We show clear evidence of non-volcanic tremor triggered by 2002 Mw7.8 Denali Fault earthquake near Parkfield. Triggered tremor is identified as bursts of high-frequency (∼2–8 Hz), non-impulsive seismic energy whose envelope is coherent among many stations and has the same periodicity as the passing surface waves. The tremor originates from at least three hypocenters near the San Andreas fault with differing frictional regimes, two in the creeping section and the other where the San Andreas is transitional between creeping and locked. All the sources originate below the seismogenic zone, suggesting that transitional frictional properties are necessary conditions for tremor generation. Tremor is excited by the Love waves when the San Andreas is sheared in a right-lateral sense, encouraging slip, and is absent when the San Andreas is sheared in a left-lateral sense, consistent with a simple frictional response to the driving stress.
 Non-volcanic tremor (NVT) is a seismic signal with long durations and no clear body wave arrivals, and with spectra depleted in high-frequency energy compared with regular earthquakes of similar amplitude. NVT was originally identified in a subduction zone southwest of Japan [Obara, 2002]. Subsequent studies have found NVT in many circum-Pacific subduction zones [e.g., Rogers and Dragert, 2003; Schwartz and Rokosky, 2007]. The tremor is often found during episodic slow-slip events, and together they are called as Episodic Tremor and Slip (ETS) [Rogers and Dragert, 2003].
 In this study, we analyze the tremor triggered by the 11/03/2002 Mw7.8 Denali Fault earthquake [Gomberg et al., 2008], and recorded by many stations near the Parkfield section of the SAF (Figure 1). We focused around Parkfield because it is one of the few places outside the subduction zone environment where ambient tremor has been identified [Nadeau and Dolenc, 2005], and there is an unprecedented density of instrumentation there, which promise to reveal details of tremor sources with unprecedented fidelity.
2. Observations of the Triggered Tremor
 We identify triggered NVT as bursts of ∼2–8 Hz, non-impulsive seismic energy whose envelope is coherent among many stations and has the same periodicity as the passing surface waves. Figure 2 shows a comparison of the broadband recordings of the Denali Fault earthquake and the band-pass-filtered seismograms from station PKD. This station contains a Streckeisen STS-2 Seismometer and is part of the Berkeley Digital Seismic Network. The energy associated with the teleseismic P waves of the Denali Fault earthquake is visible up to 3 Hz. In addition, we find bursts of higher-frequency energy during the large-amplitude surface waves. No impulsive body wave arrivals can be identified within most of the bursts. However, the high-frequency bursts are coherent for stations across an aperture of several tens of kilometers, and have hyperbola-type moveout along the SAF fault strike (Figure 3). This pattern indicates that they are not generated by instrumental noise [Hellweg et al., 2008] or dynamic triggering in the near surface [Fischer et al., 2008], nor are they waves generated outside the network, but are produced by locally triggered tremor [Gomberg et al., 2008]. The similarity in spectral shapes between the ambient and triggered tremor supports this conclusion (Figure S1). Both NVT types are deficient in high frequencies relative to regular earthquakes. The triggered tremor is larger in amplitude by a factor of ten than typical ambient tremor, consistent with triggered tremor on Vancouver Island [Rubinstein et al., 2007].
 We locate the triggered NVT with a grid search over possible locations. This is the same method used by Rubinstein et al. , and is similar to that used by Obara . In detail, we calculate cross-correlograms of envelopes from pairs of 2–8 Hz band-pass-filtered vertical-component seismograms. As has been the case for tremor in previous studies [Obara, 2002], we find the NVT has a moveout with distance matching the S-wave velocity (Figure S2). For each trial source location, we calculate the S-wave travel time difference for each station pair and determine the value of the correlation at that predicted lag time. We find the location that maximizes the sum of weighted correlations using an L-1 norm [Wech and Creager, 2007]. The S-wave velocity model is computed from a 1D P-wave velocity Vp model used by Waldhauser et al.  by assuming the Vp/Vs ratio of 1.732. We have also tried a “local” Vp/Vs ratio of 1.78 above 8 km, and 1.732 below 8 km (H. Zhang, personal communication, 2007). This result in minor decrease of the depth, but virtually no effect on the epicenter.
