Abstract

Flooding is one of the most frequent and damaging natural disasters, accounting for nearly half of global disasters and posing a major challenge in Indonesia, where floods represent approximately 77% of all nationally recorded disaster events. Rapid urbanisation, land-use change, and climate-induced extreme rainfall have intensified flood risks nationwide. However, existing vulnerability assessments remain fragmented and localised, limiting their relevance for national-scale adaptation planning. This study develops a measurable and explainable framework for assessing urban flood vulnerability across Indonesia using cloud-based geospatial data and interpretable machine learning. The approach integrates CEMS-GLOFAS (flood hazard), WorldPop (population exposure), SRTM (topography), and ESA WorldCover (land cover) datasets within Google Earth Engine (GEE). Flood vulnerability is quantified through a modified Flood Vulnerability Index (FVI) combining hazard, exposure, and physical vulnerability components. The Extreme Gradient Boosting (XGBoost) model predicts FVI values, while SHapley Additive exPlanations (SHAP) and Partial Dependence Plots (PDPs) enhance model transparency and identify the influence of key variables such as flood depth, population density, and elevation. The model achieved high predictive accuracy (R2 = 0.89; RMSE = 0.04728 FVI units, dimensionless) and revealed substantial spatial heterogeneity across 514 districts, with the highest FVI (0.75–0.85) in Banda Aceh, Mojokerto, Pasuruan, Samarinda, and Merauke. The integration of GEE and explainable AI offers a transparent, scalable framework to support data-driven flood risk mitigation and urban climate resilience in Indonesia.

IPC Classification

Keywords