At high frequencies, the acceleration spectral amplitude decreases rapidly; this has been modelled with the spectral decay factor κ. Its site component, κ₀, is used widely today in ground motion prediction and simulation, and numerous approaches have been proposed to compute it. In this study, we estimate κ for the EUROSEISTEST valley, a geologically complex and seismically active region with a permanent strong motion array consisting of 14 surface and 6 downhole stations. Site conditions range from soft sediments to hard rock. First, we use the classical approach to separate local and regional attenuation and measure κ₀. Second, we take advantage of the existing knowledge of the geological profile and material properties to examine the correlation of κ₀ with different site characterization parameters. κ₀ correlates well with V_s30, as expected, indicating a strong effect from the geological structure in the upper 30 m. But it correlates equally well with the resonant frequency and depth-to-bedrock of the stations, which indicates strong effects from the entire sedimentary column, down to 400 m. Third, we use our results to improve our physical understanding of κ₀. We propose a conceptual model of κ₀ with V_s, comprising two new notions. On the one hand, and contrary to existing correlations, we observe that κ₀ stabilizes for high V_s values. This may indicate the existence of regional values for hard rock κ₀. If so, we propose that borehole measurements (almost never used up to now for κ₀) may be useful in determining these values. On the other hand, we find that material damping, as expressed through travel times, may not suffice to account for the total κ₀ measured at the surface. We propose that, apart from material damping, additional site attenuation may be caused by scattering from small-scale variability in the profile. If this is so, then geotechnical damping measurements may not suffice to infer the overall crustal attenuation under a site; but starting with a regional value (possibly from a borehole) and adding damping, we might define a lower bound for site-specific κ₀. More precise estimates would necessitate seismological site instrumentation.

A ground-motion prediction equation (GMPE) specific to rock and stiff- soil sites is derived using seismic motion recorded on high VS30 sites in Japan. This GMPE applies to events with 4:5 ≤ Mw ≤ 6:9 andVS30 ranging from 500 to 1500 m=s (stiff-soil to rock sites). The empirical site coefficients obtained and the comparison with the simulated site functions show that seismic motion on rock and stiff-soil sites does not depend only on VS30, but also on the high-frequency attenuation site proper- ties (κ0). The effects of the site-specific κ0 on site amplification are analyzed using stochastic simulations, with the need to take into account both of these parameters for rock-site adjustments. Adding the site-specific κ0 into the GMPEs thus appears to be essential in future work. The rock-site stochastic ground-motion simulations show that the site-specific κ0 controls the frequency corresponding to the maximum response spectral acceleration (famp1). This observation is used to link the peak of the response spectral shape to κ0 in this specific Japanese dataset and then to add the effects of high- frequency attenuation into the previous GMPE from the peak ground acceleration and up to periods of 0.2 s. The inclusion of κ0 allows the observed bias to be corrected for the intraevent residuals and thus reduces sigma. However, this κ0 determination is limited to a minimum number of rock-site of less than 50 km

A key component in seismic hazard assessment is the estimation of ground motion for hard rock sites, either for applications to installations built on this site category, or as an input motion for site response computation. Empirical ground motion prediction equations (GMPEs) are the traditional basis for estimating ground motion while V_S30 is the basis to account for site conditions. As current GMPEs are poorly constrained for V_S30 larger than 1000 m/s, the presently used approach for estimating hazard on hard rock sites consists of “host-to-target” adjustment techniques based on V_S30 and κ₀ values. The present study investigates alternative methods on the basis of a KiK-net dataset corresponding to stiff and rocky sites with 500 < V_S30 < 1350 m/s. The existence of sensor pairs (one at the surface and one in depth) and the availability of P- and S-wave velocity profiles allow deriving two “virtual” datasets associated to outcropping hard rock sites with V_S in the range [1000, 3000] m/s with two independent corrections: 1/down-hole recordings modified from within motion to outcropping motion with a depth correction factor, 2/surface recordings deconvolved from their specific site response derived through 1D simulation. GMPEs with simple functional forms are then developed, including a V_S30 site term. They lead to consistent and robust hard-rock motion estimates, which prove to be significantly lower than host-to-target adjustment predictions. The difference can reach a factor up to 3–4 beyond 5 Hz for very hard-rock, but decreases for decreasing frequency until vanishing below 2 Hz.