Corresponding author: X. Meng, School of Earth and Atmospheric Sciences, Georgia Institute of Technology, 311 Ferst Dr., Atlanta, GA 30332, USA. (firstname.lastname@example.org)
 Earthquakes trigger other earthquakes, but the physical mechanism of the triggering is currently debated. Most studies of earthquake triggering rely on earthquakes listed in catalogs, which are known to be incomplete around the origin times of large earthquakes and therefore missing potentially triggered events. Here we apply a waveform matched-filter technique to systematically detect earthquakes along the Parkfield section of the San Andreas Fault from 46 days before to 31 days after the nearby 2003 Mw6.5 San Simeon earthquake. After removing all possible false detections, we identify ~8 times more earthquakes than in the Northern California Seismic Network catalog. The newly identified events along the creeping section of the San Andreas Fault show a statistically significant decrease following the San Simeon main shock, which correlates well with the negative static stress changes (i.e., stress shadow) cast by the main shock. In comparison, the seismicity rate around Parkfield increased moderately where the static stress changes are positive. The seismicity rate changes correlate well with the static shear stress changes induced by the San Simeon main shock, suggesting a low friction in the seismogenic zone along the Parkfield section of the San Andreas Fault.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Large shallow earthquakes are typically followed by increased seismic activity within several lengths of a main shock rupture, which are termed “aftershocks.” Recent studies have also shown that large earthquakes can trigger seismicity at several hundreds to thousands of kilometers away [Hill et al., 1993; Kilb et al., 2000]. While such long-range triggering is mostly due to dynamic stresses induced by passing seismic waves, it is still not clear whether earthquakes in the near field (<~100 km) are also triggered by transient dynamic stresses [Felzer and Brodsky, 2005, 2006; Gomberg et al., 2003], or static stress changes from permanent fault displacements [Harris and Simpson, 1992; Harris et al., 1995; King et al., 1994; Reasenberg and Simpson, 1992; Richards-Dinger et al., 2010; Stein, 1999; 2003]. One key means to differentiate between these triggering mechanisms is whether the static stress shadow cast by the main shock inhibits seismicity [Harris and Simpson, 1992; Harris et al., 1995; Stein, 1999, 2003]. An improved knowledge of the mechanisms of earthquake triggering is not only vital for better understanding earthquake interaction [Hill and Prejean, 2007; Stein, 1999; 2003] but also useful for seismic hazard forecasting and mitigation [Gerstenberger et al., 2005; Reasenberg and Jones, 1989].
 The 2003 Mw6.5 San Simeon earthquake struck the central California coast on 22 December 2003, 19:15:56 UTC and induced large dynamic and static stress changes in its vicinity [Hardebeck et al., 2004]. The main shock ruptured a reverse fault striking northwest and dipping northeast and was followed by numerous aftershocks [Hardebeck et al., 2004; Mclaren et al., 2008]. About 9 months later, an Mw6.0 earthquake occurred on the Parkfield section of the San Andreas Fault (SAF), ~50 km to the east of the San Simeon main shock epicenter (Figure 1a). Because the two earthquakes occurred closely in space and time, it is reasonable to speculate that the San Simeon main shock may have triggered the Parkfield earthquake. Indeed, the Parkfield section of the SAF experienced positive Coulomb stress changes on the order of 10 kPa (Figure 1a), suggesting that the static stress changes may explain their triggering relationship [Aron and Hardebeck, 2009]. However, the seismicity rate around Parkfield decreased slightly after the San Simeon earthquake [Aron and Hardebeck, 2009], which is inconsistent with either the static Coulomb hypothesis or dynamic stress changes.
