Future dynamic vulnerability to extreme snowfall under population decline: A hybrid random Forest-System Dynamics framework for disaster management policy.

Climate change and severe demographic shifts are amplifying the asymmetrical impacts of natural disasters, particularly in rural regions. However, existing vulnerability assessments largely isolate physical climate exposure from the dynamic socio-organizational adaptive capacity of local communities. To address this gap, this study proposes a novel hybrid spatial framework that couples a data-driven machine learning model (random forest regression) with a process-based simulation (System Dynamics) to evaluate the future dynamic vulnerability to extreme snowfall in Gangwon-do, South Korea, up to the year 2100. First, a random forest regressor was employed to project long-term climate exposure under four Representative Concentration Pathway (RCP) scenarios. Concurrently, System Dynamics-integrated with Monte-Carlo simulations-was utilized to model the structural degradation of Citizen Corps Active in Disaster (CAIND) under population decline. The dual outputs were spatially coupled using a Euclidean distance-based vulnerability index. The results reveal a critical turning point in the 2040s: despite up to a 33.1% decrease in extreme climate exposure over 2021-2100, the region's average dynamic vulnerability increases by 14.9-16.4% from its 2030s minimum across all RCP scenarios, a rise primarily associated with a 23.1% decline in CAIND's adaptive capacity. Furthermore, spatial analyses demonstrate a 37.2% decrease in regional vulnerability deviation (upward leveling) and a 74.4% increase in spatial autocorrelation (Moran's I) by the 2090s, indicating the formation of province-wide "high-vulnerability belts". The framework provides empirical evidence urging a paradigm shift from localized interventions to wide-area disaster governance before the 2040s turning point.