Modelling Urban Pluvial Flooding in Cincinnati, Ohio, Using Machine Learning

Urban pluvial flooding presents growing challenges for disaster risk management, yet most susceptibility studies rely on watershed-based frameworks that inadequately capture the localized dynamics of urban systems. This study proposes a city-scale flood susceptibility modeling framework for Cincinnati, Ohio. Cincinnati was chosen because it is a city with a documented history of severe urban flooding, including a once-in-a-century storm in 2016. Multi-source historical flood data were compiled from NOAA storm event records and crowdsourced reports to enhance spatial coverage. Four machine learning algorithms (Random Forest, Support Vector Machine, XGBoost, and Logistic Regression) were implemented to identify the most effective approach for urban pluvial flood prediction. Random Forest (RF) and Support Vector Machine (SVM) achieved the highest accuracy (0.84) and demonstrated strong discriminatory power. RF was selected as the optimal model because it had a higher AUC (90%) and the lowest RMSE (0.35). To assess generalizability, the RF model was validated on updated land use data and flood records from a 2020 storm event. It demonstrated robust performance (accuracy = 0.89, RMSE = 0.36, precision = 0.75, recall = 1, and AUC = 0.95), despite urban development changes. This study’s novelty lies in combining multi-source flood records with a grid-based machine learning framework and rigorously validating model robustness under evolving urban conditions. The results advance urban pluvial flood susceptibility modeling and offer actionable guidance for evidence-based flood risk management worldwide.