A Partially Non-Ergodic NGA Subduction Ground Motion Prediction Equation and Its Application to Cascadia

Date: 4/24/2019

Time: 11:30 AM

Room: Pine

We present a new NGA subduction ground-motion model (GMM), based on the 2018 PEER NGA-SUB global database. About 16,500 records from 247 events are used to estimate the model parameters for both subduction interface and intraslab events. The model takes into account regional differences in the constant, anelastic attenuation and linear site scaling. Seven regions were considered: Alaska, Cascadia, Central America and Mexico, Japan, New Zealand, South America, and Taiwan. The regional adjustment terms are modeled as regional random effects, which assumes that the regional coefficients are samples from a global coefficient distribution. The model includes a bilinear magnitude scaling, where the magnitude break point depends on the age/geometry of the subduction zone and whether the events are on the interface or within the slab. The model parameters are estimated via Bayesian inference, using Markov Chain Monte Carlo sampling. The Bayesian approach allows us to include prior information and assess uncertainty in a probabilistic way. Epistemic uncertainty associated with the model parameters and median predictions are taken into account via the posterior distribution of coefficients. To account for outliers due to low quality data, the parameters are estimated via robust regression.

For Cascadia, the available data are very limited, with only 624 records from 12 earthquakes, with no reliable data from interface events. The largest Cascadia earthquake in the data set is the Nisqually (M6.8) event. This leads to large epistemic uncertainties associated with the regional coefficients for Cascadia, which translate into large epistemic uncertainties for median predictions. The ground-motions from the Cascadia events are on average very low compared to the other regions, however, the two larger events in Cascadia have ground motions that are closer to the global average. We discuss strategies of how to adjust the model predictions for Cascadia to take these observations into account.

Presenting Author: Nicolas M. Kuehn

Authors