At many plate boundaries, conditions in the transition zone between seismogenic and stable slip produce slow earthquakes. In the Cascadia subduction zone, these events are consistently observed as slow, aseismic slip on the plate interface accompanied by persistent tectonic tremor. However, not all slow slip at other plate boundaries coincides spatially and temporally with tremor, leaving the physics of tremor genesis poorly understood. Here we analyze seismic, geodetic, and strainmeter data in Cascadia to observe for the first time a large, tremor-generating slow earthquake change from tremor-genic to silent and back again. The tremor falls silent at reduced slip speeds when the migrating slip front pauses as it loads the stronger adjacent fault segment to failure. The finding suggests that rheology and slip-speed-regulated stressing rate control tremor genesis, and the same section of fault can slip both with and without detectable tremor, limiting tremor's use as a proxy for slip.
Slow earthquakes, where a fault interface can relieve stress in discrete, but stable episodes of slip, play a fundamental role in accommodating plate convergence in the transition zones of plate boundaries worldwide [Schwartz and Rokosky, 2007]. These events typically occur adjacent to seismogenic parts of the same fault and may increase the stress on these locked regions with the potential to trigger large earthquakes as has been suggested for the Tohoku megathrust earthquake [Kato et al., 2012; Ito et al., 2013]. Monitoring where, when, and how much slip occurs provides insight into the state of stress at the base of the seismogenic zone, improving long-term and time-dependent hazard assessments. In the Cascadia subduction zone, where the Juan de Fuca plate subducts beneath North America, slow earthquakes occur as repeating episodes of geodetically observed slow slip coincident with seismically observed tectonic tremor, together called episodic tremor and slip (ETS) [Rogers and Dragert, 2003]. Though tremor only represents a tiny fraction of the total moment release [Kao et al., 2010], detailed studies of the tremor source process [Shelly et al., 2007; Wech and Creager, 2007; Bostock et al., 2012] and the spatiotemporal correlation of slip and tremor [Hirose and Obara, 2010; Bartlow et al., 2011] suggest that tremor is the seismic signature of slip at the fault interface. Thus, the well-established correlation between tremor and slip in Cascadia enlists tremor as a powerful tool for mapping out slip distributions and cataloging slip history because of the higher spatial and temporal resolution of tremor observations, compared with direct geodetic observations of slip.
But these two phenomena do not always coincide. Strike-slip boundaries show no surface deformation concurrent with deep tremor [Shelly and Johnson, 2011; Wech et al., 2012], and there is a range of smaller tremor swarms that occur in Cascadia without any associated geodetic signal [Wech et al., 2010]. These shorter duration swarms and tremor on strike-slip faults are interpreted to represent shear slip below the threshold for geodetic detection, but the factors controlling tremor genesis are less understood. Though not previously observed in Cascadia, some subduction zones have slow slip events that do not produce tremor. Slow slip beneath the Boso peninsula in Japan and offshore Cape Turnagain in New Zealand have no observed tremor but are accompanied by swarms of seismicity [Ozawa et al., 2007; Wallace et al., 2012]. Similarly, long-term slow slip events in Japan, New Zealand, and Mexico have little to no accompanying tremor signals, or tremor and slip are not colocated [Brudzinski et al., 2010; Hirose et al., 2010; Ide, 2012]. Network limitations occasionally play a role [e.g., Wech et al., 2012], but many tremorless slip events occur within networks capable of tremor detection [Hirose and Obara, 2005]. Furthermore, the type of slip is not always consistent enough to make direct comparisons. Variability in slip depth, size, and duration likely reflects distinct static plate interface properties among the different fault zones [Peng and Gomberg, 2010]. In this study we control for both static interface properties and seismic network by using data from the same network to record a Cascadia ETS event that slipped both with and without tremor on the same fault patch, allowing a direct comparison of tremorgenic and tremorless slip.
2 Adjacent ETS Events
We focus on tremor and slip activity in 2011 when two ETS events occurred within weeks and ~50 km of each other (Figure 1). The central section of the Cascadia subduction zone, where ETS repeats every ~20 months [Brudzinski and Allen, 2007], ruptured first. On 4 June 2011, approximately 21 months after the 2009 ETS, tremor and slip initiated at the downdip edge of the ETS zone in northern Oregon (Figures 1 and 2), similar to previous ETS events [Wech, 2010]. Epicenters migrated updip over the next 5 days (Figure S1 in the supporting information) before partitioning and extending both north and south along strike. The southern tremor front stopped in central Oregon on 15 June, but the northern front continued its ~8 km/day along-strike migration 150 km before stopping in southern Washington on 3 July (Figure 1).
Just 21 days later and ~50 km further north, tremor was detected beneath central Washington near the updip edge of the ETS zone on 24 July. This tremor burst marked the beginning of the northern Cascadia ETS event 10.5 months into its usual 13–16 month cycle [Miller et al., 2002; Rogers and Dragert, 2003]. The tremor continued north along strike for the next 42 days before terminating beneath Vancouver Island, and partway through this event, a south-migrating tremor front developed, filling in the gap created between the two ETS events (Figure 1).
