Scientists at Lawrence Livermore National Laboratory (LLNL) have contributed to the development of a new real-time tsunami forecasting system. The system is powered by El Capitan, which is currently the world’s fastest supercomputer and was created with support from the Advanced Simulation and Computing (ASC) program at the National Nuclear Security Administration (NNSA).
El Capitan has a theoretical peak performance of 2.79 quintillion calculations per second. Before it transitions to classified national-security work, researchers used its computing power for an offline precomputation step, generating a large library of physics-based simulations that link earthquake-induced seafloor motion to resulting tsunami waves.
The project involved more than 43,500 AMD Instinct MI300A Accelerated Processing Units (APUs) to solve acoustic-gravity wave propagation problems on an extreme scale. This process produced a dataset that enables real-time tsunami forecasting using smaller computer systems. By completing the most intensive computations in advance on El Capitan, the team solved a high-fidelity Bayesian inverse problem that allows rapid predictions in seconds during actual tsunami events using modest GPU clusters.
The system was developed in partnership with the Oden Institute at the University of Texas at Austin and the Scripps Institution of Oceanography at the University of California, San Diego. It uses real-time pressure sensor data and advanced simulations to model how seafloor earthquakes affect ocean conditions, providing forecasts with uncertainty quantification.
“This is the first digital twin with this level of complexity that runs in real time,” said LLNL computational mathematician Tzanio Kolev, co-author on the paper. “It combines extreme-scale forward simulation with advanced statistical methods to extract physics-based predictions from sensor data at unprecedented speed.”
By leveraging El Capitan for precomputation, researchers solved a billion-parameter Bayesian inverse problem in less than 0.2 seconds, achieving significant speed improvements over previous methods.
Researchers believe this capability could improve emergency response efforts and save lives by supporting next-generation early warning systems. For example, if a magnitude 8.0 or larger earthquake were to occur along the Cascadia Subduction Zone in the Pacific Northwest, destructive waves could reach shore within 10 minutes.
Current tsunami warning systems often use seismic and geodetic data but rely on simpler models that may not capture complex fault ruptures accurately. The new approach instead utilizes seafloor pressure sensors and full-physics modeling to provide faster and more reliable warnings.
As networks of seafloor sensors become more common along earthquake-prone coasts and as computational resources improve, researchers see potential for wider deployment of this approach in future tsunami warning systems.
“This framework represents a paradigm shift in how we think about early warning systems,” said senior author Omar Ghattas, professor of mechanical engineering and principal faculty in the Oden Institute at UT-Austin. “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.”
Central to this system is MFEM, LLNL’s open-source finite element library designed for scalable GPU-accelerated simulations of physical phenomena such as acoustic-gravity wave propagation in oceans. Running these simulations on all 43,520 APUs of El Capitan allowed MFEM to solve acoustic-gravity wave equations involving 55.5 trillion degrees of freedom—setting a record for unstructured mesh finite element simulation size.
“MFEM’s high-order methods and GPU readiness, developed under the ASC program at LLNL and the Department of Energy’s (DOE) Exascale Computing Project, made it possible to scale to the full machine,” Kolev said. “This was really a first-of-its-kind demonstration of how we can use that power not just for raw performance, but also for mission-relevant, time-critical decisions in many MFEM-based applications.”
Kolev explained that after completing these large-scale precomputations on El Capitan, subsequent steps like inferring seafloor motion and forecasting tsunami heights can be done on smaller GPU clusters due to efficient algorithms.
“This work is important because it shows that we can solve an inverse problem of enormous size — not for 10 or 15 variables, but for millions, or even billions of variables, very quickly,” said Kolev. “In the past, you’d either have a fast model that’s not accurate, or a full-physics model that takes hours or days. Now we’re showing that we can do both — accurate and fast — using principled mathematics and modern computing.”
He also noted that this Bayesian inversion framework could be applied beyond tsunamis—to wildfire tracking, subsurface contaminant monitoring, space weather prediction, or intelligence applications requiring quick decisions based on data.
The research team included Veselin Dobrev and John Camier from LLNL; Omar Ghattas, Stefan Henneking, Milinda Fernando and Sreeram Venkat from UT-Austin; as well as Alice-Agnes Gabriel from UC San Diego.
