flood from an image acquired
at a later stage of the event. Some GIS algorithms in ARC/INFO are promising to
perform this extrapolation from an image that captures some standing water only at
a time when the flood peak had already passed. The method of least accumulation
cost distance can provide a viable solution to this problem (Brivio et al., 2002).
This principle simulates flow direction from the river channel to the flood plain
based on the assumption that water flows through the path where the work done in
doing so is least.
This methodology yielded a remarkable accuracy of 96.7 percent when compared
with the aerial photographs of the peak of the particular flood event which
occurred in November 1994 in Italy. The main weakness of this approach is that
the least accumulation cost distance operates on certain values of the raster cells
which represent the roughness of the terrain causing frictional drag to the
overflowing
flood water. Since roughness is a function of a host of other geomorphic and
APPLICATION OF REMOTE SENSING IN FLOOD MANAGEMENT 295
lithological factors, it is very difficult to control the parameters of the experiments
in an area having diverse lithology and land use. Apart from the DEMs and remote
sensing data, field work conducted in recently flood affected area can prove useful.
Wang et al. (2002) used high-flood level marks of a recent flood on houses as a
supplementary data set to the DEMs for estimating flood depth. This kind of data
serve very well to reconstruct dimension of a past flood peak.
General studies of flood geomorphology also provide some insights to resolve
this problem. Many very high magnitude floods may leave permanent imprints in
the flood plain morphology rather than creating some transient features (Brunsden
et al., 1979).When the morphology of the basin perfectly suites with the magnitude
and frequency of the flood event, the flood created features attain a state of
permanency.
Since the frequency of such an event is high in the humid tropics, the interval
between two high magnitude floods gets reduced here. It indirectly contributes to
the permanency of the flood created characteristic morphological features in the
flood plain (Gupta, 1983). The permanency of this kind of topographic features
would be more prominent in areas of high relief and coarser valley sediments
(Gupta, 1988). Hence in the humid tropics there is enough scope to use relatively
inexpensive high resolution optical remote sensing not necessarily at the time of
flooding to demarcate the areas vulnerable to river flooding. But the analyst has to
be trained enough to identify the morphological features in the flood plain typically
associated with high magnitude floods.
5. Conclusion and Prospective
The preceding discussion regarding various facades of the prospects and constraints
of using remote sensing and GIS for flood management unveils some
significant facts. A majority of the researchers favoured multi date radar imageries
to observe a particular flood event and considered different image processing
techniques
to overcome the limitations of remotely sensed data in flood delineation.
The main weaknesses of this type of approach are manifold. Most of the
investigations
mentioned in this paper are heavily dependent on the availability of satellite
data, which is not always guaranteed for the time of peak flood. Rather than moving
towards a comprehensive flood management strategy, these papers concentrated
on
some specific issues like delineation of flooded area. Most of these projects are on
a very high budget and no attempts were made try to keep the cost low. Monsoon
floods affect the developing countries more acutely than the developed ones.
Therefore,
while framing the methodology one should be aware of its feasibility in the
operational area. Most of the studies in this field cannot achieve its desired level
of accuracy without a very high resolution DEM, therefore, we need to develop
alternative methodology to shed our dependence on high resolution terrain data.
A detailed hydro-geomorphological mapping depicting the trace of past floods
may help us in this direction. The hazard of monsoon flood and the destitution
of people associated with it are very different from the industrially developed
296 JOY SANYAL AND X. X. LU
countries. Hence, we should adopt an improvised but effective methodology of
risk and damage assessment to come out with meaningful flood hazard maps for
this region.
Application of remote sensing and GIS is convincingly a very efficient and cost
effective way of flood management. Use of very high resolution imageries like
IKONS or SPOT 5 have not been very popular yet in the field of flood management
due to its high price, but it is likely that with these imageries would be available
at a reasonable price and would be widely used for flood mapping. In the age
of internet, GIS has assumed new dimensions, especially for coping with natural
disasters like river flooding. The most important advantage of using internet based
GIS is that it has opened the door of GIS technology to the end users who would not
like to install expensive GIS software. One of the numerous examples is Arc IMS
technology. This technology has been used to develop a web enabled application
named Map Action Processing Digital Interactive Geo Resource (MAPDIGR) for
providing very recent information regarding flood risk to an analyst via internet
(Smith et al., 2002). This technology is at present at an embryonic stage of
development
but has great potential for expand the user base of GIS technology for flood
management by substantially reducing the cost of operation.
Since the problem of flood is very acute in the developing countries of monsoon
Asia, special attention should be given to deal with this problem in the regional
context. GIS models having low cost and simple data requirement are likely to
attract the local authorities in the developing countries to adopt this technology
297
Barton, I. and Bathols, J.: 1989, Monitoring floods with AVHRR, International Journal of Remote
Sensing 10(12), 18731892.
Berg, A. and Gregiore, J. M.: 1983, Use of remote sensing techniques for rice production forecasting
in West Africa, (Mali and Guinea: Niger-Bani project), ESA Satellite Remote Sensing for
Developing Countries, Ispra, Italy, pp. 161168
Beven, K. J. and Kirkby, M. J.: 1979, A physically based, variable contributing area model of basin
hydrology, Hydrol. Sci. Bull. 24, 4369.
Bhavsar, P. D.: 1984, Review of remote sensing applications in hydrology and water resource
management in India, Advances in Space Research 4(11), 193200.
Blyth, K.: 1995, A telenetwork for acquisition, processing and dissemination of Earth Observation
data for monitoring and emergency management of floods, In: Proc. First ERS Thematic Working
Group Meeting on Flood Monitoring. European Space Agency/ESRIN, Frascati, Italy, 2627
June.
Boyle, S. J., Tsams, I. K., Manber, A. S. C. E., and Kanaroglu, P. S.: 1998, Developing geographical
information system for land use impact assessment in flooding conditions, Journal of Water
Resource Planning and Management 124(2), 8998.
Brakenridge, G. R., Tracy, R. T., and Know, J. C.: 1998, Orbital SAR remote sensing of the river
flood waves, International Journal of Remote Sensing 19(7), 14391445
Brimicombe, J., and Bratlet, J. M.: 1996, Linking geographical information system with hydraulic
simulation modeling for flood risk assessment: The Honk Kong Approach, In: M. F. Goodchild
(eds), GIS and Environmental Modeling, Oxford University Press, New York, pp. 165168.
Brivio, P. A., Colombo, R., Maggi, M., and Tomasoni, R.: 2002, Integration of remote sensing data
and GIS for accurate mapping of flooded areas, International Journal of Remote Sensing 23(3),
429441.
Brouder, J. A. M.: 1994, Flood study in the Meghna-Dhonagoda Polder, Bangladesh, In: Proc. Asian
Institute of Remote Sensing, Bangalore, India, 1723 November.
Brunsden, D. and Thorns, J. B.: 1979, Landscape sensitivity and change, Trans. Inst. Br. Geogr., New
Series 4, 463484.
Chen, P., Liew, S. C., and Lim, H.: 1999, Flood detection using multitemporal Radarsat and ERS
SAR data, In: Proc. 20th Asian Conference of Remote Sensing, Hong Kong, 2225 November.
Choen, K. and Kwang-Hoon, C.: 1998, Flood damage mapping in North Korea using multi-sensor
data, In: Proc. 19th Asian Conference of Remote Sensing, Manila, Philippines, 1620 November.
Clark, M. J.: 1998, Putting water in its place: A perspective on GIS in hydrology and water
management, Hydrological Processes 12, 823834.
Collins, S. H. and Moon, G. C.: 1981, Algorithms for dense digital terrain models, Photogrammetric
Engineering and Remote Sensing 47, 7176.
Coppock, J. T.: 1995, GIS and natural hazard: an overview from a GIS perspective, In: A. Carrara
and F. Guzzetti (eds), Geographical Information System in Assessing Natural Hazard, Kluwer
Academic, Netherlands, pp. 2134.
Deutsch, M.: 1976, Optical processing of ERTS data for determining extent of 1973 Mississippi River
flood, USGS Professional Paper 929, ERTS-1, A New Window on Our Planet, pp. 209222.
Deutsch, M. and Ruggles, F.: 1974, Optical data processing and projected application of the ERTS1
imagery covering the 1973 Mississippi River Valley floods,Water Resource Bulletin 10(5), 1023
1039.
Deutsch, M., Ruggles, F., Guss, P., and Yost, E.: 1973, Mapping the 1973 Mississippi floods from
the Earth Resource Technology satellites, In: Proc. International Symposium on Remote Rensing
and Water Resource Management, American Water Resource Association, No. 17, Burlington,
Ontario, pp. 3955.
Dhakal, A. S., Amda, T., Aniya, M., and Sharma, R. R.: 2002, Detection of areas associated
with flood and erosion caused by a heavy rainfall using multi temporal Landsat TM data,
Photogrammetric Engineering and Remote Sensing 68(3), 233239.
298
Fowler, R. A.: 2002, LIDAR for flood mapping, Earth Observation Magazine 9(7)
(http://www.eomonline.com/Common/Archives/July00/robert.htm, true as par 30th May, 2003)
Gamon, J. A., Field, C. B., Goulden, M. L., Griffin, K. L., Hartley, A. E., Joel, G., Penuela, J., and
Valentini, R.: 1995, Relationship between NDVI, canopy structure and photosynthesis in their
Californian vegetation types, Ecological Applications 5, 2841.
Godschalk, D. R.: 1991, Disaster mitigation and hazard management, In: T. E. Drabek and G. J. Hoetmer
(eds), Emergency Management: Principles and Practice for Local Government, International
City Management Association, Washington, D.C. pp. 131160.
Gumley, L. and King, M. D.: 1995, Remote sensing of flooding in the US Upper Midwest during the
summer of 1993, Bulletin of American Meteorological Society 76(6), 933943.
Gupta, A.: 1983, High magnitude floods and stream channel response, Special Publication of
International Association of Sedimentology 6, 219227.
Gupta, A.: 1988, Large floods as geomorphic events in the humid tropics, In: Victor R. Barker, Craig
R. Kochel, and Peter C. Paton (eds), Flood Geomorphology, John Wiley and Sons, pp. 301321.
Hallberg, G. R., Hoyer, B. E., and Rango, A.: 1973, Application of ERTS1 imagery to flood inundation
mapping, NASA Special Publication No. 327, Symposium on Significant Results Obtained
from the Earth Resources Satellite 1, Vol. 1, Technical presentations, section A, pp. 745753.
Hausmann, P. and Weber, M.: 1988, Possible contributions of hydroinformatics to risk analysis in
insurance, In: Proc. 2nd International Conference on Hydroinformatics, Zurich, Switzerland,
913 September, Balkema, Rotterdam.
Hess, L. L., Melack, J. M., and Simonett, D. S.: 1990, Radar detection of flooding beneath the forest
canopy a review, International Journal of Remote Sensing 11(7), 13131325.
Hodgson, M. E., Jensen, J. R., Schmidt, L., Schill, S., and Davis, B.: 2003, An evaluation of LIDAR
and IFSAR derived digital elevation models in leaf-on conditions with USGS level-1 and level-2
DEMs, Remote Sensing of Environment 84, 295308.
Honda, K. C., Francis, X.J., and Sah V. P.: 1997, Flood monitoring in central plain of Thailand using
JERS-1 SAR data, In: Proc. 18th Asian Conference of Remote Sensing,Malaysia, 2024 October.
Horritt, M. S. and Bates, P. D.: 2001, Predicting flood plain inundation: Raster based modeling versus
the finite element approach, Hydrological Processes 15(5), 825842.
Huguenin, R. L., Karaska, M. A., Blaricom, D. V., and Jensen, J. R.: 1997, Subpixel classification of
bald cypress and tupelo gum trees in Thematic Mapper imagery, Photogrammetric Engineering
& Remote Sensing 63, 717725.
Huh, O. K., Ali, A., and Quadir, D. A.: 1985a, Mapping of green leaf biomass over Bangladesh
with NOAA satellite AVHRR data, Manuscript Report prepared for UN Flood and Agriculture
Organization, Coastal Studies Institute, Louisiana State University., Baton Rouge, LA.
Huh, O. K., Ali, A., and Quadir, D. A.: 1985b, Observations on the surface waters of the Bay
of Bengal with NOAA satellite AVHRR data, Manuscript Report prepared for UN Flood and
Agriculture Organization, Coastal Studies Institute , Louisiana State University, Baton Rouge,
LA.
Huh, O. K., Ali, A., and Quadir, D. A.: 1985c, Use of NOAA satellite AVHRR data to monitor
river flood hydrology in Bangladesh, Manuscript Report prepared for UN Flood and Agriculture
Organization, Coastal Studies Institute, Louisiana State University, Baton Rouge, LA.
Hunter, G. J. and Goodchild, M. F: 1995, Dealing with error in spatial databases a simple casestudy,
Photogrammetric Engineering & Remote Sensing 61(5), 529537.
Imhoff, M. L., Vermillion, C., Story, M. H., Choudhury, A. M., Gafoor, A., and Polcyn, P.: 1987,
Monsoon flood boundary delineation and damage assessment using space borne imaging radar
and Landsat data, Photogrammetric Engineering and Remote Sensing 53(4), 405413.
Islam, M. M. and Sadu, K.: 2000a, Development of flood hazard maps of Bangladesh using NOAAAVHRR
images with GIS, Hydrological Science Journal 45(3), 337355.
Islam, M. M., and Sadu, K.: 2000b, Flood hazard assessment in Bangladesh using NOAA-AVHRR
data with geographical information system, Hydrological Processes 14(3), 605620.
APPLICATION OF REMOTE SENSING IN FLOOD MANAGEMENT
299
Islam, M. M. and Sadu, K.: 2000c, Satellite remote sensing data analysis for flood damaged zoning
with GIS for flood management, Journal of Hydraulic Engineering 44, 301306.
Islam, M. M. and Sadu, K.: 2001, Flood damage and modelling using satellite remote sensing data
with GIS: Case study of Bangladesh; In: Jerry Ritchie et al. (eds), Remote Sensing and Hydrology
2000, IAHS Publication, Oxford, pp. 455458.
Islam, M. M. and Sadu, K.: 2002, Development of priority map remote sensing data for flood counter
measures by geographical information system, Journal of Hydrological Engineering 7(5), 346
355.
Jensen, J. R., Rutchey, K., Koch, M., and S. Narumalani: 1995, Inland Wetland Change Detection
in the Everglades Water Conservation Area 2A Using A Time Series of Normalized Remotely
Sensed Data, Photogrammetric Engineering & Remote Sensing 61(2), 199209.
Jin, Y. Q.: 1999, A flooding index and its regional threshold values for monitoring in china from
SSM/I data, International Journal of Remote Sensing 20(5), 10251030.
Jones, K. H.: 1998, A comparison of algorithms used to compute hill slope as a property of the DEM,
Comput. Geosci. 24(4), 315323.
Kundus, P., Karszenbaum, H., Pultz, T., Paramuchi, G., and Bava, J.: 2001, Influence of flood conditions
and vegetation status on the radar back scatter of wetland ecosystem, Canadian Journal of
Remote Sensing 27(6), 651662.
Lee, J., Snyder, P. K., and Fisher, P. F.: 1992, Modeling the effect of data errors on feature extraction
from digital elevation models, Photogrammetric Engineering & Remote Sensing 58(10), 1461
1467.
Liu, R. and Liu, N.: 2002, Flood area and damage estimation in Zhejiang, China, Journal of
Environmental Management 66, 18.
Liu, Z. L., Huang, F., Li, L.Y., and Wan, E. P.: 1999, Dynamic monitoring and damage evaluation of
flood in northwest Jilin with remote sensing, In: Proc. 20th Asian Conference of Remote Sensing,
Hong Kong, 2225 November.
Long, N. T. and Trong, B. D.: 2001, Flood monitoring of Mekong River Delta, Vietnam using ERS
SAR Data, In: Proc. 22nd Asian Conference of Remote Sensing, Singapore, 59 November.
Lowry, R. T., Langham, E. J., and Murdy, N.: 1981, A preliminary analysis of SAR mapping of
Manitoba flood, May 1979, In: Proc. Satellite Hydrology, Fifth Anniversary. William T. Pecora
Memorial Symposium on Remote Sensing 1979, AmericanWater Resource Association Technical
Publication No. TPS81-1, pp. 316323.
Lyon, J. D., Yuan, G., Lunetta, R. S., and Elvidge, C. D.: 1998, A change detection experiment using
vegetation indices, Photogrammetric Engineering and Remote Sensing, 64, 143150.
