determination of slow slip episodes and strain...

Determination of slow slip episodes and strain accumulation along the

Cascadia margin

Stephen Holtkamp1 and Michael R. Brudzinski1,2

Received 1 September 2008; revised 31 July 2009; accepted 18 September 2009; published 8 April 2010.

[1] Continuous GPS stations in the Pacific Northwest Geodetic Array network clearlyrecord subduction-related strain accumulation and slow slip episodes along the Cascadiaconvergent margin. Many of the slow slip episodes have been correlated in time and spacewith seismic evidence for nonvolcanic tremor, leading to the previous discovery ofepisodic tremor and slip (ETS). In this study, we use a hyperbolic tangent curve fittingtechnique for the identification of slow slip times and displacement magnitudes within theGPS time series, independent of seismic tremor data. We then apply this technique tostudy the patterns of strain accumulation and release associated with ETS events andcharacterize patterns of coupling associated with the locked and transition zones of theplate interface. We demonstrate the effectiveness of this automated technique for bothidentification of slow slip observations and calculation of slow slip displacements.Recurrence patterns in the distribution of GPS observations demonstrate coherence amongneighboring stations over time and apparent along-strike segmentation of the subductioninterface. When slow slip events are removed from the time series, we can estimate thetotal site velocities between slow slip events. These velocities decay as depth to thesubduction interface increases, but they diverge from the long-term trends expected fromthe interseismic cycle at �30–60 km above the interface, consistent with the locationwhere slow slip displacements occur. Forward modeling of coupling on the plate interfacereveals that in between slow slip events there is a patch of at least 30% coupling from20 to 35 km depth, which is needed to produce the observed back slip displacements.Intriguingly, our best fitting models have a decrease in coupling down to �30% at�20 km depth followed by a peak of greater than 80% coupling at �30–35 km depth,suggesting the source zone for ETS events acts as a distinct locking zone that releasesstrain more frequently than the updip seismogenic locked zone, although a zone ofconstant �30% coupling cannot be ruled out with this data set. Such a scenario indicatesthat frictional behavior with depth follows a more complex model than a simpletemperature controlled transition. We propose that coupling initially decreases with depthdue to a decrease in strength of the overriding lower crust, but then coupling increasesagain when the subducting plate comes in contact with the stronger overriding mantle.

Citation: Holtkamp, S., and M. R. Brudzinski (2010), Determination of slow slip episodes and strain accumulation along the

Cascadia margin, J. Geophys. Res., 115, B00A17, doi:10.1029/2008JB006058.

1. Introduction

[2] As the oceanic plate subducts down into the mantle,friction on the interface with the overriding plate causesstick slip behavior. The overriding plate is pulled down bythe subducting plate in areas of strong coupling, accumu-lating strain on the megathrust fault until slip occurs and theoverriding plate pops back up in the form of an earthquake.Slow slip episodes represent similar motions to the mega-

thrust stick slip behavior, but these episodes typically lastover 4 orders of magnitude longer than an earthquake[Dragert et al., 2001; Kawasaki et al., 1995; Lowry et al.,2001; Ozawa et al., 2001]. Together with nonvolcanictremor (NVT) activity, the correlated strain and seismicobservations characterize episodic tremor and slip (ETS)events, which in Cascadia recur with regular intervals thatrange from months to years [Brudzinski and Allen, 2007;Kao et al., 2006; Melbourne and Webb, 2003; Rogers andDragert, 2003; Szeliga et al., 2004]. The processes thatgovern ETS and the potential relationships to major earth-quakes remain unknown. In this study we use new methodsto identify slow slip episodes at individual stations allowingcompilation of a catalog of these observations for anentire subduction zone over a 10 year period from 1997to 2007.

JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 115, B00A17, doi:10.1029/2008JB006058, 2010ClickHere

for

FullArticle

1Geology Department, Miami University, Oxford, Ohio, USA.2Physics Department, Miami University, Oxford, Ohio, USA.

Copyright 2010 by the American Geophysical Union.0148-0227/10/2008JB006058

B00A17 1 of 21

http://dx.doi.org/10.1029/2008JB006058

[3] Previous observations of ETS in Cascadia havefocused on southern Vancouver Island, northernWashington,and northern California given the scarcity of geophysicalobservatories in other parts of Cascadia. Slow slip episodesrepresenting weeks of transient displacement are visible onGPS stations within a few hundred kilometers of the trenchand show motions of the upper plate back toward the trenchconsistent with relieving accumulated strain [Dragert et al.,2001; Melbourne et al., 2005; Miller et al., 2002; Szeliga etal., 2008]. In this study, we analyze data from 111 contin-uous GPS stations along the entire length of the 1400 kmCascadia subduction margin and into the stable NorthAmerican plate to better characterize patterns in strainaccumulation and release. We use GPS data to investigateboth long- and short-term trends in the time series by payingparticular attention to slow slip events including patterns indisplacement magnitudes, directions, and regional coher-ence. We employ an algorithm for identifying slow sliptransient displacements in a given time series to construct acatalog of slow slip observations for all of the continuousGPS stations across Cascadia (Figure 1).[4] We uncover evidence for along-strike segmentation of

slow slip events, and we identify ways to measure thestation velocities between slow slip events. Using a simpleelastic half-space forward modeling approach, we find thatlong-term velocities are consistent with previously proposedmodels for how coupling varies with depth on the plate

interface, and we find slow slip is centered at �30 km depthon the interface. However, when we examine station veloc-ities in between slow slip events, we find that coupling onthe plate interface does not transition gradually from thelocked zone to free slipping zone, but must instead have aprominent increase in coupling in the source zone of slowslip.

2. Data Analysis

[5] The continuous GPS data analyzed for slow slipepisodes is from the Pacific Northwest Geodetic Array(PANGA) network provided by the Central WashingtonUniversity clearinghouse (www.panga.cwu.edu). PANGAtime series are network solutions with phase ambiguitiesresolved and typically consist of one sample per day. Theprovided time series have been postprocessed with GIPSYand undergone a regional stabilization, with steps due toearthquakes or hardware upgrades, and annual and semi-annual sinusoidal signals simultaneously estimated andremoved [Szeliga et al., 2004]. We examine all data in theCascadia region available from PANGA at the time of thisstudy, covering 1997 to mid-2007.[6] As shown in previous studies examining transient

episodes [Brudzinski et al., 2007; Larson et al., 2004;Lowry et al., 2001], anomalous displacements during slowslip events can be estimated by fitting the GPS coordinatetime series with a function of the form

xiðtÞ ¼ x0 þ Vit þUi

2tanh

t � T0i

ti

� �� 1

� �ð1Þ

in which xi(t) are GPS site coordinates at time t in thevicinity of the ith slow slip event, x0 are coordinates at areference time, Vi is the velocity before and after the ithslow slip event, Ui is anomalous displacement during the ithtransient event, T0i is the median time of the ith event, andti scales the period over which the event occurred. If T0 andt are specified, the other parameters can be estimated fromlinear least squares inversion. To estimate anomalousdeformation during slow slip events, one can employ agrid search over T0 and t. The linear parameters of steadystate velocity and transient displacement are estimated vialeast squares minimization, weighted by the formal inversevariance of GPS coordinate estimates.[7] This technique has been particularly useful for char-

acterizing the precise timing and overall displacement whena slow slip episode has been identified [Brudzinski et al.,2007; Larson et al., 2004; Lowry et al., 2001]. In this studywe use a grid search to automate the identification processby (1) applying the hyperbolic tangent fit over a 12 monthscrolling window incremented at 0.01 years, (2) using anF test to confirm when the hyperbolic tangent fit is signifi-cantly better than a linear fit at 99% confidence within thewindow, and (3) establishing a threshold value for thetransient displacement to mark events that are larger thanbackground noise.[8] For the threshold to be consistent across the array,

we fix the t value in the final run of the grid search.This is necessary because large t values tend to promotelarger transient displacements over a fixed 12 monthwindow, so fixing t ensures that the algorithm produces

Figure 1. Experimental configuration along the Cascadiasubduction margin. Continuous GPS stations are shown assquares. Red shading indicates representative stationswhose time series are shown in Figure 2. Yellow stationDRAO effectively represents the stable North Americanplate. Dashed lines show depth contours every 20 km forthe downgoing slab [McCrory, 2006]. Thin lines andnumbers indicate apparent segmentation of ETS based onrecurring termination points of ETS events.