 Previous studies on earthquake triggering mostly utilize existing earthquake catalogs, which may be incomplete immediately following large earthquakes, mainly due to the intensive seismic activity and masking from the coda waves of the main shock and large aftershocks [Aron and Hardebeck, 2009; Enescu et al., 2007, 2009; Peng et al., 2006, 2007]. So the seismicity rate decrease around Parkfield could be due to the incompleteness of the Northern California Seismic Network (NCSN) catalog immediately after the San Simeon main shock. Inspired by recent successes in detecting earthquakes and tectonic tremors based on the waveform matched-filter technique [Peng and Zhao, 2009; Shelly et al., 2007], we apply a modified version of this technique to systematically search for missing earthquakes around Parkfield using 13 borehole stations from the High Resolution Seismic Network (HRSN), station PKD from Berkeley Digital Seismic Network (BK network), and 25 surface stations from Northern California Seismic Network (NC network) (Figures 1a and 1b). With a more complete catalog, we can elucidate the genuine seismicity rate changes along the SAF.
2 Data and Methods
2.1 Continuous and Template Waveforms
 The continuous seismic data are recorded by 39 stations from the HRSN, BK, and NC networks around Parkfield (Figures 1a and 1b). For HRSN stations, we use BP channels (short-period three-component geophone recording at 20 samples/s). For station PKD, we use BH channels (broadband three-component recording at 20 samples/s). For NC stations, we use the EHZ channel (short-period vertical component) downsampled from 100 to 20 Hz. Our study period is 90 days before to 90 days after the San Simeon earthquake. On 6 November 2003, roughly one and a half months before the San Simeon earthquake, stations CCRB, LCCB, SCYB, and SMNB from HRSN had gain value changes, which caused a significant change in background noise level (see Figure S1 in the supporting information). Since 23 January 2004, 31 days after the San Simeon earthquake, eight stations from NC network have stopped recording (Figure S2). Because the eight stations are all in the northwest side of our study region (Figure 1b), the detection ability around these stations may be reduced after that day. Hence, although we use a total of 180 days of continuous waveforms, we primarily focus on the results from 46 days before to 31 days after the San Simeon main shock. The continuous recordings are daily long segment relative to the origin time of the San Simeon main shock. When a significant data gap exists, the length of daily data is determined by the data availability (Figure S2).
 We use 3531 earthquakes listed in the relocated catalog [Thurber et al., 2006] within 5 km to the SAF as template events (Figures 1a and 1b). These templates are the same with those used in a previous study [Peng and Zhao, 2009]. Template waveforms are 1 s before and 5 s after the predicted S wave arrival for two horizontal components and 1 s before and 5 s after the predicted P wave arrival for the vertical component. The P and S wave arrival times are computed based on a 1-D velocity model in this region [Waldhauser et al., 2004]. In comparison, Peng and Zhao  used 2 s before and after the predicted S wave arrival for all three components. The motivation for such change is to enhance the ability of detecting local events along the Parkfield section by enforcing an S-P time constraint. We compare the predicted P and S wave arrival times with the ones picked by Northern California Earthquake Data Center (NCEDC) and confirm that the initial P and S wave arrivals are included in the template time window (Figure S3a).
2.2 Detection Procedure
 We first apply a two-way fourth-order band-pass filter of 2–8 Hz to both continuous and template waveforms. Next, we compute the correlation coefficient (CC) value in a 6 s time window between the template and continuous waveforms and move forward with a step of one data point (0.05 s). Then, we shift the CC values back to the origin time of the template event and compute the mean CC value of all channels at each data point. A threshold is set for each mean CC trace, above which the positive correlation is considered a detected event. The determination of threshold will be described in following sections.
 We compare the detecting time window of all positive detections, which is from the start time of the earliest correlating window to the end time of the latest correlating window among all channels. If the detecting time windows of multiple detected events overlap (Figure 2), only the one with the highest CC value is kept, and the corresponding template is referred to as the “matching template” hereafter. The hypocenter of the matching template is assigned to that of the detected event. The magnitude of the detected event is computed based on the median value of the peak amplitude ratio between the detected event and the matching template among all channels [Peng and Zhao, 2009]. Figure 3 illustrates an example of a positive detection at ~4 days after the San Simeon main shock. The template event occurred on 21 November 2002 and had a magnitude of 1.09. The estimated magnitude for the newly detected event is 0.56.