 We identify two strong tremor source regions near Parkfield and a possible weak source region further north (Figure 1). The tremor moveout south of Parkfield is well explained by a single tremor source near Cholame at (−120.28° ± 5 km, 35.68° ± 5 km), where the SAF slip behavior is transitional between creeping and locked, and where ambient tremor had already been found [Nadeau and Dolenc, 2005]. The tremor recorded by stations north of Parkfield is more complicated, suggesting the possibility of multiple sources in that region. A strong tremor source is located at (−120.84° ± 5 km, 36.36° ± 10 km) in the creeping section of the SAF between Monarch Peak and Bitterwater (Figure 1). We also find evidence of a weak tremor source occurring around stations BAV, BSG, and BSM north of Bitterwater. The tremor peaks shown in the envelope functions for these stations arrive several seconds earlier than those for stations (e.g., PHR, PJU, PPT) further south along the SAF (Figure 3). However, the absolute amplitudes of tremor bursts (Figure S3) for these stations (BAV, BSG, and BSM) are 2–3 times smaller than those stations in the south (e.g., PHR, PJU, and PPT). Hence, we suggest that a weak tremor was activated earlier and recorded only by stations BAV, BSG, and BSM, while the strong tremor occurred slightly later and was recorded by many stations north of Parkfield. Because of this, we will focus in the following sections on the two strong tremor sources south of Parkfield near Cholame, and north of Parkfield between Monarch Peak and Bitterwater.
 A few details warrant additional comment. First, it is worth noting that we did not constrain the number of source regions in our location procedure. Instead, we find from the misfit maps (Figure S4) that two strong source regions are sufficient to explain the moveout of tremor envelopes observed around Parkfield (Figure 3). The only exception is the aforementioned tremor observed near stations BAV, BSG, and BSM, which could be produced by an additional weak tremor source. As was found in previous studies, the tremor source depths are less well constrained. The best-fitting depths for the southern and northern tremor sources are 24 ± 10 km and 15 ± 10 km, respectively. Finally, our location procedure does not resolve differences between the epicenters of individual tremor pulses, although from difficulty in matching the entire envelope at all the stations, we suspect the source regions fill some volume. Attempts to analyze individual pulses separately have not yielded well-resolved location differences.
3. Correlations Between Tremor and Surface Waves
 Next, we align the tremor with the surface waves of the Denali Fault earthquake in Figure 4. This involves two corrections. We must shift (1) the surface waves to account for the time between the surface waves passing the reference broadband station PKD and the source region, and (2) the tremor envelopes to account for the shear wave travel time to the station from the source region.
 For the first correction, we shift the surface waves by an average phase velocity of 4.1 km/s, measured from seismograms recorded by nearby broadband stations (Figure S5). This has uncertainty of ±1 s due to dispersion of the surface waves, but this is similar to the potential errors from the depth uncertainty. As the phase of the surface wave varies little with depth, we do not make a depth phase correction.
 For the second correction, the time shift for the tremor envelope is calculated from the 1D velocity model and trial locations. We normalize the time-shifted tremor envelopes observed at the 7–8 nearest stations, and stack them to produce the tremor source functions for each source region (Figure S2). We observe good agreement of the phasing of the NVT with the first eight Love wave cycles for the southern region, and a less precise match for the first eight tremor peaks for the northern region (Figure 4).
