Landslide Susceptibility Mapping Using RF ‐Reconstructed SBAS ‐ InSAR Deformation and Ensemble Learning in Wanzhou District, Chongqing, China

Accurate landslide susceptibility assessment in mountainous regions is often constrained by the reliance on static environmental factors and by spatial discontinuities in SBAS‐InSAR‐derived deformation caused by decorrelation. To address this widely encountered problem in vegetated and topographically complex landslide‐prone regions, this study proposes a potentially transferable workflow for static‐deformation‐informed landslide susceptibility assessment that integrates SBAS‐InSAR deformation reconstructed using a standard random forest (RF) regression model with ensemble machine learning. Wanzhou District, Chongqing, China, located within the Three Gorges Reservoir Area, was selected as a representative mountainous study area influenced by reservoir‐related hydrological processes. Mean line‐of‐sight deformation rates were retrieved from ascending Sentinel‐1A SAR images acquired between January 2020 and December 2024. Fifteen static conditioning factors and a historical landslide inventory were compiled, and RF regression was used to reconstruct deformation in low‐coherence areas. The reconstructed deformation field was then incorporated as a multi‐year deformation‐based conditioning factor into landslide susceptibility modelling. Four models, including RF, XGBoost, LightGBM, and Stacking, were developed under static‐only and static‐deformation‐informed scenarios. The results show that incorporating InSAR‐derived deformation generally improved model performance, with the InSAR–Stacking model achieving the highest AUC of 0.9082. Frequency ratio analysis indicated that historical landslides were more strongly concentrated in high‐ and very‐high‐susceptibility classes under the static‐deformation‐informed framework. SHAP analysis further confirmed that deformation provides complementary information beyond conventional static factors. Overall, the proposed framework improves susceptibility mapping accuracy and can be adapted as a methodological reference for dynamic landslide hazard identification and risk management in mountainous and reservoir‐bank environments.