ARTICLE

The Cape Mendocino, California, Earthquakes ofApril 1992: Subduction at the Triple Junction

D. Oppenheimer, G. Beroza, G. Carver, L. Dengler, J. Eaton, L. Gee, F. Gonzalez, A. Jayko,W. H. Li, M. Lisowski, M. Magee, G. Marshall, M. Murray, R. McPherson, B. Romanowicz,

K. Satake, R. Simpson, P. Somerville, R. Stein, D. Valentine

The 25 April 1992 magnitude 7.1 Cape Mendocino thrust earthquake demonstrated thatthe North America-Gorda plate boundary is seismogenic and illustrated hazards that couldresult from much larger earthquakes forecast for the Cascadia region. The shock occurredjust north of the Mendocino Triple Junction and caused strong ground motion and moderatedamage in the immediate area. Rupture initiated onshore at a depth of 10.5 kilometers andpropagated up-dip and seaward. Slip on steep faults in the Gorda plate generated twomagnitude 6.6 aftershocks on 26 April. The main shock did not produce surface ruptureon land but caused coastal uplift and a tsunami. The emerging picture of seismicity andfaulting at the triple junction suggests that the region is likely to continue experiencingsignificant seismicity.

On 25 April 1992 at 18:06 (UTC), asurface wave magnitude (Ms) 7.1 earth-quake occurred near the town of Petrolia,California (Fig. 1). The main shock wasfollowed the next day by two MS 6.6 after-shocks at 07:41 and 11:41, located offshoreabout 25 km west-northwest of Petrolia.These three earthquakes and more than2000 recorded aftershocks illuminated theconfiguration of the Mendocino TripleJunction, where the Pacific, North Ameri-ca, and southernmost Juan de Fuca (Gorda)plates meet. The occurrence of a M 7earthquake is not unusual at the triplejunction; over 60 earthquakes of ModifiedMercalli intensity > VI (1) or M > 5.5have occurred there since 1853 (2). How-ever, this earthquake sequence may haveprovided the first direct evidence of inter-plate seismicity and thus impacts regionalhazard assessment. In this article, we de-scribe geophysical and seismological obser-vations and discuss implications for seismichazards in the Pacific Northwest.

Damage estimates ranged from $48 mil-D. Oppenheimer, J. Eaton, A. Jayko, M. Lisowski, G.Marshall, M. Murray, R. Simpson, and R. Stein are withthe U.S. Geological Survey, Menlo Park, CA 94025. G.Beroza and M. Magee are in the Geophysics Depart-ment, Stanford University, Stanford, CA 94305. G.Carver, L. Dengler, and R. McPherson are in theDepartment of Geology, Humboldt State University,Arcata, CA 95521. L. Gee and B. Romanowicz are atthe University of California, Seismographic Station,ESB 475, Berkeley, CA 94720. F. Gonzalez is with theNational Oceanic and Atmospheric Administration'sPacific Marine Environmental Laboratory, Seattle, WA98115. W. H. Li is with Geological Sciences, AJ-20,University of Washington, Seattle, WA 98195. K.Satake is in the Department of Geological Sciences,University of Michigan, Ann Arbor, Ml 48109. P. Som-erville is with Woodward-Clyde Consultants, Pasade-na, CA 91101. D. Valentine is in the Department ofGeological Sciences, University of California, SantaBarbara, CA 93106.

lion to $66 million and President Bushdeclared the region a major disaster area.Much of the damage resulted from the mainshock; however, fires triggered by the firstlarge aftershock destroyed most of theScotia shopping district, and both large,off-shore aftershocks caused additionalstructural damage. The relatively low inci-dence of injuries and structural damage

Fig. 1. Simplified tectonic map in the vicinity ofthe Cape Mendocino earthquake sequence.Stars, epicenters of three largest earthquakes;contours, Modified Mercalli intensities (values,Roman numerals) of main shock; open circles,strong motion instrument sites (22) (adjacentnumbers give peak horizontal accelerations ing). Abbreviations: FT, Fortuna; F, Ferndale; RD,Rio Dell; S, Scotia; P, Petrolia; H, Honeydew;MF, Mendocino fault; CSZ, seaward edge ofCascadia subduction zone; and SAF, San An-dreas fault.

