A Factor Model for the Estimation of Multivariate Empirical Ground-Motion Models

Abstract

A Bayesian latent factor to estimate multivariate empirical ground-motion models is proposed. In the factor model, event terms, site terms, and within-event/within-site residuals for each response oscillator period are modeled as linear combinations of a small number of latent factors, thus introducing inter-period correlation in the model. The factor model provides a lower-dimensional approximation to the full residual correlation structure. The factor model is demonstrated on Italian ground-motion data. Results indicate that the factor model leads to similar behavior as a traditional ground-motion model where each target is estimated separately.