Scientists at Lawrence Livermore National Laboratory (LLNL) have collaborated with other institutions to develop a real-time tsunami forecasting system using El Capitan, the world’s fastest supercomputer. The system is designed to improve early warning capabilities for coastal communities located near earthquake zones.
El Capitan, developed with support from the Advanced Simulation and Computing (ASC) program at the National Nuclear Security Administration (NNSA), can perform up to 2.79 quintillion calculations per second. Researchers utilized its computational power in an offline precomputation phase before the supercomputer transitioned to classified national-security tasks. This process created a large library of physics-based simulations that connect earthquake-induced seafloor movements to resulting tsunami waves.
The project relied on more than 43,500 AMD Instinct MI300A Accelerated Processing Units (APUs) to address large-scale acoustic-gravity wave propagation challenges. The data produced allows for real-time tsunami forecasting on smaller computing systems. By conducting the intensive calculations ahead of time on El Capitan, the team solved a high-fidelity Bayesian inverse problem, enabling rapid predictions during an actual tsunami event using smaller GPU clusters.
The research was conducted in partnership with the Oden Institute at the University of Texas at Austin and the Scripps Institution of Oceanography at UC San Diego. The resulting digital twin uses real-time pressure sensor data and advanced simulations to model how earthquakes affect the ocean floor and forecast tsunami behavior as events unfold.
“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’s capabilities, researchers were able to solve a billion-parameter Bayesian inverse problem in less than 0.2 seconds and predict tsunami wave heights much faster than existing methods.
According to researchers, this new capability could significantly improve emergency response efforts by providing timely warnings for events such as a magnitude 8.0 or larger earthquake along the Cascadia Subduction Zone in the Pacific Northwest, where destructive waves may reach land within ten minutes.
Current tsunami warning systems often depend on seismic and geodetic data but use simplified models that do not capture complex fault ruptures accurately, which can result in false alarms or late warnings. The team’s method instead uses seafloor pressure sensors and comprehensive physical modeling of wave propagation for more reliable forecasts.
As seafloor sensor networks expand along vulnerable coastlines and computational resources advance, researchers see potential for their approach to be integrated into future early warning systems.
“This framework represents a paradigm shift in how we think about early warning systems,” said Omar Ghattas, senior author of the study and professor at UT-Austin’s Oden Institute. “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 that supports scalable GPU-accelerated simulations like those needed for acoustic-gravity wave propagation. Running these simulations on all 43,520 APUs of El Capitan enabled researchers to break records for unstructured mesh finite element simulations.
“MFEM’s high-order methods and GPU readiness, developed under the ASC program at LLNL and DOE’s 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.”
Once initial computations are completed on El Capitan, subsequent forecasting steps can run on smaller GPU clusters because algorithms are optimized for such hardware.
“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.”
Kolev added that this Bayesian inversion framework has potential applications beyond tsunamis—such as wildfire tracking or space weather forecasting—where quick analysis is critical.
Other contributors included Veselin Dobrev and John Camier from LLNL; Omar Ghattas, Stefan Henneking, Milinda Fernando and Sreeram Venkat from UT-Austin; and Alice-Agnes Gabriel from UC San Diego.
