Generative AI for Compound Flood Forecasting

physics-based generative emulator for probabilistic hurricane flooding

Graduate research at the Climate Risk & Extreme Dynamics Lab, Georgia Tech, advised by Prof. Ali Sarhadi (Xiao & others, 2026).

Overview

Physics-based Generative AI for Probabilistic Forecasting of Hurricane-Driven Compound Flooding (in preparation for NMI)

This project builds a generative emulator that translates tropical cyclone tracks into compound flood fields for probabilistic hazard forecasting.

Key Components

  • Generative emulator: a TensorFlow-based model translating tropical cyclone tracks into compound flood fields
  • Synthetic training data: massive ensembles of physics-simulated synthetic tracks coupled with complex hydrodynamic and hydraulic simulations, overcoming historical data scarcity
  • Real-time probabilistic maps: flood maps generated from ECMWF ensembles in seconds instead of days, enabling risk-informed emergency planning
Flood field prediction. To be replaced with a figure from the project.

References

2026

  1. Preprint
    Physics-based Generative AI for Probabilistic Forecasting of Hurricane-Driven Compound Flooding
    Shijie Xiao and others
    Manuscript in preparation, 2026