This thesis takes place in the earthquake engineering context. It intends to characterize the ground motion during earthquake. Earthquake resistant constructions remain at the moment the best means for safety. This demands the knowledge of the nature of seismic loads. This work is based on two recently installed Japanese networks, since 1996, which constitute two of the best databases on the world. K-net is a surface network, whereas Kik-net is a surface/borehole network. The thesis deals with databases of crustal shallow events, depth less than 25 km, with magnitude between 4.0 and 7.3.
The Eurocode 8 currently proposes standard shapes for the design response spectra, depending on the seismicity rate of the region and concerning standard facilities. The analysis of the K-net data allowed to compute a new spectral ground motion prediction equation and to review the shape of the proposed EC8 spectra. The work then proposes new design shape for the national chapter of the Eurocode 8. In addition, the predictive equation exhibits a larger amplification at short period for Japanese data and brings in some light about the soil amplification that takes place at long period.
Some facilities require time acceleration histories in order to study nonlinear behavior
under seismic loads. This thesis develops a new empirical model for simulating synthetic stochastic nonstationary acceleration time histories. By specifying magnitude, distance and site effect, this model allows to produce many time histories, that a seismic event is liable to produce at the place of interest. The model is calibrated on a given database and is noticeable by its ability to simulate the variability of the motion. The Japanese databases K-net and Kik-net
and the European database are used to develop and calibrate this method.
In low to moderate seismicity zones, a current question that arises is how to predict the ground motion during a strong earthquake that could occur in the country. The study of near-field borehole records of the Kik-net allows to explore the validity domain of predictive equations.
In particular, this study allows us to explain what occurs by extrapolating ground motion predictions. Another problem that relies on large earthquakes deals with nonlinear behavior of soils. In this case the signature of nonlinear soil response is studied by propagating a large number of earthquake scenarios through a nonlinear medium. Our results show that nonlinearity reduces the dispersion of ground motion at the surface. In addition, independently of the computercode deamplification of the ground motion is expected for PGA’s closer to 2 m.s-2 at the surface.