SCIENCE * VOL. 261 * 23 JULY 1993

caused by this sequence is primarily theresult of low population density and thepredominance of small, wood frame struc-tures in the epicentral area. The sequencecaused 356 reported injuries, destroyed 202buildings, and caused damage to an addi-tional 906 structures primarily in the townsof Petrolia, Ferndale, Rio Dell, Scotia, andFortuna (Fig. 1) (3). It also triggered nu-merous landslides and rock falls and causedwidespread liquifaction in local river val-leys. Analysis of 1296 surveys in the northcoast area indicate that the Modified Mer-calli intensity peaked at IX in the Petroliaregion and decreased in approximately aradial pattern around the epicenter (Fig. 1).Both of the two large aftershocks producedpeak intensities of VIII, although the pat-tern was somewhat different from the mainshock.

Tectonic SettingThe Cape Mendocino earthquakes are aresponse to ongoing plate motions betweenthe Gorda, North America, and Pacificplates at the Mendocino Triple Junction.The Gorda plate is converging on theNorth America plate at about 2.5 to 3cm/year in the direction N50WE to N550E(4). The seaward edge of Gorda plate sub-duction is marked by an abrupt change insea-floor topography and by the westernlimit of the accretionary prism imaged inseismic reflection profiles (5). Active foldsand thrust faults in Franciscan Complexand Cenozoic rocks and sediments of theoverriding North America plate are parallelto the seaward edge of the Cascadia subduc-tion zone (6).

Rigid plate theory predicts oblique con-vergence of the Gorda plate with the Pacificplate at 5 cm/year in the direction N1 15VE(4). Translational motion occurs along theeast-west-trending, vertical, right-lateralMendocino transform fault, whereas theconvergence results in internal deformationof the Gorda plate. The attendant Gordaplate seismicity recorded in the 17 yearsbefore the Cape Mendocino sequence (7,8) (Fig. 2) has been concentrated in twoparallel zones with a combined thickness ofapproximately 15 km. In the region of theCape Mendocino earthquake, most seismic-

433

on S

epte

mbe

r 7, 2

011

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fro

m

ity locates at depths greater than 17 km; thehypocenter zone dips about 60 eastwardbetween 124.75 and 123.250W, at whichpoint the dip increases to about 250 (9).Most M > 5 earthquakes within the Gordaplate exhibit left-lateral motion on steepnortheast-oriented faults (7, 10) that re-lieve convergence between the Gorda andPacific plates through slip on preexistingplanes of weakness inherited at the GordaRidge (11).

The San Andreas fault marks the prin-cipal Pacific-North America plate bound-ary south of the Mendocino Triple Junc-tion. Triangulation data and observationsof ground cracks indicate the fault rupturedas far north as Point Delgada in 1906 (12),but its location farther north is uncertain.Some studies place it immediately offshore(13), but others suggest that it merges withonshore faults at the triple junction (5).Geometry requires that the Pacific plate isalso in contact with the North Americaplate along the Mendocino fault above thesubducting Gorda plate.

Until the Cape Mendocino earthquake,few earthquakes were recorded with focalmechanisms that indicated slip on the Cas-cadia subduction zone. However, compari-sons of the age, spreading rates, physiogra-

phy, and seismicity of the Juan de Fuca-Gorda plate system with other subductingplates suggest that it does not subduct aseis-mically but instead is locked and capable ofgenerating major earthquakes (14). Paleo-seismic evidence of large, late Holocenesubduction earthquakes is present along thesubduction zone in submerged and buriedwetlands (15), raised marine terraces (16),and surface displacement on thrust faultsthat may be genetically related to largesubduction events (17). Radiocarbon dat-ing indicates that at least three episodes ofseismicity of similar age are represented inthe stratigraphy from central Washingtonto northern California in the last 2000years; the last episode occurred at about1700 A.D. (17).

ObservationsSeismicity. The hypocenter of the 25 April1992 main shock was located 4 km east ofPetrolia at a depth of 10.6 km (Fig. 3). A

focal mechanism determined by the inver-sion of teleseismic mantle Rayleigh wavesand aftershock locations indicate nearlypure thrust motion on a N10TW-strikingfault plane that dips 130 to the east-north-east (18) (Table 1). The location of thehypocenter at the southeast end of theaftershock zone suggests that the fault rup-tured unilaterally to the west (19). Mostaftershocks

shock is unknown because of the paucity ofaftershocks. However, the second after-shock was located within a trend of smalleraftershocks at depths of 14 to 30 km on asoutheast-striking plane that dips about 800to the southwest (Fig. 3B, cc'); this orien-tation is consistent with the focal mecha-nism. The depths and mechanisms of thetwo large aftershocks provide evidence thatrupture took place on faults in the Gordaplate, distinct from the main shock fault.