B00A17 HOLTKAMP AND BRUDZINSKI: CASCADIA SLOW SLIP EPISODES

2 of 21

B00A17

comparable displacements within events and betweenevents. The t value of 0.025 y is found to produce thebest fit to the data across the entire subduction margin,consistent with the typical event duration of �20 daysreported in previous studies [Dragert et al., 2001; Milleret al., 2002].[9] We developed an algorithm for the identification of

slow slip event times that uses the event displacementmagnitude, F test, and associated chi squared value forthe longitudinal and latitudinal components, along with testsfor regional consistency and proper data sampling. Timesare flagged as potential event times if the F test value is>99% and near a local maximum while the associated chisquared value is low. Also, event displacement Ui must begreater than a global threshold we set to be 2.0 mm, takenfrom analyzing stations in the stable continental interior forapparent transients in their time series. For detailed infor-mation about the methodology, please see the appendix(auxiliary material).1

[10] For each station, our event detection algorithmproduces a comprehensive set of dates and sizes for slowslip episodes. Table 1 shows identified dates of observationsof slow slip events, sorted by station latitude. In total, wefind 215 observations of slow slip events recorded by60 stations across the entire Cascadia region since 1997.Out of the 61 episodes recorded at 10 representative stationsacross the subduction zone, only one false negative andeight false positives were recorded by the automated pro-cess, as defined by visual comparison for similar observa-tions at neighboring stations. Positive identification of slowslip events relies on regional consistency with other GPSstations or correlation with seismic tremor in space andtime. The false positives mentioned failed to meet thiscriteria. False negatives were identified by examining timeseries of stations near identified slow slip events andvisually determining if a consistent signal may exist.Detailed methodology on validating identified slip eventsis in the auxiliary material.[11] In Figures 2 and S1, we show the longitude and

latitude time series for the representative stations andhyperbolic tangent fits for each time identified as a slowslip episode. Each of these time series have been detrendedsuch that the overall trend thought to represent interseismicrate of strain accumulation has been removed. What remainsare the generally downward steep sloping trends that areslow slip episodes and the upward gradual sloping trendsthat are thought to represent the strain accumulation even-tually released in the form of slow slip.[12] Vertical time series were not used in the determina-

tion of ETS times due to the larger scatter for this compo-nent that is common globally. For example, when thehyperbolic tangent method was applied to the verticalcomponent of station DRAO in the stable continentalinterior, the maximum value of transient displacementestimated was 5.3 mm, compared to 2.3 mm and 1.9 mmin the east and north components, respectively. Our algo-rithm solves for vertical transients for all events using T0

identified on horizontal components, but the scatter at

DRAO suggests we should be wary of displacements lessthan �5 mm.

3. Results

3.1. Slow Slip Episodes

[13] To illustrate the regional patterns in the slow slipprocess, the final catalog of slow slip event observations isplotted as squares and associated vectors and 2-sigmauncertainty ellipses in Figure 3, with specific dates listedin Table 1. The magnitudes given for these slow slip eventobservations are surface displacements. However, we doexpect a direct relationship between the surface displace-ments and the magnitude of slip on the plate interface belowthe observations [e.g., Brudzinski et al., 2007; Melbourne etal., 2005]. All events are plotted along the entire Cascadiaregion separated into plots covering a time interval of abouthalf a year. Stations that were recording during this timeframe and have a small enough scatter in the time series tosee a transient but did not record a slow slip episode areplotted as small squares. For comparison, we also plotseismic stations where nonvolcanic tremor episodes havebeen identified independently of the GPS data [Brudzinskiand Allen, 2007]. Strong correlation between the spatial andtemporal placement of these two data sets suggest that thetimes associated with the GPS data are accurate. However,tremor data do not indicate displacement values so there isno check on the displacement magnitudes and directionscalculated from the GPS time series.[14] Szeliga et al. [2008] also analyzed the PANGA data

set up to 2006 to identify transients and inverted for slowslip for 12 of the events that had adequate sampling, 11 ofwhich were in Washington and one in northern Oregon.They report very similar results to those shown in Figures 2, 3and S1, particularly for large events. We find a few additionalsmall events (2001.3–2001.4, 2002.5–2002.6, 2002.7–2002.8, and 2003.7–2003.8, where the digit after the decimalpoint is tenths of a year) and some extra observations near thecoast and in the latitude component (e.g., Vancouver Islandobservations in July 2004). In the latter case, a few additionalconstraints near the spatial edge of transient deformationcould help reduce smearing effects in inversions but willnot likely change other model parameters significantly.Although we have not confirmed these additional slipevents through another technique, in two cases they occurrednear continuously recording seismometers that showcorresponding tremor. We feel the observation of additionalsmall events will help in estimating the overall strain releaseand the degree of coupling between ETS events.[15] ETS events in Cascadia have been shown to have

a remarkably regular recurrence interval [Rogers andDragert, 2003; Szeliga et al., 2004]. With new and existingstations recording more and more events, we set out todetermine the spatial and temporal patterns of slow slipepisodes including recurrence for stations along the entiremargin. In Figure 4, average recurrence intervals are shownfor each GPS station that shows more than one event.Recurrence intervals for each region are calculated from aweighted mean of the recurrence intervals for each stationthat shows at least 5 event observations. Error values quotedare a weighted mean of the standard deviations from eachstation in the region with 5 or more event observations.

1Auxiliary materials are available in the HTML. doi:10.1029/2008JB006058.


3 of 21

B00A17

Table 1. Summary of Slow Slip Episode Observations

Station Latitude LongitudeRecurrenceInterval SD n Episodesa

BCOV 50.544 �126.843 - - 1 2003.00WSLR 50.127 �122.921 18.96 9.49 2 2003.20, 2002.97, 2001.62ELIZ 49.873 �127.123 10.80 2.35 3 2004.57, 2003.91, 2002.77NTKA 49.592 �126.617 13.20 4.78 5 2005.98, 2004.55, 2003.93, 2002.74, 2002.34, 2001.58NANO 49.295 �124.086 14.54 4.06 6 2005.72, 2004.58, 2003.96, 2001.58, 2001.03, 1999.74, 1998.54, 1997.28UCLU 48.926 �125.542 12.67 1.97 9 2005.71, 2004.91, 2003.96, 2002.96, 2001.61,

2000.65, 1999.69, 1998.54, 1997.26SC04 48.923 �123.704 15.30 0.73 3 2007.09, 2005.74, 2004.54SC02 48.546 �123.008 14.97 1.42 5 2007.09, 2005.71, 2004.54, 2003.17, 2002.10SEDR 48.522 �122.224 15.94 1.20 6 2007.01, 2005.68, 2004.51, 2003.14, 2000.93, 1999.66, 1998.16OTIS 48.418 �122.337 15.30 0.83 3 2007.08, 2005.72, 2004.53ALBH 48.390 �123.487 14.64 1.09 9 2007.09, 2005.72, 2004.53, 2003.18, 2002.08,

2000.95, 1999.66, 1998.51, 1997.33WHD1 48.313 �122.696 14.60 1.15 7 2007.06, 2005.72, 2004.54, 2003.17, 2002.09, 2000.94, 1998.51, 1997.33NEAH 48.298 �124.625 14.79 3.68 8 2007.08, 2005.71, 2004.51, 2004.08, 2003.24, 2002.12,

2000.95, 1999.64, 1998.54, 1997.33COUP 48.217 �122.686 - - 1 2007.07ARLI 48.174 �122.142 - - 1 2007.08BLYN 48.016 �122.928 15.72 1.10 4 2007.06, 2005.68, 2004.52, 2003.13LKCP 47.944 �121.831 11.40 0.0 2 2003.02, 2002.07PFLD 47.899 �122.282 - - 1 2007.06SC03 47.817 �123.706 16.44 0.0 2 2007.06, 2005.69DWH1 47.774 �122.080 16.08 6.07 3 2005.80, 2005.30, 2004.52, 2003.12SEAW 47.687 �122.256 10.92 1.92 6 2004.53, 2003.77, 2003.08, 2002.05, 2001.15, 1999.98SEAT 47.654 �122.309 14.48 2.91 9 2007.05, 2005.31, 2003.95, 2003.09, 2002.02,

2000.65, 1999.63, 1998.56, 1997.40KTBW 47.547 �122.795 30.12 7.84 3 2007.06, 2005.35, 2002.04PUPU 47.500 �122.008 10.68 0.0 2 2004.58, 2003.69NINT 47.500 238.200 - - 1 238.20, 2007.05PRDY 47.391 �122.609 15.90 3.67 3 2007.03, 2005.33, 2004.38RPT1 47.388 �122.375 14.43 5.56 9 2007.08, 2005.32, 2004.45, 2003.84, 2003.10,

2001.22, 2000.65, 1999.95, 1998.56, 1997.46ZSE1 47.287 �122.188 16.08 3.33 3 2007.06, 2005.38, 2004.38LNGB 47.219 �122.758 20.88 0.0 2 2007.08, 2005.34PABH 47.213 �124.205 10.80 0.0 2 2002.04, 2001.14PCOL 47.172 �122.571 32.40 0.0 2 2007.04, 2004.34THUN 47.106 �122.288 33.00 0.0 2 2007.09, 2004.34TWHL 47.016 �122.923 15.60 3.81 4 2007.09, 2005.29, 2004.36, 2003.19SATS 46.966 �123.541 18.40 2.76 4 2003.16, 2001.26, 1999.98, 1998.56CPXF 46.840 �122.257 23.12 6.35 4 2007.09, 2004.38, 2003.16, 2001.31P415 46.656 �123.730 - - 1 2005.43P432 46.623 �121.683 - - 1 2005.94P420 46.589 �122.866 - - 1 2005.96, 2005.45P421 46.532 �122.429 - - 1 2006.04P702 46.300 �122.346 - - 1 2005.43JRO1 46.275 �122.218 17.08 2.58 7 2007.08, 2005.95, 2004.37, 2003.15, 2001.34, 2000.04, 1998.54TPW2 46.207 �123.768 13.32 0.0 2 2007.12, 2006.01FTS1 46.205 �123.956 24.04 6.41 4 2006.01, 2003.19, 2001.33, 2000.00KELS 46.118 �122.896 23.96 6.13 4 2005.94, 2003.14, 2001.35, 1999.95P687 46.110 �122.355 - - 1 2005.94GWEN 45.783 �121.328 - - 1 2000.04CVO1 45.611 �122.496 23.40 0.0 2 2005.92, 2003.97WACO 45.523 �122.990 23.16 0.0 2 2005.93, 2004.00CHZZ 45.487 �123.978 23.40 0.0 2 2005.94, 2003.99MCSO 44.974 �122.956 - - 1 2003.95P376 44.941 �123.102 - - 1 2005.91CORV 44.586 �123.305 21.76 7.80 4 2005.91, 2004.46, 2003.99, 2001.94, 2000.47NEWP 44.585 �124.062 21.33 2.34 5 2005.91, 2003.99, 2001.98, 2000.54, 1998.80LPSB 44.051 �123.090 32.52 0.0 2 2007.13, 2004.42PMAR 43.991 �121.687 - - 1 2003.92DDSN 43.119 �123.244 19.08 1.37 3 2005.70, 2003.97, 2002.52CABL 42.836 �124.563 33.96 0.0 2 2005.44, 2002.61PTSG 41.783 �124.255 - - 1 2006.63YBHB 41.732 �122.711 11.05 1.26 11 2006.61, 2005.70, 2004.84, 2003.83, 2002.94, 2001.89,

2001.12, 2000.23, 1999.35, 1998.55, 1997.40TRND 41.054 �124.151 21.40 4.83 4 2007.14, 2004.97, 2003.84, 2001.79

aDate format is decimal years. The first digit after the decimal point is tenths of a year, and the second digit is hundredths of a year.


