Forecasting Hurricane-Induced Compound Flooding Using Hybrid Numerical and Machine Learning Models

To address the growing demand for efficient forecasting methods in light of more frequent and intense hurricane-induced compound flooding events, this dissertation presents two novel approaches that integrate multiple models to capture the complexity of such phenomena. The first approach involves the development of a cutting-edge two-way coupled hydrological-ocean model, where the hydrological modeling extension package of the Weather Research and Forecasting model (WRF-Hydro) was dynamically coupled with the Regional Ocean Modeling System (ROMS) on the platform of the Coupled-Ocean-Atmosphere-Wave-Sediment Transport (COAWST) Modeling System. This coupled model was applied to simulate Hurricane Florence-induced compound flooding across a land-ocean continuum, demonstrating superior performance in capturing water level variations compared to traditional one-way coupled models. It accurately modeled the interaction between storm surge and inland flooding, allowing for in-depth analyses of the temporal development, water budget, and nonlinear effects in the Cape Fear River Estuary, NC. These analyses revealed a four-stage evolution pattern, quantified freshwater inputs, and highlighted the estuary's buffering effects during the event. Further improvements to the coupled model were made by introducing the Local Inertial Equation and a diagonal flow algorithm into WRF-Hydro’s overland routine, enhancing its ability to simulate backwater effects. Based on the improved coupled model’s results, an innovative framework for quantifying compound and nonlinear effects was developed, which provided a measurement of the interaction between storm surge and inland flooding, as well as the extent to which their combined effects contributed to the overall flooding. The analysis of these two effects identified the areas highly vulnerable to compound flooding and suggested lateral water transport from the channel to the floodplain during the event.\nThe second approach introduced a hybrid numerical and machine learning model, which combines the strengths of numerical models and machine learning techniques to deliver rapid, accurate, and spatially distributed flood forecasts. Trained on both observed data and outputs from the dynamically coupled hydrological-ocean model, the hybrid model achieved superior predictive accuracy while significantly increasing computational speed compared to the numerical model. This capability demonstrates a breakthrough in overcoming the typical trade-off between speed and accuracy in traditional hybrid approaches, establishing the hybrid model as a state-of-the-art tool for flood forecasting.