3 Tremorless Slip
Tremor is routinely detected across the entire subduction zone using automated methods at the Pacific Northwest Seismic Network [Wech, 2010], but the inherent limitations of unchecked automation present numerous opportunities for false negative results. We therefore perform a manual tremor search by visually inspecting bandpass-filtered waveforms and envelopes of all available seismic data in southern Washington during the 3 week quiescence between the two ETS events. We identify some continued minor tremor on 4 July at the northern edge of the Oregon ETS event, but we otherwise find no evidence for undetected tremor. Because slow slip has never been seen without tremor in Cascadia, this spatiotemporal gap in tremor would suggest an absence of slip as well. However, the relative timing and along-strike migration rate of the tremor epicenters in both events (Figure 1b) invites a causal connection between the events—perhaps tremorless slip continued in the gap between the two ETS events as observed in Shikoku [Obara et al., 2011].
To investigate the possible presence of slip, we perform a 90 day time-dependent inversion of daily GPS solutions from the Pacific Northwest Geodetic Array. Following the approach of Bartlow et al. , we invert these data using the Network Inversion Filter (NIF) [Segall et al., 1997] to obtain a time-dependent model of slip and slip rate with a nonnegativity constraint applied to slip rate [Segall et al., 1997; Miyazaki et al., 2006; Bartlow et al., 2011]. Slip is assumed to occur on the plate interface (as defined by McCrory et al. ) on a mesh of triangular dislocations in a homogenous elastic half space [Thomas, 1993]. Similar to the previous Oregon ETS event [Bartlow et al., 2011], the results indicate a spatiotemporal correlation between high slip rate and the independently identified tremor epicenters for both ETS events (Figures 2a and 3). And during the 3 weeks between events the model identifies 13 days of stationary, tremorless slip occurring with low slip speeds at the northern ETS initiation location. Slip and slip rate resolution in the NIF depend on multiple factors including the propagation rate of slip, size of slip patch, and duration of slip. For the observed stationary slip patch, signal-to-noise ratios are approximately 2 and 1.4 for slip rate and slip, respectively.
The model also produces a spatiotemporal gap in slip and, having approached the lower resolution limits of the NIF model (Figures S2–S5), it is unclear from GPS data whether slip migrated to this initiation point or the events remain separate. Slip duration from a moving source with low slip speeds could be too short to produce resolvable displacement. To address this question, we turn to borehole strainmeter data. Choosing the only two working stations near the gap ahead of the Oregon ETS slip front (Figures 1 and 2), we compare the observed updip and downdip strain history with the tremor catalog and slip model (Figure 2b). Both of these stations have recorded strain from slip in the gap in 2013 (Figure S6). We focus on the εEE component of shear strain where positive and negative strain corresponds to extension and compression, respectively, in the strike-perpendicular direction. Both stations identify clear strain beginning between the two ETS events. The observed downdip extension and updip compression are consistent with reverse slip occurring in the tremor gap, and the first onset of strain occurs on 4 July before the NIF identifies slip.
The presence of tremorless slip is also supported by the updip location of tremor after the gap (Figure S1). Tremor from large and small events in Cascadia initiates deep and migrates updip prior to along-strike migration [Wech and Creager, 2011]. Tremor from the 2011 Washington ETS event, however, initiates at the updip edge of the ETS zone and immediately begins an along-strike migration, supporting the notion that updip slip was already underway.
We interpret the seismic, geodetic, and strainmeter data to mean that low-level slip continued its northward migration from one ETS to another. This interpretation requires a tremorless continuation of low slip rate slip that eventually stops migrating and slips in place prior to the Washington ETS event (Figure 2).
Tremorless slip is observed elsewhere [Ozawa et al., 2007; Brudzinski et al., 2010; Hirose et al., 2010; Ide, 2012; Wallace et al., 2012], and its absence has been hypothesized to reflect differences in the rheological properties of the respective plate boundaries. The fact that a single event can alternate between tremorgenic and silent is therefore surprising. And what makes this observation truly remarkable is that we know this fault segment to be tremorgenic, so the absence of tremor cannot be explained by along-strike fault heterogeneity. Not only has this patch of fault generated tremor in prior ETS events (Figure 4), it does so later within this same event as part of a back-propagating slip pulse (Figures 1 and 2), and it even served as the initiation point for the 2009 ETS event [Bartlow et al., 2011] (Figure 4).