Although no large shocks ruptured theMendocino fault during this sequence,many aftershocks occurred on the eastwardprojection of the fault (Fig. 3). The after-shock activity was bounded on the southwhere the distribution of hypocenters isnear vertical and extends to a depth of 25km (Fig. 3B, bb'). If this marks the bound-ary between the Gorda and Pacific plates,then the lack of any aftershocks in thePacific plate suggests that the main shockrepresented strain release between theGorda and North America plates. Themapped location of the Mendocino fault inthis region is uncertain (5), and this east-

Fig. 4. Broad-band velocity records from fourstations (ALE: epicentral distance = 5185km, azimuth = 90; HRV: A = 4366 km, azimuth= 690; MAJO: A = 8029 km, azimuth = 3050;ISA: 1 = 745 km, azimuth = 1320) for the mainshock (top trace) and the first (middle) andsecond (lower) aftershocks. The amplitudes ofthe seismograms at ALE, HRV, and MAJO areincreased relative to ISA for display. The largeamplitudes of the second aftershock relative tothe other two events in the along-strike azimuth(ISA) is attributable to rupture directivity.

west trend of seismicity may define theposition of the Mendocino fault.

Source properties. The mechanism andlocation of the two aftershocks were simi-lar, but the second aftershock exhibited astrong variation of amplitude with azimuth(Fig. 4). The seismic moment of the secondaftershock was approximately twice that ofthe first, but amplitudes of the P wave forthis event were as much as 10 times as largenear an azimuth of 1300. This variation ismost easily attributable to enhancement ofthe amplitude in the direction of rupture,known as directivity (21). Directivity in Pwaves is surprising because it requires rup-ture velocities that are a large fraction ofthe P wave velocity. The high amplitudesand strong high-frequency content associat-

t ] 16.49 x102090/1.59 x 102 C1/s2

U

2s0

Fig. 5. Strong motion recordings and threecomponents of the peak accelerations fromstations of the California Strong Motion Instru-mentation Program (CSMIP) at Cape Mendo-cino and Petrolia (22). Discernible first motiondirections of the P and S waves for long-periodevent (Ep1 and ES1) are indicated, as is thelarge, high-frequency pulse at Cape Mendo-cino (E2).

ed with the second aftershock may explainsome of the differences in the intensitypatterns for the main shock and two after-shocks. Although the second aftershockhad 25% of the moment of the main shock,it has larger velocity amplitudes at stationsto the southeast, such as ISA. The differ-ence in both Modified Mercalli intensityand broad-band velocity records betweenthe main shock and the second aftershockwas probably enhanced by rupture propaga-tion to the west during the main shock asinferred from the location of the hypo-center at the down-dip end of the ruptureplane.

The strong ground motions of the mainshock and two aftershocks were recorded on14 instruments at epicentral distances of 5to 130 km (Fig. 1), and the peak accelera-tions were some of the highest ever record-ed (22). Recordings of the main shock atPetrolia and Cape Mendocino (Fig. 5) atepicentral distances of 5 and 10 km, respec-tively, have absolute time, facilitating theanalysis of rupture evolution on the fault.Modeling of the large, long-period pulsethat occurred 1 s after the main shockbegan (Fig. 5, Ep1 and Es1) with generalizedray theory indicates that this pulse originat-ed from slip that occurred about 5 kmup-dip from the hypocenter, beneath Petro-lia. This result is consistent with the arrivaltimes and polarities of the vertical P wavesand horizontal S waves at both stations. Inaddition, the P wave first motions wereupward and northwestward at Cape Men-docino. These motions indicate that thesource was southeast of the station. Thissource location is consistent with the west-southwest direction of rupture determinedfrom teleseismic surface waves (19), al-though that study inferred that the ruptureinitiated offshore. A large, high-frequencypulse (Fig. 5, E2) followed the long-periodpulse at Cape Mendocino and exceeded igon both horizontal components. This pulsewas not discernible at the neighboringPetrolia station; thus, it cannot be ex-plained simply by source effects but mayrepresent motions that were generated oramplified locally near the Cape Mendocinostation.

Table 1. Earthquake parameters.