4 of 21

B00A17

Figure 2. Continuous GPS time series for representative stations showing latitude and longitudecomponents. For clarity, a 10 sample running mean has been applied to the cleaned data from the PANGAdatabase. Hyperbolic tangent curve fits to slow slip episodes generated from the grid search algorithm areshown as red lines.


5 of 21

B00A17

Figure 3


6 of 21

B00A17

Individual stations may show anomalous recurrence inter-vals if events are small enough to avoid detection. In somecases, a station will show two small events that are close intime, with spacing between events being less than half thenormally observed spacing at that station. These events arealso typically much smaller than other observations at thatstation, and they commonly occur at recording stations onthe spatial edge of multiple events from neighboring sourcezones. We interpret these pairs of small motions as repre-senting the release of strain from two parts of the interfacethat would make up a single normal event, so they arecounted only once for recurrence purposes. A clear exampleof this is with station NANO in Figures 3d and 3e, whereevent observations were separated by only 0.55 years. Forthis reason, only regional recurrence intervals should begiven merit.[16] In some cases, abrupt regional variation in recurrence

intervals, like that in central Washington or northernOregon, appears to be due to the short amount of time somesites have been recording. Also, stations close to and faraway from the trench (e.g., TRND and WSLR, the south-ernmost and northernmost stations shown in Figure 4) showapparently anomalous recurrence intervals, but these stationsare recording events at or near the detection threshold, whichwe interpret as due to their distance from the expected sourcezone that limits our ability to detect all events. This results ina value that is longer than others in the region, but indicatesto us that recurrence is being affected by the detectionthreshold. Nevertheless, results from our recurrence intervalcalculation are consistent with those found in previousstudies [Brudzinski and Allen, 2007; Miller et al., 2002;Szeliga et al., 2004] and show�11.1 ± 1.3 month recurrenceintervals in northern California, �21.5 ± 2.3 month recur-rence intervals in most of Oregon and southern Washington,and �14.4 ± 1.1 month recurrence intervals in northernWashington and southern British Columbia (Figure 4). Thevalues for the longest standing stations are taken to representthe regional recurrence interval since grouping of stationsthat contain outliers (due to missing events or short stationlife) can affect calculated recurrence intervals and associatederrors.[17] For each station, the average amount and direction of

slow slip surface displacement recorded per year is calcu-lated (Figure 5a). These values represent the approximateannual rate of slow slip strain release and take into accounthow often the slow slip events occur. Modeling of thesesurface displacements is performed in section 4.4, but wefeel it is important to describe the key transient observationsthat illustrate the overall trends in slow slip. Annualamounts of transient monument displacement appear to belargest above the northern and southern parts of the marginand smaller in the middle, although the lower station densityin Oregon and northern California make these regionsdifficult to assess as displacements can vary greatly withdistance from the trench. Stations ALBH and YBHB showyearly transient displacements of 3.8 and 3.5 mm/yr,

respectively. The largest rate in the �20 month recurrenceinterval region is station NEWP, which shows a yearlytransient displacement of 1.9 mm/yr.[18] We can look at a slightly different perspective on

along-strike variability by examining the average transientdisplacement and direction per event observed at the surface(Figures 5b and 6). Because of the large degree of along-strike variability in recurrence intervals, the average eventdisplacements and annual displacements for each stationcan vary by differing amounts. Average event displacementsobserved at the surface appear to be the largest near PugetSound, with stations showing average displacements of4.8 mm (SC02), 4.7 mm (ALBH), and 7.9 mm (SC03).While ALBH has a 46% larger average event displacementthan YBHB (ALBH shows an average event displacement of4.7 mm and YBHB shows an average event displacement of3.2 mm), ALBH only has a 10% larger annual surfacedisplacement. This suggests that while the plate interfacebelow YBHB and ALBH are accumulating and releasingstrain at similar rates, less total strain is accumulated beforebeing released in northern California. In central Oregon, theaverage event displacement at NEWP is 3.4 mm, similar toYBHB, but the recurrence intervals is longer, resulting inlower annual surface displacements (Figure 5a). Whetherthis means less accumulation and release in the form of slowslip events is hard to say yet due to such limited stationcoverage in Oregon, and further investigation of this issuewill be possible with EarthScope Plate Boundary Observa-tory data.[19] Differences in monument displacement direction

from transient event to event shown in Figure 6 representone standard deviation away from the mean displacementdirection. High directional variability could be a result oferrors in calculating the displacement directions, but con-sistency within individual events argues against this beingdue to poor resolution (e.g., Figures 3a and 3n). Althoughthe variability values vary greatly from station to station, nodominant geographic patterns emerge. We do find that somestations with high variability like NANO in central Vancou-ver Island appear to move with one group of stations in aparticular direction during one event and then they movewith a different group of stations in another direction duringa subsequent event (Figures 3b, 3d, and 3e). So in this case,the variability is due to a station responding to differentsource locations over time. However, other stations like thewell-studied ALBH show very little variability, suggestingthat the slow slip path beneath southern Vancouver Island isregular in its direction and the geographic location of thispatch is regular over time. Geographic regularity of slowslip patches is evident in Figure 3 as well and will be furtherdiscussed in section 4.1 on segmentation.

3.2. GPS Site Velocities

3.2.1. Long-Term Average Velocities[20] Figure 7 shows the long-term average velocities in

the PANGA GPS time series relative to the stable North

Figure 3. Individual station observations of slow slip episodes (squares) and nonvolcanic tremor (triangles) along theentire subduction margin over successive time windows. Small symbols show when the station was operating, but aprominent event was not recorded during that time window. Color indicates relative timing in each time window in fractionsof a year. Arrows show displacement magnitudes for slow slip episodes; dashed arrows occur when only one componentsatisfies the identification algorithm. Ellipses indicate 2s uncertainties.

7 of 21

B00A17 HOLTKAMP AND BRUDZINSKI: CASCADIA SLOW SLIP EPISODES B00A17

Figure 3. (continued)


8 of 21

B00A17



9 of 21

B00A17

American plate calculated in this study, which are expectedto represent earthquake cycle strain accumulation [Dragertand Hyndman, 1995; Savage, 1983]. Best fit linear trendsare calculated from GPS time series after steps due toearthquake and hardware resets, and sinusoidal annual andsemiannual signals are removed [Szeliga et al., 2004]. Atthis point, slow slip episodes are not removed from the timeseries, so in cases where slow slip events dominate the timeseries it may take several years to produce an accuratemeasurement, even with continuously recording data. Datapreceding 1997 is not included in this calculation as manystations have significantly larger scatter. This early scattercan be seen in the ALBH longitudinal time series (Figure 2),where data after 1997 have a 74% smaller variance thandata prior to 1997.[21] To illustrate long-term velocities from GPS data

relative to the stable North American, the best fit lineartrend for the station DRAO is subtracted from all stations inthe network, as DRAO effectively represents the stableportion of the overriding plate [e.g., Dragert and Hyndman,1995]. In general, these velocities are similar to previousstudies [e.g., McCaffrey et al., 2000; Wang et al., 2003],showing larger velocities near the trench associated with theproximity of a strongly coupled subduction interface. Infact, the shape of the contours to the trend values near thecoast matches that of the depth contours to the plateinterface well (compare Figures 1 and 7). The directionsmatch that of the relative plate motion direction between thePacific and North American plate well, with the mostnoticeable exception occurring across Oregon. Following

the work of Wells et al. [1998], Figure 7b shows thevelocities with the Oregon block rotation removed for nearcoastal stations south of 46.5� latitude. Removing theOregon block rotation results in velocities that are muchcloser to the expected values of trench perpendicular short-ening, especially in northern California and southern Ore-gon [e.g.,McCaffrey et al., 2000;Wang et al., 2003;Wells etal., 2002]. When corrected for the Oregon block rotation,plate convergence rate is fastest along the northern part ofthe interface with a convergence rate of �37 mm/yr,�31 mm/yr in the Oregon area, and �34 mm/yr in thenorthern California area [Wells et al., 2002]. There is anapparent discontinuity in the long-term motion betweenstations in southern Washington (moving to the north) andstations in northern Oregon (moving to the south), but this isdue in part to the discontinuous Oregon block boundaryimposed at a specific latitude. This boundary is likely morediffuse in reality, but our current approach does not attempt toaccount for those complexities.3.2.2. Inter-ETS Velocities[22] We next turn our attention to the short-term site

velocities when transient motion is not occurring. Usingour catalog of slow slip events, we can examine the trendsin the PANGA GPS time series relative to the stable NorthAmerican plate after slow slip events have been removed(Figure 7). We denote this value as the inter-ETS velocitysince it is the velocity occurring between ETS episodes. Tocalculate this velocity, we find the mode of the best fit linearslopes of a scrolling 1.5 year window in the cleaned datawhere transient displacements have been removed using the



10 of 21

B00A17

times and displacements of slow slip episodes from ourhyperbolic tangent fitting algorithm. We discuss our tests ofother approaches to calculating the inter-ETS velocity in theappendix, but the results are quite similar (Figure 8), so weuse the mode estimate as the inter-ETS velocity for analysisin section 4. To help illustrate the different velocitiescalculated from the GPS time series, Figure 8 shows the timeseries for station ALBH after different steps in the processingand the residuals from zero before and after the hyperbolictangent fits have been removed from the time series.

