The rupture process of the M-J=7.2 1995 Hyogo-ken Nanbu (Kobe) earthquake deduced from S-wave polarization analysis

The use of S-wave polarization analysis for constraining high-frequency ( > 1 Hz) source parameters of large earthquake has been previously illustrated by Bouin & Bernard (1994) and Guatteri & Cocco (1996). In this paper, we s how the efficiency of such a methodology applied to a data set in the prese nce of strong non-linear and liquefaction effects for the particular case o f the 1995 Hyogo-ken Nanbu (Kobe) earthquake. The ii strong-motion recordin gs from the closest stations to the fault plane have been analysed in the 0 .6-2.0 Hz frequency band. We first correct the recorded ground motion data for the effect of shallow crustal anisotropy in the Kobe area, which allows us to remove part of the complexity of the observed S-wave polarigrams. Ou r results in terms of anisotropy parameters are in good agreement with thos e of previous studies performed in the same area. They show evidence, far f rom the faults, of a fast S-wave polarization oriented nearly parallel to t he regional compressive stress direction, and, close to the faults, of a fa st S-wave polarization oriented parallel to the faults' strikes. The correc ted S-wave polarigrams are then interpreted in terms of rupture propagation along an extended fault. We consider three different faulting mechanisms: a pure right-lateral mechanism and two mechanisms with 30 and 50 per cent r everse components. The polarization data and the triggering times at the se lected stations allow the identification of a set of subsources, demonstrat ing the need for a reverse slip component during the rupture of the two hig h-frequency (2.0 Hz) subevents on the Kobe fault section. The resulting loc ations of the main subevents in space and time provided an interval of poss ible values of rupture velocity ranging between 2.65 km s(-1) (that is, 80 per cent of the shear wave velocity) and 3.46 km s(-1) (that is, the value of the shear wave velocity). This study emphasizes that S-wave polarization can be a useful tool to restrict the number of reliable starting models fo r iterative waveform inversions.