We perform waveform cross correlation and high precision relocation of both background seismicity and seismicity triggered by periodic slow earthquakes at Kilauea Volcano's mobile south flank. We demonstrate that the triggered seismicity dominantly occurs on several preexisting fault zones at the Hilina region. Regardless of the velocity model employed, the relocated earthquake epicenters and triggered seismicity localize onto distinct fault zones that form streaks aligned with the slow earthquake surface displacements determined from GPS. Due to the unknown effects of velocity heterogeneity and nonideal station coverage, our relocation analyses cannot distinguish whether some of these fault zones occur within the volcanic crust at shallow depths or whether all occur on the decollement between the volcano and preexisting oceanic crust at depths of ∼8 km. Nonetheless, these Hilina fault zones consistently respond to stress perturbations from nearby slow earthquakes.
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.
 Slow (or ‘silent’) earthquakes (SEs) have now been found to occur repeatedly, and in some cases, periodically, in various tectonic settings such as subduction zones, strike-slip faults, and volcano flanks. Elucidating the fault processes and mechanical conditions that yield repeating or periodic slow slip events and associated phenomena such as tremor and triggered microseismicity are the subject of intensive ongoing research and monitoring. The observations and search for theoretical explanations have stimulated provocative hypotheses: for example, Lowry  has proposed that periodicity of subduction zone SEs could arise as a resonant response to climate-driven stress perturbations.
 In the past decade, Kilauea volcano's mobile south flank has been the site of 7 slow earthquakes identified from a continuous GPS (CGPS) network [Cervelli et al., 2002; Brooks et al., 2006; Segall et al., 2006a, 2006b]. Brooks et al.  found that one spatially distinct family of 4 SEs is periodic and separated by 774 ± 7 day periods during 1998–2005, although a subsequent 5th periodic SE failed to occur in the predicted time window in March 2007. Swarms of microearthquakes are triggered by these SEs [Brooks et al., 2006; Segall et al., 2006a, 2006b], with most occurring in a small localized region beneath the Hilina Pali (Figure 1). It is not yet known whether nonvolcanic tremor may also occur.
 The periodic SEs accrue maximum surface displacements of a few cms over several hours to 2 days and have equivalent magnitudes of ∼5.6–5.8 [Brooks et al., 2006]. These are small, but not insignificant, compared to the ∼6–10 cm/yr regional average motions from relatively stable seaward sliding on a decollement extending from a depth of ∼10 km below Kilauea volcano to where it approaches the seafloor offshore [e.g., Owen et al., 2000; Morgan et al., 2000]. A first order question then is: do the SEs occur on the basal decollement in the offshore (updip) region and reflect a transitional region between stick-slip sliding onshore and stable sliding offshore? Cervelli et al.  preferred a shallow (∼5–6 km deep) landward-dipping thrust fault located above the decollement as the source for the Nov. 2000 SE while Brooks et al.  showed that the geodetic data permit many possible SE fault plane solutions ranging from shallow thrust faults, to deeper seaward-dipping normal faults, to subhorizontal decollement planes, although solutions generally place a substantial portion of the fault surface offshore. Segall et al. [2006a, 2006b] used relocations of triggered high-frequency earthquakes of the January 2005 SE, seismicity rate theory, and Coulomb stress modeling to suggest that it occurred offshore on the decollement at ∼7–8 km depth.
 Because the estimated depth of slow slip is difficult to constrain from the geodetic data alone [Brooks and Frazer, 2005; Brooks et al., 2006], the reliability of the Coulomb stress modeling for helping to constrain SE depth critically depends on the accuracy of estimated locations and mechanisms of the triggered seismicity. In their relocations, Segall et al. [2006a, 2006b] did not use the full waveform data; rather, they used a double-difference-derived mapping with manual HVO picks, assuming a 1D velocity model, between triggered events and previous high-precision relocations and tomography from elsewhere at Kilauea [Hansen et al., 2004]. Segall et al. [2006a, 2006b] also assumed a thrust mechanism with the basal decollement. In contrast, prior high precision relocation work by Got and Okubo  using a combination of waveform correlation and handpicked travel time differences interpreted seismicity in this same region (including the events triggered by the Sep. 1998 SE) as occurring below the decollement on a steeply seaward-dipping plane and with a reverse mechanism. These two different interpretations of the seismicity patterns need to be resolved if the triggered events can be used reliably to help constrain SE location and mechanism. Here we perform double difference relocations [Waldhauser and Ellsworth, 2000] using high precision travel time differences from waveform cross correlation in order to help address this issue and to better illuminate the fine-scale characteristics of SE-triggered seismicity at the Hilina region.