4. Discussion

[23] Accurately identifying slow slip events and con-straining plate velocities are the first steps in determiningpatterns of strain accumulation and release throughout thesubduction zone. In this section, we use slow slip events anddeformation velocities to compare values of strain accumu-lation and release and highlight trench normal and trenchparallel patterns.[24] In Cascadia, the fully locked portion of the interface

has been shown to extend to a depth of �10 km [Dragert etal., 1994; McCaffrey et al., 2007; Wang et al., 2003], with

the transition zone to the stable sliding region extending tobelow 40–45 km depth [Fluck et al., 1997; McCaffrey etal., 2000; Wang et al., 2003]. Temperature has beenproposed as the primary control on the frictional propertiesof the interface and the downdip change from velocityweakening to velocity strengthening behavior [Hyndmanand Wang, 1993]. It is in the transition area from theseismogenic stick slip portion of the interface to theplastically deforming zone where ETS events are thoughtto occur [Dragert et al., 2001; Rogers and Dragert, 2003].

4.1. Episodic Tremor and Slip

[25] The intrigue surrounding slow slip events washeightened when episodes observed at stations like ALBHwere linked spatial and temporally to periods of NVT[Rogers and Dragert, 2003]. The expanded catalog of slowslip events in this study can be compared with a newexpanded catalog of NVT episodes to determine how welllinked these two phenomenon are [Brudzinski and Allen,2007]. There are 41 events during the time studied thatshow correlated observations of tremor and slip. A handfulof cases exist where observations are detected in GPS orseismic data but not in both sets of data, and we find most ofthese cases occur in areas with few recording stations (e.g.,central Washington area in Figure 3j). Correlated tremor andslip are found abundantly in every area of the margin wherethere are stations to record it, so it appears to be a processinherent to subduction in this system.[26] While ETS is common along the entire Cascadia

margin, events clearly do not occur all along the margin atthe same time [Brudzinski and Allen, 2007; McCausland etal., 2005; Szeliga et al., 2004], and we summarize theseoffsets in space and time over the whole time series inFigure 9, which shows activity with respect to along-strikedistance over time. The along-strike distance for each stationis from the southern end of the margin to a point on the 40 kmdepth-to-slab contour that is closest to the station.[27] To help highlight ETS events, we use gray bars to

connect stations that record events within 40 days and125 km of one another. This works well to both identifyETS occurrence and extent in regions of high station density(northern Cascadia), but is more limited in regions withfewer stations (southern Cascadia). In regions of lowerstation density, common timing from GPS and seismicstations indicate ETS occurrence, but spatial extent of theevents is poorly resolved. In regions of higher stationdensity, it is important to recall that we only plot stationlocations instead of source locations, so we would expectthe gray lines to extend tens of kilometers beyond the actualsource locations. Nevertheless, these observations providestrong evidence (4 or more direct geodetic or seismicobservations) for 41 ETS events between 1997 and mid-2007. In addition to this, there are several more potentialETS events that have fewer than three observations butshow similar signals in their respective time series and occurwhen expected given the well-determined recurrence inter-vals (e.g., early in the time series for northern VancouverIsland and central Oregon).

4.2. Along-Strike Segmentation

[28] Segmentation is a phenomenon already observedalong the megathrust of subduction zones throughout the

Figure 4. Recurrence intervals between slow slip eventsfor stations with at least two well-recorded events. Numbersrepresent one standard deviation from the mean recurrenceinterval, and they are absent when the interval is calculatedfrom only two events.


11 of 21

B00A17

world, as earthquakes of tend to be interrupted along strikein similar regions [e.g., Ruff, 1996; Bilek, 2010]. Megathrustsegmentation has been shown to subdivide into localizedareas of large coseismic slip called asperities [Lay et al.,1982; Burgmann et al., 2005], which may be linked to fore-arc structure through gravity and bathymetry [Fuller et al.,2006; Song and Simons, 2003; Wells et al., 2003]. Thediscovery of segmentation of ETS events suggests that theslow deformation and rapid deformation associated withearthquakes may have similar controlling factors. Relation-ships between the two types of segmentation is the nextstep, but the megathrust segmentation is not well known inCascadia due to the long times (�500 years) between greatearthquakes [e.g., Atwater and Hemphill-Haley, 1997].Brudzinski and Allen [2007] suggest that the two types ofsegmentation are linked by showing that ETS segmentboundaries outline approximate boundaries for fore-arcbasins. Further tests, such as comparing inverted slip dis-tributions to fore-arc basins or gravity lows, could strength-en this apparent connection between fore-arc structure,megathrust asperities, and ETS events.[29] In our study of individual slow slip events found in

GPS time series over the past 10 years, we have found agood deal of evidence for along-strike segmentation that issupported by similar time offsets in nonvolcanic tremorepisodes at neighboring stations [Brudzinski and Allen,

2007]. Inspection of these events together in Figure 9 showsthat the entire subduction margin can be divided into at least7 segments, defined by groups of stations that tend to movetogether at the same time. Horizontal solid lines markapproximate boundaries where events on either side of theboundary are separated by at least 40 days more than half ofthe time (estimated visually in Figure 9), and are thusthought to be significant boundaries to ETS propagation.Dotted lines mark boundaries that are separated by 40 dayson at least 3 occasions but not enough to divide episodesover half of the time, and are thus thought to represent weakboundaries. Brudzinski and Allen [2007] also show smalltiming offsets clearly in NVT signals as the source zonepropagates across segment boundaries, adding to evidencefor segmentation. The along-strike widths of segments inFigure 9 range from 100 to 255 km, but these estimates maybe longer than reality in regions with low station density.[30] It is important to underscore that the way we have

defined segments allows for neighboring segments to beactive at the same time. The 2003 event in northernCascadia that was the focus of previous study would beone such example [Melbourne et al., 2005], since thislarge event spanned three segments as we define them. Wefind a few segments are particularly ‘‘interactive’’ suchthat they are active along with neighboring segments morethan three times as often as they are active by themselves.

Figure 5. Estimates of the typical slow slip process along the Cascadia margin determined fromhyperbolic tangent fits to the GPS time series. (a) Vectors represent annual slow slip displacementmagnitudes and directions. Contour lines highlight geographic variation in magnitude. (b) Vector arrowsrepresent average event displacement magnitudes and directions.


12 of 21

B00A17

Segments 4b and 6 are interactive under this definition.For example, segment 6 moves with segment 5 in four cases(Figures 3a–3c and 3m), in two cases it moves withsegment 7 (Figures 3e and 3j), and in only one case doesit move by itself (Figure 3d). It appears that we can resolvethese interactive segments due to the relatively high stationdensity along the northern part of the interface. However,the number of stations in the Oregon and California regionsis lower, making it difficult to discern whether interactivesegments exist there as well. The ‘‘interactive’’ nature ofthese events may also have an impact on the calculatedrecurrence intervals, particularly for stations that do nothave a long history, as a segment could be influenced byactivity in adjacent segments, causing events to occur earlieror later than expected.

4.3. Strain Accumulation Rates

[31] To examine how observed surface displacements arerelated to strain accumulation, we begin by analyzing thegeographic distribution. Figures 10a and 10b show all theobserved displacements (as plotted for ALBH in Figure 8a)

with respect to height above the plate interface for stationsmore than 20 km away from strong segment boundariesdefined in section 4.2. We use height above the interfaceinstead of distance from the trench in order to correct for thevariability in slab dip angle along Cascadia and because theETS process has shown a strong depth dependence [Dragertet al., 2001; Kao et al., 2006]. Values plotted againstdistance from trench show a similar trend to those in Figure10, but the correspondence between inter-ETS velocity andthe annual slow slip displacement is more clearly resolvedusing height above the plate interface, presumably due toalong-strike variations in slab dip. In addition to plottingvalues for each station, we calculate a running mean over5 stations to help distinguish overall trends.[32] Slow slip parameters determined from stations near

the spatial edges of segments are less regular in directionand smaller in magnitude than stations geographicallycloser to the center of the segment, and since strongsegment boundaries define the spatial edge of most slipevents, we expect stations near these boundaries shouldexhibit smaller displacements. Figure 10c shows that annualslow slip displacement rates for all stations within 20 km ofsegment boundaries (circles) are lower than the runningmean line (blue line) calculated from stations further away.Furthermore, the inter-ETS velocities for most stations nearboundaries (triangles) are also lower than stations fartheraway (green line). Stations near the segment boundariesmove less before events and less during events, whichindicates that the plate interface near segment boundariesis accumulating and releasing strain at slower rates. Thissuggests the plate interface below these regions is lesscoupled than toward the center of the segment, exhibitingmore of a velocity strengthening behavior than neighboringareas of the interface at the same depth. This could meanthere are regular asperity patches that recur in ETS events.Three-dimensional modeling results from Szeliga et al.[2008] taking into account the curvature of the slab supportthis observation as their modeled fault slip appears to havegeometric regularity and agrees with our segment bound-aries. Szeliga et al. [2008] chose to restrict their modeling toevents that have high station density, while our studyattempts to expand the analysis to the whole length of themargin, in part to investigate observations of segmentation.[33] Interplate slip deficit recorded at the surface as long-

term plate velocity is expected to be caused by locking inthe seismogenic zone of the plate interface. Previousresearchers have also used long-term velocities to representinterseismic strain rate [Wang et al., 2003], but we nowfocus on the inter-ETS velocity to investigate the total strainrate that is occurring between ETS events. We find that theinter-ETS velocities are most different from the long-termvelocities at about 40–60 km above the interface. Thedivergence from the long-term velocities at these depths isnot an artifact of collapsing the data along the strike of theCascadia margin, as we find this trend in each of the threemain regions from north to south (Figure 10d). The similarbump in inter-ETS velocities at depth across all 3 regionsalso helps to confirm that these signals are not due toinappropriate treatment of the Oregon black rotation, asthis would only affect stations plotted as stars in Figure 10d.We will investigate this situation further with elastic half-space modeling in section 4.4.

