Recent advancements in the work include the development of an AI-based surrogate model for rapid and accurate flood depth estimation and the design of machine learning algorithms to classify post-disaster damage—none, minor or major or destroyed—following hurricanes and floods. Using Hurricane Melissa as a case study—a recent event lacking labeled data—the research mirrors unpredictable real-world conditions.
By developing machine learning models capable of analyzing multi-modal data in real time, the project allows authorities and insurers to conduct rapid, accurate damage assessments immediately following a catastrophe.
AI-Driven Hazard Modeling
Led by Avantika Gori, assistant professor of civil and environmental engineering at Rice, this project aims to develop a multi-scale framework to understand how changes in regional climate affect the probability and severity of damaging hurricane winds and rainfall at the local level.
“Since there is limited observational data for hurricanes, we are using a multi-pronged approach that combines reanalysis data and climate model simulations, and leverages reduced-physics models and state-of-the-art artificial intelligence models to generate large ensembles of synthetic hurricane tracks. Using these track ensembles, we can understand which climate factors drive the year-to-year differences in hurricane landfalls and hazards, “ Gori said. “Our eventual goal is to understand how changes in sea surface temperatures, and complex climate patterns such as El Niño and La Niña, determine if a tropical storm will turn into a hurricane, what its potential speed and path will be, and finally, how it will make landfall and die out.”
Recent advancements in the work include benchmarking the ability of large AI foundation models to simulate tropical cyclones, and developing self-organizing maps, a machine-learning technique, to predict hurricane paths and landfall patterns and how they vary from year to year depending on overall climate conditions.
The improved weather and hazard models resulting from this work will offer precise hyperlocal simulations that will greatly benefit communities, authorities, and decision-makers in disaster preparedness and protect lives and structures from hurricane and flood damage.
Ensuring the Future of Catastrophe Modeling and Resilience
In addition to advancing research, CERCat also supports a vital talent pipeline, ensuring that the work it has initiated will be carried forward by future generations of multidisciplinary scholars and practitioners, and propelling rising subject matter experts into leadership roles where they will shape resilience strategies across the private and public sectors.
