Abstract
This study investigates near-field ground-motion variability due to dynamic rupture models with heterogeneity in the initial shear stress. Ground velocity seismograms are synthesized by convolving the time histories of slip velocity obtained from spontaneous dynamic rupture models with Green's functions of the medium calculated with a discrete wavenumber/ finite-element method. Peak ground velocity (PGV) estimated on the synthetics generally matches well with an empirically derived attenuation relation, whereas spectral acceleration (SA) shows only an acceptable match at periods longer than 1 sec. Using the geometric mean to average the two orthogonal components leads to a systematic bias for the synthetics, in particular at the stations closest to the fault. This bias is avoided by using measures of ground motion that are independent of the sensor orientation.
The contribution from stress heterogeneity to the overall ground-motion variability is found to be strongest close to the fault and in the backward directivity region of unilaterally propagating ruptures. In general, the intraevent variability originating from the radiation pattern and the effect of directivity is on the same order or larger than the interevent variability. The interevent ground-motion variability itself is dominated by the hypocenter-station configuration and is influenced only to a lesser extent by the differences in the dynamic rupture process due to the stress heterogeneity. In our modeling approach the hypocenter location is not picked arbitrarily but is determined to be mechanically consistent with the stress heterogeneity through a procedure emulating tectonic stress loading of the fault and nucleation. Compared to the peak ground motion recorded during the 2004 Parkfield, California, earthquake our simulated seismograms show enhanced spatial correlation that may be attributed to the simplicity of the assumed crustal model or to an incomplete representation of the spatial heterogeneity of dynamic rupture parameters. Nevertheless, the intraevent PGV variability in the near-fault region determined for the Parkfield dataset is of the same order of magnitude as for our simulations.