UT-Led Team is Finalist for 2025 Gordon Bell Prize for Digital Twin Tsunami Research

A research team led by Omar Ghattas, principal faculty member at the Oden Institute for Computational Engineering and Sciences at The University of Texas at Austin has been named a finalist for the 2025 Association for Computing Machinery (ACM) Gordon Bell Prize. The team’s research – developing an advanced, real-time tsunami forecasting system using digital twin models – could dramatically improve early warning capabilities for coastal communities near earthquake zones.

In addition to Ghattas, Director of the Optimization, Inversion, Machine Learning, and Uncertainty for Complex Systems (OPTIMUS) center at the Oden Institute, team members include Stefan Henneking, research associate in the OPTIMUS center, Sreeram Venkat, student in the Computational Science, Engineering, and Mathematics (CSEM) program and member of OPTIMUS, and Milinda Fernando, research associate in the Parallel Algorithms for Data Analysis and Simulation (PADAS) group, all of the Oden Institute. The team also includes Alice Gabriel of the Scripps Institution of Oceanography at the University of California, San Diego, who brought earthquake modeling expertise to the project, and Veselin Dobrev, John Camier, and Tzanio Kolev of the Center for Applied Scientific Computing at Lawrence Livermore National Laboratory (LLNL), who are developers of the MFEM finite element library.

Their finalist submission focuses on a high-stakes challenge: Developing a real-time digital twin for tsunami early warning in the Cascadia Subduction Zone, located on the northwest coast in one of the most seismically hazardous regions in North America. The project combines high fidelity computational models, seafloor pressure sensor data, and state-of-the-art Bayesian inference to predict sea floor motion and resulting tsunami wave heights with quantified uncertainties in a fraction of a second. Delivering high fidelity forecasts to emergency managers in under a second enables them to issue targeted warnings and save countless lives.

The approach begins with acoustic pressure data acquired from seafloor sensors during an earthquake rupture event. Using 3D coupled acoustic–gravity wave equations, the team can infer the spatiotemporal motion of the seafloor immediately after an earthquake. This information is then used to forecast how the resulting tsunami will travel toward coastlines, providing a critical window for early evacuation decisions.

“This framework represents a paradigm shift in how we think about early warning systems,” said senior author of the study Ghattas, professor of mechanical engineering and Cockrell Chair in Engineering at UT. “For the first time, we can combine real-time sensor data with full-physics modeling and uncertainty quantification — fast enough to make decisions before a tsunami reaches the shore. It opens the door to truly predictive, physics-informed emergency response systems across a range of natural hazards.”

What makes this achievement remarkable is its unprecedented computational scale and speed. This research leverages the world’s most powerful supercomputer, El Capitan, at Lawrence Livermore National Laboratory (LLNL), capable of more than 2.79 exaflops per second. Funded by the National Nuclear Security Administration’s (NNSA) Advanced Simulation and Computing (ASC) program and deployed in 2024, El Capitan supports both national security missions and unclassified scientific research. The team:

●  Solved an inverse problem with ~1 billion parameters representing seafloor motion.

● Reduced what would normally require 50 years of computing time on a 512-GPU system to 0.2 seconds through novel parallel  inversion algorithms leveraging the mathematical structure of the problem and through an offline–online decomposition strategy that offloads the most expensive computations to an offline precomputation step.

“The biggest technical challenge was the enormous problem size combined with the need to provide forecasts in real time,” said Henneking. “We overcame this by developing new numerical methods, implementing them on cutting-edge GPU architectures, and leveraging large GPU supercomputers such as Perlmutter and El Capitan.”

The Cascadia Subduction Zone spans 700 miles from northern California to British Columbia, Canada, and has the potential to generate an earthquake exceeding magnitude 9.0. Scientists estimate a one-in-three chance of a major earthquake striking within the next 50 years with the potential to send a tsunami racing toward the Pacific Northwest coastline.

“At the time of an earthquake, we use pressure data from seafloor sensors to infer the seafloor motion in less than a second,” explained Henneking. “From there, the digital twin can near-instantaneously predict tsunami wave heights at key coastal locations, along with uncertainty estimates that are essential for early warning decisions,” said Dr. Henneking.

Beyond the potential for life-saving applications, this project demonstrates the broader promise of the digital twin framework developed by the team. The methods and algorithms developed here could be adapted for other real-time hazard forecasting problems, from hurricanes to volcanic eruptions, as well as for applications in wave propagation, threat detection, and treaty verification.

A preprint of the Gordon Bell paper is available here.

The winners of the 2025 ACM Gordon Bell Prize will be announced at the annual Supercomputing Conference in St. Louis, MO, in November 2025.