Origin time* (UTC) Latitude* Longitude* Depth* Centroid Monmentt Striket Dipt RaketOrigntie* (TC) (North) (West) (kin) deptht mbt Mst (dyn~cm) Srkt Dp ae(kin)

Main shock 25 April 18:06:05.16 40019.94' 124013.69' 10.6 20 to 25 6.3 7.1 4.45 x 1026 349.70 13.00 105.60First after- 26 April 07:41:39.98 40026.13' 124034.43' 19.3 20 to 25 5.9 6.6 6.35 x 1025 122.30 75.90 175.20shock

Second after- 26 April 1 1:18:25.82 40023.38' 124034.30' 21.7 30 to 35 6.5 6.6 1.20 x 1Qo26 311.50 89.60 181.80shock

*Hypocentral location determined from the local seismic network of the U.S. Geological Survey (8). tFrom surface-wave inversion (18, 20). *National EarthquakeInformation Center, Preliminary Determination of Epicenters.

SCIENCE * VOL. 261 * 23 JULY 1993

ALE (X0 ) HRV (x4)ifV~~lv'n1 I

MAJO( 0) ISA

01+0Time (s) -

--------------------------

435

on S

epte

mbe

r 7, 2

011

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fro

m

Coseismic displacement. The elastic strainreleased by the main shock caused signifi-cant horizontal and vertical deformation inthe epicentral region. The main shock ele-vated about 25 km of the coast from 3 kmsouth of Punta Gorda to Cape Mendocino(Fig. 6). Many intertidal organisms inhab-iting rocky reefs perished in the 3 weeksafter the main shock. Maximum uplift was140 + 20 cm at Mussel Rock and 40 to 50cm at the northernmost reef at Cape Men-docino (23). The lack of rocky intertidalenvironments farther north precluded theprecise location of the northern limit of theuplift, but several near-shore rocks locatedabout 7 km north of Cape Mendocinoshowed no evidence of uplift.

Coseismic horizontal and vertical sitedisplacements in a regional geodetic net-work (Fig. 6) were determined from GlobalPositioning System (GPS) surveys in 1989,1991, and 1 month after the main shock.The relative positions of most sites near theepicenter were measured shortly after the17 August 1991 Honeydew earthquake[body wave magnitude (mb) 6.0], whichoccurred 6 km south of the Cape Mendo-cino epicenter. The coseismic displace-ments were determined by comparison of the1989 to 1992 observations, except in thevicinity of the Honeydew event where the1991 survey was referenced. All displace-ments were corrected for secular strain accu-mulation estimated from Geodolite trilater-ation measurements made between 1981 and1989. A site 13 km northeast of the epicen-ter had the largest measured coseismic dis-placement, moving 40 2 cm to the west-southwest and subsiding 16 8 cm.

Our preferred uniform-slip fault model(24), estimated from the coseismic sitedisplacements and coastal uplift observa-tions, indicates 2.7 m of nearly pure thrustmotion occurred on a gently dipping faultplane. This model, chosen from a suite ofacceptable models (24), is consistent withthe main shock focal mechanism, the hy-pocenter location, and the distribution ofaftershocks (Fig. 6 and Table 2). The rangeof geodetic moment inferred from the ac-

Table 2. Displacement model.

Parameter Value

Width 16 kmLength 21.5 kmDepth to top edge 6.3 kmLatitude origin* 40018.08'NLongitude origin* 12411 .80'WDepth at epicenter 9.2 km

locationMomentt 2.79 x 1026 dyn'cmStrike/dip/rake 3500/12.00/940Slip 2.7 m

ceptable models is 2.5 x 1026 to 3.5 x 1026dyn-cm, about 60% of the main shockseismic moment (Table 1). The model pre-dicts a maximum uplift along the coast thatis consistent with but somewhat less thanthe observed uplift. More complex modelsthat use nonuniform slip to describe therupture may improve these estimates ofuplift and geodetic moment.

Tsunani. The main shock generated asmall tsunami recorded by tide gauges alongthe California, southern Oregon, and Ha-waii coastlines (Fig. 7). The largest tsunamiamplitudes were recorded at Crescent City,California, where two well-defined packetsof wave energy were recorded within thefirst 5 hours with maximum positive heightsof 35 and 53 cm. Neither the precise arrivaltime nor the polarity of the first wave areclear because of the presence of backgroundnoise. However, the first packet of waveenergy is consistent with the predicted trav-el time of 47 min for a wave ray path thattraversed deep water. The second wavepacket probably represents coastal trappedwaves, or edge waves, having much slowervelocities and amplitudes that rapidly de-crease with distance offshore. Because thetsunami arrival nearly coincided with lowtide at Crescent City, the wave did notcause any damage. The tsunami at CrescentCity had an 8-hour duration; wave heightsreached a maximum 3 to 4 hours after thefirst arrival. Tide gauges also recorded theinitial arrival and subsequent edge waves atNorth Spit (Eureka, California) (20 minand 2.5 hours), Arena Cove (35 min and

3.5 hours), and Point Reyes (65 min and 3hours).

