Scenario Prediction of the Rainstorm-Collapse-Landslide-Debris Flow Disaster Chain Based on Bayesian Networks

The rainstorm-induced collapse–landslide–debris flow (R-C-L-D) disaster chain exhibits significant cascade amplification effects, posing severe threats to socioeconomic development and human safety. Traditional approaches struggle to quantify the dynamic interactions within this multihazard system. To address this gap, this study proposes a Bayesian network (BN) based framework for probabilistic scenario prediction of the R-C-L-D disaster chain. First, the disaster formation mechanism was analyzed using the “four-ring” theory (disaster-causing, excitation, damage, and chain-breaking rings), identifying key factors including lithology, geological structure, rainfall intensity, and terrain relief. A dynamic BN model integrating 14 nodes across hazard formation, triggering, and consequence stages was then constructed. Model parameters were learned from 395 historical disaster cases in Southwest China using the Netica platform. Validation through a 2016 Guizhou disaster case confirmed model accuracy (Brier score = 0.483). Scenario prediction indicates that under conditions of topographic relief between 500 and 1,000 m and when there are water systems nearby, the probability of medium- to large-scale disasters increases, while small-scale events decrease, quantitatively proving the cascading amplification effect. The BN model enables forward prediction of hazard scales and impacts by adjusting precursor conditions. Practical strategies for chain-breaking—including river clearance, slope radar monitoring, and preventive evacuation—are proposed based on sensitivity analysis.