1) 발표자: 곽윤신

2) 학술대회명: 2025 한국기후변화학회

3) 발표 주제: Multiscale Watershed Resilience Assessment Using Explainable AI for Transparent Decision-making

4) 발표 내용

Climate change, coupled with socioeconomic complexities, has significantly disrupted watershed systems, exacerbating urban flooding. Effective responses require system-wide insights and detailed local analyses, yet existing flood resilience literature often overlooks subwatershed-specific nuances. This study integrates the Pressure-State-Response (PSR) framework with eXplainable Artificial Intelligence (XAI) to develop a multiscale watershed resilience index for Goyang, South Korea, an urban area vulnerable to flooding. Using eXtreme Gradient Boosting (XGBoost) to model flood damage across biophysical and socioeconomic criteria, we apply SHapley Additive exPlanations (SHAP) to clarify each factor’s impact at both citywide and subwatershed levels. Results indicate citywide prominence of park accessibility, rainfall, and topographic wetness, whereas localized analyses reveal critical socioeconomic factors such as land price. Variations in factor impacts highlight the necessity of context-sensitive interventions. The proposed index emphasizes spatial variability and model transparency, enhancing public trust and aiding policymakers in crafting targeted flood resilience strategies. Ultimately, this research underscores the importance of understanding flood dynamics across multiple scales, offering actionable insights for resilient urban planning and management.