 Ideally, templates that occurred during our study period should detect themselves with a CC value of approximately 1.0 (e.g., >0.98), which are referred to as perfect self-detection. Evaluating if all the perfect self-detections can be achieved is the best way to examine potential problems in data set and/or analysis procedure. A total of 198 template events occurred during our 180-day study period. Using the original daily long continuous data, however, none produces perfect self-detection (Figure 4a). The primary cause is the subtle differences in the beginning time of daily long continuous data among stations. After adding it to the predicted P or S wave arrival time, dividing by sampling rate and rounding, it may cause one data point difference (Figure S4). As a result, the best correlating window of different channels may not be aligned at the same time point, which will significantly lower the mean CC value.
 We use the following procedures to fix this problem. First, we shift the reference time of continuous data of all stations to a common time (i.e., middle night at each day). Next, we cut the data so that they have the same starting time to the nearest 0.05 s, which help to reduce the minor starting time differences among stations. With the corrected continuous data, 152 out of 198 template events can produce perfect self-detections (Figure 4a). In order to have all 198 templates achieve perfect self-detections, we allow one data point shift while stacking. That is, at each time point, the highest CC value among itself and its two neighboring points is used for stacking. This is very helpful for achieving all perfect self-detections (Figure 4a). More importantly, some local events may not be detected if the best correlating windows of all channels are not aligned at the same time point (Figures 4b, 4c, and 5). In this way, we ensure that the best correlating windows of all channels are stacked together.
2.4 False Detections
 Previous studies using the same technique generally define the threshold of positive detections as the sum of the median value and 8 or 9 times the median absolute deviation (MAD) of the mean CC trace [Peng and Zhao, 2009; Shelly et al., 2007]. However, allowing one data point shift while stacking would lower the MAD value of the mean CC trace (Figure S3b). Hence, a higher threshold should be applied in this study. We first select 12 times the MAD as threshold, which is roughly equivalent to 8 or 9 times the MAD in previous studies [Peng and Zhao, 2009; Shelly et al., 2007].
 Figure 6 illustrates a possible false detection using the sum of the median value and 12 times the MAD as threshold. The detected event occurred ~42 days after the San Simeon main shock and was detected by a template beneath Middle Mountain with a mean CC value of 0.21. The correlated window of continuous waveforms appears to be a segment of seismic signals with much longer duration, indicating a distant source. According to the NCSN catalog, an aftershock of magnitude 1.46 occurred in the San Simeon rupture zone just ~7 s prior to this detection. We confirm that the event indeed is the San Simeon aftershock by predicting the P and S wave arrival times of the San Simeon aftershock at all stations. It appears that the template correlates with the P wave train of the San Simeon aftershock. Hence, this kind of detection of a San Simeon aftershock is considered false detection.
 Figure 7 shows spatiotemporal distributions of all detected events around the San Simeon main shock by using a threshold of 12 times the MAD. Many seismic events were detected by templates beneath Middle Mountain (near station MMNB) after the San Simeon main shock. Most of them have very low CC values, suggesting that they are likely false detections. We hypothesize that a similar move out (i.e., increasing travel times with distances) between template events along the SAF and the San Simeon aftershocks is most likely the cause of false detections. To test this, we randomly select one San Simeon aftershock and compute its P and S wave arrival times at all stations based on the 1-D velocity model [Waldhauser et al., 2004]. We then measure the similarity of moveout by calculating the root mean square (RMS) of the relative arrival time differences at all stations between the San Simeon aftershock and templates. The smaller the RMS value is, the more similar the moveout between a San Simeon aftershock and templates is.
