253
Multispectral Satellite Data for Flood Monitoring and Inundation Mapping
technique algorithm (ISODATA) to retrieve the flooding pixel, which was derived
from the work of Jensen (2005) and Campbell (2007), and also cross-validated by
Brakenridge (2006) and Brakenridge et al. (2007) in their global flood monitoring
system using MODIS and ASTER.
This study presents an all-inclusive methodology to calibrate a hydrologic model,
simulate the spatial extent of flooding, and evaluate the probability of detecting
inundated areas based entirely on satellite remote sensing data. These data include
topography, land use along with land cover, precipitation, and flood inundation
extent. MODIS- and ASTER-based flood inundation maps with raster format were
derived to benchmark the distributed hydrologic model, leading to the smooth simulations and the spatial extent of flooding and associated hazards. The objective
of this research work is to combine remotely sensed multispectral estimates that
include optical and microwave data sets within a hydrologic modeling framework
to characterize the spatial extent of flooding over scarcely gauged basins. Such an
effort will potentially improve flood predictions and flood management in ungauged
catchments.
11.2 METHODOLOGY
The methodology contains three major steps. First, the data from MODIS and
ASTER sensors were archived and processed to derive flood inundation maps for the
selected events. Second, a grid-based distributed hydrologic model was implemented
and further calibrated using the satellite-derived flood inundation maps in the study
area. Finally, the performance of hydrologic prediction in the selected river basin is
evaluated by comparing the simulated flood inundation extents with those derived
from MODIS and ASTER imageries. Out of seven news-reported flood events in
the study basin, we carefully selected three events with high-quality remote sensing
imagery. In addition, the flood prediction model used in this study was well calibrated in this basin by historical data (Khan et al. 2009).
11.2.1 Satellite-BaSed flood inundation MaPPing
There are several methods of identifying flooded versus nonflooded areas using
optical remote sensing imagery (Jensen 2005; Jensen et al. 1986). The first step is
to identify spectral classes within the imagery. One of the widely used clustering
algorithms applied for this study is ISODATA, which uses the Euclidean distance
in the feature space to assign every pixel to a cluster through a number of iterations (Jensen 2005). Spectral classes identified by unsupervised classifications are
the natural, inherent groupings of spectral values within a scene of remote sensing data (Campbell 2007). ISODATA begins with either arbitrary cluster means or
means of an existing signature set, and each time the clustering repeats, the means
of these clusters are shifted. The new cluster means are used for the next iteration. To
perform ISODATA, the analyst selects the number of spectral classes, a convergence
threshold, and the number of iterations for the algorithm that introduces considerable
subjectivity into the classification process (Lang et al. 2008). This process of flood
region classification was performed using the ENVI software. The method for flood
Multispectral Satellite Data for Flood Monitoring and Inundation Mapping
technique algorithm (ISODATA) to retrieve the flooding pixel, which was derived
from the work of Jensen (2005) and Campbell (2007), and also cross-validated by
Brakenridge (2006) and Brakenridge et al. (2007) in their global flood monitoring
system using MODIS and ASTER.
This study presents an all-inclusive methodology to calibrate a hydrologic model,
simulate the spatial extent of flooding, and evaluate the probability of detecting
inundated areas based entirely on satellite remote sensing data. These data include
topography, land use along with land cover, precipitation, and flood inundation
extent. MODIS- and ASTER-based flood inundation maps with raster format were
derived to benchmark the distributed hydrologic model, leading to the smooth simulations and the spatial extent of flooding and associated hazards. The objective
of this research work is to combine remotely sensed multispectral estimates that
include optical and microwave data sets within a hydrologic modeling framework
to characterize the spatial extent of flooding over scarcely gauged basins. Such an
effort will potentially improve flood predictions and flood management in ungauged
catchments.
11.2 METHODOLOGY
The methodology contains three major steps. First, the data from MODIS and
ASTER sensors were archived and processed to derive flood inundation maps for the
selected events. Second, a grid-based distributed hydrologic model was implemented
and further calibrated using the satellite-derived flood inundation maps in the study
area. Finally, the performance of hydrologic prediction in the selected river basin is
evaluated by comparing the simulated flood inundation extents with those derived
from MODIS and ASTER imageries. Out of seven news-reported flood events in
the study basin, we carefully selected three events with high-quality remote sensing
imagery. In addition, the flood prediction model used in this study was well calibrated in this basin by historical data (Khan et al. 2009).
11.2.1 Satellite-BaSed flood inundation MaPPing
There are several methods of identifying flooded versus nonflooded areas using
optical remote sensing imagery (Jensen 2005; Jensen et al. 1986). The first step is
to identify spectral classes within the imagery. One of the widely used clustering
algorithms applied for this study is ISODATA, which uses the Euclidean distance
in the feature space to assign every pixel to a cluster through a number of iterations (Jensen 2005). Spectral classes identified by unsupervised classifications are
the natural, inherent groupings of spectral values within a scene of remote sensing data (Campbell 2007). ISODATA begins with either arbitrary cluster means or
means of an existing signature set, and each time the clustering repeats, the means
of these clusters are shifted. The new cluster means are used for the next iteration. To
perform ISODATA, the analyst selects the number of spectral classes, a convergence
threshold, and the number of iterations for the algorithm that introduces considerable
subjectivity into the classification process (Lang et al. 2008). This process of flood
region classification was performed using the ENVI software. The method for flood