Inversion Method

In our preliminary modeling we used 8 seismic stations located at distances from 3 to 328 km from the epicenter (Figure 13.3). The data was bandpass filtered between 0.02 to 0.3 Hz to emphasize the low frequency nature of the source process. The stations that were used provide excellent azimuthal coverage of the ruptured fault.

**Figure 13.3:** Location map. Seismic stations are shown as inverted triangles. The epicenter is shown as a star, and extent of fault model as a bold (red) line
$\begin{figure}\begin{center} \epsfig{file=dreger05_2_1.eps, width=8cm}\end{center}\end{figure}$

Seismic Green's functions were computed with an f-k integration code using the GIL7 velocity model, which has been found to be suitable for the California Coast Ranges (e.g. Pasyanos et al., 1996).

In addition to the seismic data we used 11 coseismic displacement measurements from near-fault continuous GPS sites (Figure 13.4). The coseismic data was obtained from the 1-second solutions by averaging from 10 minutes before to 2-10 minutes after the event, effectively removing the post-seismic signal.

The GPS Green's functions were computed for an elastic half-space (Okada, 1992) representing the average of the elastic parameters in the upper 12 km of the GIL7 seismic velocity model.

**Figure 13.4:** Observed GPS displacements (black) and predicted (green). The epicenter and assumed fault trace (blue line) are shown. The black line gives the distance scale, and the red arrow the GPS deformation scale.
$\begin{figure}\begin{center} \epsfig{file=dreger05_2_2.eps, width=8cm}\end{center}\end{figure}$

We used the inversion method of Harztell and Heaton (1983), and we performed inversions using each data set separately and combined. We performed inversions using a dipping fault (from the moment tensor solution) and a vertical fault from relocated aftershocks and modeling of post-seismic deformation. These inversions showed that better fits were obtained with the vertical fault.

Inversions were performed with variable rake, however we found that it did not significantly improve fit so subsequent inversions assumed pure right-lateral slip.

Multiple time windows were used to model spatial variation in rupture velocity and rise time.

For the best fitting combined-data inversion result we obtained a scalar seismic moment of $1.17 x 10^{25}$ dyne cm, Mw=6.0, and rupture velocity of 3.0 km/s. The rise time is spatially variable, but is on average 1.2 seconds.