Degree: Master of Science, College of Computing

Advisor: Dr. Josiah Hester

 

Title: Physics-Informed Deep Learning Emulator for Predicting Hurricane-Driven Compound Flooding

 

Date: Tuesday, April 21, 2026

Time: 10:00 AM – 11:00 AM

 

Thesis Committee:

• Dr. Josiah Hester (Advisor) – College of Computing, Georgia Institute of Technology

• Dr. Ali Sarhadi – School of Earth and Atmospheric Sciences, Georgia Institute of Technology

• Dr. Peng Chen – School of Computational Science and Engineering, Georgia Institute of Technology

 

Abstract:

This thesis presents a Transformer–Fourier Neural Operator (Trans+FNO) architecture for predicting hurricane-driven compound flooding over the New York City metropolitan area. Compound flooding, the simultaneous interaction of storm surge and heavy rainfall, poses a growing threat to coastal communities under climate change. Physics-based hydrodynamic models can simulate compound flooding at high resolution but require hours of computation per event, making them impractical for ensemble-based probabilistic forecasting. Our deep learning emulator maps variable-length hurricane track sequences to three-channel spatially resolved flood depth fields at 1024×1024 resolution, trained on 13,013 samples from a physics-based simulation framework spanning seven climate model realizations. The model achieves 97%+ wet/dry classification accuracy and sub-0.1 m RMSE. We demonstrate ensemble-based flood uncertainty quantification using ECMWF forecasts and provide SHAP-based interpretability analysis revealing physically consistent feature importance hierarchies across compound, surge, and rainfall flooding.