Integrating Time-Series InSAR Deformation and LightGBM for Landslide Susceptibility Assessment

Existing landslide susceptibility models typically rely on static predisposing factors, which effectively capture the relationship between landslides and geographic variables but neglect dynamic features like surface deformation. This study applies Time-Series Interferometric Synthetic Aperture Radar (TS-InSAR) to obtain line-of-sight deformation rates for Yunyang County, breaking these into vertical and slope-direction components as InSAR factors. These are combined with static predisposing factors to develop a LightGBM model for landslide susceptibility, and the Shapley Additive Explanations (SHAP) algorithm is used to identify key influencing factors. The results show that 28.15% of Yunyang is moderately susceptible, with high and very high susceptibility areas concentrated along the Yangtze River, in line with historical landslide distributions. SHAP analysis highlights elevation, land use, and proximity to rivers as primary factors. Incorporating InSAR data improves the model's AUC from 0.8195 to 0.8302, demonstrating its ability to enhance landslide susceptibility prediction. This study confirms the significant role of time-series InSAR deformation data in improving susceptibility assessments.