Using Bayesian Updating to Apply a Regionalized, Partially Nonergodic Ground-Motion Model to a New Region: Subduction Region as an Example

Session: Forthcoming Updates of the USGS NSHMs: Hawaii, Conterminous U.S. and Alaska [Poster]
Type: Poster
Date: 4/28/2020
Time: 08:00 AM
Room: Ballroom
Description: 

In recent years, the number of strong-motion recordings has increased significantly. This means that it is now possible to model regional differences, for example, in anelastic attenuation and linear site scaling. Typically, regional differences are modeled with a different set of coefficients for the regionalized terms for each region. Often, the regional coefficients are modeled as random effects to account for differences in the amount of data between regions. Such a model is referred to as partially nonergodic. It has the advantage of pooling together information from different regions, but allowing for regional differences in parts of the model. Since one is modeling systematic regional effects, aleatory variability is reduced.

A regionalized model means that one should only use it to calculate ground motions for regions that are used in the development of the model. However, often a model needs to be applied to new regions that were not included in the initial model development. We demonstrate how a partially nonergodic regional model can be adjusted for application to a new region. We use a recently developed ground-motion model for subduction events as an example. The model was developed for seven regions (Alaska, Cascadia, Central America and Mexico, Japan, New Zealand, South America and Taiwan), with different regional coefficients for the intercept, anelastic attenuation, linear site amplification and basin amplification.

We show how one can update the model to be applicable to a new region via Bayesian updating. The basic approach is to use the average regional coefficients and their standard deviation as a global prior distribution for the regional coefficients. This prior distribution can be updated based on local data to obtain new regional coefficients. This ensures that the epistemic uncertainty of the newly estimated regional adjustments is properly taken into account. We demonstrate the approach using simulated data and observations as an example.

Presenting Author: Nicolas M. Kuehn

Authors