Investigating the between-event variability of the site response from KiKnet data in Japan

Abstract

Uncertainties in seismic hazard analysis are very large and critical for the hazard at long return period. An important part of this uncertainty is, however, epistemic and could be reduced by improving our understanding and subsequently the models. Site effect is a major component of seismic hazard analysis, which significantly participate to the overall uncertainty. In this study, we try to understand what is controlling the between-event uncertainty of the site response. We estimate the amplification function from Standard Spectral Ratio (SSR) at three SSMNet stations in Switzerland and the surface-to-borehole spectral ratio at 60 KiKnet stations having recorded several hundreds of earthquakes. The preliminary step is the automatic quality control, phase picking, Fourier spectrum computations and signal-to-noise ratio verification of more than two million traces. Artificial outliers are identified and discarded before performing the statistical analysis. First of all, our results confirm that amplification factors at any site or frequency are well approximated by a log-normal distribution. Based on that it is possible to use the confidence interval to determine the variation of the mean computed from subset of events. Using the confidence interval, we propose an analytic equation to estimate the number of events to be recorded to have a certain stability in the mean amplification function estimation. We found that at least 20 earthquakes should be used to limit the variation of the mean amplification function below 30% for every station and frequency We finally estimate that the ground motion intensity (PGA) has no influence on the anelastic site amplification functions as far as the site behave linearly.