Landslide Susceptibility Mapping in the Garhwal Himalaya Integrating Seismic Shaking and Active‐Fire Predictors With an XGBoost Model

ABSTRACT

The Garhwal Himalaya of Uttarakhand, India, is a tectonically active, deeply dissected orogen in which landsliding is conditioned by a heterogeneous mix of topographic, geological, hydro‐climatic, seismic and anthropogenic controls. Reliable, high‐resolution susceptibility information is essential for disaster risk reduction, infrastructure sitting and climate‐adaptive land‐use planning across this densely populated mountain region. We present a data‐driven landslide susceptibility model for the region in which seismic shaking is incorporated as a continuous triggering covariate, following the operational practice established for coseismic and multi‐hazard landslide assessment, rather than as an intermediate discretised surface. Twenty‐four conditioning factors, including USGS ShakeMap‐derived Peak Ground Velocity fields for the 1991 Uttarkashi and 1999 Chamoli earthquakes, were used to train a spatially blocked XGBoost classifier on 7087 inventory points from the Geological Survey of India. The model achieved AUC‐ROC = 0.97 on the held‐out test set, with spatially blocked cross‐validation confirming robust generalisation beyond the sampling frame. The predicted susceptibility surface identifies a ‘very high’ class covering only 6.65% of the study area yet capturing 46.27% of held‐out landslides, a substantial spatial concentration offering immediate operational value for prioritising mitigation resources. TreeSHAP interpretation identifies proximity to roads, elevation, lithology and slope as the leading predictors, with the seismic layer contributing a moderate but stable share consistent with a historic‐event framing of shaking. Independent corroboration comes from three recent events, the 7 February 2021 Chamoli (Tapovan Vishnugad) rock–ice avalanche, the January 2023 Joshimath subsidence–landslide cascade and the 5 August 2025 Dharali (Uttarkashi) debris flow, all of which fall within the ‘high’ and ‘very high’ susceptibility classes on the produced map. Beyond the Garhwal case study, the framework demonstrates that treating seismic shaking as a continuous covariate yields interpretable, transferable susceptibility maps for regional‐scale hazard planning across tectonically active mountainous terrain, directly supporting evidence‐based prioritisation of engineering interventions, early‐warning system placement and land‐use zoning in the increasingly climate‐stressed Himalayan Orogen.