 The fundamental-mode Love surface waves exert strong right- and left-lateral shear stresses resolved on the SAF. The near-coincidence of the strike of the SAF fault and the great-circle path from Alaska result in the along-strike stress being proportional to the Love wave velocity. This is because the along-strike stress on a vertical strike-slip fault aligned with the propagation direction arises from radial-direction gradients in the Love wave particle displacement [e.g., Hill, 2008], which for a single frequency is equal to the ratio of the particle and Love wave phase velocities and 180° opposite in phase. In our case of waves traveling southeast, it is the velocity to the southwest that encourage NVT by boosting right-lateral stress, and velocity to the northeast that discourage it, and this is the observed correlation. In comparison, the normal stress is mainly associated with Rayleigh waves, which is proportional to a linear combination of the vertical displacement and radial velocity. Since we do not observe clear correction between the tremor envelope with the Rayleigh wave velocity (Figure 4) and displacement (Figure S6), we argue that the tremor is mainly caused by shear stresses induced by the passage of the Love waves.
 We estimate the amplitude of the shear stresses associated with the Love waves encouraging slip on the plate interface to be on the order of 10 to 20 kPa. This extrapolation from the Love wave velocity amplitude at the surface to the stress amplitude at the tremor source depth requires knowledge of the decrease in Love wave amplitude and the increase in rigidity with depth, which we compute using a locked-mode surface wave code [Gomberg and Masters, 1988] and a 1D velocity model around the Parkfield region.
Figure 1b shows a comparison of seismicity and tremor locations around the Parkfield section of the SAF, which straddles the transition between the creeping segment of the fault to the northwest and the locked segment to the southeast. These differing ambient slip behaviors likely reflect different frictional properties within the seismogenic zone, above ∼10–15 km as inferred from the maximum depth of microseismicity. Apparently these differences have little impact on tremor generation. Our best fitting depth for both tremor sources is ∼15–24 km, below the seismogenic zone and close to the depth range of 20–40 km for the ambient tremor [Nadeau and Dolenc, 2005]. This suggests tremor generation requires transitional frictional properties as well as other conditions found at depths below where earthquake typically occur. Possible conditions include temperature, pressure, rock type, frictional properties, and pore-fluid pressures. At the current stage, it is still not clear what are the necessary conditions for tremor generation.
 This study presents several complications not seen in case of Vancouver Island [Rubinstein et al., 2007]. The first complication is that tremor burst amplitudes are not proportional to the corresponding Love wave amplitudes, rather the tremor peaks grow over time while the peak stresses change much less. This is true for both the northern and southern source regions, with the southern source particularly anemic at the onset. The second complication is that the NVT does not persist as long as the surface waves. The northern source dies out after just 6 cycles, the south lasts longer, but both fade away before the strong surface waves end. A third feature is a strong stage of tremor between 1200 and 1300 s. Based on the misfit map (Figure S4), it appears that the later arriving pulses of energy originated from similar locations to the earlier triggered tremor. However, it is not clear what causes the break in the tremor activity as the surface waves were still ongoing during that time period.
 So far, triggered and ambient tremor for both subduction and strike slip regimes have much in common, and do not yet require any differences in mechanism. This mechanism, though, remains unknown. The relatively low triggering threshold (on the order of a few tens of kPa or less), together with recent observations of tidal modulation [Shelly et al., 2007; Rubinstein et al., 2008; Nakata et al., 2008], suggests that NVT is very sensitive to external stress perturbation. We note that NVT is not triggered everywhere by large teleseismic events, indicating that other conditions are needed for their occurrence. Triggered NVT observations in diverse tectonic environments will not only help to better quantify the triggering mechanisms and necessary conditions, but also improve our understanding of fundamental faulting processes.
 The data used in this study come from the High Resolution Seismic Network operated by Berkeley Seismological Laboratory, University of California, Berkeley, the Northern California Seismic Network (NCSN) operated by the U.S. Geological Survey, Menlo Park, and are distributed by the northern California earthquake data center. We thank David Oppenheimer and Doug Neuhauser for extracting the high-sampling rate NCSN data used in this study, and Bob Nadeau for sharing his ambient tremor catalog with us. We also thank the editor and two anonymous reviewers for the constructive comments. The study was supported by the National Science Foundation (grant EAR-0809834: ZP; grant EAR-0809993: JEV).