Implementing Non-Ergodic GMMs in Probabilistic Seismic Hazard Analysis

Session: Forthcoming Updates of the USGS NSHMs: Hawaii, Conterminous U.S. and Alaska
Type: Oral
Date: 4/28/2020
Time: 04:45 PM
Room: 230 + 235
Description: 

We compute non-ergodic probabilistic seismic hazard analysis (PSHA) for a few test sites to establish the methods and procedures necessary for computing location-specific hazard that accounts for systematic source, path and site effects. We make comparisons with ergodic results to gain confidence in the new approach. Implementation in PSHA requires a new type of input model that describes the non-ergodic adjustment that we must make to the median prediction of the ground motion model (GMM). We achieve this by computing location-specific adjustment terms for median predictions of ergodic GMMs based on a varying coefficient model (VCM) for an event/site constant and Vs30-scaling (Landwehr et al., 2016) and a cell-specific attenuation (Dawood and Rodriguez-Marek, 2011; Kuehn et al., 2019).

For each test site, we calculate nonergodic adjustment terms (mean and covariance) for a grid of source locations around the site. The result is a new Non-Ergodic Characterization (NEC) input to the hazard. For each grid location, we sample the distribution of the NEC terms to assess the epistemic uncertainty; the sampled NEC terms are added to the ergodic median predictions when calculating PSHA. We calculate the non-ergodic hazard using Haz45.3_NE (github.com/Abrahamson/HAZ) that accepts the standard Ground Motion Characterization (GMC) and Seismic Source Characterization (SSC) along with a new NEC model as input.

For the trial application, we compute NEC models for Bay Area sites using three ergodic base GMMs (ASK14, BSSA14 and CB14) at two spectral periods (0.02 and 1 second). The nonergodic parameters (VCM parameters and cell-specific attenuation coefficients) are estimated using Bayesian inference. All calculations for estimating the NEC models are implemented in Python. The results are non-ergodic hazard curves and fractiles at the different locations in the Bay Area; comparisons with ergodic PSHA shows wider fractiles and a different mean hazard.

Presenting Author: Nicolas M. Kuehn

Authors