 Templates with the along-strike distances between −26 and −10 km tend to have the smallest RMS value (the most similar moveout) with the San Simeon aftershock (Figure S5), where large numbers of false detections are also found following the main shock. As noted before, such false detections have very low CC values. By increasing the threshold to be the sum of the median value and 15 times the MAD of the mean CC trace, almost all the false detections can be removed. Fifteen times the MAD in this study is roughly equivalent to 11 or 12 times the MAD in previous studies. The selection of threshold is a trade-off between the number of detected events and the possibility of false detections. By applying a higher threshold, we ensure that false detections are discarded. However, some local events might be excluded as well.
 Even after removing false detections by raising the threshold to 15 times the MAD, template 20040928192605 still detected ~2800 events, most of which have very low CC values (Figure 7, triangles). Figure 8 shows one detected event by template 20040928192605, which correlates with the P wave of an Md1.78 event listed in the NCSN catalog that occurred ~80 km away. The main reason for the unusual high number of detections by this template is that its waveforms have relatively low amplitudes during most of the 6 s time window and large impulsive signals near the end. This results in false detection of the P wave from a distant event. In fact, the impulsive arrivals near the end of the correlating time window actually are from another template 20040928192610, which occurred ~5 s later and ~0.05 km away according to the NCSN catalog. By comparing the waveforms of the two templates, it is clear that template 20040928192605 is problematic (Figure S6). Therefore, we remove all detected events by template 20040928192605. We have checked other templates and do not find similar problems.
2.5 Detections in a Higher-Frequency Range
 After removing all possible false detections, 1664 detected events are left (Figure 9). We finally check all of them in a higher-frequency range using DP channels (three-component geophone recording at 250 Hz) of HRSN stations. We apply a 10–25 Hz band-pass filter to the continuous waveforms and templates recorded by the DP channels. We use this frequency range because high-frequency signals from distant events (i.e., San Simeon aftershocks) would be attenuated more as compared with local events. Hence, this frequency range would favor detections of local signals on or near the SAF. Next, for each of the 1664 events, we repeat the same detecting procedure with its matching template in DP channels. As a result, 95%, 87%, and 82% of remaining events can also be detected in the 10–25 Hz frequency range when the threshold is set to be the sum of the median value and 9, 12, and 15 times the MAD of the mean CC trace, respectively. The numbers suggest that the majority of the detected events in the 2–8 Hz band could also be detected in the 10–25 Hz band. Because higher-frequency seismic signals are less coherent, the correlation between templates and continuous data in a higher frequency range is usually lower than that in a lower-frequency range. Hence, the smaller number of positive detections is expected. For the remainder of this paper, we use the 1664 earthquakes detected by a 2–8 Hz band-pass filter.
 The CC-magnitude relationship of detected events before and after the San Simeon main shock shows essentially the same pattern (Figure 9). For same magnitude, no clear decrease of CC value can be observed for events that occurred after the main shock, suggesting no obvious change in detection ability. During the same period, 201 and 198 events were listed in the NCSN and relocated catalogs [Thurber et al., 2006], respectively.
 As briefly mentioned before, our primary study period for investigating seismicity rate changes is chosen from 46 days before the main shock, when the network change completed, to 31 days after the main shock, when eight NC stations stopped recording. Among 1664 detected events, 1046 events occurred during the primary study period. Figure 10a shows the locations of the 1046 detected events in cross section along the SAF strike. Most of the detected events occurred beneath Stone Canyon (station SCYB) and Middle Mountain (station MMNB). In contrast, there were only a few detected events farther northwest in the creeping section and southeast of Gold Hill (station GHIB) near Cholame.
