Compound multi-hazard assessment under CMIP6 climate scenarios: Tracking seasonal flood-landslide and drought-fire interactions across Nepal

This study advances multi-hazard assessment for monsoon-dominated mountain systems by moving beyond static overlays of independently modeled hazards and instead diagnosing how seasonal hazard regimes reorganize under climate change. Climate-induced hazards in mountain systems rarely occur in isolation, instead they interact through cascading and compounding pathways that single-hazard assessments fail to capture. In Nepal, extreme monsoon rainfall can trigger landslides that dam rivers and generate secondary flooding, while prolonged dry spells create drought, desiccate vegetation fuels, and elevate fire risk.

The authors show that the “best” machine-learning approach is not universal across hazards. This cross-hazard model selection is not a technical detail but is a foundation for credible compound-hazard mapping, because uncertainty and bias in any single constituent map propagate directly into compound-zone identification. Coupling these optimized models with SHAP interpretation further anchors the outputs in physically meaningful drivers, revealing consistent hydroclimatic controls (rainfall, humidity, soil moisture) and pervasive anthropogenic modifiers (population density, settlement/road proximity) that shape susceptibility across all four hazards.