Optimal Threshold Selection for Water Bodies Mapping from Sentinel-L Images Based On Sentinel-2 Water Masks

All-weather satellite Synthetic Aperture Radar (SAR) images are widely used to identify surface water bodies when clouds and shadows cause detection errors from optical data. Algorithms for mapping water bodies are often based on determining a threshold value from a backscattering coefficient histogram. The threshold cannot always be found in the case if the “water” and “land” classes are imbalanced. An approach to backscattering coefficient threshold selection is proposed and substantiated in this paper. It is based on a maximum similarity between SAR water mask and the reference mask generated from an optical multispectral satellite image. The parameters of SAR data preprocessing are selected to ensure the best compliance between the optical and SAR water masks. It has been substantiated that the optimal threshold, determined on a reference date, provides high accuracy of water bodies detection on the validation date of observation.