3.1 Spatiotemporal Changes of Seismicity
 The spatiotemporal evolution of the events shows a marked difference along the SAF strike (Figure 10b). Based on the distinct preshock behavior of seismicity, we divide our study area into three subregions and quantify the seismicity rate changes by computing β value, which is a measure of the difference between the observed number of events after the main shock and the expected number from the background rate before the main shock [Aron and Hardebeck, 2009; Kilb et al., 2002]. We also compute the seismicity ratio, which is simply the ratio between the average postmain shock rate and the premain shock rate [Shelly and Johnson, 2011]. We compute the magnitude of completeness (Mc) of −0.3, 0, and 0.6 for subregions A, B, and C using the best combined method in ZMAP [Wiemer, 2001], respectively (Figures S7b–S7d). We recognize that the obtained Mc may be inaccurate and have temporal variations (Figure S7e). Moreover, β value and seismicity ratio may strongly depend on Mc; hence, we compute β value and seismicity ratio by setting the cutoff magnitude from −1.6 to 2 to avoid potential bias.
 In subregion A (northwest of Middle Mountain, less than −19 km in the along-strike distance), we find a clear seismicity rate decrease across all magnitude bands (Figures 11b and 11c). However, such a decrease may be caused by an abrupt jump in cumulative number ~16 days before the main shock due to an Md2.77 earthquake and its aftershocks (Figure 11a). To further evaluate this, we exclude the time window from the Md2.77 earthquake to the San Simeon main shock and recalculate β value and seismicity ratio. For all magnitude bands, the decrease of seismicity rate is still significant at the 95% confidence level (i.e., β < −2), even after excluding the Md2.77 earthquake sequence (Figure 11b). Moreover, we predict the expected decay of the aftershock sequence of the Md2.77 earthquake after the San Simeon main shock by fitting the modified Omori's law [Omori, 1894; Utsu et al., 1995; Wiemer, 2001] using events that preceded the main shock (Figure 11a). The aftershock sequence appeared to be slightly stifled by the San Simeon main shock for ~30 days. We also examine the potential impact on the Parkfield seismicity patterns by the Md2.77 earthquake sequence. We compare the magnitude distributions of detected events before and after the San Simeon main shock and find no clear change caused by the Md2.77 earthquake (Figure S8).
 Finally, the significance of the seismicity rate decrease in subregion A is also evident in the smoothed β value changes using all the events listed in the NCSN catalog from 1984 to 2009 (Figure 12a). The computing time window is set to be the same as our primary study period (46 days before and 31 days after any given time). The time window moves forward by 10 data points. We obtain an Mc of 1.2 in this subregion using the best combined method in ZMAP [Wiemer, 2001] and compute the smoothed β value changes with events above Mc (Figure 12b). The β values obtained from our detected events in both cases are above two standard deviations, suggesting that the seismicity rate changes around the San Simeon main shock were indeed abnormal.
 The seismic activity changed dramatically across Middle Mountain. In subregion B (between Middle Mountain and Cholame, the along-strike distance between −19 and 8 km), only 21 and 30 seismic events occurred before and after the main shock, respectively (Figure 10b). A significant increase of seismicity rate is evident at the 90% confidence level for cutoff magnitude less than 0 (Figure 11e). For cutoff magnitude larger than 0, β values become statistically unreliable due to the small number of remaining events. The seismicity ratio increases for all cutoff magnitudes (Figure 11f).
 In subregion C (southeast of Cholame, with the along-strike distance of >8 km), the seismicity was quiescent during most of our study period. A few swarms occurred at 15 and 10 days before the main shock and completely stopped right before the origin time of the main shock (Figure 10b). A dramatic decrease of seismicity rate is evident due to the earthquake swarms (Figures 11g and 11h). However, if we remove the swarm sequences, two and six events remained before and after the main shock, respectively. We do not compute β value or seismicity ratio with only eight events. Nevertheless, a moderate increase of seismicity rate after the main shock can be inferred without the swarms.
3.2 Seismicity Rate Changes 90 Days Before and After the Main Shock
 Although four HRSN stations experienced gain value changes at ~46 days before the main shock and eight NC stations stopped recording ~31 days after the main shock (Figures S1 and S2), we still investigate whether there were significant differences in seismicity rate changes 90 days before to 90 days after the San Simeon earthquake. After extending the time window, the seismicity rate change pattern remains essentially the same (Figure 13a). In particular, the seismicity rate reduction in subregion A is still significant for most magnitude bands (Figures 13c and 13d). A minor to moderate increase of seismicity rate might be concluded in subregion B (Figures 13f and 13g).
