Rotation and Plate Locking at the Southern Cascadia Subduction Zone

GEOPHYSICALRESEARCH LETTERS, VOL. 27, NO. 19, PAGES3117-3120, OCTOBER 1, 2000

Rotation and plate locking at the southern Cascadia subduction zone

RobertMcCaffrey, 1 Maureen D. Long,1 ChrisGoldfinger, 2 Peter C. Zwick,3 JohnL. Nabelek,2 CherylK. Johnson,1 and Curt Smith4

Abstract. Global Positioning System vectors and surface using the GAMIT/GLOBK analysis software [King and tilt rates are inverted simultaneously for the rotation of west- Bock, 1999; Herring, 1998]. Site velocitiesand their covari- ern Oregon and plate locking on the southern Cascadia sub- ances were estimated by linear regression of the time series duction thrust fault. Plate locking appears to be largely of positions. Velocities were put in the North American offshore, consistent with earlier studies, and is sufficient to (NA) referenceframe by removingthe NA-ITRF96 rotation allow occasional great earthquakes inferred from geology. [DeMets and Dixon, 1999],which agrees with our solutionat Clockwise rotation of most of Oregon about a nearby pole is the mm/a level for 16 NA sites. Convergenceof the Juan de likely driven by collapse of the Basin and Range and results Fuca (JF) plate with NA is the sum of JF-Pacific [Wilson, in shortening in NW Washington State. The rotation pole 1993]and Pacific-NA [DeMetsand Dixon, 1999]poles. lies along the Olympic- Wallowa lineament and explains the The 71 horizontalGPS vectors(2 componentseach) and predominance of extension south of the pole and contraction 4 tilt rates provide 146 observations for the inversion. Misfit north of it. is defined by

1. Introduction - v-P Oblique convergenceof the young, oceanic Juan de Fuca where N is the number of observations, P is the number plate with the westernmargin of North America (Fig. 1) re- of free parameters, r• is the residual (observedminus cal- sults in both elastic deformation of coastal regions and dis- culated velocity), cr• is the formal velocity uncertainty, and tributed permanent deformation within the overriding plate. f is a scaling factor. Formal GPS velocity uncertainties are Evidence for large thrust earthquakes at the subduction well-knownto be underestimated[Mao et at., 1999]and f= 3 zoneis plentiful [Adams,1990; Atwater and Hemphill-Haley, returnsa minimumX2• • 1. The final, unscaledstandard de- 1997; Goldfingeret al., 1999] yet details of the earthquake viations of residualsare 1.2 mm/a for the north component history and potential are sketchy. Upper plate deforma- and 2.1 mm/a for the east. Velocity differencesbetween our tion, while clearly significant[Pezzopane and Weldon,1993; solutionand $ava#e et at.'s [2000]were 1.2 and 1.3 mm/a Wells et al., 1998], remainspoorly quantified. We are us- for two common sites. ing the Global PositioningSystem (GPS) to measurecrustal deformation in western Oregon from which we infer strain 3. Inversion Approach rates in the overriding plate, the increase in the potential for earthquake slip on the Cascadia subduction fault, and Our goal is to estimate the rotation of Oregon and plate how Oregon moves relative to North America. locking on the Cascadia thrust from the geodetic data. Be- cause plate locking strain extends hundreds of kilometers 2. Data and Analysis landward from the coast, the rotation pole and plate locking are estimated by a simultaneous inversion. Like many We use horizontal vectors from 50 GPS sites in northwest others, we describeplate locking as the fractional part (•) Oregon analyzed by us, 21 published vectors from southern of fault area that undergoes stick-slip and is currently stuck. Oregon [Savageet al., 2000] (Fig. 2), and four tilt rates The rate at which potential seismic moment (i.e.. mo- near the Oregon coast [Reilingerand Adams, 1982]. We ment that could later be releasedin earthquakes)builds is re-processedGPS campaign data collected by the US Geo- ]l)Io= qS!•AVwhere A is fault area, V is the long-termslip logicalSurvey (1992-1994),by the CascadesVolcano Obser- rate on the fault, and/• is the shear modulus of rocks (40 vatory (1992-1997),by us (1996-1999),and by a consortium GPaused here). Geodetic data constrain g;/o because in an of local observers under direction of the National Geodetic elastic medium surface deformation rates are proportional Survey (1998). Site positionswere calculatedin the ITRF96 to •V. referenceframe [Sillard et al., 1998]by combiningcampaign To parameterize plate locking, we specify nodes every data with regional and global sites and precise satellite orbits 10 km in depth and about every 100 km along strike on the thrust fault using the fault geometry of Hyndman and Wang [1995](Fig. 3a). The convergencevector Vi at node •RensselaerPolytechnic Institute, Troy, New York i is calculated from the Juan de Fuca (JF) - western Ore- 2OregonState University,Corvallis, Oregon gon (WO) pole of rotation which is the sum of JF - North aSeafloorSurveys International, Seattle, Washington America (NA) and NA-WO poles. Plate locking, in the di- 4National GeodeticSurvey, Salem, Oregon rection of JF-WO convergence, is •i V,. Interseismic surface deformation is estimated with an elastic, half-space dislo- Copyright2000 by theAmerican Geophysical Union. cation model [Okada, 1985] by integratingover small, finite Papernumber 2000GL011768. fault patches between nodes. • at each patch is estimated 0094-8276/00/2000GL011768505.00 by bilinear interpolation between the four enclosingnodes.

