<![CDATA[Urban flooding is a frequent hydrometeorological hazard in Indonesia, particularly in Jakarta, driven by rapid urbanization, limited drainage capacity, land cover change, and extreme rainfall. This study develops an explainable ensemble learning framework for urban flood risk mapping in Jakarta using multi-source geospatial and hydrometeorological data, including satellite-based rainfall, topography, land use/land cover, NDVI, and IoT-based river water level observations from 2023–2025. Flood occurrence labels were constructed by integrating municipal flood records with satellite-based inundation data. The framework integrates Random Forest, Gradient Boosting, and XGBoost models, with SHAP applied for interpretability and identification of dominant flood drivers. Model evaluation using ROC-AUC and RMSE indicates that XGBoost achieved the highest performance (AUC = 0.91, RMSE = 0.184), outperforming Random Forest (AUC = 0.87, RMSE = 0.221) and Gradient Boosting (AUC = 0.89, RMSE = 0.203). SHAP analysis identifies rainfall intensity, elevation, proximity to river channels, and built-up area percentage as the most influential factors. Despite uncertainties in flood labeling and the lack of high-resolution drainage data, the results demonstrate the potential of explainable ensemble learning for urban flood risk assessment and resilience planning. ]]>
Copyrights © 2026