Motion on the Plate BoundaryInterplate main shock. The main shock faultprojects to the sea floor within 5 km of theseaward edge of the Cascadia subductionzone (25) (Fig. 3), suggesting that the mainshock ruptured the Gorda-North Americaplate boundary. In contrast, the upperboundary of the pre-main shock seismicity,which is 7 km deeper than the main shockrupture plane (Figs. 2 and 3), projects tothe surface about 85 km west of the Casca-dia subduction zone and thus does notappear to define the plate boundary. Theseismicity gap between the slip plane of themain shock and the pre-main shock seis-micity is about the same thickness as theGorda crust and overlying accretionary sed-iments, as determined from refraction ex-periments 10 km east of the seaward edge ofthe subduction zone (26). The gap mayreflect a ductile subducted Gorda crust, andthe inception of seismicity at a depth of 17km may reflect brittle behavior of theGorda upper mantle (27). Tabor and Smith(28) reached a similar conclusion from theirobservations of seismicity and velocitystructure of the Juan de Fuca plate beneaththe Olympic peninsula of Washington.

However, an inversion for the three-dimensional velocity structure of the regionindicates that velocities typical of Gordacrust are evident at depths greater than 15to 20 km (29). Moreover, modeling of

Fig. 6. Observed and predictedcoseismic displacements for theCape Mendocino main shock (epi-center located at star). The vec-tors are horizontal displacementsrelative to a site located at4109.20'N, 123052.92'W. Ob-served displacements derivedfrom GPS and Geodolite mea-surements; ellipses enclose re-

40'40' - gions of 95% confidence. Predict-ed displacements are from amodel of uniform slip on the north-

(~" east-dipping rectangular faultplane, indicated by its surfaceprojection. Rounded rectanglesshow vertical displacementsmeasured by GPS that are great-er than their standard deviations.Contours are elevation changes

40 2O' - in millimeters predicted by themodel. Abbreviations: CM, CapeMendocino; MR, Mussel Rock;and PG, Punta Gorda. (Inset) Up-lift measurements and their stan-dard deviations from the die-off of

51) marine organisms at coastal sites(open circles on map) and pre-dicted uplift projected alongN1 0W.

SCIENCE * VOL. 261 * 23 JULY 1993

*Southeast corner of fault plane. tAssumes uni-form rigidity of 3 x 10l dyn/cm2.436

on S

epte

mbe

r 7, 2

011

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fro

m

thermal effects on the strength of the sub-ducting oceanic lithosphere (30) suggeststhat the double seismic layers observed atdepths of 20 and 30 km (Fig. 2) reflect,respectively, the brittle upper crust andupper mantle of the Gorda plate; the inter-vening, relatively aseismic region wouldcorrespond to the ductile lower crust. Con-sequently, these studies suggest that theCape Mendocino main shock was an intra-plate event in the North America plate.

Whether the main shock was an inter- orintra-plate event, the Cape Mendocino mainshock clearly relieved strain resulting from therelative Gorda-North America plate motion.We note, however, that the main shockruptured a region of the plate boundary thatdiffers considerably from the boundary farthernorth, as indicated by the change in itsorientation from north-northwest to north-west (5), the relatively narrow width of theplate, the likely presence of subducted sedi-ments in the region of main shock rupture,and its younger age (4). Thus, this earth-quake may not be typical of other Cascadiasubduction zone earthquakes.

Intraplate aftershocks. The location,depth, and focal mechanisms of the twolarge aftershocks indicate that they rup-tured the Gorda plate. The seismic dataindicate that right-lateral slip occurred on avertical, northwest-oriented fault plane forat least the second event. For most earlierGorda shocks, rupture occurred as left-lateral slip on a northeast-oriented plane,perhaps because this orientation may allowreactivation of normal faults formed at theGorda spreading ridge (11). From a consid-eration of stress release, either orientation

Fig. 7. Tsunami measurements at 3tide gauge stations along the E -coasts of California, Oregon, Ha- _' Hwaii, and Johnston Island. Tidal 3 2 Isignal has been removed from all * F J1data except for Crescent City, fCalifornia (inset) (note change of IL.scale). Time of main shock is 0 0hours. The vertical line marks theexpected tsunami arrival time.Contour lines (dotted) representocean depth in kilometers, exceptfor the single, unlabeled 500-mbathymetric contour line nearestthe coast.

reduces north-south compressional stressand down-dip tension in the Gorda plate,but rupture of the northwest-oriented planemay have been favored because of the staticstress changes imposed by the main shockand, perhaps, the first aftershock.

