Answers are grounded in this paper and cite their sources.
Landslides triggered by intense monsoon rainfall and active tectonics pose a persistent threat to transportation corridors and settlements in the Mandi district of the Northwestern Himalaya. Most existing susceptibility assessments for this region rely on historical inventories that do not include currently moving. This study tests whether refining a historical landslide inventory with Interferometric Synthetic Aperture Radar (InSAR) improves machine-learning susceptibility mapping in Mandi district. Sentinel-1A data from January 2023 to December 2024 (59 ascending and 43 descending acquisitions) were processed with Persistent Scatterer (StaMPS) and Small Baseline Subset (MintPy) interferometry. Active deformation areas were used to refine the Geological Survey of India’s Bhukosh inventory of 968 landslides, which consists mainly of slides, by adding 1,685 InSAR-derived landslide points, including 36 field-verified new landslides. Twelve uncorrelated topographic, hydrological, geological, climatic, and anthropogenic conditioning factors were selected using Pearson correlation and variance inflation factor analysis. Four machine learning algorithms, Extreme Gradient Boosting (XGB), Random Forest, Logistic Regression, and Multilayer Perceptron, were trained and compared using a nested spatial block cross-validation framework across block sizes of 1,000–10,000 m. XGB achieved the best performance at an optimal block size of 2000 m, and the InSAR-based XGB model generally showed higher ROC-AUC and PR-AUC, along with lower ECE and Brier scores, indicating improved discrimination, calibration, and probabilistic accuracy across spatial validation scales (at 2 km: ROC-AUC 0.887 vs 0.878; expected calibration error 0.067 vs 0.097). SHAP and Mean Decrease Accuracy (MDA) plots identified distance to roads and slope as the dominant predisposing factors. . The map was categorized into five susceptibility zones using a curvature-based break-point detection approach, with a success rate of AUC of 0.909, indicating very good discrimination of areas with different susceptibility levels. The map represents susceptibility to slide-type landslides. Areas outside the AOA accounted for only 1.1% of the study region, indicating that the model can be reliably applied across most of the study area. These results demonstrate that coupling InSAR-derived deformation with machine learning substantially improves the completeness and calibration of the model of landslide susceptibility mapping, offering a scalable approach for hazard mitigation and land-use planning in tectonically active Himalayan terrain.
No takes yet. Share an insight, caveat, or question.
Manocha et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: