Author(s): Hyeontae Moon; Kyung-Tak Kim; Gilho Kim
Linked Author(s):
Keywords: Explainable artificial intelligence (XAI); Flood influencing factors; Grid-based modeling; Hydro-geomorphology; Jeju Island
Abstract: This study applies a grid-based, explainable machine learning framework to identify key flood influencing factors by integrating historical flood trace maps with multi-layer spatial datasets. Four tree-based algorithms were employed to classify flooded and non-flooded grids at a 100 m resolution. Model performance was evaluated across subregions to examine spatial variability in predictive accuracy and the relative importance of influencing factors. The models achieved moderate-to-high predictive performance, with recall and F1-scores of up to 0.81 and 0.75, respectively. Feature-importance analyses consistently identified short- and long-duration rainfall (3-hour and 12-hour maximum), 5-day antecedent rainfall, maximum wind speed, groundwater level, and proximity to detention ponds and rivers as dominant predictors of flood occurrence. Their relative influence, however, varied across subregions. In inland and high-elevation basins, rainfall duration and groundwater dynamics were the primary controls, whereas coastal and urbanized zones were more influenced by drainage connectivity and proximity to infrastructure. These spatial contrasts reflect the heterogeneous hydro-geomorphological setting of Jeju Island and emphasize the importance of region-specific flood warning thresholds and adaptive management strategies. Overall, the findings highlight the potential of grid-based explainable machine learning for spatially adaptive flood risk assessment in volcanic island environments under changing climate conditions.
Year: 2026