To test this hypothesis, we modeled thechanges in static stress (31) imposed by themain shock (Table 2) on three vertical faultplanes: the east-west-oriented Mendocinofault, the possible N40E-oriented slipplane of the first large aftershock, and theN500W-oriented fault of the second largeGorda aftershock. Large regions of thenorthwest-oriented fault and the Mendo-cino fault received an increase in right-lateral shear stress greater than 3 bars andequally large increases in normal extensionresulting from the main shock. Both theincrease in right-lateral shear and the in-creased extension would bring both ofthese right-lateral faults closer to failureunder a Coulomb failure criterion for non-zero coefficients of friction. About 90% ofthe aftershocks within 4 km of the Men-docino fault and the northwest Gordafault occur where the modeling predictsthe stress changes should load these faultstoward failure for coefficients of frictionranging from 0.0 to 0.75. Thus, staticstress changes may have helped triggeraftershocks along these two faults.

The model also indicates that staticstress changes induced by the main shockare slightly more favorable for failure on anortheast-striking, left-lateral fault plane inthe location of the first large aftershock.primarily because of a decrease in normalstress. In consideration of the present rela-

tive hypocentral locations of the two largeaftershocks, slip on the northeast-orientedplane of the first aftershock would haveadded minor left-lateral shear to the north-west-oriented plane of the second after-shock but would have greatly decreased thenormal stress on this plane. This scenarioprovides a simple mechanism in which thefirst aftershock helped trigger the second,similar to the scenario proposed for theElmore Ranch-Superstition Hills pair ofearthquakes (32).

Hazard implications. The Cape Mendo-cino earthquake sequence provided seismo-logical evidence that the relative motionbetween the North America and Gordaplates results in significant thrust earth-quakes. In addition to the large groundmotions generated by such shocks, they cantrigger equally hazardous aftershock se-quences offshore in the Gorda plate and onthe Gorda-Pacific plate boundary. This se-quence illustrates how a shallow thrustevent, such as the one of moment magni-tude (Mv) 8.5 that is forecast for the entireCascadia subduction zone (14), could gen-erate a tsunami of greater amplitude thanthe Cape Mendocino main shock. Not onlywould this tsunami inundate communitiesalong much of the Pacific Northwest coastwithin minutes of the main shock, but itcould persist for 8 hours at some locales.The 25 April 1992 main shock rupturedonly a small part of the plate boundary andapparently did not trigger slip on any of theHolocene shallow thrust faults observedonshore in the triple junction region (17).Thus, given the high level of historicalseismicity and the emerging picture of manyactive faults, the region is likely to continueexperiencing significant seismicity.

REFERENCES AND NOTES1. H. 0. Wood and F. Neumann, Bull. Seismol. Soc.

Am. 21, 277 (1931).2. L. Dengler, G. Carver, R. McPherson, Calif. Geol.

45, 40 (1992).3. California Office of Emergency Services Situation

Report, 4 May 1992; American Red Cross Disas-ter Update, 8 May 1992.

4. R. Riddihough, J. Geophys. Res. 89,6980 (1984);C. Nishimura, D. S. Wilson, R. N. Hey, ibid., p.10283; D. S. Wilson, ibid., in press.

5. S. H. Clarke, Jr., Am. Assoc. Pet. Geol. Bull. 76,199 (1992).

6. H. M. Kelsey and G. A. Carver, J. Geophys. Res.93, 4797 (1988).

7. R. C. McPherson, thesis, Humboldt State Univer-sity (1989).

8. Earthquakes for the period August 1974 to De-cember 1984 were recorded by the Tera Corpo-ration. Since 1983, the seismicity has been re-corded by local networks operated by the U.S.Geological Survey and the University of Californiaat Berkeley. All earthquake locations are basedon a one-dimensional velocity model with stationcorrections developed from a joint hypocenter-velocity inversion of travel-time data and locatedwith Hypoinverse [F. W. Klein, U.S. Geol. Surv.Open File Rep. 89-314 (1989)]. Calibration explo-sions at Punta Gorda and Cape Mendocino werelocated to the northwest, 1.5 and 2.7 km, respec-

SCIENCE * VOL. 261 * 23 JULY 1993 437

on S

epte

mbe

r 7, 2

011

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fro

m

tively, at the surface. Locations of the offshoreearthquakes are not as accurate as those onshorebecause of the network geometry.

