Segmentation in Episodic Tremor and Slip All Along Cascadia

Michael R. Brudzinski 1, Richard M. Allen 2 1 Miami University, Oxford, OH ([email protected]) 2 University of California, Berkeley, CA ([email protected])

Submitted to Geology, January 24, 2007; Revised May 2, 2007

The recent discovery of Episodic Tremor and Slip (ETS) in subduction zones is based on slow slip episodes visible in GPS observations correlated with non-volcanic tremor signals on seismometers. ETS occurs just inboard from the region capable of great megathrust earthquakes, however, whether there is any communication between these two processes remains unknown. In this study we use new single- station methods to compile an ETS catalog for the entire Cascadia subduction zone and compare the patterns with a variety of along-strike trends for the subducting and overriding plates. Correlated ground vibrations and strain observations are found all along the subduction zone, demonstrating ETS is an inherent part of the subduction process. There are 3 broad (300-500 km), coherent zones with different recurrence intervals (14±2, 19±4, 10±2 months), where the interval duration is inversely proportional to upper plate topography and the spatial extent correlates with geologic terranes. These zones are further divided into segments of ETS that occur at times typically offset from each other. The 7 largest (100-200 km) segments appear to be located immediately landward from fore-arc basins interpreted as manifestations of megathrust asperities, implying there is a spatial link between ETS and earthquake behavior. It is not yet clear if any temporal link exists, but the regional “hold time” between ETS episodes could be controlled by strength variations due to composition of geologic terranes.

As oceanic plates subduct down into the mantle, friction on the interface with the overriding plate causes stick-slip behavior in the megathrust zone, pulling the upper plate down until it pops back up during a potentially devastating earthquake (Ruff and Kanamori, 1983). Recent observations have also revealed slow slip episodes (SSE) that occur regularly on parts of the deeper plate interface with motion indicating release of accumulated strain (Dragert et al., 2001; Lowry et al., 2001; Ozawa et al., 2002). Their frequency and amount of slip (Mw ~6.5-7.5) (Kostoglodov et al., 2003; Melbourne et al., 2005) imply they are a substantial portion of the interplate deformation budget. The duration of these episodes is much greater than earthquakes, yet they are accompanied by weeks of non-volcanic tremor (NVT) (Obara, 2002; Rogers and Dragert, 2003; Szeliga et al., 2004). As such, they represent another section of the strain rate continuum between earthquake and geologic time scales. Processes that govern ETS or potential relationships to major earthquakes and local geology remain unknown, although ETS has been proposed to impact the likelihood of megathrust earthquakes (Mazzotti and Adams, 2004). Previous observations of ETS in Cascadia have focused on southern Vancouver Island, northern Washington, and northern California given the density of geophysical observatories (Fig. 1). When observed, SSE representing weeks of transient

1 displacements are visible on GPS stations within a few hundred kilometers of the trench and show westward motions back toward the trench (Dragert et al., 2001; Melbourne et al., 2005). Corresponding NVT composed of relatively small, non-earthquake signals are most prominent in the 1-10 Hz frequency band and appear to occur in a region from the plate interface into the upper plate (Kao et al., 2005; McCausland et al., 2005; Obara, 2002). ETS was first established on southern Vancouver Island with regular recurrence at ~14 months (Fig. 2) (Rogers and Dragert, 2003). Prevalent instrumentation in Northern California has detected similar ETS signals at the other end of the subduction zone, but revealed a shorter recurrence interval of ~11 months (Fig. 2) (Szeliga et al., 2004). Previous observations from Oregon are sparse, with a few seismic stations in southern Oregon showing a period of NVT in 2005 (McCausland et al., 2005), and one GPS station in central Oregon showing SSE with a longer recurrence interval (Szeliga et al., 2004). This initial set of observations raises some important questions: 1) Is ETS a phenomenon present throughout the entire subduction zone and thus an inherent part of the subduction process? 2) What generates the spatial variability in recurrence rates of ETS despite their temporal regularity in a given region? 3) Is ETS linked in any way to properties of the megathrust seismogenic zone? Each of these questions requires a broad set of observations in space and time. We utilize a new set of ETS information generated by automated identification of NVT and SSE at individual GPS and seismic stations that circumvents the need for dense networks. These methods are not replacements for network solutions, they are simply used as surrogates to perform a uniform investigation of ETS over the entire subduction zone while network density is still quite heterogeneous. For GPS data, we identify SSE by applying a hyperbolic tangent fit (Larson et al., 2004; Lowry et al., 2001) over a scrolling window of the publicly available PANGA time series. An f-test confirms when the fit is better than a linear fit at 99% confidence, and a threshold value for the transient displacement marks events that are larger than background noise. For seismic data, we calculate a time series from the mean amplitude of filtered envelope seismograms for each nighttime hour recorded at individual stations, instead of using a station network to judge hours when NVT is present (Obara, 2002; Rogers and Dragert, 2003). After a moving average and normalization, large peaks rising above background noise at 99% confidence are identified as periods during which NVT dominates. See data repository for details on automated data processing. Figure 2 summarizes results from 7 different segments throughout Cascadia. For each region a typical seismic trace indicating the times of NVT, and a GPS time series showing the times of SSE are plotted. In regions with previous ETS estimates (McCausland et al., 2005; Rogers and Dragert, 2003; Szeliga et al., 2004), our results are in good agreement with timing of ETS (e.g., Fig. 2C and data repository). More important, NVT that correlate with SSE are apparent in several new locations along the subduction zone, particularly along central Cascadia despite more limited observatories (Fig. 2D–F). Corresponding seismic and GPS data availability ranges from 1 to 8 years, with 30 stations reporting SSE and 55 stations reporting NVT (Fig. 1), of the over 300 stations that have been investigated with our automated techniques (Brudzinski and Allen, 2006; Holtkamp et al., 2006, ms. in prep.). It is clear that ETS occurs along the entire subduction zone, meaning that localized geological conditions special to a particular site are not controlling factors that prohibit ETS. This finding is supported by a

