Using MODIS/Terra and Landsat imageries to improve surface water quantification in Sylhet, Bangladesh

Abstract

Bangladesh has experienced multiple freshwater issues including salinization from monsoonal floods and groundwater over-pumping that induces severe land subsidence. Therefore, using satellite observations to virtually build a monitoring network becomes an efficient and innovative means. We focus on the Sylhet Mymensingh haor area that has the highest annual precipitation and the largest inundation area in northeastern Bangladesh. The modified normalized difference water index is first used to extract water area from MODIS and Landsat-5/-7/-8 optical imageries. A weekly flood chance model is then created from a sequence of images to recover water extent from the cloud-covered images. Using MODIS images for water identification achieves an overall accuracy of 84% in rainy season and 41% in dry season as validated with Sentinel-1A radar images. This model can be further used to refine the Shuttle Radar Topography Mission digital elevation model (DEM). As compared with ICESat laser altimetry, the root-mean-square of the height difference is improved from 1.65 m to 1.16 m after DEM modification. By combining the recovered water area and the refined DEM, surface water volume (WV) is quantified. A comparison with the Gravity Recovery And Climate Experiment (GRACE) gravimetry retrieved equivalent water heights (EWHs) in 2002 - 2015 is conducted, where the correlation coefficient and root-mean-square of the EWH difference are 91.7% and 0.09 m, respectively.