Exciting news! We're transitioning to the Statewide California Earthquake Center. Our new website is under construction, but we'll continue using this website for SCEC business in the meantime. We're also archiving the Southern Center site to preserve its rich history. A new and improved platform is coming soon!

Site-specific Ground-Motion Predictions for Earthquake Early Warning in the LA Basin using CyberShake Simulations

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

Published December 2012, SCEC Contribution #1771

Predicting site-specific ground-motion intensity measures for large earthquakes (M>6.5) in real-time is a major challenge in earthquake early warning (EEW). Commonly, ground-motion parameters are predicted indirectly from magnitude and distance using empirical ground-motion attenuation relations. This approach though bears two major problems: (1) the extent of fault ruptures is usually unknown in real-time and the earthquake has to be approximated by a simple point-source; (2) directivity and basin effects are generally neglected. This can lead to a serious under estimation of ground-motions and result in warnings not being issued. To overcome these shortcomings, we have developed a novel approach for site-specific ground-motion prediction based on physics-based statistical models. Our models were derived from the SCEC CyberShake dataset that consists of around 400,000 full 3D wave-propagation simulations (6.5≤M≤8.5) at around 200 locations in and around the Los Angeles (LA) basin. The original purpose of this dataset was to develop a physics-based computational approach to probabilistic seismic hazard analysis (PSHA) based on southern California ruptures defined in Unified California Earthquake Rupture Forecast 2.0 (UCERF2.0). Here we extend the application of the CyberShake dataset to improve ground-motion predictions in the LA area for EEW and other purposes. The aim of our study is to provide fast and realistic site-specific ground-motion estimates for any large earthquake (M>6.5) in southern California. For this purpose we developed statistical models based on Support Vector Machines (SVMs) which were trained with source and ground-motion parameters from the CyberShake dataset to predict the expected long-period spectral acceleration (3 to 10 seconds) in the LA basin. SVMs are supervised learning models for classification and regression. Once trained, our SVM models are able to predict the spectral acceleration (SA) for any given earthquake magnitude and location in southern California. The results from our study show that due to the combined effects of fault geometries, wave propagation, and rupture directivity, it is not necessarily the closest fault ruptures that cause strongest shaking in the LA basin in the long-period range, but can be those that are considerably far away (such as those on the southern San Andreas Fault). This observation suggests that warnings could be provided some tens of seconds to around one minute before long-period shaking in the LA basin starts. Our approach does not only give a qualitative, but quantitative description of this shaking in terms of SA levels. Long-period shaking as analyzed in this study is most critical for high-rise buildings; the ground-motion predictions from our models can thus find possible application in structural control systems. However, as wave propagation simulations covering a broader frequency band become available, our ground-motion prediction models can be easily extended to shorter periods.

Boese, M., Graves, R. W., Callaghan, S., & Maechling, P. J. (2012, 12). Site-specific Ground-Motion Predictions for Earthquake Early Warning in the LA Basin using CyberShake Simulations. Oral Presentation at AGU Fall Meeting 2012.