Collapses represent a major geological threat in China's mountainous regions, which cover 65% of its land area. Accurate susceptibility assessment is crucial, yet human engineering impacts in peri-urban mountains remains poorly quantified. This study develops an interpretable Random Forest model for collapse susceptibility in Changping District, Beijing. A spatial database included ten factors, including topography, geology, environment, and human activity. The model identified DTRO as the most critical factor, contributing 29.54% of total importance. This underscores the dominant role of human engineering activities. NDVI (20.38%), DTR (10.44%), and ELV (8.65%) were also key contributors. The top four factors contributed to 69.01% of total predictive importance. The optimized model achieved outstanding performance, with an AUC of 0.981 and accuracy of 0.9315. Spatial validation confirmed reliability, as 98.01% of collapses fall within the ʻhigh’ and ʻvery high’ susceptibility zones, covering just 17.3% of the study area. The susceptibility map reveals a distinct linear pattern of high-risk zones along roads and river valleys. This demonstrates that human factors can surpass natural factors in controlling collapse distribution in developed mountainous regions. This model provides an actionable tool for risk management and transferable framework for similar regions globally.
Yang et al. (Wed,) studied this question.