Figure 6. Variability of slow slip direction from the set ofevents recorded. Pie wedge plotted behind mean slipdirection arrow represents one standard deviation of thesurface displacement direction, in degrees. Stations showvariabilities ranging from 5� to 57�. Arrow lengths representannual slow slip displacement. Only stations that recorded 3or more events are shown.


13 of 21

B00A17

[34] We then define the difference between the inter-ETSvelocity and the long-term velocity to be the additionalshort-term velocity (Figure 8). This short-term velocity is arecord of the slip deficit recorded in the inter-ETS velocitythat does not contribute to the long-term velocity. Instead ofcontributing to the long-term velocity, this should be the slipdeficit recovered during slow slip events. Figure 10a showsthat our calculations of the short-term strain accumulation(blue dashed line) and slow slip strain release (blue solidline) are similar through the ranges of 20–60 km abovethe plate interface, where larger slow slip magnitudesmake the slow slip strain release calculation more accurate.Furthermore, these observations are consistent with previousstudies that have identified the source zone of slow slip[Dragert et al., 2001; Melbourne et al., 2005] and nonvol-canic tremor [Kao et al., 2006; Rogers and Dragert, 2003] tobe roughly bounded by the 25 and 55 km depth contours.Based on these previous results, our colleagues have inter-preted this process to be associated with the plate interfacetransition zone.

4.4. Plate Interface Coupling

[35] We next assume that the cumulative surface displace-ments U at each GPS site during the slow slip events are anelastic response to dislocations along the subduction inter-

face [e.g., McGuire and Segall, 2003] and examine the fitsof a range of forward models to the observations. Likewise,the inter-ETS velocities can be investigated with a range ofcoupling models that produce elastic strain accumulationnear the plate interface. In several regions along Cascadiathere are a limited number of sites that recorded the slowslip displacements or even the long-term velocities, whichprovide too little information to reliably recover the locationand magnitude of the transient slip and preceding back slipvia a formal inversion. However, the displacements docontain sufficient information to reject or accept broadranges of potential slip distributions, particularly when weconsider continuous GPS sites from all across Cascadiarelative to their depth above the plate interface in a simple2-D configuration. We model coupling as the percentage ofthe convergence rate that is needed on the plate interface toexplain the surface motions. To accomplish this modeling,we use DISL software [Larsen, 1991], which calculatesdisplacements on the surface of a homogeneous elastic half-space in response to a user-specified slip distribution alonga fault embedded within the half-space. The analytic sol-utions embedded in this code to determine the elasticdisplacements are from Mansinha and Smylie [1971]. Thesubduction interface we use for our forward modeling has a

Figure 7. (a) Long-term average and inter-ETS GPS velocities. Best fit linear trends (long-termaverage) are calculated from PANGA GPS time series with steps due to earthquake and hardware resets,and sinusoidal annual and semiannual signals are removed. Inter-ETS velocities calculated from thelinear slope of the GPS time series after slow slip offsets have been removed. Values are plotted relativeto station DRAO, taken to represent the stable North American plate. (b) Same as Figure 7a but withOregon block rotation removed [e.g.,Wells et al., 1998]. Error ellipses in Figures 7a and 7b are 2s values.Only stations with more than 3.5 years of data are plotted, as these have considerably less uncertainty(Figure S1) and are used in subsequent analyses (Figures 10 and 11).


14 of 21

B00A17

Figure 8. (a) Longitude component of the GPS time series of ALBH relative to DRAO. Colored linesshow best fitting trends for different facets of the time series, and legend indicates terms we useto describe them. (b) Residuals of the detrended time series before hyperbolic tangent fits are removed.(c) Residuals after best fitting long-term rate and best fitting short-term rates are removed. (d) Residualsafter hyperbolic tangent fits were removed from the data for station ALBH.


15 of 21

B00A17

uniform dip of 15� and we use an average convergencevelocity of 40 mm/y.[36] Previous studies in Cascadia have used observed

GPS velocities to arrive at a locking model to describe theamount of coupling on the plate interface due to interseis-mic strain accumulation [McCaffrey et al., 2007; Wang etal., 2003]. Applying these estimates of how coupling varieswith respect to depth in our 2-D modeling approach, we findthis model works well to explain our long-term platevelocities (Figure S2). Since the interseismic strain accu-mulation is not released on the time scale of our observa-tions, the inter-ETS velocity necessarily contains the entireinterseismic signal. These velocities decays as depth to thesubduction interface increases, but they diverge from thelong-term trends expected from the interseismic cycle at�30–60 km above the interface. We test different concep-tual models of the coupling on the interface to produce theobserved inter-ETS velocities and quickly find that an areaof significant coupling near the ETS source zone is neces-sary to reproduce the observed inter-ETS velocities. Re-gardless of which model we choose for how couplingchanges with depth, a patch of at least 30% coupling from20 to 35 km depth is needed to produce the observed backslip displacements that eventually result in slow slip events.[37] To examine this deeper source region of additional

coupling, we tested a variety of models where the previ-ously accepted coupling function was adjusted to includegreater coupling at depth. First, we investigated a couplingmodel that resembles the pattern in surface velocities with adecay in coupling that flattens out before resuming the

decay at greater depth (Figure 11a). To achieve this wecombine a hyperbolic tangent function with the previouslyaccepted coupling function. The center, amplitude, andwidth of the hyperbolic tangent function were varied via agrid search, and in each case the surface displacements inbetween and during slow slip events were predicted fromthe coupling function with the elastic forward modelingcode. There are a handful of models for coupling using thisapproach that predicted surface displacements within theerror bounds of our measurements (Figures 11a and 11b,middle), while most of the models were unable to match ourobservations (Figures 11a and 11b, bottom). In order to fallwithin the error bounds, successful models need �30%coupling down to depths of about 60 km, which is deeperthan previous inversions for the source of slow slip events[Dragert et al., 2001; Melbourne et al., 2005]. This set ofmodels does not work because the shape of the predictedinter-ETS surface displacements with respect to depth of theinterface do not show the minimal change in velocitiesobserved from 40 to 60 km above the interface (Figures 11aand S2). In other words, there is still not enough coupling atdepth to temporarily halt the reduction in surface velocitiesthat we observe with increasing distance from the trench.[38] To better approximate the minimal change in surface

velocities above the slow slip zone, we next investigated acoupling model with a Gaussian peak added to the previ-ously accepted coupling function (Figure 11c). We againpredicted surface displacements for an elastic half-spacewith the depth, amplitude, and width of this peak varied viaa grid search. There is a larger set of successful models with

Figure 9. Summary of episodic tremor and slip (ETS) along the subduction margin determined fromindividual GPS and seismic observations in Figure 3. Distance along strike is from the southern end ofthe margin and runs along the 40 km depth-to-slab contour. The map is adjusted to distance along strikeversus distance from the 40 km contour. Horizontal lines propose segmentation of the margin into groupsof stations that tend to show common ETS timing. Gray bars are produced by automated code thatidentifies observations that are spatially and temporally coherent. Low station density in southern Oregonprevents precise determination of the segmentation boundary around 450 km. Segments of 4b and 6appear to be interactive, with variable timing depending on events in neighboring segments.


