Multi-scenario landslide probabilistic hazard analysis based on a single rainfall event: A case of the Zhuzhou-Guangzhou section of Beijing-Guangzhou railway in China

Abstract This study calculates the absolute probability of landslides under varying rainfall scenarios along the Beijing-Guangzhou Railway from Zhuzhou to Guangzhou, aiming to enhance railway transportation safety. Using a Bayesian sampling strategy, a Logistic Regression (LR) model was developed for landslide hazard assessment based on the geological conditions and rainfall data along the railway. The model demonstrated strong predictive performance with an AUC value of 0.86 for both training and testing sets, showing no overfitting. Results indicated that when rainfall is less than 150 mm, over 70% of the study area has an absolute landslide probability below 0.1%. However, with rainfall exceeding 150 mm, landslide hazards increase significantly, with a rapid rise in areas where the probability ranges from 0.1–1%. When rainfall reaches 500 mm, about 60% of the region exhibits a landslide probability exceeding 1%. Under real rainfall scenarios (e.g., cumulative rainfall during the 10 days before June 7, 2020), areas with probabilities greater than 1% are mainly concentrated in Fogang County, northeast of Guangzhou, and eastern Zhuzhou, aligning with heavy rainfall distributions. The relationship between rainfall and landslide occurrence is highly non-linear, with probabilities increasing exponentially as rainfall rises. These results provide an effective tool for landslide hazard assessment along the railway and offer valuable data support for disaster warning and prevention measures.