Deriving Site Effect-Free Hard-Rock Time Histories in Japan from the Generalized Inversion Technique

A key component in seismic hazard assessment is the determination of time histories for hard-rock site conditions, either as input motion for site response computations or for applications to installations built on this site type. The state of the practice is to apply physicsbased corrections for removing site effects from surface recordings to obtain the underlying bedrock motion. Here, we propose and test the use of the generalized inversion technique (GIT) for deconvolving surface recordings to hard-rock time series at the amplification-free seismic bedrock. As part of the proposed procedure, an event-specific phase scaling method is presented, which allows changes in signal duration to be considered. For validation purposes, we select a total of 90 Kiban-Kyoshin network (KiK-net) surface-downhole sites having no significant velocity contrasts below their downhole sensors and with the latter being located at sufficient depth so that they are not impaired significantly by downgoing waves. We evaluate the effectiveness of the empirical predictions by comparison with recorded time series at the downhole sensors. We find quite high correlations and small variations in both spectral shape and amplitude over the entire frequency range for the GIT deconvolution at the majority of the 90 KiK-net sites with all mean residuals less than 0.25 in contrast to empirical and 1D modeling approaches, which significantly overestimate the level of hardrock ground motion for frequencies larger than a few Hertz.