Image analysis of Sentinel 1 for flood detection in Altai Region in April 2015 and Ryazan Region in April 2016

Abstract

The present study is dedicated to flooded area detection using radar images of Sentinel 1 for two Russian rivers - Aley river (Altai Krai) in April 2015, and Moksha river (Ryazan Region) in April 2016. Multi-temporal image flood mapping involves acquiring flood and non-flood images of the same area and combining them to get an image which indicates change by colors appearing in the image. The thresholds for flooded areas and flooding under vegetated areas are determined in accordance with the method of (Long et al., 2014): the first threshold allocates a region of open water without waterlogged vegetation (dark pixels), and the second one - a region with partially waterlogged vegetation (bright pixels). Another way to determine flooded area is to use a known water surface with known backscattering coefficient, and to determine difference of two images: the water surface, which is not in non-flood image, appears in bright tone in the difference image. Another possibility to map partially waterlogged vegetation is to use the effect of double bounce scattering. This approach due to the lack of the full polarimetry data is performed by means of Haralick's texture features 'contrast' and 'entropy'.