5.1 Separation and resolution of source and path effects
Perhaps the most challenging problem with obtaining earthquake source parameters is to separate the source from the path and site effects. We apply a series of data processing strategies to improve the robustness of stress drop estimation. Our resolution benefits from the relatively low noise, and wider bandwidth of the borehole stations compared to the previous similar style analysis by AS2007. This can be seen in the lower standard deviation for stress drop variability we obtain for each of our Strategies, compared to the earlier study by AS2007 (Figure 5). We find that the assumptions and choices involved in both the attenuation and velocity corrections can lead to apparent magnitude or depth dependence (Figure 3). The final preferred strategy, which includes both factors, results in the least stress drop dependence and the lowest standard deviation of stress drop variability. AS2007 used a spatially varying ECS, with the spatial smoothing depending on the local density of seismicity, and also found no resolvable magnitude or depth dependence to their stress drop values. As in most previous studies of stress drop, we observe that the standard deviation gradually increases with decreasing magnitudes. The standard deviation increases from <0.1 (for Mw≥2.8) to 0.24 (for Mw≤1), but even the smallest magnitude bin in this study has lower standard deviation than the largest magnitude bin in AS2007, implying a more well-constrained inversion and better parameter resolution (Figure 8). This confirms that a borehole network with wider frequency bandwidth, lower noise, and less attenuated signals can retrieve better measurements of source parameters for small earthquakes.
The standard deviation in stress drop in the entire data set decreases significantly with increasing allowance for depth dependence of attenuation and rupture velocity, implying that these factors contribute to the apparent variability of stress drop. Comparison of the empirical correction spectra shows decreasing attenuation with depth, consistent with the increase in velocity and Q with depth previously observed in the region (e.g., Thurber et al., 2004; Bennington et al., 2008). These previous analyses also find that the along strike variation is less than that with depth, again consistent with our results. Unfortunately, because the spectral decomposition approach combines stations from all azimuths and distances, we cannot investigate any difference in attenuation across the fault.