GEOPHYSICAL RESEARCH LETTERS, VOL. 22, NO. 1, PAGES 13-16, JANUARY 1, 1995 Moment-tensoranalysisusingregionaldata: Applicationto the 25 March, 1993,ScottsMills, Oregon,earthquake JohnNfib•lek andGanyuanXia Collegeof Oceanic andAtmospheric Sciences, OregonStateUniversity, Corvallis Abstract. In thispaperwe outlinea procedure we usefor routine Becausethe eventwaswell recordedby theregionalshort-period moment-tensoranalysisof regionaldatafrom broadbandseismic network, as well as some teleseismic stations, we are able to stationsin northwestern North Americaandapplyit to the mo- comparetheresultsbasedon sparseregionalbroadband datawith mentmagnitude5.5, March, 1993, ScottsMills, Oregon,earth- those derived from other observations. quake.The resultscomparefavorablywith thoseobtainedfrom teleseismicdata. We found that the earthquakeoccurredat a Regional Data and Inversion Procedure depthof 13-15 km and had a mechanismwith approximately equalamountsof reverseandright-lateralstrike-slipcomponents. Theregional stations usedin thisstudyareshown in Figure1. The estimatedstressdropof 40 bar is averageon a world-wide At the time of the ScottsMills event,therewere 8 broadbandstabasis,supportingthe view that the ratherlarge damagewas tionsoperating withina radiusof 800km fromtheepicenter. We causedprimarilyby poor construction andnot by exceptional useddatafrom7 of them:COR (operated by theIncorporated propertiesof the source.The ScottsMills earthquakeis most Institutions for Seismology andthe OregonStateUniversity), likely relatedto the Mr. Angel Fault.This fault is a part of the ARC, WDC, MIN (Univ. of Californiaat Berkeley),andPGC GalesCreek-Mr.Angelstructural lineament (GCMAL) extending and WALA (CanadianNational SeismographNetwork). about150km acrossthe WillametteValley. At presentdataare AlthoughstationLON hasa broadbandsensor,its recordswere not sufficientto estimatethelikelihoodof an earthquake involv- saturated dueto thelow-dynamicrangeof therecorder.Instead, ing the entireGCMAL, but givenits lengthan earthquakeof we wereableto usedatafromthedigitallong-period seismometer magnitude7 is conceivable.The resultsof this study,together whichalsorecords at thatsite.StationHA0 (U.S. Geological with investigations of other earthquakes, suggestthat sparse Survey)is a digital strong-motion accelerograph locatedin broadbandnetworkscan be used efficiently for determining Portland, 56 kmfromtheepicenter. It recorded surprisingly good sourceparameters of earthquakes of magnitudegreaterthan4.0 in low-frequency datasuitable for ouranalysis. The inversion for the moment tensor and sourcetime function regionswith infrequentseismicity. involvesmodeling of theentire3-component seismograms. The sourceparametrization andinversionprocedure followclosely thoseusedbyNdb•lek[1984]fortheanalysis of teleseismic body waves.For a horizontallylayeredmedium,displacement as a Introduction The ScottsMills earthquake of March25 (M w= 5.5) wasthe first of severalunusuallystrongearthquakes thatrockedOregon function of time t observed at a station at a distance A and in 1993.It occurredabout35 km eastof thestatecapital,Salem, azimuth { fromtheearthquake epicenter canbeexpressed as on the eastside of the Willamette Valley, in the foothillsof theCascademountains.For its magnitudeit causedan unusually uPSV({,A,0 = {HPsv2 (A,h,t) [1/2(myy+mxx) large amountof damage[Madin et al., 1993]. The secondset of -1/2 (myy-mxx) cos2•+mxy sin2•)] M w = 6.0 earthquakes occurrednearthe town of KlamathFallsin +HPsvl (A,h,t) [my zsin•)+mxz cos•)] +HPsvø (A,h,t)mzz}*fl(t) the southernpart of the statein