3117 3118 MCCAFFREY ET AL.- ROTATION AND PLATE COUPLING IN OREGON

50*N to NA. Our preferred model includes both coupling and ro- Canada tation (Fig. 3) with 17 setsof free nodes,3 pole parameters, 49'N andgives X• - 1.07(126 degrees of freedom).All fourtilt ß ratesare fit to within onestandard error. ,.Formal errors in

ß 45'N all •b are less than 0.5. eattieWashington 4TN 4. Plate Locking

46øN The final model suggestsa double locked zone - one zone

Juan de Fuca offshore and one near the down-dip edge of the fault at 30 to 450N 40 km depth (Fig. 3a). We suspectthat the inland locked • i ß c• c•' zone might be an artifact for the following reason. Half- • i•, '• Oregon44'N space dislocation models treat the base of the lithosphere !:.t ß •, c•. effectively as a no-slip boundary in an elastic medium, re- ß ß 430N ß:ii ß sulting in unrealistically high resistance of the lithosphere Pacific to trench-normalcontraction [Thatcher and Rundle, 1984] 42'N and a steep dropoff in surface velocities above the downdip edge of the coupled zone. Preliminary finite-element models 41'N in which low basal stress is allowed produce a gentler de- 2•30:• 40•• km• a Nevada cay in velocities[Williams and McCaffrey, 1999] landward 20 mm/a• 400N C•ifomia of the locked zone as seen in the data. Hence, we suggest I100km • that the majority of plate locking along the southern Cas- cadia subduction zone occurs offshore, in general agreement 232'E 234'E 236'E 235'E 240'E 2420E

Figure 1. Pacific Northwest showingactive volcanoes(triangles) and depth contoursof the top of the subducting Juan de Fuca plate (dashed lines). Large arrows show convergenceof Juan de Fuca plate with both North America (JF-NA) and the 46•N rotating Oregon block (JF-WO) at Cascadia deformation front. Ellipse in NE Oregon shows pole and l•r uncertainty for rotation of WO relative to NA. Small arrows show velocities and uncertainties relative to NA predicted by this pole. Large op- posing arrows show predicted sense of relative motion across the 45'N Olympic-Wallowalineament (OWL).

We set •b -- 0 at the deepest nodes at 50 km depth, noting that interplate seismicity, and presumably locking, at most ,•i2 44•N subductionzones ceases at about this depth [Tichelaar and Ruff, 1993]. To increasethe sensitivityof the data for the •'3C• • .... •'•:" estimated locking parameters, in some caseswe grouped adjacent nodes into a single free parameter. For example, because land geodetic observationsare insensitive to the near- trench locking, nodes at the trench were forced to have the ...... • Calculated same •b as adjacent nodes 10-km downdip. Model edges were handled by forcing nodes outside the GPS network to ' 100km ' 42'N have the same coupling as adjacent nodes at the edge of Relative to North America the network. Other nodes were grouped if their uncertain- •..• Observed ties were large based on initial inversionsin which all nodes 236øE 237•E 238'E 239øE 240øE were free and independent. To find parameter values that minimize misfit, we apply simulated annealing to downhill SouthOregon12ø ,.. CentralOregon12øI NorthOregon] 20 41.7*N-43.4*NI ' ".. , 43.5øN-44.9øNI ['".. 44.9*N-46.3øN{ simplexminimization [Presset al., 1989]. A penalty func- tion keeps •b between 0 and i to prevent backslip or locking at a rate faster than plate convergence. At the minimum X2•,we checkfor unconstrainedparameters and calculate the covariance matrix with singular value decomposition of linearized normal equations. Longitude,øE Longitude,•E Longitude,*E We tested a range of models by varying the parameteri- zation and constraints. With no motion of WO relative to Figure 2. (a) Observedand calculatedOregon GPS site veloci- NA and no plate coupling(no free parameters),minimum ties relative to North America (NA) with 3a ellipses.Black lines X2•- 26.1;when locking alone is allowedwithout rotation of show locations of tilt lines. Dashed lines enclose profile regions. (b-d) West-to-East profilesof the North (open) and East (closed) Oregon(25 freeparameters), X2• = 7.2;and rotation with- componentsof the GPS vectors relative to NA (3(r error bars). out platelocking (3 freeparameters) gives X2• - 2.5. Hence, Curves show predictions of rotation - locking model. Triangles most of the data are explained by rotation of WO relative show where profiles cross volcanic arc. MCCAFFREY ET AL.- ROTATION AND PLATE COUPLING IN OREGON 3119