2 growing number of observations in other subduction zones showing NVT source locations and displacements from SSE over broad regions (Brudzinski et al., 2007; Hirose and Obara, 2006), ruling out hypotheses that require particular along-strike variations to produce ETS (Mitsui and Hirahara, 2006). While ETS is observed throughout the Cascadian subduction zone, the characteristics vary coherently along-strike revealing clear segmentation in the recurrence interval and relative timing of ETS events. First, there are 3 broad geographic zones with different recurrence intervals of ETS (Fig. 1). For example, in the Siletzia Zone in the central part of Cascadia, data from COR extend the history of NVT back by a factor of two to 1989 (Fig. 2E), giving a robust estimate of 19±3 months for the recurrence interval at this station. The average interval across the Siletzia Zone (19±4) is longer than those observed on Vancouver Island to the north (14±2) and is nearly twice as long as that from California to the south (10±2). The broader geographic extent of our ETS measurements relative to previous studies allows us to identify that a coherent Wrangellia Zone extends from northern Vancouver Island down to ~47.5° N, and that a Klamath Zone extends up from the southern end of the subduction zone to ~42.8° N (Fig. 1). This pattern of recurrence intervals is not tied to the overall rate of subduction which drives the earthquake cycle as a whole. Overall convergence velocities decrease slowly from the north to the south (Fig. 1) (DeMets et al., 1990), while the longest recurrence interval occurs in the middle of the subduction zone. Even if one considers the Oregon forearc as a separate rotating microplate (e.g., Wells et al., 1998), the impact on convergence rates would be gradual and not aid in matching the coherent patterns in recurrence intervals. We also find the 3 zones of relatively uniform recurrence intervals cannot be explained by age of the subducting plate, implying along strike variations in ETS are not due to temperature changes. Since age of oceanic lithosphere is proportional to distance from the mid-ocean ridge and the thermal state of oceanic plate is proportional to the square-root of its age (Parsons and Sclater, 1977), one would look for ETS recurrence to be related to distance from the ridge. However, observations along Vancouver Island and northernmost Washington run nearly perpendicular to the Explorer ridge, with station-to- ridge distances varying from 300-700 km (c.f., ~500 km in central Oregon). This means the age of oceanic plate beneath the Wrangellia zone should vary rapidly, causing both warmer and colder plate than that below central Cascadia where the longest recurrence dominates. Also, boundaries between the 3 main ETS zones occur ~100 km north of expected age/temperature boundaries in the subducting plate if the existing fracture zones are extended beneath the continent. We suggest that the recurrence interval of ETS is related to properties of the overriding continental plate instead of the subducting oceanic plate. Age and temperature of the subducting plate likely has some impact on generating ETS, because initial work has shown SSE and/or NVT are prominent in other young, warm subduction zones like southwest Japan and Mexico (Hirose and Obara, 2006; Larson et al., 2004; Lowry et al., 2001; Obara, 2002). Yet the oceanic plates subducting beneath Cascadia are relatively uniform compared to the heterogeneity of the continental plate they dive beneath. In fact, the central Siletzia Zone with an ~18 month recurrence interval corresponds to the relatively low lying and young Coastal Range Block of central and northern Oregon and southern Washington (mostly thick Siletzia terrane) (Trehu et al., 1994). The shorter-

