Probabilistic Analysis of the Landslide Hazard in Cold Regions: Considering Multiple Triggering Factors and Their Interdependence

This study reports a quantitative pattern of landslide hazard in cold regions, triggered by multiple climatic factors. Different machine learning models were first employed to analyze the stability of the disaster-pregnant environment of landslides. Subsequently, the temperature difference and precipitation weakening the geotechnical body were regarded as the triggering factors of landslides in cold regions, and their marginal distribution and interdependence were determined based on the copula theory to develop the joint distribution from the perspective of exceedance probability. Landslide hazard under different return periods (T = 5, 10, 20, and 50 years) was then identified in terms of the product of the permanent environment and the triggering factor. Results show that decision tree (DT)-boosting maintains high prediction for the disaster-pregnant environment, with AUC of 0.978 and 0.859, respectively, in the training and test phases, and the high susceptibility is concentrated in the central and southeastern parts of the study area. Further, the Gumbel copula keeps the smallest distance to the empirical copula, better portraying the interdependence of triggering factors. Specifically, the probability of landslide hazard decreases as the return period increases, while the intensity increases. This study provides a window to accurately capture the landslide hazard caused by fluctuations in multiple climate factors in cold regions.