4 Physical Models of Tremorless Slip
Understanding the connection between tremor and slip is fundamental to estimating the significance of tremor—or lack thereof—occurring adjacent to locked fault regions. The fact that slow slip is tremor-genic on some faults and silent on others provides clues to the tremor generation process. But narrowing down the variables is difficult because the plate interface conditions and observational capabilities differ everywhere. By focusing on the same fault patch with the same instrumentation, we effectively have a controlled experiment and can limit our variables to what we can infer from the data. The observations show both a pause in along-strike migration and a pronounced correlation between slip rate and tremor production. We propose a model in which tremor genesis is controlled by slip rate, and the slip rate and migration characteristics are controlled by local stress conditions.
We infer these relative local stress conditions from recent slip history. The tremor gap coincides with the segmentation boundary region between central and northern Cascadia. ETS recurs with 13–16 month intervals north of here and 20 month intervals south of here [Brudzinski and Allen, 2007], but events propagate to and sometimes through this boundary region from either side. The previous northern ETS migrated through to the southern edge of the boundary region (and observed gap) just 10.5 months prior (Figure 4b). We suggest that when the 2011 central Cascadia ETS arrived at this boundary region from the south, therefore, it encountered a fault segment that was not critically stressed and not ready to fully rupture.
Slip occurring in this understressed region would result in lower stress drops relative to large ETS-like slip. Because the ratio of propagation speed to slip speed is proportional to the peak-to-residual stress drop [Ida, 1973]
where vprop is the migration velocity of the slip front, vslip is the slip speed, μ is the shear modulus, α is constant that depends on the near-tip stress distribution, and Δσp − r is the peak-to-residual shear stress, a combination of low stress-drop slip together with a stalled slip front unable to advance could explain the decrease in slip speed [e.g., Rubin, 2011]. Eventually, this persistent, low slip rate slip in the tremor gap stressed the adjacent Washington ETS fault segment to failure, and the high stress-drop failure resulted in an increase in slip rate, and the slow earthquake advanced along strike with typical ETS behavior.
Fluids are thought to play a major role in enabling slow slip to occur by reducing the effective stress with elevated pore pressures [Audet et al., 2009]. A time-dependent pore pressure model could adjust the effective stress, thereby clamping and unclamping the fault, but there is no obvious mechanism for permeability or pore pressure changes on this timescale.
One option to explain the tremor and slip-speed correlation is that the rate of slip of the asperity itself controls tremor, but we prefer a model in which tremor genesis is controlled by the local stressing rate, which is regulated by the slip rate and stress conditions on the surrounding fault. In either case, our observations cannot distinguish between a lower stressing rate limit for tremor genesis or tremor amplitude scaling with stressing rate. Visually apparent tremor can be ruled out despite favorable noise levels (Figure S7), but whether the low slip-speed slip is truly aseismic is difficult to definitively say. Future studies using low-frequency earthquakes [e.g., Shelly et al., 2007] could help verify the presence or absence of a subtler seismic signature and elucidate the exact mechanism, but what is clear is that slip rate regulates tremor genesis.
Our observations neither validate nor eliminate the above models, but our preferred model best explains the coincidence of tremor quiescence with low slip rate, stationary slip, and the subsequent secondary Washington ETS. The fact that we control for rheology does not rule it out entirely, especially since equally high or higher slip rate observed elsewhere is not necessarily colocated with detected tremor [Hirose and Obara, 2005; Ozawa et al., 2007; Brudzinski et al., 2010; Wallace et al., 2012], but it does mean that site-specific rheologies alone cannot explain the occurrence of tremor. Based on our observations, slip speed and tremor rate are causally connected, but the slip rate, fault rheology, and stress conditions are all important factors in generating tremor.
Silent slip on a tremorgenic fault also means that the absence of tremor at other plate boundaries is not necessarily the result of data quality or undersampling. But perhaps a more important result is the effect such an observation has on our interpretation of observed tremor. While it is still likely true that tremor activity serves as a slip indicator, confirmed tremorless slip should discourage tremor's use as a slip meter. That is, if tremor, then slip is still true, but if slip, then tremor is not. In areas such as Cascadia, this means that tremor monitoring may not suffice for detecting transient events, tremor-based interpretations require caution, and discrepancies between tremor and slip distributions may be physical. This latter caveat is of particular interest in interpreting fault behavior closer to the seismogenic zone where slip distributions in Cascadia extend further updip than tremor [Wech et al., 2009]. This updip region may be devoid of tremor asperities altogether. But if not, perhaps updip tremor asperities are even stronger than their ETS zone counterparts and have not yet failed. Or the updip stressing rate has so far been too low to generate tremor. Either case emphasizes the need for routine monitoring, because updip tremor could signal even higher stresses transferred to the locked seismogenic zone.
We are grateful for discussion and suggestions from Paul Segall, Evelyn Roeloffs, Stephanie Prejean, and Joan Gomberg. We also thank Andrew Bradley and Yo Fukushima for contributing the code used to invert geodetic data and Kazushige Obara an anonymous reviewer for suggestions.
The Editor thanks Kazushige Obara and an anonymous reviewer for their assistance in evaluating this paper.