3 recurrence interval zones to the north (Wrangellia) and south (Klamath) correspond to older Pre-Tertiary blocks with higher topography consisting of a mélange of old oceanic material with later silicic intrusion in a continental environment (Harden, 1998; Jones et al., 1977). Figure 3A shows how ETS recurrence intervals are inversely proportional to onshore fore-arc topography. Correlation of these continental blocks with along-strike patterns of ETS is also consistent with the observation that NVT appears to occur throughout the continental crust at depths above the interface with the subducting oceanic crust (Kao et al., 2005). The zones of spatially coherent recurrence intervals (Fig. 1) are further divided into segments where individual events recur over roughly the same location. While the average recurrence intervals of ETS are similar within a given zone, the relative timing between ETS events shows variation with location (Fig. 2), a phenomenon that is particularly clear when comparing northern and southern Vancouver Island (Fig. 2A, 2C) (Dragert, 2004). The extent of these segments is now emerging from the increased number of ETS observations. Figure 3B illustrates the phase shift in time between different segments by displaying the timing of ETS observations all along Cascadia, with horizontal lines estimating the along-strike extent of a given episode. Since these are station locations instead of source locations, we would expect the grey lines to extend on the order of 50 km beyond the actual source locations. Dashed vertical lines are approximate boundaries defined by events on either side that are separated in time by over a month for greater than 50% of the episodes. We find 7 large segments with along- strike widths of 100-200 km (Fig. 3B). The long-term timing of events for each segment is also illustrated by representative GPS and seismic data pairs (Fig. 2 and data repository). Close inspection of NVT time series for the 2003 ETS event that appears to cross segment boundaries reveals distinct offsets in time of 1-2 weeks precisely at the proposed boundaries (Fig. 2 inset and data repository). The largest segments of ETS occur immediately landward from the proposed locations of asperities on the Cascadia megathrust (Wells et al., 2003). The asperity locations are based on large, low gravity, sedimentary basins in the forearc that have been interpreted to indicate potential seismogenic segmentation at depth. This inference is from global surveys finding most of a megathrust earthquake’s seismic moment and area of high coseismic slip (asperities) occur beneath the deep-sea forearc basin features from gravity, bathymetry, and seismic profiling (Song and Simons, 2003; Wells et al., 2003). Figure 3B shows the along-strike pattern of prominent fore-arc basins for comparison with the spatial extent of ETS segments (We find a stronger correlation using locations of fore-arc basins based on bathymetry than on gravity). The apparent correlation between segmentation of the seismogenic zone and segmentation of the ETS zone suggests that effects of locking (or lack thereof) on the megathrust are transmitted to greater depths where slow slip is believed to occur (Dragert et al., 2001). This spatially links megathrust structure and anticipated seismogenic behavior with ETS characteristics. A remaining question is whether upper plate structure controls plate interface behavior or vice versa. Both models have been proposed for fore-arc basins, with either basins developing in response to locking on the subduction interface (Song and Simons, 2003; Wells et al., 2003) or thickness of the upper plate critical wedge controlling the frictional behavior on the plate interface (Fuller et al., 2006). For ETS recurrence, the accreted terranes comprising the upper plate above ETS generate inherently sizable

4 along-strike variations in structure, composition and age that are presumably more significant than long-term effects of ETS on upper plate structure. This supports an interpretation where variations in the Wrangellia, Siletzia, and Klamath blocks control behavior of the ETS source zone. A clue to how continental blocks could be responsible for differences in ETS recurrence is geochemical evidence that the different terranes have different fluid content (Schmidt and Grunder, 2006), which is a possible catalyst for ETS (Kodaira et al., 2004; Obara, 2002). An intriguing hypothesis is that different terrane composition affects rheology of the upper plate and hence the plate interface (Kohlstedt et al., 1995). For example, the Siletzia terrane would represent denser, stronger, more oceanic-like crust, while the Klamath terrane represents lighter, weaker, more continental-like crust. Such a scenario would suggest that the low-lying Siletzia region has a longer recurrence interval because the upper plate has the strength to accumulate strain for longer periods between SSE. Although it is not yet clear whether rate- and state-dependent friction processes are the best way to explain ETS (Chen and Brudzinski, 2006), initial laboratory fault-sliding experiments suggest variable fluid pressure and rock composition would both be expected to generate coherent variations in recurrence intervals of transient slip (Liu and Rice, 2005, 2006). While it is well established that properties of the subducting plate play key roles in determining plate interface behavior, our result adds to growing evidence that the overriding plate is equally important in megathrust and ETS characteristics. More characterization of the upper plate geologic framework will be essential to assess the nature of deformation at convergent margins.

Acknowledgments We relied heavily on data from PANGA, IRIS, USGS, CNSN, PNSN, BDSN, and EarthScope. We benefited from discussions/feedback from R. Blakely, W.-P. Chen, B. Currie, K. Creager, C. DeMets, H. Deshon, W. Hart, S. Holtkamp, H. Kao, A. Lowry, W. McCausland, T. Melbourne, T. Niemi, W. Szeliga, B. Tikoff, C. Thurber, R. Wells, and an anonymous reviwer. NSF provided support for M.B. (EAR-0642765, EAR-0510810) and R.A. (EAR-0643077, EAR-0539987). UW-Madison also provided support for M.B. and deployment of the OATS array.

