Collaborative Research: CAIG: Space-time completeness of seismic ground motions via non-intrusive model order reduction

About This Grant

Large earthquakes occur infrequently, often separated from each other by decades or centuries. This makes it difficult for scientists to predict and understand large earthquakes because they have only a few historical examples to base their models on. To get around this limitation, scientists use computer simulations to create “synthetic” earthquakes that they can study. Unfortunately, these synthetic earthquakes currently require thousands of hours to compute. In this project, machine learning techniques will be developed to generate realistic, synthetic ground motions in minutes, enabling geologists to efficiently study how large earthquakes work and assess the hazards they may pose to the people of California and Nevada. The software developed will be publicly available for other researchers to use, and educational resources will be created to train future scientists and increase public awareness about large earthquakes. Large earthquakes occur on time scales of centuries or more. Earthquake prediction has proven elusive and recordings of large and damaging earthquakes are rare. Wave propagation simulations currently provide the only option for overcoming the lack of observational data, however, sufficiently realistic physics-based models are extremely computationally expensive. To address this, we will develop a physics-based machine learning model order reduction technique, Operator Inference (OpInf), to create time-dependent parametric surrogate models of seismic ground motions, fusing simulated ground motion wavefields with previously observed earthquake records. The OpInf model will be dramatically faster than traditional physics-based methods and more generalizable. The OpInf models will be applied to diverse faulting regimes in California and Nevada to quantify differences in the source and path influences on the ground motion. All software developed will be open-source and publicly available. We will train the future scientists in earthquake science and advanced model order reduction techniques. Our research will appear in events in California and Nevada to enhance public awareness of earthquake safety procedures. We will collaborate with both the U.S. Geological Survey and Statewide California Earthquake Center to assess how our research could enhance existing hazard assessment and earthquake early warning. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.