emergency relief planning and flood monitoring (Rao, 1994).
An IRS-1A imagery taken beforehand shows the river delta and
the adjacent areas. The cyclone that caused the flooding was
traced by NOAA AVHRR images. Thanks to the effective
monitoring and tracking of the cyclone, evacuation could take
place in time and damages were reduced. Eleven days after the
cyclone moved over the area, a Landsat image was acquired and
processed, fortunately without significant cloud cover over the
delta area. An analysis had been performed and areas with
different level of inundation were classified. Blue color showed
still standing water, different shades of brown indicated the
extent of flooding. When several high-resolution optical images
are available, a detailed analysis of the moving waterfront could
be performed. :
Availability of other information. For successful and reliable
analysis of flood monitoring, different information other than
satellite images are needed. Topographic maps play an
important role, but digital elevation models (DEM) and field
observations are often needed for simulation and validation
processes. In an analysis during the catastrophic flood event in
November 1994 in Northern Italy, four ERS-1 precision images,
a detailed topographic map (1:10,000) with the results of in situ
measurements of the flooding, and a DEM (230x220 m?
resolution) was used (Giacomelli et al. 1995, Boni et al. 1996).
The radar images were georeferenced, the topographic map was
scanned with a 4x4-m2 resolution and the main features were
digitized. The detection of water bodies was only possible
where the height of the water completely submerged the surface
roughness. After a first estimation of inundated areas a
threshold of 50 connected pixels had been taken as a base layer,
and the assumption was made that the water occupying the areas
highlighted by the SAR images may flow on the natural
drainage network, a process which can be simulated easily in
many GIS packages. The classification according to the base
layer was very poor, only 11 46 of the actually inundated areas
were classified as water, but with the utilization of the digital
elevation model (DEM) and the simulation of the drainage
network the results showed that 62 % of the actually inundated
areas had been classified as water.
Cloud cover. When cloud cover limits the use of optical
sensors, radar imagery can be still utilized for flood monitoring.
During the flood on the Nederrijn, Waal and Maas rivers in
Holland in January-February 1995, radar images were used
together with Landsat imagery (Wang et al, 1995). From the
Landsat image (16 July 1987), a false-color composite image
was created using the channels 4, 3 and 2 in red, green and blue.
This shows the pre-flood river channel in the dry season. The
two radar images taken during the flood (ERS-1, 30 January
1995) and before the flood (ERS-1, 21 September 1994) were
resampled with a 30-meter resolution in the same coordinate
system as the Landsat image. Inverting the radar flood image
and merging it in the blue channel with the Landsat TM channel
1 and 4 in red and green, the composite image shows the
flooded areas with light blue color, the original watercourse in
dark blue, vegetated areas in different shades of green and urban
areas in red. This analysis allows a flood damage assessment of
the agricultural zones along the flooded rivers. A second
composite image was created with the flood and pre-flood radar
images. Both the pre-flood and flood images were contrast
enhanced, inverted, and presented in green and blue channels,
respectively. For the red channel the normal pre-flood image
was taken, thus resulting in a color combination, which provides
the same information as the previous combination, only the pre-
flood water bodies are far less visible. The flooding occurred in
a part of Holland where forests and other high vegetation are
almost completely absent; thus, not causing significant errors in
the estimation of flooded areas. In addition, during the winter
flood the agricultural crops were not yet in their growing period,
so eliminating the problem of detecting water bodies under
dense vegetation.
Costs of satellite imagery. A comparison of satellite radar
imagery and aerial photography carried out by Biggin and Blyth
(1996) during the flood event on the upper Thames river in
England in November-December 1992 showed that satellite
remote sensing and aerial photography approaches to flood
monitoring have similar costs in regional scale, and the
assessment of flooded areas can be derived, in some cases, with
similar accuracy. Three ERS-1 SAR images (pre-flood: 30
October 1992; flood: 4 December 1992; after flood: 8 January
1993), a Landsat imagery (20 July 1990) and a series of aerial
photography were used in the analysis. Interpreting the three
radar images (pre-flood, flood, after flood) in the red, green and
blue channels the inundated areas appeared in magenta and the
non-flooded areas appeared mainly in different shades of gray
and brown. The Landsat imagery helped individual fields to be
more easily identified than in the SAR images. Compared to the
aerial photographs it was pointed out that the satellite images
had almost the same information content as the aerial
photographs concerning the extent and the location of the
inundated areas. A cost analysis showed furthermore that the
cost of an ERS-1 image covering about 570 km of stream
channels was similar to the cost of hiring the light aircraft used
in collecting the aerial photographs when about 180 km of river
were surveyed in varying detail, thus considering a cost-
efficient way of flood monitoring does not always mean the
necessary use of satellite imagery. However, in regional or
global areas, or in places where aerial photography and in situ
measurements are not available, satellite. remote sensing
remains the ultimate choice for flood monitoring.
Vegetation cover. If dense vegetation covers the flooded areas
and the height of the water does not submerge the vegetation, no
uniform approach for detection of water bodies with current
technologies exists. Using microwave sensors, it is theoretically
feasible for a limited extent to penetrate vegetation and detect
water bodies, depending on the wavelength of the sensor.
Current satellites carrying active microwave sensors have single
wavelengths and single view angles, with the exception of
RADARSAT for variable view angle. Experiments show that if
multiband microwave sensors with variable view angle exist,
the possibility of detecting water bodies under dense vegetation
is increased. The shuttle imaging radar (SIR) missions onboard
the space shuttle provided SAR observations with multiband,
multiple view angle sensors. With the analysis of SIR-B images
in tropical areas over Bangladesh, flood boundary delineation
through cloud and dense vegetation could be performed (Imhoff
and McCandless, 1988). The L-band SAR sensor had a HH
polarization and variable incidence angle from 25.8? to 57.8*. A
Landsat MSS scene, acquired by chance two weeks before the
SIR-B experiment in October 1984, was used for temporal
reference to observe the flood boundaries. River gauge data
were used to help verify river flood stage for the time during
which both radar and Landsat MSS data were available. Beside
the main goal of the study (monitoring disease vectors in the
tropical regions) it was possible to delineate pools of standing
water beneath a 12.5 m tall, 100% closed tree canopy in the
mangrove forests of southern Bangladesh, operating the SAR
sensor at the full range of incident angles.
782 International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998
Fk
rel
sel
fut
mc