Figure 1. Map illustrating patterns in ETS along the entire Cascadia subduction zone. Colored basemap shows topography and bathymetry. Dashed line onshore marks 40 km depth contour of the subduction interface. Arrows and associated annotations show directions and speeds of subduction relative to North America. Locations of continuous GPS stations (squares) and broadband seismometers (triangles) which exhibit ETS are shown, with colors indicating the recurrence interval when multiple ETS events were observed. Recurrence intervals establish 3 zones that are labeled based on the continental terrane block they associate with.

Figure 2. Summary of SSE and NVT measurements that characterize seven segments along Cascadia. Colored circles show processed east displacement of GPS monument and black curves are normalized, mean seismic amplitudes, where high spikes are bursts of NVT activity. Red rectangles and black arrows mark the peak of NVT and center of SSE, respectively, with clear temporal correlation in any given segment. GPS data are colored by recurrence interval of SSE as in Figure 1. Inset shows time offsets in NVT activity for the 2003 ETS event from (A) and (B), plotting stations according to distance from the Oregon-Washington border (see data repository for details).

Figure 3. Plot of along strike patterns of ETS and upper plate features. (A) Top panel shows distinct variations in ETS recurrence with symbols as in Fig 1 and vertical bars show 1σ of observed intervals. Bottom panel shows topography above the 40 km depth contour of the subduction interface, in the middle of ETS observations. Topography is inversely correlated with ETS recurrence, roughly matching the primary continental terranes of different age and composition (Wrangellia, Siletzia, Klamath). (B) Top panel shows "phase" of ETS for 7 different segments along the subduction zone from ETS timing at individual stations. Horizontal grey lines connect stations that record ETS within a month of one another. Bottom panel shows color shaded gravity anomalies and locations of offshore fore-arc sedimentary basins (white lines), features which have been correlated with megathrust asperities on the subduction interface in recent global studies (Wells et al., 2003). Vertical dashed lines show apparent edges of ETS segmentation from currently available data that seem to correlate with megathrust segmentation from the 5 largest sedimentary basins. To deal with trench curvature, station latitudes are those when projected on to 40-km contour (black curve).

8 References Brudzinski, M., Cabral, E., DeMets, C., Márquez-Azúa, B., and Correa-Mora, F., 2007, Multiple slow slip transients along the Oaxaca subduction segment from 1993-2006: Geophysical Journal International, p. in review. Brudzinski, M. R., and Allen, R. M., 2006, Segmentation in Episodic Tremor and Slip All Along Cascadia: Eos trans. AGU, v. 87, p. Abstract T53G-05. Chen, W.-P., and Brudzinski, M. R., 2006, Repeating Earthquakes, Episodic Tremor and Slip: Emerging Patterns in Complex Earthquake Cycles?: Complexity, v. in press. DeMets, C., Gordon, R. G., Argus, D. F., and Stein, S., 1990, Current plate motions: Geophysical Journal International, v. 101, p. 425-478. Dragert, H., G.C. Rogers, J. Cassidy, H. H. Kao, and K. Wang, 2004, Episodic tremor and slip in northern Cascadia: USGS Progress Report, v. 04HQGR0047, p. 1-6. Dragert, H., Wang, K. L., and James, T. S., 2001, A silent slip event on the deeper Cascadia subduction interface: Science, v. 292, p. 1525-1528. Fuller, C. W., Willett, S. D., and Brandon, M. T., 2006, Formation of forearc basins and their influence on subduction zone earthquakes: Geology (Boulder), v. 34, p. 65-68. Harden, D., 1998, California Geology, Prentice Hall, 479 p. Hirose, H., and Obara, K., 2006, Short-term slow slip and correlated tremor episodes in the Tokai region, central Japan: Geophysical Research Letters, v. 33, p. 5. Holtkamp, S., Brudzinski, M. R., and DeMets, C., 2006, Determination of Slow Slip Episodes and Strain Accumulation along the Cascadia Margin: Eos trans. AGU, v. 87, p. Abstract T41A-1541. Jones, D. L., Silberling, N. J., and Hillhouse, J., 1977, Wrangellia; a displaced terrane in northwestern North America: Canadian Journal of Earth Sciences, v. 14, p. 2565-2577. Kao, H., Shan, S. J., Dragert, H., Rogers, G., Cassidy, J. F., and Ramachandran, K., 2005, A wide depth distribution of seismic tremors along the northern Cascadia margin: Nature, v. 436, p. 841-844. Kodaira, S., Iidaka, T., Kato, A., Park, J. O., Iwasaki, T., and Kaneda, Y., 2004, High pore fluid pressure may cause silent slip in the Nankai Trough: Science, v. 304, p. 1295-1298. Kohlstedt, D. L., Evans, B., and Mackwell, S. J., 1995, Strength of the Lithosphere - Constraints Imposed by Laboratory Experiments: Journal of Geophysical Research-Solid Earth, v. 100, p. 17,587-17,602. Kostoglodov, V., Singh, S. K., Santiago, J. A., Franco, S. I., Larson, K. M., Lowry, A. R., and Bilham, R., 2003, A large silent earthquake in the Guerrero seismic gap, Mexico: Geophysical Research Letters, v. 30. Larson, K. M., Lowry, A. R., Kostoglodov, V., Hutton, W., Sanchez, O., Hudnut, K., and Suarez, G., 2004, Crustal deformation measurements in Guerrero, Mexico: Journal of Geophysical Research-Solid Earth, v. 109, p. B04409, doi:10.1029/2003JB002843. Liu, Y. J., and Rice, J. R., 2005, Aseismic slip transients emerge spontaneously in three- dimensional rate and state modeling of subduction earthquake sequences: Journal of Geophysical Research-Solid Earth, v. 110. Liu, Y. J., and Rice, J. R., 2006, Recurrence interval and magnitude of aseismic deformation transients: an investigation using rate- and state-dependent friction: EOS Trans. Am. Geophys. Union, v. 87, p. T41A-1543. Lowry, A. R., Larson, K. M., Kostoglodov, V., and Bilham, R., 2001, Transient fault slip in Guerrero, southern Mexico: Geophysical Research Letters, v. 28, p. 3753-3756.