McGinnis, D. F. and Rango, A.: 1975, Earth Resource Satellite System for flood monitoring,
Geophys. Res. Lett. 2(4), 132135.
Melack, J. M., Hess, L. L., and Sippel, S.: 1994, Remote sensing of lakes and floodplains in the
Amazon Basin, Remote Sensing Review 10, 127142.
Michener, W. K. and Houhoulis, P. F.: 1997, Detection of vegetation changes associated with extensive
flooding in a forested ecosystem, Photogrammetric Engineering & Remote Sensing 63(12),
13631374.
Miller, L. D., Yang, Y., Matthews, M., and Irons, R. T.: 1983, Correlations of rice yields to radiometric
estimates of canopy biomass as a junction of growth stage, In: Proc. 4th Asian Conference of
Remote Sensing, 1015 November, Colombo, Srilanka, A.6.1A 6.21.
Moore, I. D., Grayson, R. B., and Ladson, A. R.: 1991, Digital terrain modelling: A review of
hydrological, geomorphological, and biological applications, Hydrological Processes 5, 330.
Morrison, R. B. and Cooley, M. E.: 1973, Assessment of flood damage in Arizona by means of
ERTS-1 imagery, In: Proc. Symposium on Significant Result Obtained from the Earth Resource
Satellite 1, Vol. 1, New Carrollton, Maryland, pp. 755760.
Morrison, R. B. and White, P. G.: 1976, Monitoring flood inundation, USGS Professional Papers No.
929, ERTS-1, A New Window on Our Planet, pp. 196208.
300
Nico, G., Pappalepore, M., Pasquariello, G., Refice, A., and Samarelli, S.: 2000, Comparison of
SAR amplitude vs. coherence flood detection methods- a GIS application, International Journal
of Remote Sensing 21(8), 16191631.
Oberstadler, R., Honsch, H., and Huth, D.: 1997, Assessment of the mapping capabilities of ERS-1
SAR data for flood mapping: A case study of Germany, Hydrological Processes 10(10), 1415
1425.
Okamoto, K. and Fukuhara,M.: 1996, Estimation of paddy field area using the area ratio of categories
in each pixel of Landsat TM, International Journal of Remote Sensing 17, 17351749.
Okamoto, K., Yamakawa, S., and Kawashima, H.: 1998, Estimation of flood damage of North Korea
in 1995, International Journal of Remote Sensing 19(2), 365371.
Patel, N. K., Singh, T. P., Sahai, B., and Patel, M.: 1985, Spectral response of rice crops and its
relation to yield and yield attributes, International Journal of Remote Sensing 6, 657664.
Pope, K. O., Rejmankova, E., Paris, J.F. et al.: 1997, Detecting seasonal flooding cycles in marshes
of the Yucatan Peninsula with SIR-C polarimetric radar imagery, Remote Sensing of Environment
59(2), 157166.
Profeiti, G. and MacIntosh, H.: 1997, Flood management through Landsat TM and ERS SAR data:
a case study, Hydrological Processes 11(10), 13971408.
Quan,W.,Watanabe, M., Hayashi, S., and Murakami, S.: 2003, Using NOAA AVHRR data to assess
flood damage in China, Environmental Monotoring and Assessment 82, 119148.
Rango, A. and Anderson, A. T.: 1974, Flood hazard studies in the Mississippi River Basin using
remote sensing, Water Resource Bulletin 10(5), 10601081.
Rango, A. and Solomonson, V. V.: 1974, Regional flood mapping from space, Water Resource
Research 10(3), 473484.
Rango, A. and Solomonson, V. V.: 1977, The utility of short wavelength (<1 mm) remote sensing
techniques for the monitoring and assessment of hydrological parameters, In: Proc. 11th Int.
Symp. on Remote Sensing of Environ., Ann Arbor, MI, 2529 April 1977, pp. 5564.
Rashid, H. and Pramanik, M. A. H.: 1993, Areal extent of the 1988 flood in Bangladesh: How much
did the satellite imagery show? Natural Hazards 8, 189200.
Reed, S., Johnson, D., and Sweeney, T.: 2002, Application of national geographic information system
data base to support two-year flood and the threshold run-off estimate, Journal of Hydrological
Engineering 7(3), 209219.
Rejesk, D.: 1993, GIS and risk: A three cultural problem, In: M. F. Goodchild, B. O. Parks, and L.
T. Steyaerts (eds), Environmental Modeling With GIS, Oxford University Press, New York, pp.
49267.
Rocha, J. S.,Marques, Z., Ramos, I., and Ameida, R.: 1994, Simulation of risk flood areas on GIS, In:
Tsakiris, G. and Santos, M. A. (eds), Advances in Water Resource Technology and Management,
Balkema, Rotterdam, pp. 375383.
Rosenqvist, A., Shimada, M., Chapman, B., Freeman, A., Grande, G. De, and Saatchi, S.: 2000,
The global rainforest mapping a review, International journal of Remote Sensing 21(6 & 7),
13751387.
Rosenqvist, A., Forsberg, B. R., Pimentel, T., Rauste, Y. A., and Riche, J. E.: 2002, The use of space
borne radar data to model inundation patterns and trace gas emission in the central Amazon flood
plain, International Journal of Remote Sensing 23(7), 13031328.
Ruangsiri, P., Sripumin, R., Polongam, S., Kanjanasuntorn, P., and Wongparn, S.: 1984, State of
flooding in the Mun-Chi River Basin area, N. E. Thailand by digital Landsat data analysis,
Report: Remote Sensing Division, National Resource Council of Thailand, Bangkok, Thailand.
Sado, K., and Islam, M. M.: 1997, Satellite remote sensing data analysis for flooded area and weather
study: Case study of Dhaka city, Bangladesh, Journal of Hydraulic Engineering 41, 945950.
Saikh, M., Green, D., and Cross, H.: 2001, A remote sensing approach to determine environmental
flows for wet lands of the Lower Darling River, New South Wales, Australia, International
Journal of Remote Sensing 22(9), 17371751.
APPLICATION OF REMOTE SENSING IN FLOOD MANAGEMENT
301
Sharma, P.: 1999, Flood risk mapping of Dikrong subbasin of Assam, Proc. Map Asia 1999. Source
: http://www.gisdevelopment.net/proceedings/mapindia/1999/index.htm
Shibayama, M. and Akiyama, T.: 1989, Seasonal visible, near-infrared and multi infrared spectra
of rice canopies in relation to LAI and above- ground dry phytomass, Remote Sensing of the
Environment 27, 119127.
Singh, A.: 1989, Digital change detection techniques using remotely-sensed data, International
Journal of Remote Sensing 10, 9891003.
Smith, L. C.: 1997, Satellite remote sensing of river inundation area, stage and discharge: A review,
Hydrological Processes 11, 14271439.
Smith, C. and Price, J.: 2002, A new generation of flood management, Arc User 5(4), 2022.
Tappen, G. D., Tyler, D.,Wehde,M., andMoor, D.: 1992, Monitoring rangeland dynamics in Senegal
with AVHRR data, Geocarto International 1, 8798.
Tholey, N., Clandillon, S., and De Fraipont, P.: 1997, The contribution of space borne SAR and
optical data in monitoring flood events; examples in Northern and Southern France, Hydrological
Processes 11(10), 14271439.
Toutin, T.: 2002, DEMfrom stereo Landsat 7 ETM+ data over high relief areas, International Journal
of Remote Sensing 23(10), 21332139.
Townsend, P. A. and Walsh, S. J.: 1998, Modelling flood plain inundation using integrated GIS with
radar and optical remote sensing, Geomorphology 21(98), 295312.
Wadge, G., Wislocki, A. P., Pearson, J., and Whittow, J. B.: 1993, Mapping natural hazards with
spatial modeling System, In: P. M. Mather (eds), Geographic Information Handling Research
and Applications, John Wiley and Sons Inc., New York.
Wang, W., Hess, L. L., Filoso, S., and Melack, T. M.: 1995, Understanding the radar back scattering
from flooded and non-flooded Amazonian Forests: result from canopy back scatter modelling,
Remote Sensing of Environment 54(3), 324332.
Wang, Y., Colby, J. D., and Mulcahy, K. A.: 2002, An efficient method for mapping flood extent in
a coastal flood plain using Landsat TM and DEM data, International Journal of Remote Sensing
23(18), 36813696.
Wiesnet, D. R., McGinnis, D. F., and Pritchard, J. A.: 1994, Mapping of the 1973 Mississippi River
floods by the NOAA-2 satellite, Water Resource Bulletin 10(5), 10401049.
Wolock, D. M.: 1995, Effects of subbasin size on topographic characteristics and simulated flow
paths in sleepers river watershed, Vermont, Water Resour. Res. 31(8), 19891997.
Yamagata, Y. and Akiyama, T.: 1988, Flood damage analysis using multi temporal Landsat TM data,
International Journal of Remote Sensing 9(3), 503514.
Yang, C., Zhou, C., and Wan, Q.: 1999, Deciding the flood extent with Radarsat SAR data and Image
Fusion, In: Proc. 20th Asian Conference of Remote Sensing, Hong Kong, 2225 November.
See discussions, stats, and author profiles for this publication at:
http://www.researchgate.net/publication/226259443
CITATIONS
127
READS
702
2 AUTHORS:
Joy Sanyal
Presidency University
"X.X." Lu
National University of Singapore
144 PUBLICATIONS 2,470 CITATIONS
SEE PROFILE
Available from: "X.X." Lu
Retrieved on: 09 December 2015
283
Department of Geography, Faculty of Arts and Social Science, National University of Singapore, 1
Arts Link, Singapore 117570
(Received: 20 June 2003; accepted: 4 November 2003)
Abstract. The conventional means to record hydrological parameters of a flood often fail to record
an extreme event. Remote sensing technology along with geographic information system (GIS) has
become the key tool for flood monitoring in recent years. Development in this field has evolved
from optical to radar remote sensing, which has provided all weather capability compared to the
optical sensors for the purpose of flood mapping. The central focus in this field revolves around
delineation of flood zones and preparation of flood hazard maps for the vulnerable areas. In this
exercise flood depth is considered crucial for flood hazard mapping and a digital elevation model
(DEM) is considered to be the most effective means to estimate flood depth from remotely sensed or
hydrological data. In a flat terrain accuracy of flood estimation depends primarily on the resolution
of the DEM. River flooding in the developing countries of monsoon Asia is very acute because of
their heavy dependence on agriculture but any flood estimation or hazard mapping attempt in this
region is handicapped by poor availability of high resolution DEMs. This paper presents a review of
application of remote sensing and GIS in flood management with particular focus on the developing
countries of Asia.
Key words: flood, remote sensing, geographic information system, digital elevation model, hazard
mapping, monsoon Asia
1. Introduction
Among all kinds of natural hazards of the world flood is probably most devastating,
wide spread and frequent. In the humid tropics and subtropical climates, especially
in the realms of monsoon, river flooding is a recurrent natural phenomenon. Floods
resulting from excessive rainfall within a short duration of time and consequent
high river discharge damage crops and infrastructures. They also result in siltation
of the reservoirs and hence limit the capacity of existing dams to control floods.
For formulating any flood management strategy the first step is to identify the
area most vulnerable to flooding. This step is even more critical for the developing
countries in the monsoon Asia as the funding available for developmental activities
is very limited. Thus these funds need to be utilized optimally for the areas that
_
prospects and constraints of using the technology of remote sensing and GIS
in flood management. This paper presents recent developments on delineation of
flooded areas and flood hazard mapping using remote sensing and GIS. In particular
this paper draws attention to some of the issues associated with application of
remote sensing in combating floods in extremely flat flood plains of monsoon Asia.
Our review includes three aspects. First, we focus on the development of remote
sensing as a tool of flood delineation. Second, we emphasize the assessment of the
intensity of flood hazards and damage. Third, we highlight some of the issues in
the application of the technology with particular reference in the monsoon Asia, an
agricultural area with lack of high resolution spatial data bases.
2. Remote sensing as a tool of flooded area delineation
2.1. APPLICATION OF OPTICAL REMOTE SENSING
In the initial stages of satellite remote sensing the data available was from Landsat
Multi Spectral Scanner (MSS) with 80 m resolution. The pioneering investigations
in the field of application of remote sensing in flood mitigation were predominantly
concentrated on the flood prone regions of USA. MSS data were used to deal with
the flood affected areas in Iowa (Hallberg et al., 1973; Rango et al.,1974), Arizona
(Morrison et al., 1973), and Mississippi River basin (Deutsch et al., 1973; Deutsch
et al., 1974; Rango et al., 1974; McGinnis et al., 1975; Morrison et al., 1976). MSS
APPLICATION OF REMOTE SENSING IN FLOOD MANAGEMENT 285
band 7 (0.81.1 m) has been found particularly suitable for distinguishing water
or moist soil from dry surface due to strong absorption of water in the near infrared
range of the spectrum (Smith, 1997).
From the early 1980s, Landsat Thematic Mapper (TM) imageries with 30 m
resolution became the prime source of data for monitoring floods and delineating
the boundary of inundation. Special attention was given to dealing with monsoon
flooding in the developing countries like West Africa (Berg et al., 1983), India
(Bhavsar, 1983) and Thailand (Ruangsiri et al., 1984). Landsat TM band 4 proves
to be very useful in discriminating water from the dry land surface because it is a
near equivalent of MSS band 7. But Landsat TM NIR band cannot be used optimally
in developed land use areas such as downtown commercial or industrial areas.
The main reason is that the NIR band reflects very little energy from asphalt areas,
appearing black in the imageries. Therefore makes it easy to confuse developed
areas with water. Wang et al. (2002) successfully solved this problem by adding
Landsat TM band 7 to the NIR (band 4) band to delineate the inundated areas.
In TM band 7 (2.082.35 micrometre) images the reflectance from water, paved
road surfaces, and roof tops differ significantly and therefore in the Band4+Band7
image, it becomes easier to choose the density slice for extracting the flood water.
But in some cases a simple density slice or supervised classification is not enough
to identify the inundated area accurately.
During later stages SPOT multi spectral imageries, were also used for flood
delineation with the similar assumption that water has very low reflectance in
the near infrared portion of the spectra. SPOT imageries, for example, were used
along with a DEMfor delineation of monsoon flood in Bangladesh (Brouder, 1994;
Oberstadler et al., 1997; Profeiti et al., 1997; Sado et al., 1997).
Apart from these medium resolution imageries, coarse resolution imageries like
Advanced Very High Resolution Radiometer Radiometer (AVHRR) data have been
also found useful for floods of a regional dimension ( Wiesnet et al., 1974; Huh et
al., 1985ac; Ali et al., 1987; Islam et al., 2000ac, 2001, 2002). Although AVHRR
imageries are coarse in resolution and frequently contaminated by cloud cover their
merit lies in their high temporal resolution. This advantage enables us to monitor
the progress of a flood in near real-time.
To use the capability of the near infra red band more effectively to detect water
Normalized Difference Vegetation Index (NDVI) can be used to monitor river
inundation
from AVHRR images. It is well known that water has a unique spectral
signature in the near infrared which is very different from other surface features.
Therefore, when a surface feature is inundated its NDVI value changes considerably
from the normal situation. Wang et al. (2002) observed that in the lower
reaches of the Yangtze River, the NDVI value for inundated surface features remains
negative while the value for non inundated surface is commonly greater than
0. But choice of this threshold is critical because natural condition of river flooding
varies greatly from place to place. The main difficulties of selecting an appropriate
threshold arise from two facts. Firstly the albedo of water bodies increases signi286
JOY SANYAL AND X. X. LU
ficantly due to high concentration of sediment in the flooded water and secondly,
albedo of bare soil decreases considerably due to its high moisture content during
the monsoon season. These two factors collectively reduce the difference in NDVI
value between inundated and dry surface. In some studies, NDVI values of flood
water were found to be significantly positive (Barton et al., 1989) Thus, a straight
forward approach of using simple NDVI values might not be universally effective
in delineation of inundated area. Moreover, many other factors such as atmospheric
condition, cloud cover and satellite viewing angle also influence NDVI values and
attempts should be made to minimize these effects before calculating the NDVI.