September. The analysisof the KlamathFallsearthquakes is presented in a companion paperby Braunmiller et al. [1994b]. For its locationin a subductionzone, 12, ;'W Oregonover the last hundredyearshashad anomalously low seismicity[Jacobsen,1986]. The recentearthquakes, however, indicatethat the earthquakepotentialin this regioncannotbe ignored. The ScottsMills earthquakeoccurredat the time whenseveral newbroadband digitalseismicstations hadrecentlybegunoperating in the Pacific Northwest.Anticipatingthis technicaldevelopment,we hadbegunto deviseprocedures for routineanalysis of suchdataandwereableto applythemto thiseventas oneof the first of applications.Using regionalbroadbandstationswe now routinelyestimatethe sourcemomenttensor,depth,and sourcetime function from low-frequencywaves for eventsof magnitude4.0 and larger. Our efforts in the Pacific Northwest parallel those in other parts of the world (e.g., California [Romanowicz et al., 1993; Ritsemaand Lay, 1993] or Europe [Giardini et al., 1993;Braunmilleret al., 1994a]). 48'N 44'N 120'W 116'W 112'W .co., .. . .co...u.s ?, ß -\ . 45' 12'N . 45' O0'N 40'N In this paperwe outlinethe techniquewe useto analyzethe regionalwaveformsandapplyit to the ScottsMills earthquake. - , 122' 48• 1•' 36'W Figure 1. (Background)Map of stationdistributionusedin the regionalwaveanalysis.(Foreground) Seismicity(mb>l.5, 19901993, WRSN data) in the area of the ScottsMills earthquake. Mainshockis indicatedby the star.The mappedtraceof the Mr. Angelfault is indicated.Topography,75-m contourinterval. Copyright1995by the AmericanGeophysical Union. Paper number 94GL02760 0094-853 4/95/94GL-02760503.00 13 N*BgLEK AND XIA: MOMENT-TENSOR ANALYSIS USING REGIONAL DATA 14 uSH((D,A,t ) = œI-I sH2(A,h,t)[1/2(myy-mxx) sin2•+mxy cos2q)] +Hsm (A,h,t)[myz cos(D-mxz sin(Dl) * •(t) z whereu?svsignifiesdisplacement (verticalandradialcompo- nents)resulting fromtheP-SV coupled seismic waves,andusI• signifiesdisplacement (transverse component) dueto SH coupled waves.mijarethesource moment tensor components. H's are excitationfunctionsfor a sourceat a depthh with a unit-stepslip history,and•'l(t) is the far-fieldsourcetime function. FollowingNdb•ek [1984], we parameterizethe sourcetime function LON -" V¾v - -- 'v'"'vV-- - - as •'l(t) = Z akT2x(t-[k- 1Ix); k = 1, 2..... n; whereT2xis a seriesof n isosceles-triangle functionsof a unit area,duration2x, andoverlappedby x, andakarethe corresponding amplitudeweights(requiredto sumto 1). The resultingtime functionhasamplitudesspecifiedat an equaltimeintervalx, with the interveningsamplesinterpolatedby the trapezoidalrule. By varyingx andn we cancontrolthe time resolutionandthe total ARC- v"'"VV- - W 'Vv,,v duration of the source time function. If we know the crustal structure, we can calculate the excita- tion functionsand determinethe best fitting (we use the least- squares inversion)momenttensorandsourcetime function.The depthcanbe determined by observing thefits for a suiteof trial depths.We calculatethe excitationfunctionsusingthe method discrete wavenumber summation of Bouchon [1982] for the crustal model shown Table 1. This model was derived from a refractionstudyin westernOregonandWashington[Tr•hu et aL, 1994], adjustingthe Poisson'sratio so that major phasesmatch thearrivaltimesat distantstations. Usingthesamecrustalmodel for the entireregionlimits the frequencybandfor whichthe theoretical seismogramscan adequatelydescribethe observeddata. We used 0.1-0.01 Hz frequencyband for the nearestthree stations, and 0.05-0.01 Hz band for the others.In thesefrequency bandsseismograms are