9. R. S. Cockerham, Bull. Seismol. Soc. Am. 74, 569(1984); S. R. Walter, ibid. 76, 583 (1986).

10. T. V. McEvilly, Nature220, 901 (1968); J. P. Eaton,Eos 62, 959 (1981); T. J. Lay, J. W. Given, H.Kanamori, Bull. Seism. Soc. Am. 72, 439 (1982).

11. D. S. Wilson, J. Geophys. Res. 94, 3065 (1989).12. A. C. Lawson, Camegie Inst. Washington Publ.

87, 54 (1908); W. Thatcher and M. Lisowski, Eos68,1507 (1987).

13. J. R. Curray and R. D. Nason, Geol. Soc. Am. Bull.78, 413 (1967); L. Seeber, M. Barazangi, A. Now-roozi, Bull. Seismol. Soc. Am. 60,1669 (1970).

14. T. H. Heaton and H. Kanamori, Bull. Seismol. Soc.Am. 74, 993 (1984); T. H. Heaton and S. H.Hartzell, ibid. 76, 675 (1986).

15. B. F. Atwater, Science 236, 942 (1987); M. E.Danenz and C. D. Peterson, Tectonics 9, 1 (1990);A. R. Nelson, Quat. Res. 38, 74 (1992); G. S. Vick,thesis, Humboldt State University (1989); D. W.Valentine, thesis, Humboldt State University (1992).

16. D. Merritts and W. B. Bull, Geology 17, 1020(1989).

17. S. H. Clarke, Jr., and G. A. Carver, Science 255,188 (1992).

18. We inverted for the main shock source parame-ters using the formalism of B. Romanowicz and T.Monfret [Ann. Geophys. 4, 271 (1986)] and thespherical-Earth model PREM [A. M. Dziewonskiand D. L. Anderson, Phys. Earth Planet. Inter. 25,297 (1981)] to correct for propagation effects.Rayleigh waves with periods between 140 and320 s from a global distribution of stations weresampled at 10 frequencies.

19. An analysis of broad-band surface waves indi-cates that the main shock ruptured to the south-west (azimuth = 2300) with the largest momentrelease beginning 5 km from and extending to 20km offshore [C. J. Ammon, A. A. Velasco, T. Lay,Geophys. Res. Lett. 20, 97 (1993)].

20. For the aftershocks, we computed path-depen-dent amplitude and phase corrections based onanalysis of the main shock. The approach ofpath-dependent corrections is advantageous foruse of data at regional distances as well asshorter periods [M. Pasyanos and B. Romano-wicz, Eos 73, 372 (1992)].

21. H. Benioff, Calif Div. Mines Geol. Bull. 171, 199(1955); A. Ben Menahem, Bull. Seismol. Soc. Am.51, 401 (1961).

22. A. Shakal et al., Calif. Div. Mines Geol. Rep.OSMS 92-05 (1992).

23. Where colonies of organisms were completelykilled, the vertical range of the organisms repre-sents a minimum measure of the uplift. On somereefs, only the upper parts of some colonies ofplants and sessile animals, such as mussels,died. The vertical extent of mortality in thesecolonies provided the best basis for estimates ofthe amount of uplift. The mean and standarddeviation of the vertical extent of mortality for eachcolony (Fig. 6, inset) is estimated from multiplemeasurements made by a laser total station withan instrument precision of 1 mm.

24. We assumed that slip is uniform on a rectangular,planar fault embedded in an elastic half space.We performed a Monte Carlo search of the pa-rameters describing the geometry of the faultplane and, for each geometry, used the horizontaland vertical GPS and coastal-uplift data to esti-mate the magnitude and rake of the slip vector.The scatter of the residuals for the best fittingmodels, including our preferred model (Table 2),is 2.2 times the a priori observational errors. AnF-ratio test indicates other models with residualscatter less than 2.6 times the a priori errors donot differ significantly from the best fitting modelsat the 95% confidence level.

25. In the projections, we assumed that faults areplanar, the ocean depth is 2.3 km at the seawardedge of the Cascadia subduction zone at thelatitude 40'27'N, and the mean elevation of theseismic network is 0.9 km. Main-shock parame-ters are from Table 1. We assumed the upper

boundary of the background Gorda seismicitydips 60 east at a depth of 17.5 km at the coordi-nates of the main shock hypocenter.

26. G. G. Shor, P. Dehlinger, H. K. Kirk, W. S. French,J. Geophys. Res. 73, 2175 (1968); S. W. Smith, J.S. Knapp, R. C. McPherson, ibid. 98, 8153 (1993).