2.2. APPLICATION OF MICROWAVE REMOTE SENSING
The existence of cloud cover appears as the single most important impediment to
capture the progress of floods in bad weather condition (Rango et al., 1977; Lowry
et al., 1981; Imhoff et al., 1987; Rashid et al., 1993; Melack et al., 1994). The
development of microwave remote sensing, particularly radar imageries, solve the
problem because the radar pulse can penetrate cloud cover. Currently the most
common approach to flood management is to use synthetic aperture radar (SAR)
imagery and optical remote sensing imagery simultaneously in one project (Honda
et al., 1997; Liu et al., 1999; Chen et al., 1999). Apart from its all weather capability
the most important advantage of using SAR imagery lies in its ability to sharply
distinguish between land and water.
Thresholding is one of the most frequently used techniques in active remote
sensing to segregate flooded areas from non flooded areas in a radar image (Liu
et al., 1999; Townsend et al., 1998; Brivio et al., 2002). Commonly, a threshold
value of radar back scatter is set in decibel (dB) and a binary algorithm is followed
to determine whether a given raster cell is flooded or not. Radar back scatter is
computed as a function of the incidence angle of the sensor and digital number
(DN) (Chen et al., 1999). The threshold values are determined by a number of
processes depending on the study area and overall spectral signature of the
imagery.
Change detection can be used as a powerful tool to detect flooded area in SAR
imagery. It is generally performed by acquiring two imageries taken before and
after the flood. Coherence and amplitude change detection techniques are widely
applied in SAR domain. In the amplitude approach, areas are estimated as flooded
where the radar back scatter is observed to be in considerable decline from before
flood to after flood imagery. In the coherence approach areas are generally
These two basic characteristics of overland flow have significant implications for
formulating the GIS model. In the case of non-source floods, all the raster cells or
vector points having an elevation below the water level are considered as
inundated
while for source floods, it is necessary to simulate the path of over bank flow
from the main channel to the adjacent flood plain to accurately estimate the flood
affected area.
The concept of topographical convergence or wetness index (Beven et al., 1979;
Moore et al., 1991; Wolock et al., 1995) was used to measure the depth of
inundation
(Townsend et al., 1998). The concept of wetness index is based on the
assumption that the accumulation of water in a particular cell of a raster depends
upon the area of the upslope region contributing water to that particular cell. The
main problem of using this index is that when a slope tends to zero the wetness
index becomes undefined. Thus this index is not very useful for modelling in
extremely flat flood plains. Townsend et al. (1998) also used another model for
simulating source flooding. The model assumes that the potential for any site to
be inundated is directly related to the difference in elevation between that site and
the river at its nearest hydrological link.
Perhaps the most innovative, simple and cost effective study regarding flood
hazard management has been conducted by Islam et al. (2001). He assessed the
flood depth from NOAA AVHRR imageries simply by the tonal difference of the
flood water. In this study, the flood affected area was subdivided into different flood
depth zones using supervised classification. To accurately identify the training sets,
the AVHRR data was superimposed over a DEM. Flood hazard has been assessed
by calculating a weighted score for each land use, physiographic and geologic
division of the country. The highlight of this methodology is that it assigns greater
weight to the categories of deeper flood depth in an exponential manner. In other
words, for the shallow depth, the weight increases in a progressive manner but
beyond a certain depth the increment of weight is much higher than the previous
depth categories. This process ensures that areas having higher depth of inundation
will be assigned a high hazard denomination. The rationale is that after a certain
depth, the flood water becomes very destructive and identifying this critical depth
is crucial for mapping hazard zone. This depth of water, or more precisely the
critical river stage, is likely to vary from region to region depending on the local
topography, building materials, settlement pattern etc.
Population density and hazard ranking have been multiplied to produce land
development
priority maps. Zones of this type of maps indicate development priority
APPLICATION OF REMOTE SENSING IN FLOOD MANAGEMENT 291
for flood counter measures (Islam et al., 2002). This method is heavily dependent
upon the skill of the analyst who chooses the training sites for different depth of
inundation. The process of choosing the training sets is likely to be hampered by
forest cover and varying sediment load of the flood water which may alter the
natural
relationship of depth of the water and its tonal appearance. The main limitation
of this approach is that AVHRR imageries are too coarse to be used for analyzing a
local flood and cloud free images are very rare, especially in the monsoon season.
Boyle et al. (1998) used an even more diverse dataset to estimate flood risk.
Land use, hydraulic characteristics and human activities like demography, property
values and land ownership have been taken into consideration to determine the
type, location and severity of the hazard produced by the floods. Maps have been
prepared showing the properties under potential risk in the event of floods having
500, 100 to 52 years return period.
Flood hazard maps derived from coarse to moderate resolution imageries is
misfit in the developed countries. This kind of flood hazard maps can hardly make
any improvement to the existing flood insurance infrastructure in North America
or Western Europe. But in contrast, these maps are extremely useful for the
developing
countries of monsoon Asia, countries like China, India and Bangladesh etc.
These countries often suffer from devastating monsoon floods and a large
proportion
of the population in these countries live in the flood prone areas. In this part
of the world flood insurance maps are often unheard of. Here flood hazard maps
are primarily required for saving the lives and livelihood of millions of marginal
farmers. Thus, these flood hazard maps can be used meaningfully by planners to
formulate strategies to combat this natural hazard.
Developed countries also need to evaluate the accuracy of flood risk mapping.
A very accurate flood risk map will exclude those individuals, found to be located
at a very high flood risk area, from a flood insurance scheme. On the other hand,
the insurance sector would lower the premium for those individuals who would be
found at minimum flood risk zone. Thus, a very high resolution flood hazard map
is likely to reverse the basic principle of insurance that loss of few should fall upon
many (Clark, 1998). Therefore, generalizing the risk zones to a certain extent is
likely to be more beneficial to the community as a whole.
Hazard maps should have a good visual effect so that the end user can have an
overview of the entire situation at a glance. In recent times hazard maps have
become
so technical in nature that they convey very little information to the planners
and policy makers (Rocha et al., 1994). Unless the experts of hydrology and GIS
overcome this handicap, hazard maps cannot be used optimally by all the relevant
users.
292 JOY SANYAL AND X. X. LU
4. Some Issues of Remote Sensing Applications with Special Reference to
Monsoon Asia
Having briefly given an overview of the development and methods of the
application
of remote sensing in flood management, this section tries to highlight the
constraints in applying this technique. Some of the problems, for example,
agricultural
damage assessment, are typical for monsoon Asia which is heavily dependent
on agriculture
4.1. DEPENDENCY OF DIGITAL ELEVATION MODELS IN FLOOD MANAGEMENT
In a majority of the studies dealing with the application of remote sensing in
inundated
area delineation and flood risk assessment, digital elevation models (DEM)
are used to visualize the interface of flood water with the terrain. Flood depth is
normally calculated by subtracting the elevation of each cell in a raster from its
water level (Brouder, 1994; Townsend et al., 1998; Ali et al., 2001; Islam et al.,
2002). DEMs are also used to simulate the flood depth from discharge data and
very often the result is compared with actual flooded area derived from satellite
imageries. As the spatial extent of inundation is subjected to a method of cross
checking this methodology is likely to yield a more accurate flood map as compared
to one derived from pure hydrological modelling. The main drawback of this
approach is that it is fully dependent on the accuracy of the DEM (Jones et al.,
1998). In the largely featureless plains of monsoon Asia, the accuracy required of
the DEMs is almost unattainable.
In a flat flood plain, where a vertical error of 1 m in the DEM may lead to an
error of 100s of square kilometres in flood estimation, recognition of the magnitude
of errors in the DEM is important in hydrological modelling. This issue has been
addressed from the view point of its significance in flood plain mapping (Lee
et al., 1992; Hunter et al., 1995). In the near absence of high resolution DEMs
required for flood mapping in a very flat terrain, multi-date SAR imageries can
serve as a very potent alternative data set in monsoon Asia. For example, multidate
Radarsat imageries were used in monsoon Asia to depict progress of a flood from
its inception to the peak (Liu, 1999; Chen, 1999). This particular operation can
create a visualization of the course of inundation from the river channel to the
adjacent low lying areas of the flood plain and complement the method of flow
direction simulation. Although multi-date imageries can serve as an alternative to
flow direction simulation, it has severe limitation of determining the flood depth.
A very accurate terrain data base regarding the local flood plain morphology is an
essential prerequisite for such an operation. The flood water intersecting with the
slope is taken as the primary indicator of determining flood depth. Thus for the
gently sloping topography, the resolution of the terrain data actually controls the
accuracy of the estimated flood depth (Brakenridge et al., 1998). High resolution
satellite imageries or aerial photographs are needed for preparing an accurate DEM
which can meet the precision level of a flood depth investigation.
APPLICATION OF REMOTE SENSING IN FLOOD MANAGEMENT 293
One of the recent developments in the application of remote sensing to flood
related problems is the use of LIDAR (Light Detecting and Ranging) sensor. In the
developed countries, especially in USA, this technology has become very popular
for creating DEMs for flood prone areas. LIDAR sensors can readily identify the
vertical differences in the landforms and can be exploited as powerful instruments
to create DEMs of exceptional accuracy. This sensor can also detect the flood depth.
Although LIDAR sensor can attain the vertical accuracy of 5 cm or better, it is
difficult to map it in that resolution. Due to the limitation of GPS systems to locate
an aircraft/sensor functionally LIDAR generated DEMs are released at an accuracy
of 15 to 25 cm RMSE. Moreover the accuracy decreases gradually with increase
in the density of vegetation cover of the ground (Hodgson et al., 2003). Although
LIDAR data is even more expensive than the SAR imageries, sometimes it provides
the only appropriate option to do flood mapping in the extremely flat flood plains.
Fowler (2002) maintained that the resolution of the LIDAR data depends upon the
intensity of the laser pulses and any attempt to make the survey more intensive by
increasing the laser pulses would increase the cost of the data exponentially.
4.2. AGRICULTURAL DAMAGE ASSESSMENT
Boyle et al. (1998) classified flood damages into two categories; tangible and
intangible.
Tangible damage occurs due to direct contact with the flood water whereas
intangible damage is exemplified by the loss of historical monuments, heritage sites
etc. A disease assuming a form of an epidemic due to flood is also categorized as
place during that time. With the current Radarsat resources it is very difficult to
capture the spatial extent of a flood at its peak. Thus, attempts have been made to
extrapolate the extent of inundation at the peak of a flood from an image acquired
at a later stage of the event. Some GIS algorithms in ARC/INFO are promising to
perform this extrapolation from an image that captures some standing water only at
a time when the flood peak had already passed. The method of least accumulation
cost distance can provide a viable solution to this problem (Brivio et al., 2002).
This principle simulates flow direction from the river channel to the flood plain
based on the assumption that water flows through the path where the work done in
doing so is least.
This methodology yielded a remarkable accuracy of 96.7 percent when compared
with the aerial photographs of the peak of the particular flood event which
occurred in November 1994 in Italy. The main weakness of this approach is that
the least accumulation cost distance operates on certain values of the raster cells
which represent the roughness of the terrain causing frictional drag to the
overflowing
flood water. Since roughness is a function of a host of other geomorphic and
APPLICATION OF REMOTE SENSING IN FLOOD MANAGEMENT 295
lithological factors, it is very difficult to control the parameters of the experiments
in an area having diverse lithology and land use. Apart from the DEMs and remote
sensing data, field work conducted in recently flood affected area can prove useful.
Wang et al. (2002) used high-flood level marks of a recent flood on houses as a
supplementary data set to the DEMs for estimating flood depth. This kind of data
serve very well to reconstruct dimension of a past flood peak.
General studies of flood geomorphology also provide some insights to resolve
this problem. Many very high magnitude floods may leave permanent imprints in
the flood plain morphology rather than creating some transient features (Brunsden
et al., 1979).When the morphology of the basin perfectly suites with the magnitude
and frequency of the flood event, the flood created features attain a state of
permanency.
Since the frequency of such an event is high in the humid tropics, the interval
between two high magnitude floods gets reduced here. It indirectly contributes to
the permanency of the flood created characteristic morphological features in the
flood plain (Gupta, 1983). The permanency of this kind of topographic features
would be more prominent in areas of high relief and coarser valley sediments
(Gupta, 1988). Hence in the humid tropics there is enough scope to use relatively
inexpensive high resolution optical remote sensing not necessarily at the time of
flooding to demarcate the areas vulnerable to river flooding. But the analyst has to
be trained enough to identify the morphological features in the flood plain typically
associated with high magnitude floods.
5. Conclusion and Prospective
The preceding discussion regarding various facades of the prospects and constraints
of using remote sensing and GIS for flood management unveils some
significant facts. A majority of the researchers favoured multi date radar imageries
to observe a particular flood event and considered different image processing
techniques
to overcome the limitations of remotely sensed data in flood delineation.
The main weaknesses of this type of approach are manifold. Most of the
investigations
mentioned in this paper are heavily dependent on the availability of satellite
data, which is not always guaranteed for the time of peak flood. Rather than moving
towards a comprehensive flood management strategy, these papers concentrated
on
some specific issues like delineation of flooded area. Most of these projects are on
a very high budget and no attempts were made try to keep the cost low. Monsoon
floods affect the developing countries more acutely than the developed ones.
Therefore,
while framing the methodology one should be aware of its feasibility in the
operational area. Most of the studies in this field cannot achieve its desired level
of accuracy without a very high resolution DEM, therefore, we need to develop
alternative methodology to shed our dependence on high resolution terrain data.
A detailed hydro-geomorphological mapping depicting the trace of past floods
may help us in this direction. The hazard of monsoon flood and the destitution
of people associated with it are very different from the industrially developed
296 JOY SANYAL AND X. X. LU
countries. Hence, we should adopt an improvised but effective methodology of
risk and damage assessment to come out with meaningful flood hazard maps for
this region.
Application of remote sensing and GIS is convincingly a very efficient and cost
effective way of flood management. Use of very high resolution imageries like
IKONS or SPOT 5 have not been very popular yet in the field of flood management
due to its high price, but it is likely that with these imageries would be available
at a reasonable price and would be widely used for flood mapping. In the age
of internet, GIS has assumed new dimensions, especially for coping with natural
disasters like river flooding. The most important advantage of using internet based
GIS is that it has opened the door of GIS technology to the end users who would not
like to install expensive GIS software. One of the numerous examples is Arc IMS
technology. This technology has been used to develop a web enabled application
named Map Action Processing Digital Interactive Geo Resource (MAPDIGR) for
providing very recent information regarding flood risk to an analyst via internet
(Smith et al., 2002). This technology is at present at an embryonic stage of
development
but has great potential for expand the user base of GIS technology for flood
management by substantially reducing the cost of operation.
Since the problem of flood is very acute in the developing countries of monsoon
Asia, special attention should be given to deal with this problem in the regional
context. GIS models having low cost and simple data requirement are likely to
attract the local authorities in the developing countries to adopt this technology
as an essential input towards a comprehensive flood management system. In the
age of all embracing flood plain management, these sophisticated technologies can
be very useful for the planners to formulate effective strategy for combating the
perpetual natural disaster of river flooding.
Acknowledgement
This study has been done as part of an ongoing project titled Application
of Remote Sensing and GIS in Flood Hazard Management: A Case Study of
West Bengal, India, funded by National University of Singapore research grant
R-109-000-049-112.
References
Ali, A. and Quadir, D. A.: 1987, Agricultural hydrologic and oceanographic studies in Bangladesh
with NOAA AVHRR data, International Journal of Remote Sensing 8(6), 917925.
Ali, S., Hassan, A.,Martin, T. C., and Hassan Q. K.: 2001, Geospatial tools for monitoring flood plain
water dynamics, In: M. Owe, K. Brubaker, J. Ritchie, and A. Rango, A. (eds), Remote Sensing
and Hydrology 2000, IAHS Publication, Oxford, pp. 465471.
Andre, G., Guillande, R., and Bahoken, F.: 2002, Flood mapping using spatial radar and optical
imagery and digital elevation model: Limits and capacities, Houille Blanche-Revue Int. De LEAU
1, 4954.
APPLICATION OF REMOTE SENSING IN FLOOD MANAGEMENT
297
Barton, I. and Bathols, J.: 1989, Monitoring floods with AVHRR, International Journal of Remote
Sensing 10(12), 18731892.
Berg, A. and Gregiore, J. M.: 1983, Use of remote sensing techniques for rice production forecasting
in West Africa, (Mali and Guinea: Niger-Bani project), ESA Satellite Remote Sensing for
Developing Countries, Ispra, Italy, pp. 161168
Beven, K. J. and Kirkby, M. J.: 1979, A physically based, variable contributing area model of basin
hydrology, Hydrol. Sci. Bull. 24, 4369.
