Lawrence Livermore National Laboratory (LLNL) scientists are pushing the boundaries in tsunami forecasting by developing a real-time warning system that uses El Capitan, the world's fastest supercomputer. LLNL announced that this new system will drastically improve early warning capabilities for communities near quake-prone coastal regions.
The supercomputer, which performs at an incredible 2.79 quintillion calculations per second, was part of a collaboration with the Oden Institute at the University of Texas at Austin and the Scripps Institution of Oceanography at UC San Diego, aimed at enhancing the timeliness and accuracy of tsunami predictions. "This is the first digital twin with this level of complexity that runs in real time," LLNL computational mathematician Tzanio Kolev said, per the Lawrence Livermore National Laboratory.
El Capitan utilized a whopping 43,500 AMD Instinct MI300A Accelerated Processing Units (APUs) to precompute the extensive simulations necessary for forecasting tsunamis with precision. These simulations linked seafloor earthquake motion to ensuing tsunami waves. This process, called front-loading, allowed the researchers to process the data quickly – within seconds – using significantly smaller GPU clusters in real time during an actual tsunami event.
The advanced system uses real-time pressure sensor data, paired with intricate physics-based simulations to form a "digital twin" that can model the impact of an underwater earthquake on the seafloor and predict, with uncertainty quantification, the resulting tsunami's behavior. It's a sharp move away from conventional methods, which rely on seismic and geodetic data and have been known to provide false alarms or dangerously late warnings due to oversimplified models. The Bayesian inverse problem tackled by the scientists, which had a billion parameters, was solved in less than 0.2 seconds, a 10-billion-fold speedup over previous methods, illustrating that significant advancements in computational capabilities can directly translate into life-saving technologies.
With the spread of seafloor sensor networks and improvements in computational infrastructure, there's a clear path forward for deploying this method in future tsunami warning systems globally. Omar Ghattas, a senior author of the study and professor at UT-Austin, pointed out the significance of the advancement, saying, per LLNL, "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." This could mean faster, smarter, more reliable emergency alerts for natural disasters, marking a new era in emergency response systems.
The core technology, an open-source finite element library known as MFEM, enabled this milestone. According to LLNL, the simulations' preliminary phase utilized an immense 55.5 trillion degrees of freedom to precompute the ocean-floor motion and sensor data mappings. The achievements not only demonstrate the raw power of a supercomputer like El Capitan but also how it can be leveraged to make real-time, life-or-death decisions in emergencies.
