PEER has just published Report No. 2020/03: "NGA-Subduction Global Ground-Motion Models with Regional Adjustment Factors." It was authored by Grace A. Parker, formerly of UCLA and now at the U.S. Geological Survey; Jonathan P. Stewart, University of California, Los Angeles; David M. Boore, Geophysicist (Emeritus, formerly of the U.S. Geological Survey) Los Altos, California; Gail M. Atkinson, Western University London, Canada; Bezhad Hassani, BC Hydro, Canada.
Visit the PEER publications page to download a free color pdf of the document.
Abstract
Next Generation Attenuation Subduction (NGA-Sub) is a multi-year, multidisciplinary project with the goal of developing an earthquake ground-motion database of processed time series and ground-motion intensity measures (IMs), as well as a suite of ground-motion models (GMMs) for global subduction zone earthquakes. The project considers interface and intraslab earthquakes that have occurred in Japan, Taiwan, New Zealand, Mexico, Central America, South America, Alaska, and Cascadia. This report describes one of the resulting GMMs, one important feature of which is its ability to describe differences in ground motions for different event types and regions.
We use a combination of data inspection, regression techniques, ground-motion simulations, and geometrical constraints to develop regionalized models for IMs for peak ground acceleration, peak ground velocity, and 5%-damped pseudo-spectral acceleration at 26 oscillator periods from 0.01 to 10 sec. We observe significant differences in ground-motion scaling for interface and intraslab events; therefore, the model terms for source and path effects are developed separately. There are complex distance-scaling effects in the data, including regional variations and forearc and backarc effects. No differences in site effects between the event types were observed; therefore, a combined site term is developed that is taken as the sum (in natural log units) of a linear term conditioned on the time-averaged shear-wave velocity in the upper 30 m (VS30), and an empirically constrained nonlinear term. Basin sediment depth terms are developed for Cascadia and Japan that are conditioned on the depth to the 2.5 km/sec shear-wave velocity horizon (Z2.5).
Our approach to model development was to first constrain a path term capturing the observed effects, then to subsequently investigate magnitude scaling, source-depth scaling, and site effects. Regionalized components of the GMM include the model amplitude, anelastic attenuation, magnitude-scaling corner, VS30-scaling, and sediment depth terms.
Aleatory variability models are developed that encompass both event types, with different coefficients for each IM. Models are provided for four components of ground-motion variability: (1) between-event variability, Φ ; (2) within-event variability, Φ ; (3) single-station within-event variability, ΦSS ; and (4) site-to-site variability,ΦS2S. The aleatory variability models are magnitude independent. The within-event variability increases with distances beyond 200 km due to complexities in path effects at larger distances. Within-event variability is VS30-dependent for distances less than 200 km, decreasing for softer soils with VS30 less than 500 m/sec. These reductions are attributed to soil nonlinearity. An ergodic analysis should use the median GMM and aleatory variability computed using the between-event and within-event variability models. An analysis incorporating non-ergodic site response (i.e., partially non-ergodic) should use the median GMM at the reference-rock shear-wave velocity (760 m/sec), a site-specific site amplification model, and aleatory variability computed using the between-event and single-station within-event variability models. Epistemic uncertainty in the median model is represented by standard deviation terms on region-dependent model constant terms, which facilitates scaled-backbone representations of model uncertainty in hazard analyses.
Model coefficients are available in the electronic supplement to this report (Tables E1–E4), and coded versions of the model are available in Excel, MatLab, R, and Python from Mazzoni et al. [2020(b)].