Bhavsar, P. D.: 1984, Review of remote sensing applications in hydrology and water resource
management in India, Advances in Space Research 4(11), 193200.
Blyth, K.: 1995, A telenetwork for acquisition, processing and dissemination of Earth Observation
data for monitoring and emergency management of floods, In: Proc. First ERS Thematic Working
Group Meeting on Flood Monitoring. European Space Agency/ESRIN, Frascati, Italy, 2627
June.
Boyle, S. J., Tsams, I. K., Manber, A. S. C. E., and Kanaroglu, P. S.: 1998, Developing geographical
information system for land use impact assessment in flooding conditions, Journal of Water
Resource Planning and Management 124(2), 8998.
Brakenridge, G. R., Tracy, R. T., and Know, J. C.: 1998, Orbital SAR remote sensing of the river
flood waves, International Journal of Remote Sensing 19(7), 14391445
Brimicombe, J., and Bratlet, J. M.: 1996, Linking geographical information system with hydraulic
simulation modeling for flood risk assessment: The Honk Kong Approach, In: M. F. Goodchild
(eds), GIS and Environmental Modeling, Oxford University Press, New York, pp. 165168.
Brivio, P. A., Colombo, R., Maggi, M., and Tomasoni, R.: 2002, Integration of remote sensing data
and GIS for accurate mapping of flooded areas, International Journal of Remote Sensing 23(3),
429441.
Brouder, J. A. M.: 1994, Flood study in the Meghna-Dhonagoda Polder, Bangladesh, In: Proc. Asian
Institute of Remote Sensing, Bangalore, India, 1723 November.
Brunsden, D. and Thorns, J. B.: 1979, Landscape sensitivity and change, Trans. Inst. Br. Geogr., New
Series 4, 463484.
Chen, P., Liew, S. C., and Lim, H.: 1999, Flood detection using multitemporal Radarsat and ERS
SAR data, In: Proc. 20th Asian Conference of Remote Sensing, Hong Kong, 2225 November.
Choen, K. and Kwang-Hoon, C.: 1998, Flood damage mapping in North Korea using multi-sensor
data, In: Proc. 19th Asian Conference of Remote Sensing, Manila, Philippines, 1620 November.
Clark, M. J.: 1998, Putting water in its place: A perspective on GIS in hydrology and water
management, Hydrological Processes 12, 823834.
Collins, S. H. and Moon, G. C.: 1981, Algorithms for dense digital terrain models, Photogrammetric
Engineering and Remote Sensing 47, 7176.
Coppock, J. T.: 1995, GIS and natural hazard: an overview from a GIS perspective, In: A. Carrara
and F. Guzzetti (eds), Geographical Information System in Assessing Natural Hazard, Kluwer
Academic, Netherlands, pp. 2134.
Deutsch, M.: 1976, Optical processing of ERTS data for determining extent of 1973 Mississippi River
flood, USGS Professional Paper 929, ERTS-1, A New Window on Our Planet, pp. 209222.
Deutsch, M. and Ruggles, F.: 1974, Optical data processing and projected application of the ERTS1
imagery covering the 1973 Mississippi River Valley floods,Water Resource Bulletin 10(5), 1023
1039.
Deutsch, M., Ruggles, F., Guss, P., and Yost, E.: 1973, Mapping the 1973 Mississippi floods from
the Earth Resource Technology satellites, In: Proc. International Symposium on Remote Rensing
and Water Resource Management, American Water Resource Association, No. 17, Burlington,
Ontario, pp. 3955.
Dhakal, A. S., Amda, T., Aniya, M., and Sharma, R. R.: 2002, Detection of areas associated
with flood and erosion caused by a heavy rainfall using multi temporal Landsat TM data,
Photogrammetric Engineering and Remote Sensing 68(3), 233239.
298
Fowler, R. A.: 2002, LIDAR for flood mapping, Earth Observation Magazine 9(7)
(http://www.eomonline.com/Common/Archives/July00/robert.htm, true as par 30th May, 2003)
Gamon, J. A., Field, C. B., Goulden, M. L., Griffin, K. L., Hartley, A. E., Joel, G., Penuela, J., and
Valentini, R.: 1995, Relationship between NDVI, canopy structure and photosynthesis in their
Californian vegetation types, Ecological Applications 5, 2841.
Godschalk, D. R.: 1991, Disaster mitigation and hazard management, In: T. E. Drabek and G. J. Hoetmer
(eds), Emergency Management: Principles and Practice for Local Government, International
City Management Association, Washington, D.C. pp. 131160.
Gumley, L. and King, M. D.: 1995, Remote sensing of flooding in the US Upper Midwest during the
summer of 1993, Bulletin of American Meteorological Society 76(6), 933943.
Gupta, A.: 1983, High magnitude floods and stream channel response, Special Publication of
International Association of Sedimentology 6, 219227.
Gupta, A.: 1988, Large floods as geomorphic events in the humid tropics, In: Victor R. Barker, Craig
R. Kochel, and Peter C. Paton (eds), Flood Geomorphology, John Wiley and Sons, pp. 301321.
Hallberg, G. R., Hoyer, B. E., and Rango, A.: 1973, Application of ERTS1 imagery to flood inundation
mapping, NASA Special Publication No. 327, Symposium on Significant Results Obtained
from the Earth Resources Satellite 1, Vol. 1, Technical presentations, section A, pp. 745753.
Hausmann, P. and Weber, M.: 1988, Possible contributions of hydroinformatics to risk analysis in
insurance, In: Proc. 2nd International Conference on Hydroinformatics, Zurich, Switzerland,
913 September, Balkema, Rotterdam.
Hess, L. L., Melack, J. M., and Simonett, D. S.: 1990, Radar detection of flooding beneath the forest
canopy a review, International Journal of Remote Sensing 11(7), 13131325.
Hodgson, M. E., Jensen, J. R., Schmidt, L., Schill, S., and Davis, B.: 2003, An evaluation of LIDAR
and IFSAR derived digital elevation models in leaf-on conditions with USGS level-1 and level-2
DEMs, Remote Sensing of Environment 84, 295308.
Honda, K. C., Francis, X.J., and Sah V. P.: 1997, Flood monitoring in central plain of Thailand using
JERS-1 SAR data, In: Proc. 18th Asian Conference of Remote Sensing,Malaysia, 2024 October.
Horritt, M. S. and Bates, P. D.: 2001, Predicting flood plain inundation: Raster based modeling versus
the finite element approach, Hydrological Processes 15(5), 825842.
Huguenin, R. L., Karaska, M. A., Blaricom, D. V., and Jensen, J. R.: 1997, Subpixel classification of
bald cypress and tupelo gum trees in Thematic Mapper imagery, Photogrammetric Engineering
& Remote Sensing 63, 717725.
Huh, O. K., Ali, A., and Quadir, D. A.: 1985a, Mapping of green leaf biomass over Bangladesh
with NOAA satellite AVHRR data, Manuscript Report prepared for UN Flood and Agriculture
Organization, Coastal Studies Institute, Louisiana State University., Baton Rouge, LA.
Huh, O. K., Ali, A., and Quadir, D. A.: 1985b, Observations on the surface waters of the Bay
of Bengal with NOAA satellite AVHRR data, Manuscript Report prepared for UN Flood and
Agriculture Organization, Coastal Studies Institute , Louisiana State University, Baton Rouge,
LA.
Huh, O. K., Ali, A., and Quadir, D. A.: 1985c, Use of NOAA satellite AVHRR data to monitor
river flood hydrology in Bangladesh, Manuscript Report prepared for UN Flood and Agriculture
Organization, Coastal Studies Institute, Louisiana State University, Baton Rouge, LA.
Hunter, G. J. and Goodchild, M. F: 1995, Dealing with error in spatial databases a simple casestudy,
Photogrammetric Engineering & Remote Sensing 61(5), 529537.
Imhoff, M. L., Vermillion, C., Story, M. H., Choudhury, A. M., Gafoor, A., and Polcyn, P.: 1987,
Monsoon flood boundary delineation and damage assessment using space borne imaging radar
and Landsat data, Photogrammetric Engineering and Remote Sensing 53(4), 405413.
Islam, M. M. and Sadu, K.: 2000a, Development of flood hazard maps of Bangladesh using NOAAAVHRR
images with GIS, Hydrological Science Journal 45(3), 337355.
Islam, M. M., and Sadu, K.: 2000b, Flood hazard assessment in Bangladesh using NOAA-AVHRR
data with geographical information system, Hydrological Processes 14(3), 605620.
APPLICATION OF REMOTE SENSING IN FLOOD MANAGEMENT
299
Islam, M. M. and Sadu, K.: 2000c, Satellite remote sensing data analysis for flood damaged zoning
with GIS for flood management, Journal of Hydraulic Engineering 44, 301306.
Islam, M. M. and Sadu, K.: 2001, Flood damage and modelling using satellite remote sensing data
with GIS: Case study of Bangladesh; In: Jerry Ritchie et al. (eds), Remote Sensing and Hydrology
2000, IAHS Publication, Oxford, pp. 455458.
Islam, M. M. and Sadu, K.: 2002, Development of priority map remote sensing data for flood counter
measures by geographical information system, Journal of Hydrological Engineering 7(5), 346
355.
Jensen, J. R., Rutchey, K., Koch, M., and S. Narumalani: 1995, Inland Wetland Change Detection
in the Everglades Water Conservation Area 2A Using A Time Series of Normalized Remotely
Sensed Data, Photogrammetric Engineering & Remote Sensing 61(2), 199209.
Jin, Y. Q.: 1999, A flooding index and its regional threshold values for monitoring in china from
300
Nico, G., Pappalepore, M., Pasquariello, G., Refice, A., and Samarelli, S.: 2000, Comparison of
SAR amplitude vs. coherence flood detection methods- a GIS application, International Journal
of Remote Sensing 21(8), 16191631.
Oberstadler, R., Honsch, H., and Huth, D.: 1997, Assessment of the mapping capabilities of ERS-1
SAR data for flood mapping: A case study of Germany, Hydrological Processes 10(10), 1415
1425.
Okamoto, K. and Fukuhara,M.: 1996, Estimation of paddy field area using the area ratio of categories
in each pixel of Landsat TM, International Journal of Remote Sensing 17, 17351749.
Okamoto, K., Yamakawa, S., and Kawashima, H.: 1998, Estimation of flood damage of North Korea
in 1995, International Journal of Remote Sensing 19(2), 365371.
Patel, N. K., Singh, T. P., Sahai, B., and Patel, M.: 1985, Spectral response of rice crops and its
relation to yield and yield attributes, International Journal of Remote Sensing 6, 657664.
Pope, K. O., Rejmankova, E., Paris, J.F. et al.: 1997, Detecting seasonal flooding cycles in marshes
of the Yucatan Peninsula with SIR-C polarimetric radar imagery, Remote Sensing of Environment
59(2), 157166.
Profeiti, G. and MacIntosh, H.: 1997, Flood management through Landsat TM and ERS SAR data:
a case study, Hydrological Processes 11(10), 13971408.
Quan,W.,Watanabe, M., Hayashi, S., and Murakami, S.: 2003, Using NOAA AVHRR data to assess
flood damage in China, Environmental Monotoring and Assessment 82, 119148.
Rango, A. and Anderson, A. T.: 1974, Flood hazard studies in the Mississippi River Basin using
remote sensing, Water Resource Bulletin 10(5), 10601081.
Rango, A. and Solomonson, V. V.: 1974, Regional flood mapping from space, Water Resource
Research 10(3), 473484.
Rango, A. and Solomonson, V. V.: 1977, The utility of short wavelength (<1 mm) remote sensing
techniques for the monitoring and assessment of hydrological parameters, In: Proc. 11th Int.
Symp. on Remote Sensing of Environ., Ann Arbor, MI, 2529 April 1977, pp. 5564.
Rashid, H. and Pramanik, M. A. H.: 1993, Areal extent of the 1988 flood in Bangladesh: How much
did the satellite imagery show? Natural Hazards 8, 189200.
Reed, S., Johnson, D., and Sweeney, T.: 2002, Application of national geographic information system
data base to support two-year flood and the threshold run-off estimate, Journal of Hydrological
Engineering 7(3), 209219.
Rejesk, D.: 1993, GIS and risk: A three cultural problem, In: M. F. Goodchild, B. O. Parks, and L.
T. Steyaerts (eds), Environmental Modeling With GIS, Oxford University Press, New York, pp.
49267.
Rocha, J. S.,Marques, Z., Ramos, I., and Ameida, R.: 1994, Simulation of risk flood areas on GIS, In:
Tsakiris, G. and Santos, M. A. (eds), Advances in Water Resource Technology and Management,
Balkema, Rotterdam, pp. 375383.
Rosenqvist, A., Shimada, M., Chapman, B., Freeman, A., Grande, G. De, and Saatchi, S.: 2000,
The global rainforest mapping a review, International journal of Remote Sensing 21(6 & 7),
13751387.
Rosenqvist, A., Forsberg, B. R., Pimentel, T., Rauste, Y. A., and Riche, J. E.: 2002, The use of space
borne radar data to model inundation patterns and trace gas emission in the central Amazon flood
plain, International Journal of Remote Sensing 23(7), 13031328.
Ruangsiri, P., Sripumin, R., Polongam, S., Kanjanasuntorn, P., and Wongparn, S.: 1984, State of
flooding in the Mun-Chi River Basin area, N. E. Thailand by digital Landsat data analysis,
Report: Remote Sensing Division, National Resource Council of Thailand, Bangkok, Thailand.
Sado, K., and Islam, M. M.: 1997, Satellite remote sensing data analysis for flooded area and weather
study: Case study of Dhaka city, Bangladesh, Journal of Hydraulic Engineering 41, 945950.
Saikh, M., Green, D., and Cross, H.: 2001, A remote sensing approach to determine environmental
flows for wet lands of the Lower Darling River, New South Wales, Australia, International
Journal of Remote Sensing 22(9), 17371751.
APPLICATION OF REMOTE SENSING IN FLOOD MANAGEMENT
301
Sharma, P.: 1999, Flood risk mapping of Dikrong subbasin of Assam, Proc. Map Asia 1999. Source
: http://www.gisdevelopment.net/proceedings/mapindia/1999/index.htm
Shibayama, M. and Akiyama, T.: 1989, Seasonal visible, near-infrared and multi infrared spectra
of rice canopies in relation to LAI and above- ground dry phytomass, Remote Sensing of the
Environment 27, 119127.
Singh, A.: 1989, Digital change detection techniques using remotely-sensed data, International
Journal of Remote Sensing 10, 9891003.
Smith, L. C.: 1997, Satellite remote sensing of river inundation area, stage and discharge: A review,
Hydrological Processes 11, 14271439.
Smith, C. and Price, J.: 2002, A new generation of flood management, Arc User 5(4), 2022.
Tappen, G. D., Tyler, D.,Wehde,M., andMoor, D.: 1992, Monitoring rangeland dynamics in Senegal
with AVHRR data, Geocarto International 1, 8798.
Tholey, N., Clandillon, S., and De Fraipont, P.: 1997, The contribution of space borne SAR and
optical data in monitoring flood events; examples in Northern and Southern France, Hydrological
Processes 11(10), 14271439.
Toutin, T.: 2002, DEMfrom stereo Landsat 7 ETM+ data over high relief areas, International Journal
of Remote Sensing 23(10), 21332139.
Townsend, P. A. and Walsh, S. J.: 1998, Modelling flood plain inundation using integrated GIS with
radar and optical remote sensing, Geomorphology 21(98), 295312.
Wadge, G., Wislocki, A. P., Pearson, J., and Whittow, J. B.: 1993, Mapping natural hazards with
spatial modeling System, In: P. M. Mather (eds), Geographic Information Handling Research
and Applications, John Wiley and Sons Inc., New York.
Wang, W., Hess, L. L., Filoso, S., and Melack, T. M.: 1995, Understanding the radar back scattering
from flooded and non-flooded Amazonian Forests: result from canopy back scatter modelling,
Remote Sensing of Environment 54(3), 324332.
Wang, Y., Colby, J. D., and Mulcahy, K. A.: 2002, An efficient method for mapping flood extent in
a coastal flood plain using Landsat TM and DEM data, International Journal of Remote Sensing
23(18), 36813696.
Wiesnet, D. R., McGinnis, D. F., and Pritchard, J. A.: 1994, Mapping of the 1973 Mississippi River
floods by the NOAA-2 satellite, Water Resource Bulletin 10(5), 10401049.