2. Waveform Cross Correlation and Relocation Methodology
 We obtained waveform data from the USGS Hawaiian Volcano Observatory (HVO) in the Hilina Pali region of SE-triggered seismicity (Figure 1) for 1203 earthquakes spanning the entire years of 2004 and 2005, as well as for periods of ±1 month around the dates of all identified Kilauea SEs [Brooks et al., 2006]. We focus our study on the limited region where triggered earthquakes predominantly occur on short (∼1 day) time periods around SEs (Figure 1). The cross correlation methodology for measured travel time differences generally followed the procedure by Wolfe et al.  and double difference relocations used the method of Waldhauser and Ellsworth . P and S waves were sliced in windows of ±1.5 s around the predicted arrival time on vertical component seismometers. For cases where the predicted S minus P wave arrival time was less than 1.5 s and the P and S windows would overlap, only the P-wave data were correlated (alternatively excluding these cases was found to have little effect on the results). To reduce outliers from noisy stations, the correlations used only the best 30 stations in the HVO seismic network, as indicated by handpicked travel times (Figure 1a). This process yielded 132,535 travel time delays for 32,109 pairs of earthquakes.
 Tests demonstrate that the assumed 1-D velocity model has a significant influence on the absolute depths of the relocations (Text S1 and Figures S1 and S2 of the auxiliary material). The relative depths between separate clusters of earthquakes are also not well constrained. Our preferred velocity model contains slower-than-typical mid-crustal velocities (Figure S3), as has been observed beneath the Hilina Pali [Hansen et al., 2004; Park et al., 2007], but as discussed below the effects of unknown velocity heterogeneity may have important influence.
Figures 2 and 3 display the results of high precision relocation of 895 earthquakes, along with the patterns of relocated triggered seismicity on the days of 3 slow earthquakes on 1/26/2005, 9/19/1998, and 11/9/2000 (the slow earthquake of 12/16/2002, with smallest magnitude, produced few triggered earthquakes). The triggered seismicity consistently relocates on several distinct clusters, or fault zones, with depths varying from 5 km to 7 km (see auxiliary material demonstrating how the absolute depths can vary with differing 1-D velocity models). While the epicenters for the original HVO locations are diffuse clouds (Figure 2), after relocation, epicenters collapse into several distinct clusters that form streaks aligned nearly parallel to the geodetically defined slip direction of SEs from GPS data [Brooks et al., 2006]. After relocation, the depths of earthquakes also collapse from a diffuse cloud at 4–9 km depth to several compact clusters that are each more localized in depth (Figure 3). The absolute depth of each cluster can vary with assumed velocity model.
 The most active fault zone, and on which triggered seismicity also dominantly occurs, was studied by Got and Okubo , who imaged a strongly southeast dipping fault zone and suggested that this fault is not the decollement but rather a deeper reverse fault with conjugate sense faulting. Our relocations do not image such a dipping fault (Figure 3), but rather result in a horizontally aligned band of earthquakes that tends to be shallow (4–6 km depth). A composite focal mechanism for earthquakes on this fault zone using first motion polarities displays one fault plane with seaward slip on a low-angle plane (Figure 2). These observations suggest that this fault plane may be near horizontal with seaward slip, and not strongly dipping.
 However, we are concerned that poor station geometry as well as velocity heterogeneity may be biasing the absolute depth of our relocations (and may also affect focal mechanism) and our analyses may not be capable of distinguishing between a fault zone located at shallow depths within the volcanic crust or a fault zone located at the decollement interface between the volcano and preexisting oceanic crust (near 8 km depth). Because this region is near the coast, the azimuthal distribution of stations is not optimal (Figure 1) and there are no stations to the south of this region. In addition, nearby station coverage to the east is poor, because station KAE (Figure 1b) was noisy and provided few travel time differences (stations HLP and AHU yielded ∼10,000 travel time difference measurements, POL yielded ∼6,000, whereas KAE yielded only ∼800).
