Identification, Analysis, and Prediction of Landslide Deformation Based on InSAR and an Explainable Neural Network Model: A Case Study in the Ili River Basin, Xinjiang, China

The Ili River basin is situated at the intersection of China and Central Asia. Due to the Tien Shan’s complex terrain and geological structures, frequent and widespread landslides occur in this region, accounting for nearly 60% of all geological hazards in Xinjiang Province. Although satellite-based interferometric monitoring (InSAR) is an effective approach for identifying potential landslides, challenges remain regarding the interpretability of observed deformation signals. In this study, wide-area InSAR processing was employed to detect the distribution of potential landslides. An explainable artificial intelligence (XAI) model—LSTM-SHAP—was then proposed to analyze deformation mechanisms and elucidate landslide types. Notably, the SHAP map provided a quantitative and detailed explanation of landslide attributions, revealing how controlling factors vary during deformation evolution. By training on historical deformation patterns, future scenarios can be generated for more accurate deformation prediction and landslide risk assessment. Our research is expected to provides a new technical reference for landslide monitoring. Moreover, these findings suggest that XAI-based methods can offer civil protection agencies a data-driven perspective for understanding deformation evolution and implementing precautionary measures.