Wolock, D. M.: 1995, Effects of subbasin size on topographic characteristics and simulated flow
paths in sleepers river watershed, Vermont, Water Resour. Res. 31(8), 19891997.
Yamagata, Y. and Akiyama, T.: 1988, Flood damage analysis using multi temporal Landsat TM data,
International Journal of Remote Sensing 9(3), 503514.
Yang, C., Zhou, C., and Wan, Q.: 1999, Deciding the flood extent with Radarsat SAR data and Image
Fusion, In: Proc. 20th Asian Conference of Remote Sensing, Hong Kong, 2225 November.
ABSTRACT
The growing availability of multi-temporal satellite data has increased opportunities for monitoring
large rivers from
space. A variety of passive and active sensors operating in the visible and microwave range are
currently operating, or
planned, which can estimate inundation area and delineate ood boundaries. Radar altimeters show
great promise for
directly measuring stage variation in large rivers. It also appears to be possible to obtain estimates of
river discharge
from space, using ground measurements and satellite data to construct empirical curves that relate
water surface area to
discharge. Extrapolation of these curves to ungauged sites may be possible for the special case of
braided rivers.
Where clouds, trees and oating vegetation do not obscure the water surface, high-resolution
visible/infrared sensors
provide good delineation of inundated areas. Synthetic aperture radar (SAR) sensors can penetrate
clouds and can
also detect standing water through emergent aquatic plants and forest canopies. However, multiple
frequencies and
polarizations are required for optimal discrimination of various inundated vegetation cover types.
Existing singlepolarization, xed-frequency SARs are not sucient for mapping inundation area in all riverine
environments. In the
absence of a space-borne multi-parameter SAR, a synergistic approach using single-frequency, xedpolarization SAR
and visible/infrared data will provide the best results over densely vegetated river oodplains. #1997
John Wiley &
Sons, Ltd.
Hydrol. Process., Vol. 11, 14271439 (1997)
(No. of Figures: 3 No. of Tables: 0 No. of Refs: 103)
KEY WORDS
remote sensing; river inundation; review; synthetic aperture radar, Landsat, oods
INTRODUCTION
Satellite-derived ood inundation maps produced in near-real time are invaluable to state or
national
agencies for disaster monitoring and relief eorts. New facilities are being developed that
will utilize
Internet/World Wide Web technology to disseminate satellite data rapidly during ood events
(Biasutti and
Lombardi, 1995; Blyth 1995). Precise mapping of the maximum ood extent is also required
for detecting
deciencies in existing ood control measures and for arbitrating damage claims later.
Remote sensing has proved useful in ecological, hydrological and geomorphological river
studies. Its
value in remote regions has been demonstrated in the Amazon Basin (Sipple et al., 1992;
Koblinsky et al.,
1993; Hess et al., 1995), where seasonal to interannual variations in stage and oodplain
inundation area are
are potential breeding grounds for mosquitoes that carry the dangerous Rift Valley Fever
virus have been
mapped in Kenya using Landsat TM and airborne polarimeter data (Pope et al., 1992).
Floodplain boundaries and land surface types can be delineated with Landsat (Rango et al., 1975; Hollyday,
1976; Sollers et al.,
1978); Nagarajan et al. (1993) used Landsat images and aerial photographs over the Rapti
River in India to
identify areas vulnerable to channel migration and oods.
The sensors used in these and other river studies may be classied into two types: (1)
passive, in which the
sensor receives energy naturally reected by or emitted from the earth's surface; and (2)
active, in which the
sensor provides its own illumination and records the amount of incident energy returned
from the imaged
surface. Passive sensors include all of the visible and infrared instruments such as the
Landsat Thematic
Mapper (TM) and Multi-Spectral Scanner (MSS), the Advanced Very High Resolution
Radiometer
(AVHRR), the Satellite Pour l'Observation de la Terre (SPOT) and the anticipated Advanced
Spaceborne
Thermal Emission and Reection Radiometer (ASTER), Moderate-Resolution Imaging
Spectroradiometer
(MODIS) and Landsat-7 sensors. Passive microwave radiometers such as the Special Sensor
Microwave/
Imager (SSM/I) measure the amount of microwave energy naturally emitted from the Earth's
surface.
However, the poor spatial resolution of spaceborne microwave radiometers (ca. 27 km at 37
GHz) limits
their use to very large areas. This problem has been mitigated through use of spectral
mixing models, which
extract subpixel proportions of spectrally distinct end members (Sipple et al., 1992;
Hamilton et al., 1996).
The active sensors described in this review consist of imaging radars and radar altimeters.
Radars can
penetrate clouds, darkness and, at the longer wavelengths, tree canopies. Cloud penetration
is particularly
important for monitoring ood events, as they commonly occur during periods of extended
rainfall.
However, interpretation of synthetic aperture radar (SAR) imagery is less straightforward
than for the
visible/infrared range. In addition, the presence of wind-induced waves or emergent
vegetation can roughen
the surface of open water bodies, making them dicult to discriminate from other, nonooded land surface
types.
The aim of this paper is to review briey eorts to use active and passive remote sensing to
estimate water
surface area, stage and discharge. Methods for mapping surface area are by far the best
developed. More
recently, improvements in satellite orbital precision and the increasing availability of multitemporal satellite
data have enabled the estimation of river stage and discharge from space. While these
techniques are largely
in their infancy and not yet used operationally, three general approaches have emerged: (1)
direct measurement of water surface level from radar altimeter waveform data; (2) determination of water
surface
elevations at their point of contact with the land surface using high-resolution satellite
imagery and
topographic data; and (3) correlation of satellite-derived water surface areas with ground
measurements of
stage or discharge. It should be noted that river ow velocity cannot be directly measured
from space.
Satellite estimates of discharge therefore require the use of ground-based empirical
relationships between
discharge and inundation area or stage.
Studies that have used passive (visible, infrared and microwave range) and active
(microwave) sensors to
delineate inundation area are reviewed in the rst section, and eorts to estimate river
stage and discharge
from space are reviewed in the second section.
REMOTE SENSING OF INUNDATION AREA WITH PASSIVE AND ACTIVE SENSORS
Visible/infrared Remote Sensing of Inundation Area
Much of the pioneering work on the remote sensing of oods was accomplished using the
Multi-Spectral
Scanner (MSS) sensor on ERTS-1 (the rst Earth Resources Technology Satellite, later
renamed Landsat-1),
launched on 23 July 1972. With a spatial resolution of about 80 m, MSS data were used to
map the extent of
ooding in Iowa (Hallberg et al., 1973; Rango and Salomonson, 1974), Arizona (Morrison and
Cooley,
1973), Virginia (Rango and Salomonson, 1974) and along the Mississippi River (Deutsch et
al., 1973;
Deutsch and Ruggles, 1974; Rango and Anderson, 1974; McGinnis and Rango, 1975;
Deutsch, 1976;
Morrison and White, 1976). Maximum ood boundaries derived from the MSS imagery were
shown to
1428 L. C. SMITH
HYDROLOGICAL PROCESSES, VOL. 11, 14271439 (1997) #1997 John Wiley & Sons, Ltd.
agree well with those derived from aerial photography (Hallberg et al., 1973; Morrison and
Cooley, 1973)
and US Army Corps of Engineers and US Geological Survey ood hazard maps (Rango and
Anderson,
1974). In one case, the MSS-derived ood hazard map performed best, near a major
tributary where
backwater ooding occurred (Rango and Anderson, 1974). In all studies, MSS band 7
(0.81.1 mm) was
most useful for discriminating water from dry soil or vegetated surfaces owing to the strong
absorption of
water in the near-infrared range. This was further conrmed by comparing panchromatic
and colour
infrared aerial photographs over inundated areas (Hallberg et al., 1973; Moore and North,
1974), and by
analysing MSS band 5 (0.60.7 mm), band 7 and eld spectral radiometer data along
shoreline waterwet
soildry soil transitions (Gupta and Banerji, 1985). Errors in MSS-derived ooded areas have
been
estimated to be less than 5% (Rango and Salomonson, 1974); a similar error estimate has
been reported for
lakes (Gupta and Banerji, 1985). Chidley and Drayton (1986) suggested that the SPOT
system can detect
water bodies smaller than 0.5 ha. France and Hedges (1986) found minimum detectable lake
areas of 0.6 ha
for TM and 2.4 ha for MSS. However, only a 64% accuracy was reported for MSS
classications of ooded
land in Bangladesh, where the 80 m spatial resolution failed to resolve adequately the raised
dike system that
separates rice paddies and serves as infrastructure for transport and dwellings (Imho et al.,
1987). Even
larger errors can occur in ooded forests, because trees are highly reective in the visible
and near-infrared
range (Hallberg et al., 1973; Moore and North, 1974). Floating emergent macrophytes are
also a problem in
many riverine environments, particularly in tropical systems such as the Amazon (Melack et
al., 1994; Hess,
1995). Vila da Silva and Kux (1992) noted a signicant underestimation of inundation area
over such
aquatic habitats when using Landsat TM.
Perhaps the greatest problem with visible/infrared sensors is their inability to image the
Earth's surface
during cloudy conditions (Rango and Salomonson, 1977; Lowry et al., 1981; Van den Brink,
1986; Blyth
and Biggin, 1993; Rasid and Pramanik, 1993; Melack et al., 1994). For the purpose of
determining
maximum ood extent, this diculty is somewhat mitigated by the fact that residually wet
soils and stressed
vegetation can be mapped even after ood stage recession (Rango and Anderson, 1974;
Deutsch, 1976). This
eect can last from one to two weeks (Hallberg et al., 1973; Rango and Salomonson, 1974;
Morrison and
White, 1976; Salomonson, 1983).
Other studies have continued the methodology rst developed with MSS, using TM and
SPOT data
(France and Hedges, 1986; Jensen et al., 1986; Watson, 1991; Blasco et al., 1992; Pope et
al., 1992; Vila
da Silva and Kux, 1992). There has also been some success studying very large rivers or
oods with the
coarser resolution (ca. 1 km) US National Ocean and Atmospheric Administration Very High
Resolution
Radiometer (VHRR) (McGinnis and Rango, 1975; Dey et al., 1977), the Advanced Very High
Resolution
Radiometer (AVHRR) (Ali et al., 1989; Barton and Bathols, 1989; Gale and Bainbridge, 1990;
Rasid and
Pramanik, 1993) and the Nimbus-7 Coastal Zone Color Scanner (CZCS) (Wiesnet and
Deutsch, 1987).
However, since the 1970s there has been little change in the way visible/infrared satellite
data are used to
map open water bodies. The approach is relatively straightforward and now considered
operational where
trees and vegetation do not obscure the water surface and clear skies prevail to permit data
acquisition.
Passive Microwave Remote Sensing of Inundation Area
Natural thermal emission of microwaves from the Earth's surface may be measured by
passive microwave
radiometers such as the Scanning Multichannel Microwave Radiometer (SMMR) which ew on
board the
Nimbus-7 satellite from 1978 to 1987, and the Special Sensor Microwave/Imager (SSM/I)
currently
operating on a DMSP (Defense Meteorological Satellite Program) platform. Brightness
temperatures (K)
measured at the sensor are proportional to the product of the eective surface temperature
and the
emissivity of the emitting medium (Choudhury, 1989). At the 37 GHz frequency, atmospheric
water vapour
content and temperature inuence the received brightness temperatures. To mitigate these
eects, the
dierence between the vertically and horizontally polarized brightness temperatures (DT) is
often used
instead of the actual brightness temperatures. Large values of DT are found over surfaces
that emit a
strongly polarized microwave signal (such as open water). Depolarization by scattering from
vegetation or
REMOTE SENSING OF RIVER INUNDATION 1429
#1997 John Wiley & Sons, Ltd. HYDROLOGICAL PROCESSES, VOL. 11, 14271439 (1997)
1430
L. C. SMITH
HYDROLOGICAL PROCESSES, VOL. 11, 14271439 (1997) #1997 John Wiley & Sons, Ltd.
1994). Rasid and Pramanik (1993) cited pervasive cloud cover as the single greatest
limitation to their study of
ooding in Bangladesh using AVHRR imagery. Van den Brink (1986) reported the same
diculty in Kenya.
An early attempt to map ood inundation extent with active microwaves used a side-looking
airborne
radar (SLAR) at the X- and L-bands (Lowry et al., 1981). X-band SLAR imagery, acquired in
the late 1970s
was also used to map oodplain lakes along the Amazon River, Brazil (Sipple et al., 1992). At
the First ERS
Thematic Working Group Meeting on Flood Monitoring (ESA/ESRIN, 1995), numerous
investigators
presented ood inundation maps derived from C-band ERS-1 synthetic aperture radar (SAR)
images. A
general observation was that for smooth open water bodies without vegetation, especially
trees, radar
returns are normally low due to specular reection from the water surface. This
characteristic permits ood
boundaries to be determined with a good level of accuracy under many conditions. However,
turbulence,
wind, emergent vegetation and trees can all cause signicant increases in radar backscatter, making
inundation extent dicult or impossible to determine.
When a micro-scale element (at the scale of the radar wavelength) of an imaged surface is
subjected to an
incident microwave, the ratio of incident energy reected away from the surface to the
energy refracted
downwards into the media is proportional to the dielectric constant and the incidence angle
at which the
microwave intersects the imaged surface. The complex dielectric constant of a material is
proportional to the
strength of its dipole moment in the presence of a time-varying external electric eld, while
the incidence
angle (at the micro-scale) is determined by the surface roughness. Over a larger area (such
as a pixel), the
angular radiation pattern reected by the surface is also aected by the local slope, which
exerts a net
directionality to the radiation pattern. For smooth water bodies, surface roughness and the
local slope are
both nearly at. Consequently, most incident microwave energy impacts the surface
obliquely and is
specularly reected away from the satellite, yielding low radar returns. Wind reduces this
eect by roughening the water surface, increasing radar back-scattering to the satellite. Since water bodies
are perfectly
saturated, dielectric contrast eects are less important than surface roughness eects. This
is not the case for
unsaturated soils, where the presence of liquid water causes increased radar back-scatter
owing to an
increased complex dielectric constant of the soilwater mixture.When the soil becomes
saturated and water
is ponded, specular reection is enabled (in the absence of wind or emergent vegetation)
and back-scatter to
the satellite is dramatically reduced.
Where vegetation or trees are present, a wavelength-dependent increase in radar backscatter is commonly
observed. Surface roughness elements of the order of the radar wavelength exert maximum
scattering eect.
For example, X-band (2.43.8 cm) and C-band (3.87.5 cm) signals are eectively scattered
by the leaves,
twigs and branches within a forest canopy. L-band (15.030.0 cm) and P-band (30.0100.0
cm) display
greater canopy penetration but interact strongly with trunks and large branches. Radar
impulses are
transmitted and received in plane-polarized form by the radar antenna. Polarizations may be
transmitted
and received horizontally (HH), vertically (VV) or cross-polarized (HV or VH). Radars with
multiple
frequencies and polarizations provide much more information about the imaged surface than
singlepolarization, xed-frequency radars. L-band radar back-scatter is commonly increased over
ooded forests
through a double-bounce mechanism between the water surface and inundated tree trunks
or branches
(Hoer et al., 1985; Harris and Digby-Argus, 1986; Richards et al., 1987; Hess et al., 1990;
Hess andMelack,
1994). An HV phase dierence of 1538 was found in airborne polarimeter data over a
forested swamp
(Durden et al., 1989), agreeing closely with a phase dierence of 1528 modelled for the site
(a pure doublebounce return is dened as having a 1808 phase dierence between the vertical and
horizontal polarizations).
The magnitude of the back-scatter increase is signicantly aected by vegetation type
(Krohn et al., 1983;
Evans et al., 1986; Harris and Digby-Argus, 1986; Pope et al., 1994) and may be suppressed
altogether by a
dense undergrowth between the tree canopy and water surface (Waite et al., 1981).
Airborne multipolarization radar studies have generally indicated that ooding beneath forest canopies is
best detected
with co-polarizations rather than cross-polarizations (Wedler and Kessler, 1981; Hoer et al.,
1985; Evans
et al., 1986; Pope et al., 1992; Hess et al., 1995), although Wu and Sader (1987) favour use
of a co- and crosspolarization ratio.
