Mechanism-guided Bayesian diagnosis of progressive toppling in an interbedded sandstone–mudstone dangerous rock mass using multisource monitoring

Progressive toppling of dangerous rock masses in steep mountain cliffs is governed by coupled water-induced weakening, toe-support retreat, deformation of overhanging sandstone, and rear-edge fracture development. Because monitoring signals operate at different spatial and temporal scales, isolated peaks or fixed thresholds cannot reliably distinguish environmental disturbances from progressive stability deterioration. This study investigates a toppling-prone sandstone–mudstone rock mass in the Three Gorges Reservoir area, southwestern China, using field geological investigation, laboratory mechanical and acoustic-emission tests, and multisource monitoring. A mechanism-guided Bayesian diagnostic framework is developed to organize rainfall and humidity as hydraulic evidence, earth pressure and crack displacement as support-adjustment evidence, and tilt and in-situ acoustic-emission observations as deformation–damage evidence. Laboratory tests show that increasing mudstone water content reduces cohesion and friction angle, while repeated wetting–drying promotes slaking and loss of durability. Field observations indicate weak, asynchronous or lagged relationships among rainfall, humidity, pressure, deformation, and acoustic-emission responses. Within the investigated 0–7 d lag window, the relative association between rainfall and acoustic-emission energy is highest at a 2 d lag, but remains weak overall. The Bayesian system therefore provides conditional posterior states under laboratory-informed parameter scenarios rather than uniquely identified failure probabilities. Posterior predictive checks, terminal-period temporal holdout, leave-one-sensor-out analysis, and comparison with external proxy indicators are used to examine temporal stability and evidential consistency. The framework provides a practical reference for long-term inspection and tiered risk management of unstable rock masses in mountainous canyon regions. It is a site-specific diagnostic prototype, not a fully validated failure-prediction or early-warning model; absolute stability indices and warning thresholds require additional geometric, mechanical, and documented failure-event data.