, 40• the largestearthquake [Rundle, 1989], is approximately1-/3 [McCaffrey,1997]. Hence, the total momentreleased dur- ingtime T equalsboth Tl•o (if/¾/oremains constant) and MoTM / (1 -/•), giving

(2) ß+-400 -200 ; 200 400O• Distancein km northof 44.1øN r• Great historic subduction earthquakes are thought to have occurred every 400 to 650 years, based on geologic

ß ß evidence, along the southern 2/3 of the Cascadia thrust 126øW [Adams, 1990; Atwater and Hemphill-Haley, 1997]. The + 45'N + + magnitudes of these events are not well-known, yet even if onlyM,o• 9 (i.e., Mom• = 3.5 x 1022N-m and/3=0)events occur and account for all the moment released, as thought + + ß + by some[Atwater and Hemphill-Haley,1997; Goldfingeret ß al., 1999], the modern geodeticmoment rate, if constant throughtime, is sufficient(T = 500-970years). Alternatively, at seismically active subduction zones /3 +ß ß+ + ß rangesfrom 0.56 to 0.80 [Kagan,1997] and is in theory2/3 [Rundle, 1989] meaningthat earthquakesof various sizes A release the moment. If southern Cascadia has a /3 value within this range and M m• = 9, the current moment rate is sufficient for a M•o- 9 approximately every 1000 to 2900 L4 years, considering the range in/3 and Mo. Also, during this - 1.15 +• .• • - 1.2 time T there should be on average 7 to 16 M•o>_ 8 earth- 1.0 quakes, equivalent to one about every 150-930 years along 14 14 i 19.5019.7520100 C20.25 0.8 the margin from 40øN to 46ø N. If each of these earthquakes log (Moment rate in N-m/a) ruptures on average a 200-km segment of the 750-km long subduction zone, then each part of the fault will rupture Figure 3. (a) Best-fitplate lockingmodel. Gray dotsshow node every 550 years on average, also similar to the observed re- locations on thrust fault. Nodes are aligned along depth contours currence time. It should be noted that this latter view is - depths are given above southern set of nodes. Numbers below inconsistent with recent interpretations by Goldfinger et al., nodes are parameter indices, nodes sharing an index have same [1999]who imply a one-to-onecorrespondence between tur- values of qbin inversion. Arrows show inferred locking rates at nodes. Darker shading indicates greater locking. Triangles show bidites, that occur every 655 years, and M•o = 9 subduction GPS sites and black bars show tilt lines. Tic spacingis 1ø. (b) zone earthquakes. While our geodetic data do not constrain Variation in lockingrates qbV(4-ey)alongthe margin at different past or future earthquake size distributions for Cascadia, depths. (c) Misfits for combinationsof pole and node valuesand the modern surface strain rate implies a rate of moment ac- correspondingtotal geodetic moment rates. Large dots show best- cumulation(as interpretedwith a dislocationmodel) that is fit solutions for various parameter sets; small dots are from grid sufficient for either scenario. searches.The uncertaintyin momentrate, 1019N-m/a, is the 99% confidencelevel, i.e., an increasein X2•of 0.08 (AX2 -- 9 for 120degrees of freedom)above the minimumX2• =1.07 (shownby 5. Rotation of Oregon line at X2•-- 1.15). The best-fit pole of rotation for WO-NA (at 45.9ø -9 0.6øN, 241.3ø -9 0.7 øE, with a clockwise rotation rate of 1.05 -9 0.16ø/Ma, Fig. 1) is closeto a pole inferred from with inferencesmade from thermal and uplift data [Hynd- geology[Wells et al., 1998]but far from one estimatedfrom man and Wang, 1995]. Couplingbeneath land is possible southernOregon GPS data [Savageet al.,'2000]. The east- but not required by the data. Because we had to smooth ward decreasein the northward motion of Oregon (Figs. the model offshore, we cannot address whether the fault is 2b-d) is matchedsufficiently by rotation about a pole to the locked out to the trench or not. Our results suggesta north- east and does not require slip on faults in eastern Oregon. ward increasein the average'offshorelocking (Fig. 3b). Nevertheless,slip may occur at the few mm/a level- below The geodetic moment rate estimated by integrating over the uncertainties in the GPS measurements- as indicated