Satellite Data Predicts Himalayan Glacial Lake Outbursts and Landslides

Researchers have investigated the potential of free satellite data to predict hazardous events like glacial lake outbursts, landslides, and ice floods in the Nepal Himalaya. The study utilized radar interferometry to detect moraine dam deformation and satellite weather data to identify periods of stress on primed lakes. While a previous feasibility study indicated that deformation points to "which" lake is destabilizing and weather to "when" it's at risk, this paper develops and evaluates predictive models for both aspects. The findings show that antecedent weather data is a strong indicator for timing triggers across all three hazard types, achieving high ROC scores. Terrain data, however, is less effective at ranking susceptibility when compared against similar nearby sites, performing only slightly better than chance for smaller floods. Interestingly, deep learning models did not significantly outperform simpler gradient-boosted baselines, suggesting that straightforward decision trees based on ruggedness and monsoon rainfall can be quite effective for lake hazards.