A Deep Learning Approach to Map Himalayan Glacial Lakes in Monsoon Using Synthetic Aperture Radar Datasets

The rapid melting of glaciers in response to the changing climate is leading to the rapid formation of new proglacial lakes and the expansion of existing ones. These lakes pose a significant threat in the form of Glacial Lake Outburst Floods. Thus, it is crucial to map and monitor these glacial lakes. The existing methods employ optical images to map glacial lakes which has limited utility during the monsoon. In this study, a convolutional neural network model has been used with an encoder decoder structure to perform the semantic segmentation of the glacial lakes using SAR imagery in tandem with topographic information. The results indicate a high training accuracy of 97.29% and a testing accuracy of 94.35%. The proposed novel deep learning approach has the potential to be operationalized as a vital tool in monitoring the glacial lake to study any devastating outburst floods during the monsoon period.