16 of 21

B00A17

this approach, and the best fitting models provide a bettermatch to the shape of predicted surface displacements,reducing the variance by more than 30% compared tocoupling models using the hyperbolic tangent function.Intriguingly, our best fitting models have a decrease incoupling down to �30% at �20 km depth followed by apeak of greater than 80% coupling at �30–35 km depth,which corresponds to the location of slip on the interfaceinverted for from GPS observations. For the Cascadia

margin, 80% coupling corresponds to an �32 mm/yr slipdeficit rate, a value that is similar to inversions of slipmagnitude during ETS events. These models indicate adistinct depth separation in strongly coupled regions sug-gesting that the source zone for ETS events acts somewhatlike a second locked zone that releases strain more fre-quently than the updip seismogenic locked zone.[39] While our modeling shows a strong preference for

significantly increased coupling at depth, we cannot yet ruleout a zone of constant �30% coupling. This is a result ofthe wide error bounds on our surface displacements due tothe station spacing of GPS observatories during our studytime frame and the act of collapsing these stations alongstrike to simulate a 2-D situation. However, preliminaryinversions of these data taking into account the 3-D natureof the subducting slab do reveal more spatial heterogeneity,but they support the conclusion that patches of strongercoupling occur downdip and distinct from the seismogeniczone [Holtkamp et al., 2007]. Szeliga et al. [2008] also used3-D modeling to invert for fault slip and show that eventshave 2–3 cm of slip, with some cases displaced signifi-cantly inland from the predicted seismogenic zone (e.g.,July 1998 and May 2004). Furthermore, a study of a smallerregion in the Oaxacan segment of the Middle AmericanSubduction Zone combining dense campaign measurementswith some continuously recording sites finds an area ofnearly full coupling downdip from the locked zone [Correa-Mora et al., 2008]. There is currently no evidence to suggestthis feature is continuous along strike in Mexico, suggestingthat the patch of stronger coupling at depth we are inter-preting in Cascadia may be discontinuous along strike.Nevertheless, the detailed study from Mexico supports the

Figure 10. (a) Comparison of GPS-derived velocitiesrelative to the height above the subduction interface forstations more than 20 km from a segment boundary.Colored lines are the same as in Figure 8. Since slab dipvaries along the margin, height above slab is moresignificant than distance from trench. Yearly slow slip rateand short-term strain accumulation rate should match in ourcalculations, and both show peaks from �25 to �55 km,consistent with a source zone along the deeper, transitionalzone of the plate interface. The 3 top curves that representcalculations of velocities between slow slip events alsomatch, and they also show a distinct drop in displacementsafter �55 km. (b) Symbols indicate individual stationmeasurements for stations >20 km from a segmentboundary, colored lines are running means, and dottedcurves are standard deviations for the inter-ETS velocity(triangles) and the annual slow slip magnitude (circles).(c) Symbols show values for stations within 20 km of asegment boundary indicating rates between �25 and�55 km are consistently smaller for these stations, consis-tent with an asperity model for slow slip events that isgeographically regular over time. (d) Comparison of inter-ETS velocity and annual slow slip displacements for the3 main geographic regions of the Cascadia subduction zone.Each region shows divergence from the expected long-termvelocity at stations above the 30–60 km depth to slabcontours.

17 of 21

B00A17 HOLTKAMP AND BRUDZINSKI: CASCADIA SLOW SLIP EPISODES B00A17

idea that the coupling can be significantly higher than onewould expect from typical models with a gradual decline.

4.5. Thermal andMechanical Models of the Lithosphere

[40] Evidence for a large degree of coupling at 30–35 kmdepth and the distinct separation between the locked and ETSsource zones are surprising, particularly given the temper-atures involved. Temperature estimates from heat flow datashow that the 450� isotherm occurs near 25–30 km depth[Fluck et al., 1997; Hyndman and Wang, 1993], indicatingthat large amounts of strain accumulation are occurring attemperatures above 450�. Meanwhile, a local minimum in thecoupling ratio occurs at depths where the temperature isbetween 350� and 450�. These opposing relationships indi-cate that temperature is not the only control on frictionalcoupling, and other factors need to be considered. Newresults from New Zealand showing slip at distinctivelydifferent depths (and thus temperatures) support the hypoth-esis that temperature is not the only control on frictionalcoupling [McCaffrey et al., 2008].[41] A change in material composition along the fault is

an attractive hypothesis for the apparent increase in cou-pling at depth, and the largest compositional change alongthe subducting interface is expected to be crossing the upperplate Moho. As such, we propose that the mechanical

strength of the overriding lithosphere is the other key factorthat is controlling the frictional properties on the plateinterface. Studies of intraplate earthquakes and laboratorymeasurements of representative geologic materials haveproduced evidence that is consistent with reduced strengthin the lower crust due to the thermally induced brittle-ductile transition near 350� [e.g., Chen and Molnar, 1983;Kohlstedt et al., 1995]. In our study, we find a reduction incoupling ratios at depths near 20 km where the temperatureis above 350�, consistent with a reduction in strength of theoverriding lower crust. However, identification of subcrust-al earthquakes and lab measurements of olivine rheologyare consistent with the brittle-ductile transition occurring athigher temperatures in mantle compositions. The increase incoupling ratios we find below 30 km depth could beexplained by the subducting plate encountering the strongerupper mantle of the overriding plate (Figure 12).[42] Although precise measurements have not yet been

made, we estimate the depth where the plate interfacecrosses the Moho to be �33 km from layering establishedin scattered teleseismic wavefield studies [e.g., Bostock etal., 2002; Nicholson et al., 2005]. Based on this crossoverdepth, it does appear the majority of slow slip (this paper)and NVT locations [Brudzinski and Boyarko, 2010] occuradjacent to the overriding mantle. This model is also

Figure 11. Results from forward modeling of surface displacements using models that combine thepreviously determined coupling function with a hyperbolic tangent function. (a) Running average of theinter-ETS velocity (green solid curve) and standard deviation of observations (green dotted curve)compared with predicted velocities from the best fitting model for coupling. (b) Running average of theannual slow slip displacements (blue solid curve) and standard deviation of observations (blue dottedcurve) compared with predicted velocities when the interseismic coupling function is subtracted from thebest fitting model. (c, d) Same as Figures 11a and 11b except using modeling of surface displacementswith a Gaussian peak. The best fitting models with a peak in coupling at depth are better able to match theshape of the observed inter-ETS velocity and result in significantly lower variance. The middle plotsshow coupling functions (gray curves) that predicted displacements within the error bounds, where theblack line represents the best fitting model. The bottom plots show coupling functions that cannot predictdisplacements within the error bounds.


18 of 21

B00A17

consistent with new results in southwest Japan where denseinstrumentation has been used to determine high-precisionsource locations of NVT and slow slip events that occur attemperatures of 400–500 along the plate interface [Obara,2002; Obara and Hirose, 2006; Shelly et al., 2006;Yoshioka and Murakami, 2007], which is below wherethe plate interface meets the upper plate Moho [Kurashimoet al., 2002; Nakamura, 2002; Murakoshi, 2003; Salahand Zhao, 2004].[43] An intriguing implication of this hypothesis is that

the upper mantle creates higher coupling on the interfacedespite evidence for significant serpentinization [e.g.,Bostock et al., 2002]. Serpentinization is often thought tolower strength and has been proposed as a control on thedowndip limit of great thrust earthquakes [e.g., Escartın etal., 2001]. However, measurements on serpentine faultgouges at progressively higher temperatures and pressuressuggest the coefficient of friction increases dramatically[Moore et al., 1997] and the average shear strength ofantigorite fault gouge can approach 50 MPa [Moore et al.,1996]. The idea that serpentine is weak may have more todo with serpentinite-bearing fault sections being conduciveto the generation of high fluid pressures, through theformation of permeability barriers that trap fluids withinthe fault zone [e.g., Escartın et al., 1997].[44] The variation in strength of serpentine from strong to

weak depending on the pore fluid pressures could helpexplain the contradiction of the high inter-ETS coupling weobserve and independent evidence that suggests low effec-tive stress is necessary for slow slip to occur. The evidence

for low effective stress includes rate-and-state frictionmodels that need to rationalize the large spatial dimensionand short (�1 yr) recurrence of slow slip [Liu and Rice,2005, 2007] and observed modifications to NVT activityfrom small stress changes due to tides or passing seismicwaves [Gomberg et al., 2008; Rubinstein et al., 2008, 2007;Shelly et al., 2007]. Recent studies examining the velocitystructure near the plate interface at these depths also findanomalies in Vp, Vs, and Vp/Vs, consistent with high porefluid pressures in the subducting crust [Audet et al., 2009;Kodaira et al., 2004; Shelly et al., 2006; Song et al., 2009].Frictional models that incorporate the tendency for porevolume to increase with shear, decreasing pore pressure andincreasing frictional resistance, can explain how slip insta-bilities nucleate but become quenched before reaching fastearthquake slip speeds, as pore pressures drop faster thanthey can be replenished [Segall and Rice, 1995; Segall etal., 2008].

5. Conclusion

[45] Using continuous GPS data recorded over the last10 years, we examine slow slip events along the entireCascadia subduction margin in an attempt to better under-stand the dynamics associated with subduction deformation.We use a hyperbolic tangent curve fitting algorithm to scanthe GPS time series and extract slow slip event times andmagnitudes. To demonstrate the effectiveness of our tech-nique, we investigated 10 representative stations where61 cases of slow slip were visually determined, and the

Figure 12. Comparison of (left) a schematic diagram for the variation of the mechanical strength of thecrust and uppermost mantle with depth and (right) the best fitting model for coupling ratio with depthfrom Figure 11. Mechanical strength is from Chen and Molnar [1983] and is estimated for cold (solidcurve), intermediate (dashed curve), and warm (dotted curve) geotherms. The top part of the curverepresents stick slip or brittle failure behavior and is based on a linear relationship between the shearstress and normal stress and Byerlee’s law for friction. The strength in the lower crust and upper mantle iscontrolled by flow laws of crustal and mantle materials, respectively. The dashed curve is smoothed toindicate possible gradual changes of strength in the brittle-ductile transition zone and the crust-mantleboundary. This curve matches the overall pattern of coupling with depth found in our study.


