Use Of Machine Learning On SAR Data To Classify Submerged And Flooded Regions

In the past few years very much, attention has been given to flood monitoring. Along with flood detection, it is also essential to classify the fully merged, submerged, and without merged regions, which are helpful in various decision-making and environmental studies, but it is onerous to classify due to the mixed scattering response from these classes. However, there is some difference in the scattering response, which is hard to classify using only ordinary techniques. Machine learning (ML) is apt to deal with such mixed-class problems and classify them with acceptable accuracy. Therefore, in this paper, ML (Xgboost and Radom Forest) models are applied to the extracted optimum features from the Sentinel-1 synthetic aperture radar (SAR) data to classify the submerged and flooded regions. By combining ML algorithms with time series SAR data, flood monitoring systems can provide accurate and timely information about flooded and submerged areas, enhancing flood management strategies and reducing the impact on vulnerable communities.