4 Comparisons With the Static Stress Changes
 We calculate the static stress changes produced by the San Simeon earthquake based on an updated finite fault model [Ji et al., 2004]. Because the focal mechanisms of more than 95% of the earthquakes around Parkfield are pure right lateral on the fault within their uncertainties [Thurber et al., 2006], the effects of variant geometry of “receiver” faults can be largely neglected [Toda et al., 2012]. Here we calculate the shear, normal, and Coulomb stress changes at 7.5 km depth in map view (Figure 1) and along the strike in cross section (Figures 7a and 10a) with an effective coefficient of friction of 0.4 [Aron and Hardebeck, 2009]. The 7.5 km depth is chosen to be close to the average depth of the detected events.
 No apparent consistency between the Coulomb stress and seismicity rate changes is found. Instead, static shear stress changes alone explain the behavior of seismicity across our study region. In particular, a clear negative static shear stress (i.e., stress shadow) is found in the creeping section of the SAF (subregion A). We also observe a significant decrease of seismicity rate across all magnitude bands (Figures 11b, 11c, 13c, and 13d) and a stifled aftershock sequence of an Md2.77 earthquake (Figure 11a), consistent with the stress shadow effect. An abrupt recovery at ~30 days after the main shock can be seen for our detected events (Figures 11a and 13b). In comparison, between Middle Mountain and Cholame, the static shear stress increased after the San Simeon main shock, and we find a moderate increase of seismic activity for most magnitude bands (Figures 11d–11f and 13e–13g).
 In this study, we applied the waveform matched-filter technique to detect earthquakes around the Parkfield section of the SAF following the nearby San Simeon main shock. After removing all possible false and duplicated detections, we have detected ~8 times more events than listed in the NCSN catalog. The seismicity rate beneath the creeping section of the SAF showed a statistically significant reduction following the San Simeon main shock, which correlates well with the negative shear stress changes. Interestingly, deep tectonic tremors in that subregion were also stifled for 3–6 weeks and followed by an accelerated recovery from ~30 days after the main shock, which agrees well with estimated duration of the stress shadow effect [Shelly and Johnson, 2011]. However, the recovery of earthquake activity in this study was transient, and the seismicity rate remained low near the end of our study period (Figure S9). The seismicity rate increased moderately in subregion B around the rupture zone of the 2004 Parkfield earthquake (Figures 11e and 11f), where positive static shear stress changes were induced (Figure 10a), suggesting that the San Simeon main shock loaded this section of the SAF. This is also consistent with the triggered creep events near Parkfield on the SAF observed by the USGS creepmeters after the San Simeon main shock [Aron and Hardebeck, 2009] and a minor increase of tremor activity in the lower crust [Shelly and Johnson, 2011]. In subregion C, the turnoff of earthquake swarms is inconsistent with positive static stress changes or oscillatory dynamic stresses. According to the focal mechanism catalog from NCSN, the very first event of the earthquake swarms is a normal faulting event, instead of right lateral (Figure 10b, beach ball). The computed static shear and Coulomb stress increase for a receiver fault with normal faulting in this subregion, which still cannot explain the turnoff. However, the swarms completely stopped about half-day prior to the main shock (Figure 11h). Therefore, while the coincidence of the turnoff of the swarms at the time of the San Simeon earthquake may suggest a causal connection, the Coulomb stress changes do not explain the turnoff.
