Multi-scenario probabilistic hazard assessment of flash flood disasters in Chengkou County based on a multi-state system model

Flash flood disasters pose a serious threat to lives, property, and regional sustainable development in mountainous areas. Existing hazard assessment methods often rely on static indicator systems, which suffer from excessive subjectivity in weight assignment and difficulties in integrating dynamic triggering factors effectively. To address these limitations, this study introduces multi-state system theory and, using Chengkou County in Chongqing as a case study, constructs a multi-scenario probabilistic hazard assessment model that couples a disaster-forming environment subsystem (static background) with a rainfall-triggering subsystem (dynamic driver). The disaster-forming environment subsystem is characterized by a parallel structure to represent baseline susceptibility, while the triggering subsystem uses a series structure to simulate the activating effect of rainfall. Based on GIS and multi-source data, each system component is classified into four states—from Level I (high hazard) to Level IV (very low hazard)—and a multi-state system probability model is applied to quantitatively analyze the hazard levels of 25 townships under different rainfall scenarios. Results show that, without rainfall triggers, the disaster-forming environment in all townships remains at Level IV (very low hazard), indicating low baseline susceptibility. However, heavy rainfall significantly alters the spatial pattern of hazard. During the extreme rainfall event on May 28, 2023 (precipitation ≥ 180 mm), 18 townships were classified as high-hazard, and their spatial distribution showed strong consistency with recorded local disaster data. Because both the rainfall inputs and disaster records are derived exclusively from local sources within Chengkou County, this evaluation should be regarded as an internal/local consistency check rather than independent external validation; cross-regional transferability remains unproven. In contrast, lower-intensity rainfall events resulted in a marked reduction in hazard magnitude. By adopting a multi-state system model, this study reduces subjective weight assignment and quantifies flash flood hazard probabilistically across multiple rainfall scenarios, offering a new probabilistic paradigm for county-scale hazard assessment in mountainous regions. The model’s outputs represent relative hazard probabilities rather than deterministic flood predictions, and its thresholds require site-specific recalibration when applied to other regions.