19 of 21

B00A17

algorithm finds only 1 false negative and 8 false positives.Applying the algorithm to the entire data set, we haveconstructed a catalog of 215 observations of slow sliprecorded on 60 stations across the entire margin. About halfof these stations have enough events to calculate reliablerecurrence intervals, with a geographic pattern of 3 regionsof coherent intervals consistent with previous studies.[46] In our study of individual ETS events, we find that

the entire subduction margin could be divided into at least 7groups of stations, or segments, which tend to move witheach other during slow slip events. Segment boundaries aredefined by areas where slow slip displacements fail topropagate through at least half of the time. These segmentsvary in size from 100 to 255 km along strike, and can differfrom other segments in terms of recurrence interval andconsistency with timing in neighboring segments. By re-moving the slow slip displacements from the GPS timeseries, we were then able to calculate the station velocitiesbetween slow slip events for comparison with the long-termaverage velocities. Comparison of these values gives anestimate of the slip deficit rate that is released in the form ofslow slip events, and allows us to quantitatively compareearthquake cycle strain accumulation and slow slip cyclestrain accumulation at points all along the Cascadia margin.[47] Investigating a range of simple 2-D models of

coupling on the plate interface, we find the inter-ETSvelocities are best explained by a large, distinct peak incoupling at �33 km depth. However, enough scatterremains in the current set of inter-ETS velocities that wecannot yet rule out a model with a zone of constant �30%coupling. This result suggests that the source region for ETScannot be described by simple temperature-dependent fric-tional coupling, but instead the deeper transition zone actsmore like a second, independent locked zone that slips morefrequently. We propose that coupling initially decreaseswith depth as the subducting plate encounters the weakerlower crust of the overriding plate but that the couplingincreases again when it encounters the stronger uppermantle of the overriding plate.

[48] Acknowledgments. This project was possible thanks to thePANGA data set availability. Our analysis also utilized seismic data fromthe PNSN, BDSN, GSN, PBO, and OATS obtained via the IRIS DMC. Webenefited from discussions with C. DeMets, H. Dragert, K. Feigl, T. Lowry,R. McCaffrey, T. Melbourne, W. Szeliga, and M. Pritchard. Two reviewersand the associate editor provided beneficial comments which improved ourmanuscript. This project is supported by the National Science Foundationgrants EAR-510812 and EAR-642765 and by the Undergraduate SummerScholars program at Miami University.

ReferencesAtwater, B. F., and E. Hemphill-Haley (1997), Recurrence intervals forgreat earthquakes of the past 3,500 years at northeastern Willapa Bay,Washington, U.S. Geol. Surv. Prof. Pap., 1576, 108 pp.

Audet, P., M. G. Bostock, N. I. Christensen, and S. M. Peacock (2009),Seismic evidence for overpressured subducted oceanic crust and mega-thrust fault sealing, Nature, 457, 76–78, doi:10.1038/nature07650.

Bilek, S. L. (2010), Seismicity along the South American subduction zone:Review of large earthquakes, tsunamis, and subduction zone complexity,Tectonophysics, doi:10.1016/j.tecto.2009.02.037, in press.

Bostock, M. G., R. D. Hyndman, S. Rondenay, and S. M. Peacock (2002),An inverted continental Moho and serpentinization of the forearc mantle,Nature, 417, 536–538, doi:10.1038/417536a.

Brudzinski, M. R., and R. M. Allen (2007), Segmentation in episodictremor and slip all along Cascadia, Geology, 35, 907–910, doi:10.1130/G23740A.1.

Brudzinski, M. R., and D. C. Boyarko (2010), Spatial and temporal patternsof nonvolcanic tremor along the southern Cascadia subduction zone,J. Geophys. Res., doi:10.1029/2008JB006064, in press.

Brudzinski, M., E. Cabral-Cano, F. Correa-Mora, C. DeMets, andB. Marquez-Azua (2007), Slow slip transients along the Oaxaca subduc-tion segment from 1993 – 2007, Geophys. J. Int., 171, 523 – 538,doi:10.1111/j.1365-246X.2007.03542.x.

Burgmann, R., M. G. Kogan, G. M. Steblov, G. Hilley, V. E. Levin, andE. Apel (2005), Interseismic coupling and asperity distribution alongthe Kamchatka subduction zone, J. Geophys. Res., 110, B07405,doi:10.1029/2005JB003648.

Chen, W.-P., and P. Molnar (1983), Focal depths of intracontinental andintraplate earthquakes and their implications for the thermal and mechan-ical properties of the lithosphere, J. Geophys. Res., 88, 4183–4214,doi:10.1029/JB088iB05p04183.

Correa-Mora, F., C. DeMets, E. Cabral-Cano, B. Marquez-Azua, andO. Diaz-Molina (2008), Interplate coupling and transient slip alongthe subduction interface beneath Oaxaca, Mexico, Geophys. J. Int.,175, 269–290, doi:10.1111/j.1365-246X.2008.03910.x.

Dragert, H., and R. D. Hyndman (1995), Continuous GPS monitoring ofelastic strain in the northern Cascadia subduction zone, Geophys. Res.Lett., 22, 755–758, doi:10.1029/95GL00469.

Dragert, H., R. D. Hyndman, G. C. Rogers, and K. Wang (1994), Currentdeformation and the width of the seismogenic zone of the northernCascadia subduction thrust, J. Geophys. Res., 99, 653–668, doi:10.1029/93JB02516.

Dragert, H., K. L. Wang, and T. S. James (2001), A silent slip event on thedeeper Cascadia subduction interface, Science, 292, 1525 – 1528,doi:10.1126/science.1060152.

Escartın, J., G. Hirth, and B. Evans (1997), Nondilatant brittle deformationof serpentinites: Implications for Mohr-Coulomb theory and the strengthof faults, J. Geophys. Res., 102, 2897–2913, doi:10.1029/96JB02792.

Escartın, J., G. Hirth, and B. Evans (2001), Strength of slightly serpenti-nized peridotites: Implications for the tectonics of oceanic lithosphere,Geology, 29, 1023 –1026, doi:10.1130/0091-7613(2001)029<1023:SOSSPI>2.0.CO;2.

Fluck, P., R. D. Hyndman, and K. Wang (1997), Three-dimensionaldislocation model for great earthquakes of the Cascadia subduction zone,J. Geophys. Res., 102, 20,539–20,550, doi:10.1029/97JB01642.

Fuller, C. W., S. D. Willett, and M. T. Brandon (2006), Formation of forearcbasins and their influence on subduction zone earthquakes, Geology, 34,65–68, doi:10.1130/G21828.1.

Gomberg, J., J. L. Rubinstein, Z. G. Peng, K. C. Creager, J. E. Vidale, andP. Bodin (2008), Widespread triggering of nonvolcanic tremor inCalifornia, Science, 319, 173, doi:10.1126/science.1149164.

Holtkamp, S., M. Brudzinski, and M. Pritchard (2007), Constrainingtransient vertical deformation associated with slow slip events in Cascadia,Eos Trans. AGU, 88(52), Fall Meet. Suppl., Abstract T21A-0366.

Hyndman, R. D., and K. Wang (1993), Thermal constraints on the zone ofmajor thrust earthquake failure: The Cascadia subduction zone, J. Geo-phys. Res., 98, 2039–2060, doi:10.1029/92JB02279.

Kao, H., S.-J. Shan, H. Dragert, G. Rogers, J. F. Cassidy, K. Wang,T. S. James, and K. Ramachandran (2006), Spatial-temporal patterns ofseismic tremors in northern Cascadia, J. Geophys. Res., 111, B03309,doi:10.1029/2005JB003727.

Kawasaki, I., Y. Asai, Y. Tamura, T. Sagiya, N. Mikami, Y. Okada,M. Sakata, and M. Kasahara (1995), The 1992 Sanriku-oki, Japan,ultra-slow earthquake, J. Phys. Earth, 43, 105–116.

Kodaira, S., T. Iidaka, A. Kato, J. O. Park, T. Iwasaki, andY. Kaneda (2004), High pore fluid pressure may cause silent slip in theNankai Trough, Science, 304, 1295–1298, doi:10.1126/science.1096535.

Kohlstedt, D. L., B. Evans, and S. J. Mackwell (1995), Strength of thelithosphere: Constraints imposed by laboratory experiments, J. Geophys.Res., 100, 17,587–17,602, doi:10.1029/95JB01460.

Kurashimo, E., et al. (2002), Geometry of the subducting Philippine Seaplate and the crustal and upper mantle structure beneath eastern ShikokuIsland revealed by seismic refraction/wide-angle reflection profiling,J. Seismol. Soc. Jpn., 54, 489–505.

Larsen, S. (1991), Geodetic measurements of deformation in southernCalifornia, Ph.D. thesis, Calif. Inst. of Technol., Pasadena.

Larson, K. M., A. R. Lowry, V. Kostoglodov, W. Hutton, O. Sanchez,K. Hudnut, and G. Suarez (2004), Crustal deformation measurementsin Guerrero, Mexico, J. Geophys. Res., 109, B04409, doi:10.1029/2003JB002843.

Lay, T., H. Kanamori, and L. Ruff (1982), The asperity model and thenature of large subduction zone earthquakes, Earthquake Predict. Res.,1, 3–71.

Liu, Y. J., and J. R. Rice (2005), Aseismic slip transients emerge sponta-neously in three-dimensional rate and state modeling of subduction earth-


20 of 21

B00A17

quake sequences, J. Geophys. Res., 110, B08307, doi:10.1029/2004JB003424.

Liu, Y., and J. R. Rice (2007), Spontaneous and triggered aseismic defor-mation transients in a subduction fault model, J. Geophys. Res., 112,B09404, doi:10.1029/2007JB004930.

Lowry, A. R., K. M. Larson, V. Kostoglodov, and R. Bilham (2001),Transient fault slip in Guerrero, southern Mexico, Geophys. Res. Lett.,28, 3753–3756, doi:10.1029/2001GL013238.