the entirefault is 7+1 x 1019N-m/a (Fig. 3c), whichis by paleoseismology[Pezzopane and Weldon,1993]. -•35% of full locking to 50 km depth. If this moment is The WO-NA pole lies along the Olympic- Wallowa lin- released in earthquakes, in theory the earthquakes will be eament (OWL) (Fig. 1) at the northern edgeof the Basin distributedas N(Mo) = aM•-• whereN is the numberof and Range extensional regime and the southeastern edge of earthquakes of moment Mo or greater. The constant a de- the Yakima fold-thrustbelt [Wells et al., 1998]. Pezzopane pends on rates of earthquake activity,/3 governsthe relative and Weldon[1993] show largely extensional structures along numbers of large and small quakes, and a maximum event the OWL in Oregon and contractional structures along it in size, M• ax, is specified to keep the total moment finite. Washington, consistent with our pole location and clock- Dividing MoTM by the total moment with N(MoTM) = 1 wise rotation. Similarity of GPS vectors in SW Washington shows that the ratio of the largest quake's moment to the [Khazaradzeet al., 1999] to thosein NW Oregonsuggests total moment during the time T, the average repeat time of that it rotates with Oregon (Fig. 1). The southernedge of 3120 MCCAFFREY ET AL.: ROTATION AND PLATE COUPLING IN OREGON the WO block is likely in N. California but our data do not Goldfinger, C., H. Nelson, and J. E. Johnson, Holocene recurrence cover it. of Cascadia great earthquakes based on the turbidite event Lack of an offset in the GPS north componentsat the vol- record, Eos Trans. AGU, 80, F1024, 1999. Herring, T. A., GLOBK: Global Kalman filter VLBI and GPS canic arc (Fig. 2b-d) showsthat it doesnot act as a block analysis program, v.4.1, MassachusettsInstitute of Technology, boundary, suggestingthe north-moving Sierra Block at the Cambridge, 1998. south edge of the Oregon forearc is not driving the block Humphreys, E., and M. A. Hemphill-Haley, Causesand character- rotation. If it is, easternOregon lithosphere (including the istics of Western U.S. deformation, Geol. $oc. Am. Abstracts, magmaticarc) would have to be very strongunder tension 28, 1996. to move coherently under a northwestward push at its SW Hyndman, R. D., and K. Wang, The rupture zone of Cascadia great earthquakes from current deformation and the thermal corner. For the same reason, coast-parallel traction asso- regime, J. Geophys. Res., 100, 22133-22154, 1995. ciated with oblique subduction of the Juan de Fuca plate Kagan, Y. Y., Seismic moment-frequency relation for shallow under the forearc is probably not alone driving the block ro- earthquakes: Regional comparison, J. Geophys. Res., 102, tation. Oregon's senseof rotation and its lack of rapid inter- 2835-2852, 1997. nal deformation support the suggestionby Humphreys and Khazaradze, G., A. Qamar, and H. Drageft, Tectonic deformation Hemphill-Haley[1996] that it is movingin responseto Basin in western Washington from continuous GPS measurements, Geophys. Res. Left., 26, 3153-3156, 1999. and Range extension to the southeast. However, in contrast King, R. W. and Y. Bock, Documentation for the GAMIT GPS to their idea that the NW-directed extension of the Basin analysis software, Release 9.82, Massachusetts Institute of and Range is a responseto a weak Cascadia subduction zone Technology, Cambridge, 1999. that allows westward escape of Oregon, the nearby pole of Mao, A., C. G. A. Harrison, and T. H. Dixon, Noise in GPS rotation ultimately results in N-S shortening in Washing- coordinate time series, J. Geophys. Res., 104, 2797-2816, 1999. McCafrey, R., Statistical significance of the seismic coupling co- ton or Canada and little change in the overall subduction efficient, Bull. $eismol. $oc. Am., 87, 1069-1073, 1997. rate at the Cascadia thrust. That Oregon rotates about a Okada, Y., Surface deformation due to shear and tensile faults in nearby pole suggeststhat the force resisting shortening at a half-space, Bull. $eismol. $oc. Am., 75, 1135-1154, 1985. the subduction zone is sufficient to deflect Oregon's motion Pezzopane, S. K., and R. Weldon, Tectonic role of active faulting northward where it is instead accommodated by shortening in central Oregon, Tectonics, 12, 1140-1169, 1993. Press, W. H., B. P. Flannery, S. A. Teukolsky, and W. T. Vetter- in the continental lithosphere. ling, Numerical Recipes, Cambridge Univ. Press, Cambridge, 1989. 