Abstract
Precise flood mapping is a prerequisite for efficient flood disaster management. Flood induced losses can be minimized by means of acquiring precise flood disaster information through mapping of flood prone areas. In this paper, radar remote sensing (RADARSAT) data is used to accurately delineate the distribution of open water inundation during the biggest flood of 1998 in Greater Dhaka City, Bangladesh. Seven flood time imageries and one dry season image were used to identify the precise and widest flooding area that is expected to be helpful for developing apposite flood countermeasures. The radar images were digitally processed to geometrically rectify the pixel geometry and were filtered to suppress speckle noise. The imageries were classified according to flooded (open water) and non-flooded categories using the Maximum Likelihood Decision Rule. The classification accuracy of the radar data were then evaluated using ground reference data. Spatial-temporal analysis of flooding revealed that flood water started to increase from early July 1998 (inundating 34.65%) and reached its peak on 25 August (inundating 51.29%). This was mainly attributed to heavy monsoonal downpour and early peaking of river water levels. The eastern part of the city suffered severe flood damage. In contrast, most of the western part of Greater Dhaka was not severely flooded due to the existing flood control works and relatively higher elevation. Classification accuracy of imageries show that all image in the dataset produced satisfactory results. The highest overall accuracy attained was 88.57% for 31 July image, followed by 7 July and 10 September images, respectively (82.86% and 81.90%). Kappa statistics for all the images were also in good agreement except the August images. The results showed that SAR data is of value for real-time flood monitoring for a fast growing city like Dhaka and can provide the necessary information for flood disaster information and management.
