Predicting catastrophic geohazard cascades: Integrating machine learning and field evidence for shal-low landslide clusters in North China

Shallow landslides are among the slope failures most sensitive to rainfall and account for a large proportion of landslide occurrences. From July 29 to August 2, 2023, extreme rainfall in North China triggered numerous shallow landslides in the western mountains of Beijing, many of which evolved into catastrophic geohazard chains. In Zhaitang Town, Men-tougou District, we integrated field surveys and remote-sensing interpretation with linear and nonlinear ma-chine-learning (ML) models for slope-scale susceptibility assessment. Shapley Additive Explanations (SHAP) quanti-fied the dominant controls, and a concise framework was developed for geohazard-chain assessment. High-susceptibility zones are concentrated in northeastern Zhaitang. Lithology, elevation, landcover, and normalized difference vegetation index (NDVI) are the dominant controls, whereas event-cumulative rainfall shows a positive but nonlinear relationship with susceptibility at the township scale. The nonlinear models outperformed the linear models (AUC > 0.93), and their high-susceptibility zones generally agreed with observed landslide distributions and field evi-dence. Where favorable topography, abundant loose material, and concentrated surface runoff coexist, clustered shallow landslides can converge into channelized flows and evolve into geohazard chains. Integrating landslide clustering, chan-nel morphology, material-supply proxies, and chain-evolution indicators enables catchment-scale susceptibility screen-ing and identification of conditions conducive to cascading geohazards. These results support hazard assessment and risk-reduction planning in the bedrock mountains of monsoonal northern China.