Abstract: Glacial Lake Outburst Floods (GLOFs) are an escalating hazard in the Himalayas, where climate-driven glacier retreat is rapidly expanding high-altitude lakes. These sudden floods can devastate downstream communities, infrastructure, and ecosystems. This study develops a scalable framework for forecasting GLOF susceptibility using remote sensing and machine learning. Multi-temporal satellite imagery and terrain data were used to map glacial lakes, monitor their expansion, and extract features such as area, elevation, expansion rate, and proximity to glaciers. To address sparse and incomplete records, features were derived directly from satellite imagery and merged with global glacial lake inventories, while missing data and class imbalance were handled with imputation and resampling strategies to ensure robust model performance. Validation showed that automated NDWI-derived lake areas correlated well with documented inventories (r = 0.67), and expansion statistics revealed a steady increase in lake growth across the region. Supervised machine learning models trained on historical GLOF events distinguished hazardous lakes from stable ones and generated probabilistic susceptibility maps highlighting areas of elevated risk. By integrating Earth observation with data-driven modeling, this research advances understanding of GLOF hazards and provides an automated framework for large-scale monitoring, early warning, and climate adaptation in vulnerable Himalayan landscapes.
Nima Jangbu Sherpa (Thu,) studied this question.