Slope unit-based evaluation of climate change impacts on landslide susceptibility in the Nepal Himalaya

The limited integration of geomorphologically coherent terrain units and climate projections into landslide susceptibility modeling has hindered region-specific risk assessments in mountainous areas like Seti River Basin in Nepal. This study integrates slope unit (SU)–based terrain partitioning with a machine learning framework to evaluate both current and future landslide susceptibilities. A total of 36 models were developed by testing six machine learning algorithms across six covariate combinations derived from SU statistics. The Random Forest model consistently outperformed others, so it was selected for final mapping, achieving high predictive performance (AUC = 0.83). Future precipitation predictors were derived from an ensemble of Coupled Model Intercomparison Project Phase 6 (CMIP6) Global Climate Models for two Shared Socioeconomic Pathway scenarios (SSP2-4.5 and SSP5-8.5) across near-future (2025–2054) and far-future (2070–2099) windows. The projections reveal a spatial expansion in moderate and high susceptibility zones, along with a notable redistribution of landslide-prone areas. Contrary to expectations of intensified localized risk, however, a decline in the very high susceptibility class suggests a broader spatial diffusion of hazard exposure. This outcome underscores the crucial role of climate-informed, SU-based modeling for wide-area landslide risk reduction and provides a scalable methodology for other high-relief, data-scarce regions.