Hybrid Numerical and Machine-Learning Modeling Framework for Hurricane-Induced Compound Flood Prediction

Hurricane-induced compound flooding, driven by the combination of storm surges and heavy rainfall, poses significant threats to lives, property, and economies. Traditional prediction approaches, including process-based numerical models and data-driven machine learning (ML) models, each have their strengths and limitations. Numerical models provide accurate spatial predictions but are computationally intensive, while ML models offer faster predictions but are often limited by data availability.

To address these challenges, we developed a novel framework that combines numerical models with ML models for precise spatial flood predictions. This framework includes a prediction model, a numerical model, and a mapping model. The prediction model, based on Long Short-Term Memory (LSTM) networks, is trained on time series data of water depth, discharge, and wind. The numerical model, a dynamically two-way coupled hydrological-ocean model, integrates WRF-Hydro and ROMS via the COAWST Modeling System, capturing the complex interactions between hydrological and oceanic processes. This integration significantly improves the accuracy of simulating compound flooding events. The mapping model utilizes ML to map spatial water depth distributions.

When applied near Galveston Bay, TX, using data from 2013-2023 and validated with Hurricane Harvey data, our hybrid approach demonstrated highly accurate predictions with a remarkable level of computational efficiency. It produced flood maps with average differences of 0.31 m to 0.36 m compared to traditional numerical simulations, and did so in just 4 seconds using a single core, as opposed to 15 minutes with 480 cores in conventional models.

This hybrid approach exemplifies the potential for rapid, accurate spatial flood predictions. As part of the Gulf of Mexico Coastal Hazards Forecast system, this hybrid approach is gradually being adopted into an operational mode. The ultimate goal is to provide timely and accurate predictions of hurricane-induced compound flooding, supporting effective flood preparation and risk mitigation efforts across the Gulf Coast.