Landslide susceptibility assessment along China’s railways assisted by machine learning

Landslides pose significant threats to the safety and reliability of railway systems, yet existing studies lack a unified nationwide framework and rarely consider future climate-driven changes. Using a 1 km buffer along China’s railway network, this study develops a national-scale susceptibility assessment framework integrating topographic, geological, and meteorological factors. Based on historical landslide and non-landslide samples, five machine learning models were evaluated, with backpropagation neural network (BPNN), random forest (RF), and extreme gradient boosting (XGBoost) showing superior performance (AUC = 0.8877–0.9034). A stacking ensemble further improved prediction accuracy (AUC = 0.9135). Under climate change, three shared socioeconomic pathways (SSP1-2.6, SSP3-7.0, SSP5-8.5) were incorporated, with precipitation as the key driver, to simulate future susceptibility for 2021–2060 and 2061–2100. Results indicate that high-susceptibility areas (probability > 0.4) expand from 23.5% historically to 27.1–28.8% across scenarios, with greater increases under higher emission scenarios. In addition, regions transitioning from low or moderate to high risk expand markedly, with the most pronounced increases observed in the Changbai Mountain region of Northeast China (up to 0.74). This framework provides a robust basis for understanding climate-driven risk evolution and supports targeted mitigation and early warning strategies for railway landslides.