Evolution of landslide susceptibility in the Three Gorges Reservoir area over the three decades from 1991 to 2020

Despite more than 30 years of construction of the Three Gorges Project, comprehensive studies on the spatiotemporal evolution of landslides in the Three Gorges Reservoir area are still limited. This study aims to analyze changes in landslide susceptibility from 1991 to 2020 and identify the key factors driving these changes. We constructed 24 datasets based on 4860 landslide events, 13 static factors, and 7 dynamic factors, covering 12 time periods for analysis. We applied two machine learning models—Light Gradient Boosting Machine (LightGBM) and eXtreme Gradient Boosting (XGBoost)—along with the isolation Forest (iForest) algorithm. The iForest-LightGBM model achieved the highest accuracy and demonstrated efficient training performance. Temporal analysis showed that high-susceptibility areas expanded along the Yangtze River, peaking in 2020, with notable anomalies from 2001 to 2010, followed by stabilization between 2011 and 2020. Using the SHapley Additive exPlanation (SHAP) algorithm, we quantified the importance of the influencing factors over time. This study establishes a multi-temporal evaluation framework for landslide susceptibility and introduces a method for quantitatively analyzing the evolution of influencing factors. The findings provide valuable insights into landslide risk management in the Three Gorges Reservoir area and contribute to understanding the spatial evolution of landslides in dynamic environments.