9 Mazzotti, S., and Adams, J., 2004, Variability of near-term probability for the next great earthquake on the Cascadia subduction zone: Bulletin of the Seismological Society of America, v. 94, p. 1954-1959. McCausland, W., Malone, S., and Johnson, D., 2005, Temporal and spatial occurrence of deep non-volcanic tremor: From Washington to northern California: Geophysical Research Letters, v. 32. Melbourne, T., Szeliga, W. M., Miller, M. M., and Santillan, V. M., 2005, Extent and duration of the 2003 Cascadia slow earthquake: Geophysical Research Letters, v. 32, p. L04301, doi:10.1029/2004GL021790. Mitsui, N., and Hirahara, K., 2006, Slow slip events controlled by the slab dip and its lateral change along a trench: Earth and Planetary Science Letters, v. 245, p. 344-358. Obara, K., 2002, Nonvolcanic deep tremor associated with subduction in southwest Japan: Science, v. 296, p. 1679-1681. Ozawa, S., Murakami, M., Kaidzu, M., Tada, T., Sagiya, T., Hatanaka, Y., Yarai, H., and Nishimura, T., 2002, Detection and monitoring of ongoing aseismic slip in the Tokai region, central Japan: Science, v. 298, p. 1009-1012. Parsons, B., and Sclater, J. G., 1977, An analysis of the variation of ocean floor bathymetry and heat flow with age: J. Geophys. Res., v. 82, p. 803-827. Rogers, G., and Dragert, H., 2003, Episodic tremor and slip on the Cascadia subduction zone: The chatter of silent slip: Science, v. 300, p. 1942-1943. Ruff, L. J., and Kanamori, H., 1983, Seismic coupling and uncoupling subduction zones: Tectonophysics, v. 99, p. 99-117. Schmidt, M. E., and Grunder, A. L., 2006, Segmentation of the Cascade Arc Based on Compositional and Sr and Nd Isotopic Variations in Primitive Volcanic Rocks: EOS Trans. Am. Geophys. Union, v. 87, p. T53G-02. Song, T. R. A., and Simons, M., 2003, Large trench-parallel gravity variations predict seismogenic behavior in subduction zones: Science, v. 301, p. 630-633. Szeliga, W., Melbourne, T. I., Miller, M. M., and Santillan, V. M., 2004, Southern Cascadia episodic slow earthquakes: Geophysical Research Letters, v. 31, p. L16602, doi:10.1029/2004GL020824. Trehu, A. M., Asudeh, I., Brocher, T. M., Luetgert, J. H., Mooney, W. D., Nabelek, J. L., and Nakamura, Y., 1994, Crustal Architecture of the Cascadia Fore-Arc: Science, v. 266, p. 237-243. Wells, R. E., Blakely, R. J., Sugiyama, Y., Scholl, D. W., and Dinterman, P. A., 2003, Basin- centered asperities in great subduction zone earthquakes: A link between slip, subsidence, and subduction erosion?: Journal of Geophysical Research–Solid Earth, v. 108, p. 2507, doi:10.1029/2002JB002072. Wells, R. E., Weaver, C. S., and Blakely, R. J., 1998, Fore-arc migration in Cascadia and its neotectonic significance: Geology (Boulder), v. 26, p. 759-762.