Mansinha, L., and D. E. Smylie (1971), The displacement fields on inclinedfaults, Bull. Seismol. Soc. Am., 61, 1433–1440.

McCaffrey, R., M. D. Long, C. Goldfinger, P. C. Zwick, J. L. Nabelek,C. K. Johnson, and C. Smith (2000), Rotation and plate locking alongthe southern Cascadia subduction zone, Geophys. Res. Lett., 27,3117–3120, doi:10.1029/2000GL011768.

McCaffrey, R., A. I. Qamar, R. W. King, R. Wells, G. Khazaradze,C. A. Williams, C. W. Stevens, J. J. Vollick, and P. C. Zwick (2007),Fault locking, block rotation and crustal deformation in the Pacificnorthwest, Geophys. J. Int., 169, 1315–1340, doi:10.1111/j.1365-246X.2007.03371.x.

McCaffrey, R., L. M. Wallace, and J. Beavan (2008), Slow slip and fric-tional transition at low temperature at the Hikurangi subduction zone,Nat. Geosci., 1, 316–320, doi:10.1038/ngeo178.

McCausland, W., S. Malone, and D. Johnson (2005), Temporal and spatialoccurrence of deep non-volcanic tremor: From Washington to northernCalifornia, Geophys. Res. Lett., 32, L24311, doi:10.1029/2005GL024349.

McCrory, P. A. (2006), Depth to the Juan de Fuca slab beneath the Cascadiasubduction margin: A 3-D model for sorting earthquakes, U.S. Geol.Surv. Data, 91.

McGuire, J. J., and P. Segall (2003), Imaging of aseismic fault slip tran-sients recorded by dense geodetic networks, Geophys. J. Int., 155,778–788, doi:10.1111/j.1365-246X.2003.02022.x.

Melbourne, T. I., and F. H. Webb (2003), Slow but not quite silent, Science,300, 1886–1887, doi:10.1126/science.1086163.

Melbourne, T. I., W. M. Szeliga, M. M. Miller, and V. M. Santillan (2005),Extent and duration of the 2003 Cascadia slow earthquake, Geophys. Res.Lett., 32, L04301, doi:10.1029/2004GL021790.

Miller, M. M., T. Melbourne, D. J. Johnson, and W. Q. Sumner (2002),Periodic slow earthquakes from the Cascadia subduction zone, Science,295, 2423, doi:10.1126/science.1071193.

Moore, D. E., D. A. Lockner, R. Summers, M. Shengli, and J. D. Byerlee(1996), Strength of chrysotile-serpentinite gouge under hydrothermalconditions: Can it explain a weak San Andreas fault?, Geology, 24,1041–1044, doi:10.1130/0091-7613(1996)024<1041:SOCSGU>2.3.CO;2.

Moore, D. E., D. A. Lockner, M. Shengli, R. Summers, and J. D. Byerlee(1997), Strengths of serpentinite gouges at elevated temperatures, J. Geo-phys. Res., 102, 14,787–14,801, doi:10.1029/97JB00995.

Murakoshi, T. (2003), Seismic structure of the crust and uppermost mantlebeneath Kyushu as inferred from receiver function analysis, Ph.D. thesis,105 pp., Kyushu Univ., Fukuoka, Japan.

Nakamura, M. (2002), Estimate of Moho discontinuity by using convertedwave in seismograms of intermediate earthquakes beneath the Kyushudistrict, Japan, M.Sc. thesis, Kyushu Univ., Fukuoka, Japan.

Nicholson, T., M. Bostock, and J. F. Cassidy (2005), New constraints onsubduction zone structure in northern Cascadia, Geophys. J. Int., 161,849–859, doi:10.1111/j.1365-246X.2005.02605.x.

Obara, K. (2002), Nonvolcanic deep tremor associated with subductionin southwest Japan, Science, 296, 1679 – 1681, doi:10.1126/science.1070378.

Obara, K., and H. Hirose (2006), Non-volcanic deep low-frequency tremorsaccompanying slow slips in the southwest Japan subduction zone,Tectonophysics, 417, 33–51, doi:10.1016/j.tecto.2005.04.013.

Ozawa, S., M. Murakami, and T. Tada (2001), Time-dependent inversionstudy of the slow thrust event in the Nankai trough subduction zone,southwestern Japan, J. Geophys. Res., 106, 787 –802, doi:10.1029/2000JB900317.

Rogers, G., and H. Dragert (2003), Episodic tremor and slip on the Cascadiasubduction zone: The chatter of silent slip, Science, 300, 1942–1943,doi:10.1126/science.1084783.

Rubinstein, J. L., J. E. Vidale, J. Gomberg, P. Bodin, K. C. Creager, andS. D. Malone (2007), Non-volcanic tremor driven by large transientshear stresses, Nature, 448, 579–582, doi:10.1038/nature06017.

Rubinstein, J. L., M. La Rocca, J. E. Vidale, K. C. Creager, and A. G. Wech(2008), Tidal modulation of nonvolcanic tremor, Science, 319, 186–189,doi:10.1126/science.1150558.

Ruff, L. J. (1996), Large earthquakes in subduction zones: Segment inter-action and recurrence times, in Subduction Top to Bottom, Geophys.Monogr. Ser., vol. 96, edited by G. E. Bebout et al., pp. 91 – 104,AGU, Washington, D. C.

Salah, M. K., and D. Zhao (2004), Mapping the crustal thickness in south-west Japan using Moho-reflected waves, Phys. Earth Planet. Inter., 141,79–94, doi:10.1016/j.pepi.2003.10.002.

Savage, J. C. (1983), A dislocation model of strain accumulation andrelease at a subduction zone, J. Geophys. Res., 88, 4984–4996, doi:10.1029/JB088iB06p04984.

Segall, P., and J. R. Rice (1995), Dilatancy, compaction, and slip instabilityof a fluid-infiltrated fault, J. Geophys. Res., 100, 22,155 – 22,171,doi:10.1029/95JB02403.

Segall, P., A. Rubin, and J. Rice (2008), Thoughts on mechanism of slowslip events, non-volcanic tremor, and earthquake triggering, paperpresented at Aseismic Slip, Non-Volcanic Tremor, and EarthquakesWorkshop, U.S. Geol. Surv., Sidney, B. C., Australia.

Shelly, D. R., G. C. Beroza, S. Ide, and S. Nakamula (2006), Low-frequency earthquakes in Shikoku, Japan, and their relationship to episo-dic tremor and slip, Nature, 442, 188–191, doi:10.1038/nature04931.

Shelly, D. R., G. C. Beroza, and S. Ide (2007), Complex evolution oftransient slip derived from precise tremor locations in western Shikoku,Japan, Geochem. Geophys. Geosyst., 8, Q10014, doi:10.1029/2007GC001640.

Song, T.-R. A., and M. Simons (2003), Large trench-parallel gravityvariations predict seismogenic behavior in subduction zones, Science,301, 630–633, doi:10.1126/science.1085557.

Song, T.-R. A., D. V. Helmberger, M. R. Brudzinski, R. W. Clayton,P. Davis, X. Perez-Campos, and S. K. Singh (2009), Subducting slabultra-slow velocity layer coincident with silent earthquakes in southernMexico, Science, 324, 502–506.

Szeliga, W., T. I. Melbourne, M. M. Miller, and V. M. Santillan (2004),Southern Cascadia episodic slow earthquakes, Geophys. Res. Lett., 31,L16602, doi:10.1029/2004GL020824.

Szeliga, W., T. Melbourne, M. Santillan, and M. Miller (2008), GPS con-straints on 34 slow slip events within the Cascadia subduction zone,1997–2005, J. Geophys. Res., 113, B04404, doi:10.1029/2007JB004948.

Wang, K., R. Wells, S. Mazzotti, R. D. Hyndman, and T. Sagiya (2003), Arevised dislocation model of interseismic deformation of the Cascadiasubduction zone, J. Geophys. Res., 108(B1), 2026, doi:10.1029/2001JB001227.

Wells, R. E., C. S. Weaver, and R. J. Blakely (1998), Fore-arc migration inCascadia and its neotectonic significance, Geology, 26, 759 –762,doi:10.1130/0091-7613(1998)026<0759:FAMICA>2.3.CO;2.

Wells, R. E., R. J. Blakely, and C. S. Weaver (2002), Cascadia microplatemodels and within-slab earthquakes, in The Cascadia Subduction Zoneand Related Subduction Systems—Seismic Structure, Intraslab Earth-quakes and Processes, and Earthquake Hazards, edited by S. Kirby, K.Wang, and S. Dunlop, U.S. Geol. Surv. Open File Rep., 02-328, 17–23.

Wells, R. E., R. J. Blakely, Y. Sugiyama, D. W. Scholl, and P. A. Dinterman(2003), Basin-centered asperities in great subduction zone earthquakes: Alink between slip, subsidence, and subduction erosion?, J. Geophys. Res.,108(B10), 2507, doi:10.1029/2002JB002072.

Yoshioka, S., and K. Murakami (2007), Temperature distribution of theupper surface Philippine Sea plate along the Nankai Trough, from athree-dimensional subduction model: Interplate and low-frequency earth-quakes, Geophys. J. Int., 171, 302–315, doi:10.1111/j.1365-246X.2007.03510.x.

��M. R. Brudzinski and S. Holtkamp, Geology Department, Miami

University, Oxford, OH 45056, USA. ([email protected])


21 of 21

B00A17

determination of slow slip episodes and strain...

Documents