EXTRACTION OF LAND COVER FEATURES IN A FLOOD PLAIN AREA USING MICROWAVE DATA

Abstract

This study focuses on the use of Microwave data for mapping the extent of flooding events. A small test area in the middle Brahmaputra River valley in the western part of Guwahati is studied. Two TerraSAR-X images pre-flood (10 March 2008) and post-flood (7 July 2008) have been used in the analysis for delineation of flooded areas. The potential of thresholding to delineate a group of pixels was used with intermediate processing of the images. Five different land cover classes were found to have a distinct backscatter pattern, and could therefore easily be identified in the images. The pre-flood TerraSAR-X image was classified into five different land cover classes while the post-flood TerraSAR-X image was classified into two land cover classes, with a assumption that other land cover classes would not be changed during the time gap (four months) between the two images. The classification result for the TerraSAR-X image before flood has been compared to the LISS-III (28 March 2008) image for validation purposes. LISS-111 is being used as a reference image in the study. The agreement between the TerraSAR-X and the LISS-III based mapping is high with an overall classification accuracy of 83.70%. Four change detection maps were generated for the purpose of evaluation of extent of damage for different land cover classes by the flood. The maps showed the percentage of the land cover classes been merged into the flood water.