REMOTE SENSING OF RIVER INUNDATION 1431
#1997 John Wiley & Sons, Ltd. HYDROLOGICAL PROCESSES, VOL. 11, 14271439 (1997)
From L-HH Shuttle Imaging Radar B (SIR-B) SAR data collected over the Altamaha River in
Georgia,
Hess and Melack (1994) conrmed the observation of Hoer et al. (1985) that L-band
enhancement is
aected by radar incidence angle. Flooded pine forest could be discriminated from dry forest
at incidence
angles of 188 and 458 from vertical, but not 588. Horizontally polarized (HH) C-, L-, and Pband NASA/JPL
AIRSAR data were also acquired over this site, and illustrate superbly the features that can
be observed in
wetlands and ooded forests with dierent frequencies. Increased returns at C-band over
marshes were
interpreted as double-bounce reections between emergent plant stalks and a smooth water
surface. These
stalk dimensions were too small to increase returns at L- and P-band, so marshes appeared
dark at these
longer wavelengths. The converse was true in ooded bottomland forests, where L- and Pbands penetrated
the upper canopy but experienced double-bounce reections between the water surface and
inundated tree
trunks. C-band was attenuated by the leaves and twigs of the forest canopy, except in areas
defoliated by
caterpillar damage and permanent inundation by beaver ponds (Hess and Melack, 1994).
Comprehensive
reviews of the interaction of various radars with a wide range of ooded vegetation types are
provided by
Hess et al. (1990) and Melack et al. (1994).
Recent eorts have used fully polarimetric SAR data to classify various inundated land
surface types.
Amplitudes at multiple frequencies and polarizations vary with vegetation structure and
species. Multiparameter SAR can thus be used to detect ooding in vegetated riverine environments to an
extent not
possible with single-frequency, xed-polarization radars. Pope et al. (1994) presented four
biophysical
indices that can be used to discriminate thickets and areas of forest regrowth, marshes,
swamp forests and
upland forests. Hess et al. (1994, 1995) used a decision-tree model to classify polarimetric
SIR-C (Shuttle
Imaging Radar C) acquired over the Negro and Amazon rivers near Manaus, Brazil in April
and October,
1994. Both C- and L-bands were used to distinguish open water, oating macrophytes and
ooded forest
from upland forest and clearings. While the total area of open water did not change
signicantly between
data acquisitions, the extent of ooded forest was nearly 50% less in October. It is clear that
multiparameter SAR oers a distinct advantage over single-frequency, xed-polarization SAR
where diverse
vegetation cover is found.
REMOTE SENSING OF RIVER STAGE AND DISCHARGE
In the previous section studies that used passive or active sensors to measure river
inundation area or ood
extent were reviewed. In this section, a small group of studies that attempted to use satellite
data to
estimate river stage or discharge are discussed. Such eorts will likely become more
common in the future
as satellite orbital errors decrease and high-resolution multi-temporal data become more
plentiful. Data
availability has already increased with the launch of the ERS-1, ERS-2, JERS-1 and RADARSAT
spaceborne SARs, which are not limited by weather conditions or darkness. Even more data
are anticipated from future high-resolution SARs (JERS-2, ENVISAT-1, RADARSAT-2) and visible/infrared
sensors (Landsat-7, ASTER, SPOT 5 and 6).
Radar Altimetry of Water Surface Elevation
Radar altimetry has emerged as a promising method for directly measuring stage variations
in large rivers.
Radar altimeters emit a short, nadir-directed radar pulse to the Earth's surface. The two-way
return time is
used to calculate the range between the satellite and the target. Because the pulse is in the
form of a curved
wave front, a time-varying waveform is returned to the satellite. An on-board tracker, which
assumes a
gradually changing surface (i.e. ice sheets and water bodies), is used to estimate when the
return echo will
arrive. Interactions with the surface cause the waveform to be distorted. If the tracker
receives a rapidly
changing echo, the tracker may not `lock' properly and range estimates are lost. For this
reason, radar
altimeters generally do not work well over land. However, where a lock on the land power
return is
maintained, it is possible to identify peaks in the waveform data that result from specular
returns over large
rivers (Guzkowska et al., 1990). These spikes progress through the waveform as the satellite
passes over the
1432 L. C. SMITH
HYDROLOGICAL PROCESSES, VOL. 11, 14271439 (1997) #1997 John Wiley & Sons, Ltd.
river. Koblinsky et al. (1993) used this approach with Geosat waveform data to estimate river
stage at four
sites along the Amazon. The derived water surface levels are presented with the
corresponding ground
measurements in Figure 2. Altimeter ranges were calculated by manually selecting specular
returns from the
waveform data. An average root mean square error of 0.7 m was found between the ground
and altimeter
estimates of water surface level. Uncertainties in the ground stage measurements (ca. 10
cm or more) and the
radial component of the satellite orbit (estimated at 50 cm) were major contributors to this
error.
Birkett (1994) obtained lake level changes using a similar analysis of Geosat altimeter
waveforms. Her
results suggest that radar altimetry can be used to measure relative water level changes in
lakes within 10 cm.
For very large lakes, waveform analysis is not required to obtain good estimates of water
level: Morris and
Gill (1994) used average height estimates from the Geosat altimeter to estimate levels in
the ve Great Lakes
of the United States, with an average root mean square error of 11.1 cm. Average heights
are a standard
altimeter product and represent the average water level over the altimeter footprint, which
typically has a
radius of several kilometres.
Figure 2. Comparisons in river stage at four sites in the Amazon basin, as estimated from river gauges (solid line)
and Geosat altimeter
waveforms (pluses). Standard deviations of the dierences between the altimeter and river stage data are also
shown. Period of
observation is from November 1986 to November 1988. (Adapted from Koblinsky et al., 1993)
REMOTE SENSING OF RIVER INUNDATION
1433
#1997 John Wiley & Sons, Ltd. HYDROLOGICAL PROCESSES, VOL. 11, 14271439 (1997)
Errors in orbital precision pose the greatest problem to using satellite altimetry to measure
stage in rivers
or lakes. The Geosat orbital error has been estimated at about 50 cm (Koblinsky et al.,
1993), although it
has been subsequently reduced through recomputation of some of the 17-day exact repeat
mission (ERM)
orbits (Morris and Gill, 1994). However, this problem is steadily diminishing as satellite
tracking systems
improve. The radial component of the ERS-1 orbit is determined to within 15 cm; the
equivalent error is
only 3 cm for TOPEX/Poseidon (Le Traon et al., 1995). ERS-1 orbital errors have been reduced
even
further by using TOPEX/Poseidon data to minimize the variation in measured sea surface
heights at points
where the two satellite tracks cross (Le Traon et al., 1995). As altimeters with small orbital
errors (such as
TOPEX/Poseidon) become more common, radar altimetry over inland water surfaces may
become routine
for those sites where a lock on the land surface can be maintained.
Watersurface Elevations from Satellite Imagery and Topographic Data
River stage can be estimated at the landwater contact, using high-resolution satellite
imagery in
combination with topographic maps or a digital elevation model (DEM). Requirements for
this approach
are high image and topographic resolution and an unobscured water edge boundary along
an area of gently
sloping relief. Gupta and Banerji (1985) used topographic maps and Landsat MSS data to
obtain water
surface elevations for a large reservoir in India. Landwater contact elevations were chosen
along two lowgradient, inundated stream channels near their conuences with the reservoir. The derived
water surface
elevations agreed closely with values estimated from measurements surveyed on the
ground.
Miller (1986) used Landsat MSS and 1 : 50 000 topographic maps to obtain spot water
surface elevations
over the Belize River in Central America. Flood volumes subsequently estimated by
assuming mean
inundation depths of 1.53.0 m were in general agreement with volumes calculated from
hydrograph
analysis. However, signicant parts of the landwater boundary were obscured by
vegetation, limiting the
spot elevations to areas near permanent lagoons. Brakenridge et al. (1994) used ERS-1 SAR
images and
1 : 24 000 US Geological Survey topographic maps to pick water edge elevations at gently
sloping alluvial
fans along the Mississippi River during the `Great Flood' of 1993. Instantaneous ood stages
diered by as
much as 2.4 m at opposite sides of the Mississippi River, illustrating the value of this
approach for
estimating ood proles and monitoring ood wave dynamics.
Correlation of Inundated Area with Ground Measurements of Stage or Discharge
A third method for estimating river stage or discharge from space is by correlating ground
measurements
of these variables with satellite-derived estimates of inundation area. Until recently, the
major obstacle to
this approach has been the requirement of a large number of satellite images to construct
empirical rating
curves relating inundation area to stage or discharge. Kruus et al. (1981) compared total
inundation areas
from seven Landsat MSS scenes over the St John River in New Brunswick, Canada with
simultaneous
measurements of river stage. They found a general increase of stage with inundation area,
but at some times
an increase in stage was associated with a decrease in inundation area. Usachev (1983)
used a similar
approach to determine the relationship between inundation area and ground measurements
of stage and
discharge in the Ob River of Siberia. In all of the Ob River analyses, increasing stage or
discharge was
associated with increasing inundation area. The correlation was quasi-linear and quite
similar for three
study sections at Mogochin, Kolpashevo and Kargasok. Xia et al. (1983) presented a strongly
linear
correlation between water surface area and ground measurements of water level for
Dongting Lake, China's
second largest freshwater body.
Until additional empirical rating curves relating inundation area to ground measurements of
stage or
discharge are made, it is dicult to assess their potential for extrapolation to other rivers of
similar
morphology. However, it seems likely that such curves will vary signicantly between rivers
and therefore
must be constructed for each site. An exception may be the special case of braided rivers,
which display an
extreme spatial sensitivity of water surface area to changing discharge, and also share some
common
morphological properties. Smith et al. (1995) used multi-temporal ERS-1 SAR data and
simultaneous
1434 L. C. SMITH
HYDROLOGICAL PROCESSES, VOL. 11, 14271439 (1997) #1997 John Wiley & Sons, Ltd.
used to parameterize empirically derived rating curves for extrapolation to other ungauged
braided rivers.
CONCLUSIONS
High-resolution visible/infrared sensors such as Landsat provide good delineation of ood
extent where
clouds, trees or oating vegetation do not obscure the water surface. Passive microwave
radiometers show
promise for acquiring estimates of inundation area over very large rivers or wetlands, but
sensor resolution
is coarse and the landwater boundary cannot be located accurately. Spaceborne synthetic
aperture radars
(SARs) are not limited by weather conditions or darkness. Single polarization, C-band SARs
are eective
for mapping smooth, open water bodies. However, emergent vegetation, trees, wind or ow
turbulence can
all increase radar back-scatter returns, making delineation of inundated areas problematic.
L-band SARs
can penetrate forest canopies and often display increased returns over ooded forest
produced by a doublebounce mechanism between inundated tree trunks and the water surface. Multi-frequency
polarimetric SAR
provides much more information and can be used to classify dierent types of inundated
terrain to an extent
not possible with single-frequency SARs and visible/infrared sensors. However, there are no
current plans to
launch a multi-frequency, fully polarimetric SAR on a satellite platform. As a consequence, a
synergistic
Figure 3. Relationship between satellite-derived eective width and river discharge for a large braided river in
Canada. Eective width
is the total water surface area contained within a 10 km _ 3 km river reach, divided by the reach length. Each point
is determined from
a single ERS-1 SAR image and a ground measurement of discharge. The outlier is an extreme ood event; maximum
annual ows do
not normally exceed 2000 m3/s. (From Smith et al., 1995)
REMOTE SENSING OF RIVER INUNDATION
1435
#1997 John Wiley & Sons, Ltd. HYDROLOGICAL PROCESSES, VOL. 11, 14271439 (1997)
remote sensing approach utilizing both visible/infrared and SAR data may be the only
eective way
to monitor inundation area in vegetated riverine environments. Visible/infrared data will
provide good
discrimination of ood extent in open areas, identify places where vegetation is a problem
and help with
interpretation of SAR data. SARs will provide excellent temporal coverage and, in certain
situations, will be
able to determine the ood extent through emergent plants and forest canopies.
As satellite orbital precision and data availability improve, promising methods for estimating
river
stage and discharge should continue to develop. Radar altimetry shows potential for
measuring stage in
large rivers to within 10 cm. Orbital laser altimeters such as GLAS (Geoscience Laser
Altimeter System),
planned for launch in 2003, should also permit monitoring of stage variation in large rivers.
Landsat and
ERS-1 SAR data have been combined with topography to obtain water surface elevations at
the point of
contact between water and land. This approach can also be used to determine lateral
asymmetries in river
stage and estimate the ood prole. Satellite-derived estimates of water surface area have
been correlated
with ground measurements of stage or discharge to obtain rating curves that can
subsequently be used to
estimate these ow variables from satellite data alone. Transfer of such empirical rating
curves to estimate
discharge in ungauged basins appears feasible in the special case of braided rivers, which
are spatially
sensitive to changing discharge and commonly share some morphological similarities
between sites.
ACKNOWLEDGEMENTS
This work was carried out at the Department of Geological Sciences, Cornell University, with
funding from
NASA through Graduate Student Researchers Program Fellowship NGT-51223, Mission to
Planet Earth/
Earth Observing System (EOS) grant NAGW-2638 and Spaceborne Imaging Radar (SIR-C)
grant 958745.
Helpful comments and assistance were provided by A. Bloom, B. Isacks, J. Liebeskind and R.
Forster of the
Department of Geological Sciences, Cornell University. S. Hamilton of the W.K. Biological
Station,
Michigan State University, and C. Koblinsky of NASA/Goddard Space Flight Center provided
Figures 1
and 2.
REFERENCES
Ali, A., Quadir, D. A., and Huh, O. K. 1989. `Study of river ood hydrology in Bangladesh with AVHRR data', Int. J.
Remote Sens.,
10(12), 18731891.
Asselmann, I. and Crutzen, P. J. 1989. `Global distribution of natural freshwater wetlands and rice paddies, their net
primary
productivity, seasonality, and possible methane emissions', J. Atmos. Chem., 8, 307358.
Astaras, T. 1985. `Drainage network analysis of Landsat images of the Olympus-Pieria mountain area, Northern
Greece', Int. J.
Remote Sens., 6, 673686.
Astaras, T., Lambrinos, N., and Soulakellis, N. 1990. `A drainage system analysis evaluation of, and comparison
between,
Landsat-3 RBV, Landsat-5 TM and SPOT PA imageries covering the Central Macedonia district, Greece', Int. J.
Remote
Sens., 11(9), 15491559.
Barrett, E. C., Herschy, R. W., and Stewart, J. B. 1988. `Satellite remote sensing requirements for hydrology and
water management
from the mid 1990's, in relation to the Columbus Programme of the European Space Agency', Hydrol. Sci. J., 33(1),
117.
Barton, I. J., and Bathols, J. M. 1989. `Monitoring oods with AVHRR', Remote Sens. Environ., 30, 8994.
Bates, P. D. and Anderson, M. G. 1995. `Issues of hydraulic model validation and design using remotely sensed
data', Proceedings,
First ERS Thematic Working Group Meeting on Flood Monitoring, European Space Agency/ESRIN, Frascati, Italy,
2627 June
1995.
Bhavsar, P. 1984. `Review of remote sensing applications in hydrology and water resources management in India',
Adv. Space Res.,
4(11), 193200.
Biasutti, R. and Lombardi, A. 1995. `EARTHWATCH: a service provided by EURIMAGE the ood case', Proceedings,
First ERS
Thematic Working Group Meeting on Flood Monitoring, European Space Agency/ESRIN, Frascati, Italy, 2627 June
1995.
Birkett, C. M. 1994. `Radar altimetry: a new concept in monitoring lake level changes', EOS Trans., AGU, 75(24),
273275.
Blasco, F., Bellan M. F., and Chaudhury, M. U. 1992. `Estimating the extent of oods in Bangladesh using SPOT data',
Remote Sens.
Environ., 39, 167178.
Blyth, K. 1995. `FLOODNET a telenetwork for acquisition, processing, and dissemination of Earth Observation data
for
monitoring and emergency management of oods', Proceedings, First ERS Thematic Working Group Meeting on
Flood Monitoring,
European Space Agency/ESRIN, Frascati, Italy, 2627 June 1995.
Blyth, K. and Biggin, D. S. 1993. `Monitoring oodwater inundation with ERS-1 SAR', Earth Obs. Quart., 42, 68.
1436
L. C. SMITH
HYDROLOGICAL PROCESSES, VOL. 11, 14271439 (1997) #1997 John Wiley & Sons, Ltd.
Brakenridge, G. R., Knox, J. C., Paylor, E. D., and Magilligan, F. J. 1994. `Radar remote sensing aids study of the great
ood of 1993',
EOS Trans., AGU, 75(45), 521527.
