Diagnosing region-specific flood influencing factors in a volcanic island using explainable machine learning

Abstract This study applies a grid-based, explainable machine learning framework to identify dominant flood-influencing factors under hydrologically heterogeneous conditions, using Jeju Island as a representative volcanic environment. Historical flood trace maps were integrated with multi-layer spatial datasets to classify flooded and non-flooded grid cells at a 100 m resolution using five tree-based algorithms. The models achieved recall and F1-scores of up to 0.86 and 0.78, respectively. Feature-importance and explainability analyses demonstrate that the dominance of flood-influencing factors varies across spatial contexts, with hydrometeorological conditions acting as the primary drivers and spatial exposure and proximity-based factors contributing in a region-specific manner. The results highlight diagnostic insights into the primary drivers of flooding and suggest that factors which may be underrepresented in conventional rainfall-centric assessments can exert substantial influences on flood occurrence.