10 Data Repository: Methods for Automated Data Processing and Observations of ETS in Cascadia

Michael R. Brudzinski 1, Richard M. Allen 2 1 Miami University, Oxford, OH ([email protected]) 2 University of California, Berkeley, CA ([email protected])

A.1. Slow Slip Episodes In this section, we discuss the relevant information for identification of slow slip episodes, with additional details presented elsewhere (Holtkamp et al., 2006, ms. in prep.). The GPS data analyzed for slow slip episodes is from the PANGA network provided by the Central Washington University clearinghouse (2005). PANGA time series are network solutions with phase ambiguities resolved and typically consist of one sample per day. Provided time series have been postprocessed with GIPSY, and have linear trends, steps due to earthquakes or hardware upgrades, and annual and semi-annual sinusoids signals simultaneously estimated and removed (Szeliga et al., 2004). In this study, the signal to noise ratio of the GPS data is further enhanced by taking a 10 sample running average of the time series that reduces the scatter among neighboring data points. As shown in previous studies examining transient episodes (Brudzinski et al., 2007; Larson et al., 2004; Lowry et al., 2001), anomalous displacements during slow slip events can be estimated by fitting the GPS coordinate time series with combination of a linear and hyperbolic tangent function. By using a grid search over hyperbolic tangent function parameters (timing and duration of transient), the linear parameters of steady-state velocity and transient displacement are estimated via least squares minimization, weighted by the formal inverse variance of GPS coordinate estimates. This technique has been particularly useful for characterizing the precise timing and overall displacement when a slow slip episode has been identified (Brudzinski et al., 2007; Larson et al., 2004; Lowry et al., 2001). Slow slip episodes are automatically identified by 1) applying the hyperbolic tangent fit over a scrolling window of the time series (6 months before event time and 6 months after) incremented at 0.01 years, 2) using an f-test to confirm when the combined linear and hyperbolic tangent fit is significantly better than a linear fit at 99% confidence within the window, and 3) establishing a threshold value for the transient displacement (~2.0 mm) to identify potential events that are larger than background noise from examination of stations within the stable continental interior (DeMets et al., 2004). Slow slip episodes are verified and event times are selected with an algorithm that first identifies one of the components with an event magnitude that exceeds the displacement threshold and the chi squared value associated with that component must be low. Precise times are selected when the ftest value is near its maximum relative to neighboring times, and the chi squared value is minimized relative to neighboring times. If multiple events within 0.25 years are identified, we take the time with the lowest ratio of the chi squared value to the average chi squared value for all times in the time series. Of these potential events, each one goes through a series of tests (density of data samples, chi squared relative to all time series, and coherence with other neighboring stations) to determine whether it should be included or excluded. Lastly, we perform a manual closer inspection of events, including a handful of extra events that are clear in the time series but only met a smaller displacement threshold (1.5 mm) and excluding a small set

11 of events that appear as noise spikes in the time series. Further details of this analysis will be presented in separate work (Holtkamp et al., 2006, ms. in prep.).

A.2. Non-volcanic Tremor (NVT) In this section, we discuss the relevant information for identification of episodes of NVT, with additional details presented elsewhere (Brudzinski and Allen, 2006, ms. in prep.). The seismic data for analysis of NVT comes from a variety of networks that spread across the Cascadia subduction zone including the PNSN, USGS-NCSN, CNSN, OATS, IRIS-GSN, BDSN, and the new EarthScope Transportable Array. These networks primarily consist of short period vertical component seismometers, but these are ample instruments to identify NVT which has characteristic frequencies between 1 and 10 Hz (Obara, 2002). Although signals appear to be more pronounced on horizontal components when they are available, vertical components have been sufficient to clearly identify NVT episodes. PNSN data comprise the majority of data analyzed in this study, but continuous data only goes back to 2002, so longer running stations of other networks are critical for identifying longer term trends. Typically a permanent broadband seismic station uses 24-bit digitizers to continuously record all three components of ground velocity at a rate of 20 samples per second. To fully analyze this voluminous dataset which lasts up to 15 years, we use an automated scheme which begins by removing the instrument response and applying a filter with a pass-band of 2-6 Hz. We then take absolute values of the time series and calculate the envelope. Up to this point, our analysis is the same as other researchers (Obara, 2002; Rogers and Dragert, 2003; Szeliga et al., 2004). Next, instead of judging whether tremors are present in each segment of data by the hour, we calculate the mean amplitude of the envelope (Figure A1a). Since various non-tectonic factors can influence the amplitude, we limit our analysis to nighttime hours when culture-generated noise is at a minimum and exclude hours during which signals from known, large earthquakes or other spikes, such as those used for instrument calibration, are present (Figure A1b). To further avoid seismic events, we exclude hours with a high max-to-mean ratio. To help reduce longer-term noise trends that exist at a few stations, we high-pass filter the envelope time series to signals with periods less than about a week (Figure A1c). Lastly, we average amplitudes over a four-day moving window to help accentuate sustained tremor activity, and we normalize to values between 0 and 1 (Figure A1d). In the final time series, large peaks are identified as periods during which tremors dominates. The first cut for identifying NVT periods is by marking peaks that rise above background noise at greater than the 3σ confidence interval. In some cases, analyst intervention is necessary to identify NVT periods due to data loss. To confirm NVT, individual seismograms are investigated as well as mean amplitude time series from neighboring stations. For station VGZ, there is data loss in early 2003 preventing large enough amplitudes to meet the threshold value (Figure A1 and A2), but an NVT period is confirmed by further analysis. We also find that in some cases there are peaks that do not meet the 3σ confidence interval threshold, but the timing corresponds to peaks that meet the threshold at neighboring stations. In these cases, we also find that it is typical for this smaller peak to be accompanied by another smaller peak that is close in time. For these situations, we have added a second criteria that can meet the threshold for an event: at least two successive peaks near in time which rise above background noise at greater than 2σ confidence interval. When calculating the recurrence, we take the mean of the times for the smaller peaks that comprise the event. Our rationale for this