Brooner, W. G. and Binaghi, C. M. V. 1983. `Landsat monitoring of temporal hydrological variations on the Pilcomayo
River,
19721981', Proceedings, 17th International Symposium on Remote Sensing of the Environment., Ann Arbor, MI,
pp. 399407.
Chidley, T. R. E. and Drayton, R. S. 1986. `The use of SPOT-simulated imagery in hydrological mapping', Int. J.
Remote Sens., 7(6),
791799.
Choudhury, B. J. 1989. `Monitoring global land surface using Nimbus-7 37 GHz data: theory and examples', Int. J.
Remote Sens.,
10(10), 15791605.
Choudhury, B. J. 1991. `Passive microwave remote sensing contribution to hydrological variables', Surv. Geophys.,
12, 6384.
Cihlar, J., Pultz, T. J., and Gray, A. L. 1992. `Change detection with synthetic aperture radar', Int. J. Remote Sens.,
13(3), 401414.
Deutsch, M. 1976. `Optical processing of ERTS data for determining extent of the 1973 Mississippi River ood', U.S.
Geol. Surv. Prof.
Pap. 929, ERTS-1, a New Window on Our Planet, pp. 209222.
Deutsch, M. and Ruggles, F. 1974. `Optical data processing and projected applications of the ERTS-1 imagery
covering the 1973
Mississippi River Valley oods', Wat. Resour. Bull., 10(5), 10231039.
Deutsch, M., Ruggles, F. H., Guss, P., and Yost, E. 1973. `Mapping of the 1973 Mississippi river oods from the Earth
Resources
Technology Satellite', Proceedings, International Symposium on Remote Sensing and Water Resources Management,
American Water
Resources Association, Proceedings no. 17, Burlington, Ontario, pp. 3955.
Dey, B., Moore, H., and Gregory, A. F. 1977. `The use of satellite imagery for monitoring ice break-up along the
Mackenzie River,
N.W.T', Arctic, 30(4), 234242.
Durden, S. L., Van Zyl, J. J., and Zebker, H. A. 1989. `Modeling and observation of the radar polarization signature of
forested areas',
IEEE Trans. Geosci. Remote Sens., 27(3), 290301.
ESA/ESRIN, 1995. First ERS thematic Working Group Meeting on Flood Monitoring, Frascati, Italy, 2627 June 1995.
Evans, D. L., Farr, T. G., Ford, J. P., Thompson, T. W., and Werner, C. L. 1986. `Multipolarization radar images for
geologic
mapping and vegetation discrimination', IEEE Trans. Geosci. Remote Sens., GE-24(2), 246257.
Ferguson, H. L., Deutsch, M., and Kruus, J. 1980. `Applications to oods of remote sensing from satellites', Adv.
Space Res., 9,
195206.
France, M. J. and Hedges, P. D. 1986. `A hydrological comparison of Landsat TM, Landsat MSS, and black and white
aerial
photography', in Damen, Smit, and Verstappen (Eds), Proceedings, 7th International Symposium, ISPRS (Internat.
Soc. Photogram.
and Remote Sens.) Commission VII, Enschede, 2529 August 1986, pp. 717720.
Gale, S. J. and Bainbridge, S. 1990. `The oods in eastern Australia', Nature, 345, 767.
Giddings, L. and Choudhury, B. J. 1989. `Observation of hydrological features with Nimbus-7 37 GHz data, applied to
South
America', Int. J. Remote Sens., 10(10), 16731686.
Gupta, R. P. and Banerji, S. 1985. `Monitoring of reservoir volume using Landsat data', J. Hydrol., 77, 159170.
Gupta, R. P. and Bodechtel, J. 1982. `Geotechnical applications of Landsat image analysis of Bhakra dam reservoir,
India', Remote
Sens. Environ., 12, 313.
Guzkowska, M. A. J., Rapley, C. G., Ridley, J. K., Cudlip, W., Birkett, C. M., and Scott, R. F. 1990. `Developments in
inland water
and land altimetry', European Space Agency Contract Report 7839/88/F/FL.
Hallberg, G. R., Hoyer B. E., and Rango, A. 1973. `Application of ERTS-1 imagery to ood inundation mapping', NASA
Special Pub.
No. 327, Symposium on signicant results obtained from the Earth Resources Satellite-1, Vol. 1, Technical
presentations, section A,
pp. 745753.
Hamilton, S. K., Sipple, S. J., and Melack, J. M. 1996. `Inundation patterns in the Pantanal wetland of South America
determined
from passive microwave remote sensing.' Arch. Hydrobiol., 137, 115141.
Harris, J. and Digby-Arbus, S. 1986. `The detection of wetlands on radar imagery', Proceedings, 10th Canadian
Remote Sensing
Symposiums Edmonton, Alberta, 58 May, 1986. Canadian Aeronautics and Space Institute, Ottawa, pp. 529543.
Hess, L. L. and Melack, J. M. 1994. `Mapping wetland hydrology and vegetation with synthetic aperture radar', Int. J.
Ecol. Environ.
Sci., 20, 197205.
Hess, L. L., Melack, J. M., and Simonett, D. S. 1990. `Radar detection of ooding beneath the forest canopy: a
review', Int. J. Remote
Sens., 11(7), 13131325.
Hess, L. L., Melack, J. M., and Davis, F. W. 1995. `Mapping of oodplain inundation with multi-frequency polarimetric
SAR: use of a
tree-based model', Proceedings of the 1994 Int. Geosci. and Remote Sens. Symp (IGARSS '94), pp. 10721073.
Hess, L. L., Melack, J. M., Filoso, S., and Wang, Y. 1995. `Delineation of inundated area and vegetation along the
Amazon oodplain
with the SIR-C synthetic aperture radar', IEEE Trans. Geosci. Remote Sens., 33(4), 895903.
Hockey, B., Richards, T., and Osmaston, H. 1990. `Prospects for operational remote sensing of surface water',
Remote Sens. Rev., 4(2),
265283.
Hoer, R. M., Mueller, P. W., and Lozano-Garcia, D. F. 1985. `Multiple incidence angle shuttle imaging radar data for
discriminating
forest cover types', Proceedings, ASCM-ASPRS Fall Technical Meeting, Indianapolis, 813 September 1985, pp.
476485.
Hollyday, E. F. 1976. `Improving estimates of streamow characteristics by using Landsat-1 imagery', J. Res. US
Geol. Surv., 4(5),
517531.
Imho, M. and Gesch, D. 1988. `The derivation of sub-canopy surface terrain models of coastal forests using
synthetic aperture radar',
Proceedings, 1988 Int. Geosci. and Remote Sens. Symp (IGARSS '88), pp. 613617.
Imho, M. L., Vermillion, C., Story, M. H., Choudhury, A. M., Gafoor, A., and Polcyn, F. 1987. `Monsoon ood boundary
delineation and damage assessment using space borne imaging radar and Landsat data', Photogramm. Eng.
Remote Sens., 53(4),
405413.
Jensen, J. R., Hodgson, M. E., Christensen, E., Mackey, H. E., Jr., Tinney, L. R., and Sharitz, R. 1986. `Remote sensing
inland
wetlands: a multispectral approach', Photogramm. Eng. Remote Sens., 52, 87100.
REMOTE SENSING OF RIVER INUNDATION
1437
#1997 John Wiley & Sons, Ltd. HYDROLOGICAL PROCESSES, VOL. 11, 14271439 (1997)
Koblinsky, C. J., Clarke, R. T., Brenner, A. C., and Frey, H. 1993. `Measurement of river level variations with satellite
altimetry',
Wat. Resour. Res., 29(6), 18391848.
Krohn, M. D., Milton, N. M., and Segal, D. B. 1983. `Seasat synthetic aperture radar (SAR) response to lowland
vegetation types in
eastern Maryland and Virginia', J. Geophys. Res., 88(C3), 19371952.
Kruus, J., Deutsch, M., Hansen, P. L., and Ferguson, H. L. 1981. `Flood applications of satellite imagery' in Satellite
Hydrology,
Proceedings, Fifth Ann. William T. Pecora Memorial Symp. on Remote Sens., 1979, Amer. Wat. Resour. Assoc. Tech.
Pub. No. TPS
811, pp. 292301.
Le Traon, P. Y., Gaspar, P., Ogor, F., and Dorandeu, J. 1995. `Satellites work in tandem to improve accuracy of data',
EOS Trans.,
AGU, 76(39), 385389.
Lowry, R. T., Langham, E. J., and Mudry, N. 1981. `A preliminary analysis of SAR mapping of the Manitoba Flood,
May 1979',
in Satellite Hydrology, Proceedings, Fifth Ann. William T. Pecora Memorial Symp. on Remote Sens., 1979, Amer. Wat.
Resour.
Assoc. Tech. Pub. No. TPS81-1, pp. 316323.
Macdonald, H. C., Waite, W. P., and Demarcke, J. S. 1980. `Use of Seasat satellite radar imagery for the detection of
standing
water beneath forest vegetation', Tech. Papers of the Amer. Soc. of Photogramm. Fall Tech. Meeting, Niagara Falls,
New York,
710 October, 1980, RS3B1RS3Bl3.
McGinnis, D. F. and Rango, A. 1975. `Earth Resources Satellite systems for ood monitoring', Geophys. Res. Let.,
2(4), 132135.
Melack, J. M., Hess, L. L., and Sippel, S. 1994. `Remote sensing of Lakes and Floodplains in the Amazon Basin',
Remote Sens. Rev.,
10, 127142.
Mertes, L. A. K., Smith, M. O., and Adams, J. B. 1993. `Estimating suspended sediment concentrations in surface
waters of the
Amazon River wetlands from Landsat images', Remote Sens. Environ., 43, 281301.
Miller, S. T. 1986. `The quantication of oodplain inundation by the use of Landsat and metric camera information,
Belize, Central
America', Proceedings, 7th International Symposium, ISPRS (Internat. Society Photogrammetry and Remote
Sensing) Commission
VII, Enschede August 1986, pp. 733738.
Moore, G. K. and North, G. W. 1974. `Flood inundation in the southeastern United States from aircraft and satellite
imagery',
Wat. Resour. Bull., 10(5), 10821096.
Morris, C. S. and Gill, S. K. 1994. `Variation of Great Lakes water levels derived from Geosat altimetry', Wat. Resour.
Res., 30(4),
10091017.
Morrison, R. B. and Cooley, M. E. 1973. `Assessment of ood damage in Arizona by means of ERTS-1 imagery',
Proceedings,
Symposium on Signicant Results Obtained from the Earth Resources Satellite-1, Vol. 1, New Carrollton, Maryland,
pp. 755760.
Morrison, R. B. and White, P. G. 1976. `Monitoring ood inundation', U.S. Geol. Surv. Prof. Pap. 929, ERTS-1, A New
Window on
Our Planet, pp. 196208.
Mouchot, M. C., Alfoldi, T., DeLisle, D., and McCullough, G. 1991. `Monitoring the water bodies of the Mackenzie
Delta by remote
sensing methods', Arctic, 44 (suppl. 1), 2128.
Muller, E., Decamps, H., and Dobson, M. K. 1993. `Contribution of space remote sensing to river studies',
Freshwater Biol., 29,
301312.
Murray, A. B. and Paola, C. 1994 `A cellular model of braided rivers', Nature, 371, 5457.
Nagarajan, R., Marathe, G. T., and Collins, W. G. 1993. `Identication of ood prone regions of Rapti river using
temporal remotelysensed data', Int. J. Remote Sens., 14(7), 12971303.
Ormsby, J. P., Blanchard, B. J., and Blanchard, A. J. 1985. `Detection of lowland ooding using active microwave
systems',
Photogramm. Eng. Remote Sens., 51(3), 317328.
Philipson, W. R. and Hafker, W. R. 1981. `Manual versus digital Landsat analysis for delineating river ooding',
Photogramm. Eng.
Remote Sens., 47(9), 13511356.
Pope, K. O., Shener, E. J., Linthicum, K. J., Bailey, C. L., Logan, T. M., Kasichke, E. S., Birney, K., Njogu, A. R., and
Roberts, C. R.
1992. `Identication of Central Kenyan Rift Valley Fever virus vector habitats with Landsat TM and evaluation of
their ooding
status with airbourne imaging radar', Remote Sens. Environ., 40, 185196.
Pope, K. O., Rey-Benayas, J. M., and Paris, J. F. 1994. `Radar remote sensing of forest and wetland ecosystems in the
Central
American tropics', Remote Sen. Envion., 48, 205219.
Rango, A. and Anderson, A. T. 1974. `Flood hazard studies in the Mississippi River basin using remote sensing', Wat.
Resour. Bull.,
10(5), 10601081.
Rango, A. and Salomonson, V. V. 1973. `Repetitive ERTS-1 observations of surface water variability along rivers and
low-lying areas',
Amer. Wat. Resour. Assoc. Proc. no. 17, Remote Sensing and Water Resources Management, pp. 191199.
Rango, A. and Salomonson, V. V. 1974. `Regional ood mapping from space', Wat. Resour. Res., 10(3), 473484.
Rango, A. and Salomonson, V. V. 1977. `The utility of short wavelength (51 mm) remote sensing techniques for the
monitoring and
assessment of hydrological parameters' Proceedings, 11th Int. Symp. on Remote Sens. of Environ., Ann Arbor, MI,
2529 April 1977,
pp. 5564.
Rango, A., Foster, J., and Salomonson, V. V. 1975. `Extraction and utilization of space acquired physiographic data
for water
resources development', Wat. Resour. Bull, 11(6), 12451255.
Rapley, C. G., Baker, S. G., Birkett, C. M., Laxon, S. W., Ridley, J. K., Scott, R. F., and Strawbridge, F. 1992. `ERS-1
altimetry of inland
water and land', Proceedings, First ERS-1 Symp., ESA SP-359 (March, 1993), Cannes, France, 46 November, 1992,
pp. 539542.
Rasid, H. and Pramanik, M. A. H. 1993. `Areal extent of the 1988 ood in Bangladesh: how much did the satellite
imagery show?',
Natural Hazards, 8, 189200.
Rast, M., Jaskolla, F., and Arnason, K. 1991. `Comparative digital analysis of Seasat-SAR and Landsat-TM data for
Iceland', Int. J.
1438
L. C. SMITH
HYDROLOGICAL PROCESSES, VOL. 11, 14271439 (1997) #1997 John Wiley & Sons, Ltd.
Richards, J. A., Woodgate, P. W., and Skidmore, A. K. 1987b. `An explanation of enhanced radar backscattering from
ooded
forests', Int. J. Remote Sens., 8(7), 10931100.
Richey, J. E., Mertes, L. A. K., Dunne, T., Victoria, R. L., Forsberg, B. R., Tancredi, A., and Oliveira, E. 1989. ` Sources
and routing
of the Amazon River ood wave', Global Biogeochem. Cycles, 3, 1991204.
Richey, J. E., Victoria, R. L., Salati, E., and Forsberg, B. R. 1991. `The biogeochemistry of a major river system: the
Amazon case
study', in Degens, E. T., Kempe, S., and Richey, J. E. (eds), Biogeochemistry of Major World Rivers, John Wiley &
Sons, Ltd,
Chichester. pp. 5774.
Ringrose, S., Matheson, W., and Boyle, T. 1988. `Dierentiation of ecological zones in the Okavango Delta,
Botswana, by classication and contextural analysis of Landsat MSS data', Photogramm. Eng. Remote Sens., 54(5), 601608.
Salomonson, V. V. (Ed.), 1983. `Water resources assessment', in Manual of Remote Sensing, Chap. 29, Amer. Soc. of
Photogramm.
pp. 14971570.
Salomonson, V. V. and Rango, A. 1974. `ERTS-1 applications in hydrology and water resources', J. Amer. Wat. Works
Assoc., 66(3),
168172.
Schmugge, T. 1987. `Remote sensing applications in hydrology', Rev. Geophys., 25(2), 148152.
Schmugge, T. J., O'Neill, P. E., and Wang, J. R. 1986. `Passive microwave soil moisture research', IEEE Trans. Geosci.
Remote Sens.,
24, 1222.
Schultz, G. A. 1988. `Remote sensing in hydrology', J. Hydrol., 100, 239265.
Sipple, S. J., Hamilton, S. K., and Melack, J. M. 1992. `Inundation area and morphometry of lakes on the Amazon
River oodplain,
Brazil', Arch. Hydrobiol., 123(4), 385400.