 Figure S10 shows that the rupture of the Parkfield earthquake ended close to the boundary of positive static shear stress changes [Murray and Langbein, 2006]. Hence, it is tempting to conclude that the negative static shear stress changes from the San Simeon main shock may have stopped the rupture propagation of the 2004 Parkfield event. However, the negative shear stress change also coincided with the creeping section of the SAF, where the far-field tectonic loading is mostly released by creeps and numerous small earthquakes. Hence, the overlapping between static stress changes and the main rupture zone of the 2004 Parkfield event could be just by coincidence.
 Finally, we check whether the observed seismicity rate changes were resolvable using the NCSN catalog alone. In subregion A, we obtain an Mc of 1.5 for the NCSN catalog (Figure S7b). β value and seismicity ratio in subregion A at cutoff magnitude 1.5 are −1.55 and −0.57, respectively (Figure S11). In subregion B, only one and four events listed in the NCSN catalog before and after the San Simeon earthquake during the primary study period, respectively (Figure 10b). Mc and β value cannot be measured due to the small data set. Hence, without newly detected events, one might still conclude the aforementioned seismicity rate change pattern, but statistical significance can no longer be established.
 In summary, the seismicity rate changes at Parkfield correlates well with the static shear stress changes induced by the San Simeon earthquake, suggesting that static stress change is an important, if not the only, agent for near-field triggering in this region. Our results are consistent with a recent finding of a stress shadow in the aftershock zone of the 1992 Mw6.3 Joshua Tree earthquake cast by the 1992 Mw7.3 Landers earthquake [Toda et al., 2012], indicating that stress shadows do exist and may play an important role in changing seismic activities in the near field. Transient dynamic triggering may have occurred in our study region but was not detected by our technique because of the close distance between Parkfield and the San Simeon epicenter and clipping of the stations in the first ~100 s immediately after the San Simeon main shock.
 In the stress shadow region, we found that β values increase with increasing cutoff magnitude (Figure 11b). Hence, one may argue that the seismicity rate drop is simply due to the reduction in detection ability after the main shock, especially for small-magnitude events, because large San Simeon aftershocks could contaminate the continuous data and increase the noise level at the stations. However, it appears that the San Simeon main shock and its aftershocks only cause significant increases of the noise levels at the stations for ~3 days. The median noise level decayed back to preshock level in 3–5 days after the main shock (Figure S1). Moreover, the relationship between the mean CC values and the magnitude of detected events before and after the main shock shows essentially the same pattern (Figures 9b and 9c), suggesting a consistent detection through our study period. Finally, if detection ability indeed lowered after the main shock, we should observe the stifled seismicity rate and β values increase with cutoff magnitude in subregion B as well, which is not the case (Figure 11e). Excluding the possibility of changing detecting ability, β value declining significance with cutoff magnitude could be mainly due to the smaller data set with increasing cutoff magnitude.
 The fact that static shear stress, instead of Coulomb stress, changes were consistent with the seismicity rate changes, as well as the behavior of tectonic tremors [Shelly and Johnson, 2011], argues for a very low friction along the Parkfield section of the SAF through the entire crust. This was also suggested by the laboratory measurement from SAFOD drilling samples [Lockner et al., 2011] and previous study on the response of the Parkfield section of the SAF to large earthquake [Toda and Stein, 2002]. Such fault weakness may attribute to the fact that the mature SAF developed a thick, impermeable gauge zone that reduces the effective coefficient of friction [Felzer and Brodsky, 2005; Zoback et al., 1987].
 The seismic data used in this study are recorded by the High Resolution Seismic Network (HRSN), Berkeley Digital Seismic Network (BK), and Northern California Seismic Network (NC) operated by Berkeley Seismological Laboratory, University of California, Berkeley, and are distributed by the Northern California Earthquake Data Center (NCEDC). We thank David Shelly and Peng Zhao for useful suggestions and Ross Stein for valuable comments. X.M. and Z.P. were supported by the National Science Foundation through award EAR-0956051, the USGS NEHRP program through award G12AP20090, and Georgia Tech's Institute for Data and High-Performance Computing (IDH) Seed Grant.