CyberShake-Derived Ground-Motion Prediction Models for the Los Angeles Region with Application to Earthquake Early Warning

Maren Boese, Robert W. Graves, David Gill, Scott Callaghan, & Philip J. Maechling

Published 2014, SCEC Contribution #1799

Real-time applications such as earthquake early warning (EEW) typically use empirical ground-motion prediction equations (GMPEs) along with event magnitude and source-to-site distances to estimate expected shaking levels. In this simplified approach, effects due to finite-fault geometry, directivity, and site and basin response are often generalized, which may lead to a significant under- or overestimation of shaking from large earthquakes (M>6.5) in some locations. For enhanced site-specific ground-motion predictions considering 3D wave-propagation effects, we develop Support Vector Regression models from the SCEC CyberShake low-frequency (<0.5 Hz) and broadband (0-10 Hz) datasets. CyberShake encompasses 3D wave-propagation simulations of >415,000 finite-fault rupture scenarios (6.5≤M≤8.5) for southern California defined in UCERF 2.0. We use CyberShake to demonstrate the application of synthetic waveform data to EEW as a ‘proof of concept’, being aware that these simulations are not yet fully validated and might not appropriately sample the range of rupture uncertainty. Our regression models predict the maximum and the temporal evolution of instrumental intensity (MMI) at 71 selected test sites using only the hypocenter, magnitude, and rupture ratio, which characterizes uni- and bilateral rupture propagation. Our regression approach is completely data-driven (where here the CyberShake simulations are considered data) and does not enforce predefined functional forms or dependencies among input parameters. The models were established from a subset (~20%) of CyberShake simulations, but can explain MMI values of all >400k rupture scenarios with a standard deviation of about 0.4 intensity units. We apply our models to determine threshold magnitudes (and warning times) for various active faults in southern California that earthquakes need to exceed to cause at least ‘moderate’, ‘strong’, or ‘very strong’ shaking in the Los Angeles (LA) basin. These thresholds are used to construct a simple and robust EEW algorithm: to declare a warning, the algorithm only needs to locate the earthquake and to verify that the corresponding magnitude threshold is exceeded. The models predict that a relatively moderate M6.5-7 earthquake along the Palos Verdes, Newport-Inglewood/Rose Canyon, Elsinore, or San Jacinto faults with a rupture propagating towards LA could cause ‘very strong’ to ‘severe’ shaking in the LA basin; however, warning times for these events could exceed 30 seconds.

Boese, M., Graves, R. W., Gill, D., Callaghan, S., & Maechling, P. J. (2014). CyberShake-Derived Ground-Motion Prediction Models for the Los Angeles Region with Application to Earthquake Early Warning. Geophysical Journal International, 198(3), 1438-1457. doi: 10.1093/gji/ggu198.