Sipple, S. J., Hamilton, S. K., Melack, J. M., and Choudhury, B. J. 1994. `Determination of inundation area in the
Amazon River
oodplain using the SMMR 37 GHz polarization dierence', Remote Sens. Environ., 48, 7076.
Smith, L., Isacks, B. L., Forster, R. R., Bloom, A. L., and Preuss, I. 1995. `Estimation of discharge from braided glacial
rivers using
ERS-1 SAR: rst results', Wat. Resour. Res., 31, 13251329.
Smith, L. C., Isacks, B. L., Bloom, A. L., and Murray, A. B. 1996. `Estimation of discharge from three braided rivers
using synthetic
aperture radar (SAR) satellite imagery: potential for application to ungaged basins', Water Resour. Res., 32(7),
20212034.
Sollers, S. C., Rango, A., and Henninger, D. L. 1978. `Selecting reconnaissance strategies for oodplain surveys',
Wat. Resour. Bull.,
14(2), 359373.
Solomon, S. I. 1992. `Methodological considerations for use of ERS-1 imagery for the delineation of river networks in
tropical
forest areas', Proceedings, First ERS-1 Symp., ESA SP-359 (March, 1993), Cannes, France, 46 November, 1992, pp.
595600.
Usachev, V. F. 1983. `Evaluation of ood plain inundations by remote sensing methods', Proceedings of the Hamburg
Symposium,
IAHS Publ., 145, 475482.
Van den Brink, J. W. 1986. `A study with NOAA-7 AVHRR imagery in monitoring ephemeral streams in the lower
catchment area of
the Tana River, Kenya', Damen, Smit, and Verstappen (Eds.), Proceedings, 7th International Symposium, ISPRS
(Internat. Soc.
Photogram. and Remote Sens.) Commission VII, Enschede, 2529 August 1986, pp. 783785.
Van Es, E., Bomez, H., and Soeters, R. 1975. `An inundation study of the lower Magdalena-Cauca River Basin', NASA
Tech. Memo
X-58168, Vol. 1-D, pp. 22952297.
Vila da Silva, J. S. and Kux, H. J. H. 1992. `Thematic mapper and GIS data integration to evaluate the ooding
dynamics within the
Panatal, Mato Grosso do Sul State, Brazil', Proceedings, 1992 Int. Geosci. Remote Sens. Symp. (IGARSS '92), pp.
14781480.
Waite, W. P., MacDonald, H. C., Kaupp, V. H., and Demarcke, J. S. 1981. `Wetland mapping with imaging radar',
Proceedings, 1981
Int. Geosci. and Remote Sens. Symp. (IGARSS '81), pp. 794799.
Watson, J. P. 1991. `A visual interpretation of a Landsat mosaic of the Okavango Delta and surrounding area',
Remote Sens. Environ.,
35, 19.
Wedler, E. and Kessler, R. 1981. `Interpretation of vegetative cover in wetlands using four-channel SAR imagery',
Tech. Papers of the
Amer. Soc. of Photogramm., ASP-ASCM Convention, Washington, D.C., 2227 February, 1981, pp. 111124.
Whitelaw, A., Howes, S., Fletcher, P. and Rast, M. 1995. `Remote sensing applications in hydrological modelling',
SPIE (Soc. of
Photo-Optical Instrumentation Engineers) vol. 2314, Multispectral and Microwave Sensing of Forestry, Hydrology,
and Natural
Resources, Rome, Italy, 2630 September, 1994, pp. 618627.
Wiesnet, D. R. and Deutsch, M. 1987. `Flood monitoring in South America from the Landsat NOAA and NIMBUS
Satellites',
Adv. Space Res., 7(3), 7784.
Wu, S. T. and Sader, S. A. 1987. `Multipolarization SAR data for surface feature delineation and forest vegetation
characterization',
IEEE Trans. Geosci. Remote Sens., GE-25(1), 6776.
Xia, Liu, Shulin, Z., and Xianglian, L. 1983. `The application of Landsat imagery in the surveying of water resources
of Dongting
Lake', Proceedings of the Hamburg Symposium, IAHS Publ., 145, 483489.
REMOTE SENSING OF RIVER INUNDATION
1439
#1997 John Wiley & Sons, Ltd. HYDROLOGICAL PROCESSES, VOL. 11, 14271439 (1997)
Abstract
Satellite remote sensing is providing a systematic, synoptic framework for advancing scientific
knowledge of the Earth as a
complex system of geophysical phenomena that, directly and through interacting processes, often lead
to natural hazards.
Improved and integrated measurements along with numerical modeling are enabling a greater
understanding of where and when
a particular hazard event is most likely to occur and result in significant socioeconomic impact.
Geospatial information products
derived from this research increasingly are addressing the operational requirements of decision
support systems used by policy
makers, emergency managers and responders from international and federal to regional, state and
local jurisdictions. This forms
the basis for comprehensive risk assessments and better-informed mitigation planning, disaster
assessment and response
prioritization. Space-based geodetic measurements of the solid Earth with the Global Positioning
System, for example,
combined with ground-based seismological measurements, are yielding the principal data for modeling
lithospheric processes
and for accurately estimating the distribution of potentially damaging strong ground motions which is
critical for earthquake
engineering applications. Moreover, integrated with interferometric synthetic aperture radar, these
measurements provide
spatially continuous observations of deformation with sub-centimeter accuracy. Seismic and in situ
monitoring, geodetic
measurements, high-resolution digital elevation models (e.g. from InSAR, Lidar and digital
photogrammetry) and imaging
spectroscopy (e.g. using ASTER, MODIS and Hyperion) are contributing significantly to volcanic hazard
risk assessment, with
the potential to aid land use planning in developing countries where the impact of volcanic hazards to
populations and lifelines
is continually increasing. Remotely sensed data play an integral role in reconstructing the recent
history of the land surface and
in predicting hazards due to flood and landslide events. Satellite data are addressing diverse
observational requirements that are
imposed by the need for surface, subsurface and hydrologic characterization, including the delineation
of flood and landslide
zones for risk assessments. Short- and long-term sea-level change and the impact of oceanatmosphere processes on the coastal
land environment, through flooding, erosion and storm surge for example, define further requirements
for hazard monitoring
and mitigation planning. The continued development and application of a broad spectrum of satellite
remote sensing systems
and attendant data management infrastructure will contribute needed baseline and time series data,
as part of an integrated
global observation strategy that includes airborne and in situ measurements of the solid Earth. Multihazard modeling
0924-2716/$ - see front matter D 2005 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS).
Published by Elsevier B.V.
All rights reserved.
doi:10.1016/j.isprsjprs.2005.02.002
T Corresponding author. Tel.: +1 818 354 1835.
E-mail address: david.m.tralli@jpl.nasa.gov (D.M. Tralli).
ISPRS Journal of Photogrammetry & Remote Sensing 59 (2005) 185 198
www.elsevier.com/locate/isprsjprs
capabilities, in turn, will result in more accurate forecasting and visualizations for improving the
decision support tools and
systems used by the international disaster management community.
D 2005 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS). Published by
Elsevier B.V. All rights
reserved.
Keywords: Satellite remote sensing; Imaging spectroscopy; Interferometric synthetic aperture radar (InSAR); Global
positioning system (GPS);
risk mitigation; Disaster response; Decision support system (DSS); Natural hazards
1. Introduction
The last two decades have witnessed the increasing
use of remote sensing for understanding the geophysical
phenomena underlying natural hazards. The
scientific knowledge gained along with the ability to
disseminate timely geospatial information that can be
integrated with demographic and socioeconomic data
are contributing to comprehensive risk mitigation
planning and improved disaster response. As the
discipline of Earth science began to recognize the
interactions between the hydrosphere, atmosphere,
biosphere and solid Earth as a complex system, satellite
remote sensing uniquely has provided the synoptic
perspective for in situ measurements on local spatial
scales and variable temporal resolution. Observations
from Earth orbiting satellites are complementary to
local and regional airborne observations, and to traditional
in situ field measurements and ground-based
sensor networks in seismology, volcanology, geomorphology
and hydrology. The contributions of
satellite remote sensing to solid Earth science, ranging
from high-resolution topography (using e.g. Interferometric
SAR, Lidar and digital photogrammetry) and
geodesy to passive hyperspectral (such as ASTER,
MODIS and Hyperion) and active microwave imaging,
Abstract
The aim of this project was to determine the accuracy of using
simple digital image processing techniques to map riverine
water bodies with Landsat 5TM data. This paper quantifies
Introduction
Accurate information on the extent of water bodies is important
for flood prediction, monitoring, and relief (Smith, 1997;
Tholey et al., 1997; Baumann, 1999); production of wetland
inventories (Bennett, 1987; Johnston and Barson, 1993; Blackman
et al., 1995; Shaikh et al., 1998; Phim et al., 1999); and the
evaluation of water resources (Morse et al., 1990; Manavalan et
al., 1993). Often this information is difficult to produce using
traditional survey techniques because water bodies can be fast
moving as in floods, tides, and storm surges or may be inaccessible.
Remotely sensed data provide a means of delineating
water boundaries over a large area at a given point in time. To
capture fast moving hydrological features, the data need to be
either of a high temporal resolution or in a substantial archive
to cover a range of hydrological conditions. Landsat MSS and
TM provide high spatial resolution data at 16-day intervals over
a long archival history, exceeding 25 years in most locations.
The long archive period and repetitive capture make the data
useful for mapping water bodies at a regional scale over a range
of hydrological conditions.
Since Landsat data became available in 1972, they have
been used to map water extent. Smith (1997) cites several early
studies where Landsat MSS data, in particular Band 7, were
used to distinguish water bodies from surrounding dry soil or
vegetation. Comparison with aerial photography gave error
estimates of less than 5 percent.
Bennett (1987) used density slicing of Landsat MSS band 7
to map water bodies to the west of Griffith, New South Wales,
Australia (NsW). He compared the area of MSS-derived water
bodies with that derived from digitized aerial photography and
found that the MSS data underestimated the area of water by
around 40 percent.
ment is needed.
This project set out to determine the classification accuracy
of density slicing and multispectral image classification
of Landsat 5 TM data to map the water bodies of the Murrumbidgee
River and floodplain near Wagga Wagga, Australia (Figure
1). Landsat TMdata were classified using single-band
density slicing and multispectral maximum-likelihood classification.
These classified data were then compared to high resolution
aerial photography to determine how well the Landsat
data could be used to detect and delineate water bodies.
Specifically this project set out to
Quantify the accuracy of simple digital techniques for the detection
and delineation of riverine floodplain water bodies using
Landsat TMdata,
Recommend a suitable technique for mapping riverine flood@)
Y1.E
PI'%
(C)
over 400 mg/l (Olive and Olley, 1997). On the day of data capture
the river at Wagga Wagga had a flow of 24,000 Mlld (Pinneena,
1998) with a turbidity of 90 mgll (Southern Riverina
County Council, water quality archive, 26 October 1990).
Methods
In order to assess the accuracy of using Landsat TM data to accurately
locate and delineate water bodies, classified Landsat TM
data were compared with manually mapped aerial photography
of the same area. Luckily for this study, both the Landsat TM
data and the aerial photography were captured on the same date
(26 October 1990), thus ensuring identical hydrological
conditions.
Materials
The color aerial photography was acquired by the NSW Central
TABLE1.CLASSESA ND DESCRIPTIONSO F DIGITIZEDW ATERB ODIES.
the original data. This procedure resulted in the creation of an selected as close as possible to the margins of the
water bodies
image of 510 pixel by 187 lines with bands 1,2,3,4,5, and 7 (Fig- on each photograph in order to maximize the
likelihood of a
ure 1). ERMapper software Version 5.7 was used for all image good match to the TMdata.
processing, polygon digitizing, and polygon manipulation. The first attempt at registration gave an average rootmeansquare
(RMS)error of 10 pixels (21 m). The images were then
Aedal photograph lnterpretatlon warped using a linear geometric transformation, nearest-neighThe aerial photographs were scanned at 300 DPI using a Micro- borhood intensity interpolation, and a pixel
resampling size of
3 m. Examination of the fit between the photographs and the TM
data indicated that there were significant registration errors.
Each photograph was then re-registered using an additional
seven GCPs on the margins of the water bodies, achieving an
average RMSerror of 23 pixels (9 m). This time overlaying the
photograph onto the TMimage indicated a good fit.
All water bodies greater than 625 m2 (lmm by lmm) in size
were then highlighted directly on the hardcopy prints of photographs
143,145, and 147 using magnified stereoscopic examination,
On the photographs, water bodies appear as uniform
regions of dark green to bright khakibrown in the low lying
channels and depressions. Each water body was then digitized
directly onto the scanned, registered aerial photograph using
the ERMapper regions tool. The polygons were then labeled as
River, Lagoon, Small Pool, or Dam (Table 1, Figure 2). The polygons
were then converted into a five-class raster image for comparison
with the Landsat TM classification results.
Landsat Image CIassHkation
1-
~ ~ z ~ q ~ ~ g ~ ~ ~ ~ s~~s$>~g2>~- ~- -$
s~:a:~~
t-4
Band 1
Band 2 Band 7
Band 3
Key to densty sl~c~ng
Water Not water
Ground truth Image
Band 4
I
I
Key to ground truth Image
R~ver Dam Not Water
Lagoon Small
9 00 5 ; K i
Q5"m
.I 0 h
0 255 0 255
Band 4
g5;m
95q-J
.I P
rn
0 255 0255
Band 6
5 0 255
9yrl
8 0 255
-0
8K
Range ofwater training pixels
26.496.74.8
38.096.54.9
493.677.142.6
596.981.764.5
794.474.445.4
I!
Classification Data
Percentage
Band 5Error Matrix Water Not Water Total Correct
Reference Small Pool 21222844048.2%
Dam 1642919385.0%
River 2626419304586.2%
Lagoon 77717395081.8%
Not Water 2079886639074297.7%
Total 58588951295370Percentage Correct 64.5%99.1%- 96.9%
TABLE6 . SUPERVMISAEXIDM UML IKELIHOOCD LASSIFICATIEORNR ORM ATRIX.
Classification Data
Maximum-Likelihood Percentage
Classification Error Matrix Water Not Water Total Correct
Reference Small Pool
Dam
River
Lagoon
Not Water
Total
Percentage Correct
lrr~gatedP asture
II
mm
r
Small unclassi
polygons
Conclusion
From this study it is clear that Landsat TMdata can be used to
map water bodies associated with the Murrumbidgee River
and its floodplain in the Wagga Wagga region. A multispectral
maximum-likelihood classification was able to produce an
overall classification accuracy of 97.4percent. This classification
was able to locate all of the major water bodies but underestimated
the number of water pixels present on the image,
achieving a producer's accuracy of only 59.6percent.
Importantly, a simple density slice classification of the
mid-infrared band 5produced a classification accuracy (96.9
percent) similar to that of the maximum-likelihood classification.
Acknowledgments
The authors thank Rodney Rumbachs for preparation of the
aerial photography and the production of the location map and
John Louis, David Roshier, David Lamb, and the anonymous
reviewers from PE&RS for their comments on the manuscript.
References
Baumann, P., 1999. Flood Analysis, http://www.research.umbc.edu/
tbenja~/baumann/mod2.html1,0 February 1999.
Bennett, M.W.A., 1987. Rapid monitoring of wetland water status using
density slicing, Proceedings of the 4th Australasian Remote Sensing
Conference, [14-18 September], Adelaide, pp. 682-691.
Blackman, J.G., S.J. Gardiner, and M.G. Morgan, 1995. Framework for
biogeographic inventory, assessment, planning and management
of wetland systems: the Queensland approach, Workshop Proceedings,
Wetland Research in the WetlDry Tbopics of Australia, (C.M.
Finlayson, editor), Jabiru, NT,Australia, Supervising Scientist
Report No. 101, Department of Environment and Heritage, Canberra,
pp. 114-122.
Brady, A., M. Shaikh, A. King, and P. Sharma, 1999. Remote sensing
and the Great Cumbung Swamp, Wetlands Australia, 7:596-606.
Jensen, J.R., 1996. Introductory Digital Image Processing: A Remote
Sensing Perspective. Prentice Hall, Upper Saddle River, New Jersey,
316 p.
Johnston, R., and M. Barson, 1993. Remote sensing of Australian wetlands:
An evaluation of Landsat TM data for inventory and classification,
Australian Journal of Marine and Freshwater Research,
44235-252.
Kingsford, R.T., R.F. Thomas, P.S. Wong, and E. Knowles, 1997. GIs
prs. org