Glacial Lake Outburst Floods (GLOFs) in high mountain Asia (HMA) contains large-scale hazards which threaten downstream populations, infrastructure and ecosystems. Current automated self-supervised glacial lake mapping methods mainly rely on single-sensor data or visual interpretation, which result in boundary inaccuracies and the dif iculty of hazard dif erentiation. We present the Multi-Sensor Multi-Sensor Geo-Spatial Deep Learning Model (MS-GDLM) (MS-GFM): a joint encoder-decoder model which learns from Sentinel-1 synthetic aperture radar (SAR) and optical Sentinel-2/Landsat imagery, Copernicus DEM terrain variables, and glacier inventory data. The model utilizes a multi-task head for performing pixel-wise lake facet segmentation and GLOF susceptibility prediction in parallel. The observations described include a well characterized synthetic dataset of 320 glacial lakes across eight HMA subregions—Karakoram, Hindu Kush, Western and Eastern Himalayas, Tibetan Plateau, Pamir, Tian Shan and Hengduan—over years 2008-2017 with an associated set of 35 spectral, terrain, morphometric and temporal features. Experimental evaluation shows a mean Intersection-over-Union (IoU) of 0.82, Dice/F1 score of 0.90, general classification accuracy of 66.25% and an ROC-AUC of 0.90 in predicting susceptibility detection. Random Forest analysis with a baseline detects that glacier retreat rate, dam slope, and lake area are among the most influential predictors of GLOF. The proposed MS-GFM is able to perform well with good generalization across regions, and outperforms all single-sensor benchmarks by a considerable margin.