Integrating data-driven and physical models for urban flood prediction in a single framework

Urban flooding, driven by rapid urbanization and climate change poses critical challenges globally. This research develops an innovative framework, combining diverse models, data and methods to address flood susceptibility, intensity prediction, and inundation simulation across multiple scales. The framework includes: (1) A machine learning based method to assess flood susceptibility, using social media data and environmental factors. It provides low-cost and real-time insights into flood-prone areas. (2) The Log-Gaussian Cox Process (LGCP) model as a spatial statistical model, for predicting flood intensity while capturing unexplained spatial variability; (3) A coupled 1D-2D hydrodynamic model that integrates a 1-dimensional flooding model with a 2D spatial model to simulate inundation. The framework was applied in the rapidly urbanizing regions of Chengdu and Haining, China. Key flood drivers were identified, vulnerable areas were highlighted, and actionable insights for urban flood mitigation were provided. By integrating data-driven models, spatial analysis, and physical simulations into a single framework, this research offers a scalable and comprehensive approach for urban flood management, with potential applications to other natural hazards globally.