A Two-Stage Regression Framework for Assessing Municipal Flood Risks and Mitigation Policy Effectiveness under Climate Change

Climate change has intensified extreme rainfall events, increasing flood risks at the local level. To support evidence-based flood management, this study develops a flood risk model based on a two-stage regression structure. The first stage develops a nonlinear flood damage function using daily maximum rainfall as the independent variable. The second stage employs machine learning to relate the coefficients of the flood damage function to flood mitigation policy options, including retention reservoir ratio, pumping capacity ratio, and river channel improvement ratio. This second-stage function operates as a policy evaluation module, enabling assessment of how policy interventions affect flood damage mitigation. The model was developed for 228 municipalities across South Korea using 24 years of historical flood records from 1998 to 2021. The model offers two key capabilities: estimating economic flood damage from rainfall input and comparing economic damage across different policy options. To assess climate change impacts and mitigation effects of policy options, future rainfall projections from the WRF climate model under SSP2-4.5 and SSP5-8.5 scenarios were applied. The analysis indicates that integrated policy interventions could reduce future economic losses by approximately 34.92% under SSP2-4.5 and 1.62% under SSP5-8.5 compared to baseline scenarios. Model development is expected to be completed by 2026, with a web-based platform scheduled for deployment in 2027–2028. Once operational, the platform will enable local governments to assess flood risks and evaluate policy options tailored to their specific conditions, providing practical decision support for climate-resilient flood management. AcknowledgementThis work was supported by Korea Environment Industry & Technology Institute(KEITI) through Climate Change R&D Project for New Climate Regime, funded by Korea Ministry of Environment(MOE)(grant number RS-2022-KE002152)