A MaxEnt-TRIGRS hybrid model with dynamic safety factor mapping for enhanced debris flow susceptibility assessment in...

Traditional statistical models for debris-flow susceptibility often overlook critical triggering mechanisms and geotechnical parameters. To address this, we propose an innovative framework that couples the Maximum Entropy (MaxEnt) statistical model with the TRIGRS physical model, which simulates transient rainfall infiltration and grid-based regional slope stability. Focusing on seven towns in Beichuan County, China, we integrated thirteen environmental factors, geotechnical parameters, and historical hazard records to build a dual-driven "statistical-physical" evaluation framework. Our methodology consists of three steps: (1) Use TRIGRS to compute rainfall-induced safety factors (FS) and identify unstable zones (FS < 1), which serve as the positive-sample database for MaxEnt; (2) Employ the MaxEnt model-using the TRIGRS-derived positive samples and historical debris-flow factors-to predict the spatial distribution of susceptibility; (3) Integrate both outputs spatially in GIS using dynamic weighting. Validation shows that the hybrid model improves prediction accuracy by 21% compared to MaxEnt alone (AUC = 0.845). Its susceptibility map corrects 34.7% of the overpredicted areas from the statistical model and enlarges stable zones by 1.8 times. Additionally, to determine the optimal weighting between machine learning and the physical model, we tested three weight combinations and found that a 0.55:0.45 ratio (MaxEnt: TRIGRS) yields the best performance. Using an independent validation set from another study area, we correctly identified 83.6% of the historical debris-flow events in Changtan, demonstrating the framework's ability to integrate geostatistical patterns with geomechanical processes. This coupled framework offers a paradigm for multi-hazard chain assessment in complex terrain and can be directly applied to debris-flow early warning and regional disaster mitigation planning.