Derivation of consistent hard rock (1000 < Vs < 3000 m/s) GMPEs from surface and down-hole recordings: analysis of KiK-net data

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 kappa(0) 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.