Automated mapping of glacial lakes in Himachal Pradesh using multi source remote sensing data and machine learning

The Himalayas are undergoing active climate-induced changes, resulting in glacial lake formation and expansion. Glacial lakes are important reservoirs of freshwater but also run the risk of glacial lake outburst floods (GLOFs). As sentinels of climate change, dynamic monitoring of glacial lakes is necessary. This research introduces an automated method for mapping glacial lakes in Himachal Pradesh based on multi-source remote sensing data and a random forest (RF) classifier. The model was tested under various scenarios using spectral bands and remote sensing indices extracted from Sentinel-2 and Planet images. The combination of Sentinel-1 SAR, Sentinel-2 MSI, and SRTM DEM data resulted in a classification accuracy of 93.69%, which increased to 94.44% with the addition of high-resolution Planet images. Although the method was effective in identifying glacial lakes, it faced difficulties in distinguishing glaciers from supraglacial lakes. Postprocessing methods were used to enhance the results. Model performance was evaluated using statistical measures, such as recall, precision, F1-score, and overall accuracy. The RF classifier performed well robustly, identifying its reliability in glacial lake mapping even being a machine learning method.