 Strong (as much as ∼15%) velocity heterogeneity observed beneath the Hilina Pali can vary over horizontal distances on the order of 5 km [Hansen et al., 2004], which means that the entire ray paths from the source region to stations outside the Hilina Pali are likely not well approximated by any 1-D velocity model. Because the fault zone of interest spans several kilometers distance, and correlated earthquake pairs link this entire fault zone into a single cluster, double difference relocations have some sensitivity to absolute location [Wolfe, 2002; Menke and Schaff, 2004], but coupling this sensitivity with a velocity model that does not correctly account for strong heterogeneity may possibly lead to erroneous absolute depths. We have recently deployed a temporary network of 12 additional station sites in this region and expect that future work involving double difference tomography [Zhang and Thurber, 2003] will be able to resolve these issues.
4. Discussion and Conclusions
 Although our high precision relocations of triggered seismicity at the Hilina region cannot constrain the absolute depths and whether triggered seismicity occurs on the decollement or within the volcano, earthquake patterns nonetheless collapse into several streaks of seismicity that are aligned parallel to the GPS vectors of SE slip (Figure 2) and have narrow depth extent (Figure 3). Interestingly, the orientations of streaks are generally more closely aligned with the GPS measured surface displacement directions of intermittent SEs than with the surface displacement directions inferred to result from long-term decollement creep (Figure 2). It should be noted that current geodetic models of SEs [Brooks et al., 2006; Segall et al., 2006a, 2006b] indicate that the triggered microearthquakes are too far inland to be directly on the SE fault planes (which mostly occur offshore but may have some onshore component) so that stresses arising from SE slip are the most likely cause for the triggering.
 Inspection of less accurate catalog locations suggests that these Hilina seismicity streaks are features that have persisted for at least 40 years, since the beginning of high quality HVO monitoring for this region in the late 1960s. Similar types of concentrated streaks of microseismicity aligned in the slip direction have been observed in many regions [e.g., Gillard et al., 1996; Rubin et al., 1999]. Streaks are generally interpreted as reflecting heterogeneity in the frictional properties of the fault and the patterns of locked (stick slip/velocity weakening) versus creeping (stable sliding/velocity strengthening) regions. One suggestion is that seismicity streaks are caused by locked regions completely surrounded by larger regions of creep [e.g., Rubin et al., 1999]; another suggestion is that seismicity streaks occur along a boundary between a locked and creeping section [e.g., Sammis and Rice, 2001]. Geodetic analyses to date have given differing results regarding the issue of whether the microearthquake streaks we observe at the Hilina region could reflect the geometry of locked and creeping sections of the decollement. Consistent with this interpretation, Cayol et al.  model geodetic observations between 1975–1983 with the decollement having a creeping zone near the rift zones but a locked zone towards the coast, with the locked regions corresponding to regions of decollement seismicity. But in contrast, Owen et al.  model GPS observations with the decollement having a single large creeping section with no locked regions.
 The underlying cause of this type of structural organization of a fault surface and how it affects rupture dynamics remains a topic of much interest, with modeling studies indicating that the interaction between velocity strengthening and velocity weakening regions is also important in the occurrence of slow earthquakes [Kato, 2004; Liu and Rice, 2005]. Eventual constraint on the depth of Hilina streaks will be key to further understanding their tectonic and frictional implications. For example, if it is later demonstrated that these microearthquakes do occur on the decollement, then why is there a seaward protruding band of earthquakes here, unlike the decollement seismicity to the east that occurs in a narrow onshore strip parallel to the east rift zone (Figure 1)? One possibility is that the Hilina protrusion of seismicity might be associated with a more complex stress field (and stress history) created by the intersection of the east rift and south west rift zones. Another possibility is that the seismicity may be related to a physical anomaly on the seafloor interacting with the overriding plate, such as a seamount, as it underthrusts the Hilina slump.
 Regardless of the tectonic characterization of the Hilina streaks, Figure 2 demonstrates that they respond to the stress perturbations produced by the nearby slow earthquakes. This behavior is consistent with the previous results of Dieterich et al. , who compared rate and state friction stress predictions and independent estimates to demonstrate that Kilauea's decollement seismicity east of our study region is a reliable stress meter of nearby diking events and moderate earthquakes.
 This research was funded by the Geophysics and Petrology and Geochemistry Programs of the US National Science Foundation and the USGS. We thank Jean-Luc Got, an anonymous reviewer, and the GRL editor for their constructive comments.