Ground-Motion Prediction Equation Development Using Random Vibration Theory, Bayesian Regression, and Finite Fault Simulations: The PEER NGA-East Example
Justin Chow Hollenback1, Nicolas Martin Kuehn
1, Christine A. Goulet
1 and Norm A Abrahamson
2, (1)University of California, Pacific Earthquake Engineering Research Center, Berkeley, CA, United States, (2)Pacific Gas and Electric Company, San Francisco, CA, United States
Contact First Author: Justin Chow Hollenback; jhollenback@berkeley.edu
English Abstract:
Central and eastern North America (CENA) presents a challenging problem in regards to ground-motion modeling as there is a lack of data in areas critical to engineering design. The two cases where data is both sparse and necessary are for large magnitude and short source-to-site distance. Here, we present a methodology used to develop ground-motion prediction equations (GMPEs) for pseudo spectral acceleration (PSA) that utilizes both finite fault simulations and regionalized data from around the world. This methodology allowed us to produce several alternative models for predicting PSA all the way up to magnitude 8.2. Since there is little to no useable amplitude data in CENA for magnitude greater than 6.0 many alternative models are necessary to capture the epistemic uncertainty of the median prediction. This methodology was implemented for development of GMPEs as part of the Pacific Earthquake Engineering Research center (PEER) NGA-East Project.
There are two other relatively uncommon steps taken to develop these alternative models: the majority of the model development is performed on the Fourier Amplitude Spectrum (FAS) of acceleration, and a Bayesian inference scheme was used to estimate both fixed and random effects of the GMPE. In general, GMPEs for PSA are developed using PSA values themselves. However, working in FAS space allowed for better integration of the finite fault simulations and other seismological models for extrapolation. Using the Bayesian inference scheme allowed for the integration of data from regions thought to have different seismological properties. The hierarchical nature of the model accounts for regional difference in how ground motions scale with distance. The way in which our model was set up inherently assumes that the scaling of ground motion with magnitude is similar across regions. The methodology enabled us to produce alternative median GMPEs in CENA that behave reasonably well up to magnitude 8.2.
