A new approach to site classification: Mixed-effects Ground Motion Prediction Equation with spectral clustering of site amplification functions

Abstract

With increasing amount of strong motion data, Ground Motion Prediction Equation (GMPE) developers are able to quantify empirical site amplification functions (2) from GMPE residuals, for use in sitespecific Probabilistic Seismic Hazard Assessment. In this study, we first derive a GMPE for 5% damped Pseudo Spectral Acceleration (g) of Active Shallow Crustal earthquakes in Japan with 3.4 ≤ ≤ 7.3 and 0 ≤ < 600. Using k-mean spectral clustering technique, we then classify our estimated 2 (= 0.01-2) of 588 well-characterized sites, into 8 site clusters with distinct mean site amplification functions, and within-cluster site-to-site variability ~50% smaller than the overall dataset variability (φS2S). Following an evaluation of existing schemes, we propose a revised data-driven site classification characterized by kernel density distributions of Vs30, Vs10, H800, and predominant period (TG) of the site clusters

Re-thinking site amplification in regional seismic risk assessment

“…The sheer volume of records within the KiK-net data set, with many sites having the number of repeated observations necessary for robust estimation of δ S 2 S S , make this the optimum test case for exploring correlations between δ S 2 S S and other parameters. In their analysis of the spectral properties of δ S 2 S S for the KiK-net data, Kotha et al (2018) derive a GMM from the KiK-net data using the flatfile of Dawood et al (2016), and it is their resulting estimates of δ S 2 S S for 641 recorded sites used herein. The approach to determine δ S 2 S S adopted by Kotha et al (2018) differs from that of preceding authors by virtue of the fact that the GMM itself does not include an explicit site amplification term, instead:…”

“…In their analysis of the spectral properties of δ S 2 S S for the KiK-net data, Kotha et al (2018) derive a GMM from the KiK-net data using the flatfile of Dawood et al (2016), and it is their resulting estimates of δ S 2 S S for 641 recorded sites used herein. The approach to determine δ S 2 S S adopted by Kotha et al (2018) differs from that of preceding authors by virtue of the fact that the GMM itself does not include an explicit site amplification term, instead:…”

“…One can illustrate the impacts of using inferred V S 30 by considering the KiK-net site data and locally calibrated site-to-site residual of Kotha et al (2018) that will be described in further detail in the next section. Figure 2 shows that while a relatively strong correlation can be seen between the δ S 2 S S term and the observed V S 30 for both 0.2 and 1.0-s spectral acceleration, both the correlation with inferred V S 30 from slope (using the Wald and Allen (2007) approach calibrated for active tectonic regions) and with the slope itself show similar predictive capacities, with weaker correlation and greater uncertainty.…”

“…Correlation between observed δ S 2 S S from Kotha et al (2018) data set and observed V S 30 , inferred V S 30 from Wald and Allen (2007), and slope, fit using a non-parametric LOESS regression (line with shading).…”

“…This invaluable data set is used not for the purpose of refining models of V S 30 from mappable geological and geomorphological proxies, but to construct models of amplification directly from such proxies. Building on the work of Kotha et al (2018), who constructed a database of site-to-site amplification with respect to a GMM (hereafter δ S 2 S S , explained in the next section) for each of the KiK-net stations, we explore correlations between this amplification and mappable proxies, specifically focusing on slope and geology. This site-to-site amplification term is a relatively recent feature of ground motion modeling and arises from the growth of ground motion data allowing for repeated seismic observations across many stations in a seismic network.…”