A combination of artificial intelligence (AI), machine learning and digital twin technologies can help predict severe weather events, according to researchers at the Oden Institute for Computational Engineering & Sciences at the University of Texas at Austin.

More than half of the US population lives in coastal watershed counties or parishes. Coastal communities along the Gulf of Mexico are among the most heavily populated, and this is also a region where high concentrations of energy resources have made it a national hub for many large-scale carbon-to-capture storage facilities.

The proximity to the ocean of local communities and energy infrastructures makes both extremely vulnerable to the devastation that can be caused by flooding and wind damage from severe weather events in the Gulf, which are increasing in both frequency and intensity with every hurricane season.

Clint Dawson, director of the Computational Hydraulics Group at the Oden Institute for Computational Engineering & Sciences at UT Austin, is working to make storm surge predictions for hurricanes more accurate.

Thanks to a grant from the US Department of Energy (DoE), Dawson will lead an interdisciplinary research project to develop a computational digital-twin framework that bridges the gap between multi-physics simulations and knowledge discovery through AI and machine learning (ML) technologies, called Musikal.

Simply put, a digital twin is a virtual representation of an object or system that spans its entire lifecycle through regular real-time data updates provided by sensors spread across the object or system. Using simulations, machine learning and other decision-making technologies, digital twins can help predict future performance and behaviour.

Dawson’s team has been modelling storm surge predictions for two decades, from Hurricane Katrina, Rita, Ike and Harvey to this season’s biggest storm to date, Hurricane Ida. And the storm surge expert says each brings its own set of unique characteristics. But lessons can still be learned from each that could inform future responses.

Currently when a hurricane model is running, measurements are being collected at very discrete places – along the coastline and in the ocean, for example – but these points don’t represent every point in every region that might be impacted.

“We need to have a model that provides additional information,” Dawson said. “If we have those data available to use, they can better inform the models we are currently running. And then we can go back and compare the models to the data for a more accurate picture.”

Digital twins have already been developed for a variety of situations, from modern aircraft design to systems assisting in the management of entire cities. In the context of extreme weather modelling, the technology could enable even faster predictions of storm behaviour in real time by combining knowledge about previous storms with the aid of AI and ML.

“These models are very complex and can take hours to simulate on a supercomputer,” Dawson said. “If we can use machine learning based on data that have been collected from prior hurricanes that are very similar, then we could perhaps give faster predictions in real- time.”

Through the Advanced Scientific Computing Research (ASCR) programme, the DoE will support a collaborative team of experimental and computational scientists from UT Austin, Louisiana State University, University of Notre Dame and Pacific Northwest National Laboratory. They will be led by Dawson alongside fellow professor and Oden Institute core faculty member Tan Bui-Thanh.

Oden Institute affiliated faculty member Dev Niyogi from the Jackson School of Geosciences is also part of UT’s research team along with his colleague Zong-Liang Yang. In addition, Bridget Scanlon and Alexander Sun from UT’s Bureau of Economic Geology will collaborate on the project.

The DoE has recently been investing in the development of Earth system models for climate research. Dawson said he looked forward to working on research that was directly related to climate predictions.

“I think this is going to be a ground-breaking project, and aligns well with the expertise we’ve been building up for 20 years,” Dawson said. “Connecting with the Department of Energy to develop longer scale projections of what’s going to happen to the energy sector and to society as a whole because of future climate is very exciting.”

The DoE’s fund for integrated computational and data infrastructure for science research will provide $5.2m overall to the project with UT Austin receiving $3m.