One-Dimensional Site Response Analysis: Model Error Estimation

Abstract

ABSTRACT Two essential elements of partially nonergodic probabilistic seismic hazard analysis are the estimation of local site effects and the quantification of their uncertainty. Local site effects usually are evaluated using numerical site response analyses, often assuming the 1D site response. This research addresses two questions regarding the estimation of local site effects: for which stations may 1D site response analysis be beneficial? What is the model uncertainty in 1D site response analysis? In answering these two questions, we compare empirical and theoretical transfer functions (TFs) at KiK-net stations and apply state-of-the-art statistical tools. Regarding the first question, previous research associated the applicability of 1D site response to the topography of the site, geological features, and the shape and frequency of the horizontal-to-vertical spectral ratio (HVSR) peak. Among 22 candidate parameters, we find the combination of topographical position index (also known as relative elevation) and the ratio of theoretical TF predominant frequency over HVSR predominant frequency to be the most efficient predictor. However, the accuracy of predictions made by these two parameters is only 60%, highlighting the need for additional research. To address the second question, we use theoretical and empirical TFs to obtain the model error in 1D site response analyses. Model error uncertainty (in natural log units) in predicting TF ranges from 0.1 at low frequencies up to 0.4 around 10 Hz.