Enhancing flood risk assessment in urban areas by integrating hydrodynamic models and machine learning techniques

Abstract

Urban flood risks have intensified due to climate change and dense infrastructural development, necessitating innovative assessment approaches. This study aimed to integrate advanced hydrodynamic models with machine learning (ML) techniques to improve urban flood prediction and hazard analysis. Integrating 1D and 2D hydrodynamic models calibrated with precise parameters demonstrated exceptional predictive accuracy for flood dynamics. The high-resolution Lidar images and sophisticated modeling were used to simulate flood scenarios for the River Thames in West London. The socioeconomic data were incorporated to map vulnerable zones accurately. The findings revealed that 2D hydrodynamic models, while computationally intensive, offered superior accuracy in predicting flood dynamics. In contrast, 1D models could have been more accurate in predicting inundation depth and extent in adjacent urban areas. Advanced ML models, such as the Extra Trees-Principal Component Analysis (ET-PCA) model, further instilled confidence in the reliability of the research findings, demonstrating near-perfect predictive reliability with an R2 of 0.999. This model provided near-perfect predictions and enabled precise flood risk zoning. The integration of socioeconomic data further highlighted the vulnerability of certain urban areas. It emphasized the importance of targeted flood mitigation strategies. Beyond the specific case study, this research demonstrated the potential of combining hydrodynamic simulations with ML to enhance urban flood resilience globally. This framework helps urban planners and policymakers devise effective flood mitigation strategies and improve infrastructure resilience against future flood events.