12 second criteria is that these cases predominantly occur at stations near the boundary between regions, suggesting they are recording signals from events in neighboring segments. The technique used here has two distinct advantage over previous analyses: 1) it does not requiring the scientist to examine each record to evaluate whether NVT has occurred and 2) it does not require an array of stations to identify NVT through station coherence. The first advantage results in the ability to automate the process and analyze data over large ranges of time and location. Nevertheless, in each case of a large peak, individual seismograms have been examined to confirm the presence of NVT. The second advantage is important for investigating NVT in regions where station density is significantly reduced, like central Cascadia.

A.3. Comparison with Previous Studies In order to confirm the validity of the automated seismic and GPS single-station methods, we compare the results of the techniques in this study with previous identifications of ETS. Figure A2 compares the results from a previous for southern Vancouver Island (Rogers and Dragert, 2003) with analysis from this study for stations ALBH and VGZ. Despite the different analysis approaches, there is one-to-one consistency among identified tremor peaks and slow slip episodes between the two studies. This confirms that for single stations with a low enough noise level, the automated techniques produce results very similar to that of previous studies relying on inspection of a network of observations. These methods are not replacements for network solutions, they are simply used as surrogates to perform a uniform investigation of ETS over the entire subduction zone while network density is still quite heterogeneous. We also demonstrate the ability to resolve accurate timing of ETS by illustrating our identification of NVT for the early 2003 event in northern Cascadia that has been characterized in previous papers (e.g., Melbourne, 2004). Figure A3 shows a zoomed in time frame for our seismic time series around the 2003 event, with stations plotted according to distance along strike. Our seismic time series show the same progression to the north and south away from the Puget Sound region seen in the previous GPS analyses, demonstrating that our single station data have the resolution to accurate timing of ETS. Furthermore, our data show that the propagation of ETS is somewhat discontinuous, with changes in timing of the maximum NVT activity by 1-2 weeks around 150 km and 310 km. These locations correspond to the northern and southern boundaries of segment C defined from offsets in timing of several months between other ETS episodes (47.5° and 48.4° in Figure 3 of the main text). These offsets in time during the 2003 event reinforce our proposed segmentation boundaries for ETS.

A.4. Observations of Episodic Tremor and Slip in Cascadia To illustrate the new set of ETS observations in Cascadia, we plot time series for a total of 43 GPS and seismic stations all along the subduction zone (Figures A4-A5). These time series were selected to show an equal mixture of slow slip episode and non-volcanic tremor events over an along-strike station spacing of ~25 km. The coherence in timing between stations in each segment provide convincing evidence of common recurrence intervals and segmentation of ETS along strike. Recurrence intervals for each station are calculated from the mean of each interval between times of slow slip or NVT. Then for each segment, we take the mean of all observed intervals for both slow slip and NVT within that segment. Uncertainties on the recurrence shown in Figure 2 are the 1s standard deviation of observed intervals from the mean within that segment.

13 We have also constructed a movie of ETS occurrence at stations along Cascadia (Figure A6), to help readers to visualize the segmentation of ETS episodes over time and space. The images in this movie color the station symbols by the time since the last ETS event when data is available. The movie progresses from 1997 to 2006 by 5 day increments, and the time differences between different segments can be seen in when a group of stations “reset” to a dark red color. Also, one can begin to see the propagation of ETS onsets during a given event from our dataset.

