Split-Window Based Flood Mapping with L-Band ALOS-2 SAR Images: A Case of Kerala Flood Event in 2018

The Kerala state of India experienced a devastating flood during Aug 2018, which incurred huge socio-economic losses and human fatalities. ALOS-2 L-band SAR image acquired during the peak flood was used in this study. A split-window approach combined with thresholding algorithm was used for analyzing the 2018 flood event of Kerala, India. The SAR image splitting and tile selection was carried out based on two parameters, namely the Coefficient of Variation (CV) and ratio to the scene. Kittler and Illingworth's thresholding algorithm was implemented on the selected split images. Euclidean distance was used to shortlist the split images with large variation in the data representing both thematic classes (flood/non-flood). An independent split based analysis (ISBA) was implemented in which respective threshold values obtained from the split images are averaged to get an optimum threshold value. From the results, we observe that an underestimation of flood area in urban land use due to double bounce, volume scattering and shadow effects. Validation is carried out on a small subset area for which the field data was available, and an accuracy of 73 % was obtained.