6. Conclusions Reilinger, R., and J. Adams, Geodetic evidence for active landward tilting of the Oregon and Washington coastal areas, Geo- Inversion of GPS and tilt data from western Oregon re- phys. Res. Left., 9, 401-403, 1982. veal a clockwise rotation of Oregon relative to North Amer- Rundle, J., Derivation of the complete Gutenberg-Richter ica about a nearby pole superimposedon ENE-directed con- magnitude-frequency relation using the principle of scale in- traction arising from locking with the subducting Juan de variance, J. Geophys. Res., 94, 12337-12342, 1989. Savage, J. C., J. L. Svarc, W. H. Prescott, and M. H. Murray, De- Fuca plate. Geodetic estimates of rotation and locking are formation across the forearc of the Cascadia subduction zone similar to those made from geologic, paleomagnetic, ther- at Cape Blanco, Oregon, J. Geophys. Res., 105, 3095-3102, mal, and uplift data. The extent of the Oregon block and 2000. its sense of rotation suggest that it is driven by Basin and Sillard, P., Z. Altamimi, and C. Boucher, The ITRF96 realization Range extension and not by localized forces along its west- and its associatedvelocity field, Geophys. Res. Lett., œ5,3223- ern edge. The NE boundary of the Oregon block, including 3226, 1998. Thatcher, W., and Rundle, J. B., A viscoelastic coupling model SW Washington, appears to be along the Olympic - Wallowa for cyclic deformation due to periodically repeated earthquakes lineament. at subduction zones, J. Geophys. Res., 89, 7631-7640, 1984. Tichelaar, B. W., and L. J. Ruff, Depth of seismic coupling along Acknowledgments. B. Atwater, C. Williams, R. Wells, subduction zones, J. Geophys. Res., 98, 2017-2037, 1993. and two GRL reviewers made useful comments. C. Stevens, R. Wells, R. E., C. S. Weaver, and R. J. Blakely, Forearc migration in King, T. Herring, and M. Murray advised on data processing. Cascadia and its neotectonic significance, Geology œ6,759-762, Many people assistedin field measurements. The U.S. Geological 1998. Survey and Cascades Volcano Observatory provided GPS data. Williams, C. A., and R. McCaffrey, Modeling of surface deforma- Supported by National Science Foundation grant EAR-9814926 tion at subduction zones: alternatives to dislocation models, and by US Geological Survey under award 99HQGR0014. The Eos Trans. AGU, 80, F266, 1999. views and conclusions contained in this document are those of the Wilson, D. S., Confidence intervals for motion and deformation of authors and should not be interpreted as official policies of the the Juan de Fuca plate, J. Geophys. Res., 98, 16,053-16,071, US Government. GPS velocities used in this paper are available 1993. at www.rpi.edu/•mccafr/gps/gr100_or.dat. C. Johnson, M. Long, and R. McCaffrey, Department of Earth References and Environmental Sciences, Rensselaer Polytechnic Institute, Troy, NY 12180. (e-mail: [email protected]) Adams, J., Paleoseismicity of the Cascadia subduction zone: Evi- C. Goldfinger and J. Nabelek, College of Oceanic and Atmo- dence from turbidites off the Oregon-Washington margin, Tec- spheric Sciences,Oregon State University, Corvallis, OR 97330. tonics 9, 569-583, 1990. C. Smith, National Geodetic Survey, P.O. Box 12114, Salem, Atwater, B. F., and E. Hemphill-Haley, Recurrence intervals OR 97309. for great earthquakesof the past 3,500 years at northeastern P. Zwick, Seafloor Surveys International, 2727 Alaskan Way, Willapa Bay, Washington, USGS Professional Paper 1576, Pier 69, Seattle, WA 98121. 1997. DeMets, C. and T. H. Dixon, New kinematic models for Pacific- North America motion from 3 Ma to present, I: Evidence for steady motion and biases in the NUVEL-1A model, Geophys. (ReceivedMay 10, 2000; revisedJuly 14, 2000; Res. Left., 26, 1921-1924, 1999. acceptedJuly 21, 2000.)