Landslide Hazard Assessment Based on Ensemble Learning Model and Bayesian Probability Statistics: Inference from Shaanxi Province, China

The geological and environmental conditions of the northern Shaanxi Loess Plateau are highly fragile, with frequent landslides and collapse disasters triggered by rainfall and human engineering activities. This research addresses the limitations of current landslide hazard assessment models, considers Zhuanyaowan Town in northern Shaanxi Province as a case study, and proposes an integrated model combining the information value model (IVM) with ensemble learning models (RF, XGBoost, and LightGBM) employed to derive the spatial probability of landslide occurrences. Adopting Pearson’s type-III distribution with the Bayesian theorem, we calculated rainfall-induced landslide hazard probabilities across multiple temporal scales and established a comprehensive regional landslide hazard assessment framework. The results indicated that the IVM coupled with the extreme gradient boosting (XGBoost) model achieved the highest prediction performance. The rainfall-induced hazard probabilities for the study area under 5-, 10-, 20-, and 50-year rainfall return periods are 0.31081, 0.34146, 0.4, and 0.53846, respectively. The quantitative calculation of regional landslide hazards revealed the variation trends in hazard values across different areas of the study region under varying rainfall conditions. The high-hazard zones were primarily distributed in a belt-like pattern along the Xichuan River and major transportation routes, progressively expanding outward as the rainfall return periods increased. This study presents a novel and robust methodology for regional landslide hazard assessment, demonstrating significant improvements in both the computational efficiency and predictive accuracy. These findings provide critical insights into regional landslide risk mitigation strategies and contribute substantially to the establishment of sustainable development practices in geologically vulnerable regions.