27. W. P. Chen and P. Molnar, ibid. 88, 4183 (1983);P. Molnar and P. England, ibid. 95, 4833 (1990).

28. J. J. Tabor and S. W. Smith, Bull. Seismol. Soc.Am. 75, 237 (1985).

29. D. Verdonck and G. Zandt, Lawrence LivermoreNati. Lab. Rep. UCRL-JC-1 11629(1992).

30. K. Wang and G. Rogers, Eos 73, 504 (1992).31. We estimated static stress changes caused by

32.

33.

the main shock using a rectangular dislocationsurface (Table 2) in an elastic half-space [Y.Okada, Bull. Seismol. Soc. Am. 82, 1018 (1992)]to stimulate the earthquake rupture. The results ofthe model are insensitive to small variations in theassumed fault geometries or slip distributions,although the maximum values and spatial detailsof the stress fields are.K. Hudnut et al., Bull. Seismol. Soc. Am. 79, 282(1989).We thank B. Ellsworth, C. Weaver, D. Wilson, andD. Merritts for reviews of the manuscript. Themeasurement of coastal uplift was supported by agrant from the Pacific Gas and Electric Company.

* RESEARCH ARTICLE

NMR Structure of a Specific DNAComplex of Zn-Containing DNA

Binding Domain of GATA-1James G. Omichinski, G. Marius Clore,* Olivier Schaad,

Gary Felsenfeld, Cecelia Trainor, Ettore Appella,Stephen J. Stahl, Angela M. Gronenborn*

The three-dimensional solution structure of a complex between the DNA binding domain ofthe chicken erythroid transcription factor GATA-1 and its cognate DNA site has beendetermined with multidimensional heteronuclear magnetic resonance spectroscopy. TheDNA binding domain consists of a core which contains a zinc coordinated by four cysteinesand a carboxyl-terminal tail. The core is composed of two irregular antiparallel (3 sheets andan a helix, followed by a long loop that leads into the carboxyl-terminal tail. The amino-terminal part of the core, including the helix, is similar in structure, although not in sequence,to the amino-terminal zinc module of the glucocorticoid receptor DNA binding domain. In theother regions, the structures of these two DNA binding domains are entirely different. TheDNA target site in contact with the protein spans eight base pairs. The helix and the loopconnecting the two antiparallel P sheets interact with the major groove of the DNA. Thecarboxyl-terminal tail, which is an essential determinantof specific binding, wraps around intothe minor groove. The complex resembles a hand holding a rope with the palm and fingersrepresenting the protein core and the thumb, the carboxyl-terminal tail. The specific inter-actions between GATA-1 and DNA in the major groove are mainly hydrophobic in nature,which accounts for the preponderance of thymines in the target site. A large number ofinteractions are observed with the phosphate backbone.

The erythroid-specific transcription factorGATA- 1 is responsible for the regulation oftranscription of erythroid-expressed genesJ. G. Omichinski, G. M. Clore, 0. Schaad, and A. M.Gronenborn are in the Laboratory of Chemical Phys-ics, Building 5, National Institute of Diabetes andDigestive and Kidney Diseases, National Institutes ofHealth, Bethesda, MD 20892. G. Felsenfeld and C.Trainor are in the Laboratory of Molecular Biology,Building 5, National Institute of Diabetes and Digestiveand Kidney Diseases, National Institutes of Health,Bethesda, MD 20892. E. Appella is in the Laboratoryof Cell Biology, Building 37, National Cancer Institute,National Institutes of Health, Bethesda, MD 20892. S.J. Stahl is in the Protein Expression Laboratory, Build-ing 6B, Office of the Director, National Institutes ofHealth, Bethesda, MD 20892.*To whom correspondence should be addressed.

and is an essential component required forthe generation of the erythroid lineage (1).GATA-1 binds specifically as a monomer tothe asymmetric consensus target sequence(T/A)GATA(AIG) found in the cis-regu-latory elements of all globin genes and mostother erythroid-specific genes that havebeen examined (2). GATA-1 was the firstmember of a family of proteins, which nowincludes regulatory proteins expressed inother cell lineages, characterized by theirrecognition of the GATA DNA sequenceand by the presence of two metal bindingregions of the form Cys-X-X-Cys-(X)17-Cys-X-X-Cys separated by 29 residues (2,3). Mutation and deletion studies on

SCIENCE * VOL. 261 * 23 JULY 1993

......

438

on S

epte

mbe

r 7, 2

011

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fro

m