Figure A1. Processed time series for seismic activity at station VGZ on southern Vancouver Island illustrating the effects of different components of our processing. (a) Time series are created by taking the mean amplitude of each hour segment of bandpass filtered envelope seismograms (a). Time series are limited to nighttime hours and those without significant earthquake or instrument glitches (b). Time series are high-pass filtered to amplify signals with periods less than about a week (c). Amplitudes are averaged over a four-day moving window to accentuate sustained tremor activity, and normalized to values between 0 and 1 (d).

Figure A2. Comparison of slow slip and non-volcanic tremor measurements made in this study with previous analysis of episodic tremor and slip. (a) Plot taken from Rogers and Dragert (2003) showing GPS time series for station ALBH relative to North America including both the long-term trend of eastward motions and the slow slip episodes, and hours of tremor activity from a network of seismic stations. (b) Plot showing GPS time series provided by the PANGA network for station ALBH relative to North America with long-term trend removed and stations corrections made. We have further reduced the scatter by using a 10 sample running mean to help accentuate slow slip episodes. Tremor activity is identified for station VGZ by the single- station method of this study. Despite the different analysis approaches, there is one-to-one consistency among identified tremor peaks and slow slip episodes between the two studies.

Figure A3. Zoomed view of NVT activity for the 2003 ETS event showing stations from north- central Cascadia plotted according to their distance along strike from the Oregon-Washington border. Distance is estimated by projecting stations onto the 40 km contour of the slab interface. Vertical bars mark week-long increments. Note that the progression to the north and south from the central Puget Sound region is discontinuous, with jumps in timing around 150 km and 310 km. These locations correspond to the northern and southern boundaries of this segment defined from offsets in timing of several months between other ETS episodes (Figure 3, main text).

Figure A4. Summary map of seismic (triangles) and GPS (squares) stations in the Cascadia subduction zone whose processed time series and ETS event times are illustrated in Figure A5. The expected source zone is near the dashed 40-km contour of the plate interface.

Figure A5. Examples of slow slip and non-volcanic tremor measurements from stations all along the Cascadia margin. Stations are sorted by latitude from the north to the south, separated into the segments of common ETS timing determined in this study. Station locations are shown in Figure A4.

(File: ets.97-06.c.gif) Figure A6. Movie illustrating ETS occurrence at stations along Cascadia, to help visualize the segmentation of ETS episodes over time and space. Individual images color station symbols by time since the last ETS event, when data is available. The movie progresses from 1997 to 2006 by 5 day increments, and the time differences between different segments can be seen in when a group of stations “reset” to a dark red color when an ETS event occurs. Also, one can see some indications of the propagation of ETS along strike during a given event as illustrated in Figure A3.

20 References

Brudzinski, M., Cabral, E., DeMets, C., Márquez-Azúa, B., and Correa-Mora, F., 2007, Multiple slow slip transients along the Oaxaca subduction segment from 1993-2006: Geophysical Journal International, p. in review. Brudzinski, M. R., and Allen, R. M., 2006, Segmentation in Episodic Tremor and Slip All Along Cascadia: Eos trans. AGU, v. 87, p. Abstract T53G-05. Central Washington University, 2005, Geodesy Laboratory & PANGA Data Analysis Facility, www.panga.cwu.edu. DeMets, C., Brudzinski, M., Cabral, E., Márquez-Azúa, B., and Correa-Mora, F., 2004, Large- scale seismic and aseismic deformation patterns associated with subduction: Constraints from continuous GPS measurements in Mexico (Abstract): EOS Trans. Am. Geophys. Union, v. 85. Holtkamp, S., Brudzinski, M. R., and DeMets, C., 2006, Determination of Slow Slip Episodes and Strain Accumulation along the Cascadia Margin: Eos trans. AGU, v. 87, p. Abstract T41A-1541. Larson, K. M., Lowry, A. R., Kostoglodov, V., Hutton, W., Sanchez, O., Hudnut, K., and Suarez, G., 2004, Crustal deformation measurements in Guerrero, Mexico: Journal of Geophysical Research-Solid Earth, v. 109, p. B04409, doi:10.1029/2003JB002843. Lowry, A. R., Larson, K. M., Kostoglodov, V., and Bilham, R., 2001, Transient fault slip in Guerrero, southern Mexico: Geophysical Research Letters, v. 28, p. 3753-3756. Obara, K., 2002, Nonvolcanic deep tremor associated with subduction in southwest Japan: Science, v. 296, p. 1679-1681. Rogers, G., and Dragert, H., 2003, Episodic tremor and slip on the Cascadia subduction zone: The chatter of silent slip: Science, v. 300, p. 1942-1943. Szeliga, W., Melbourne, T. I., Miller, M. M., and Santillan, V. M., 2004, Southern Cascadia episodic slow earthquakes: Geophysical Research Letters, v. 31, p. L16602, doi:10.1029/2004GL020824.