dominatedby guidedwavesin the crust (Pnl,Rayleigh,Love) andwide-anglereflections. The momenttensorwas constrainedto be purelydeviatoric.In order to give roughly equal weight to all stations,we usedthe samenumberof samples(200) from eachseismogram (the near stationswere sampledat a rate of 2 spsandthe distantstationsat 1 sps)andthe amplitudesof the seismograms were correctedto a referencedistanceassumingcylindrical geometricalspreading. Even after correctingfor the geometricalspreading,becauseof the differencein pass-bands, the power of the signalat the near stationswas higherthan at the distantstations;we correctedfor thisdiscrepancy by down-weighting thenearstations by 40%. In orderto counter-actsomeof the phasemisalignmentdue to Time(s) Time(s) Time(s) Max.amplitude: 676Izm(top3 rows);204Izm(bottom 5 rows) Figure 2. Observed(solid lines) andbest-fittingsynthetic(dashed lines) regionalseismograms of the ScottsMills earthquake.Note the differencein scalesbetweenthe top tracesfor stationsclosest to the epicenterand the lower tracesfrom distantstations. Resultsof the Regional Wave Analysis Figure 2 showsthe observedand theoreticalseismograms for the best-fit model. The data showlarge variationsin the amplitudesof the three componentsfrom stationto stationreflecting the radiationpatternof the sourcemechanism(momenttensor). Table 2 liststhe results.Data are generallywell matched;the misfit is understandably largerfor the distantstations.It appearsthat someof the mismatch,in particularat WALA, can be attributed to L/R anisotropy. Figure 3 (top) showsthe variancereductionas a functionof sourcedepth.The minimumis found for 12 and 13 km. If we assume10% varianceincreasebeing significant,the uncertainty extendsfrom about 11 to 15 km. The figure showsthat the esti- deviations from the assumed crustal model and errors in the ori- mated moment tensor varies as a function of the assumed source gin time and epicentrallocation,we realignedthe observedand syntheticseismograms to maximize the correlation.The maximum shift was 3 s. Becausewe requiredall 3-componentsat a given stationto shift together,this processdid not removemisalignmentscausedby Love/Rayleighwaveanisotropy. depthfrom predominantlythrustmechanisms at a shallowdepth to predominantlystrike-slipmechanisms for the lower of the calculateddepths.The northeast-dipping planeof the focal mechanism is relatively stable (its dip increasesslightly with depth) while the southeast-dipping plane progressivelychangesthe dip directionfrom southerlyto easterly.As the depthincreases,the estimatedseismicmoment increases.For the best fitting depth, the mechanismis mid-way betweenthrustand strike-slip,andthe Table 1. Crustal layer parameters used for calculation of synthetic seismograms Thickness, Vp, vs, Density, km kfia/s km/s 1.0 5.0 3.0 3.0 5.0 5.0 4.0 9.0 6.0 4.00 5.50 6.35 6.47 6.67 6.75 6.93 7.16 6.50 2.25 3.09 3.56 3.63 3.74 3.79 3.89 4.02 3.65 2.14 2.56 2.80 2.83 2.89 2.91 2.96 3.02 2.84 8.10 4.55 3.29 half-space seismic moment is 2.4xl0•7Nm (M• = 5.5).Nearthebest-fitting depththe solutionis 90% double-couple. g/cm• Table 2. Source parameters of the ScottsMills earthquake determined in this study Data Strike, deg Dip, deg Rake, deg Depth, km Mo, Nm Regional 302 51 145 12-13 2.5x10n Teleseismic 297 48 138 14.3 2.4x 10•7 Duration: 2.0 s Fault radius:3.0 km; Av. slip:22 cm; Av. stressdrop:40 bar N•B•LEK AND XIA: MOMENT-TENSOR ANALYSIS USING REGIONAL DATA 15 0.5 ale P mbc 0.4 1.8, 92 0.3 ..• thj yss 0.2 2.•,• •_ •.,•_•_.•. • kip 2.4, 90•[Y • 2.5, 89 0.1 ,. o.... .... .... .... io.... J frb , . .• , .• Imq gac ,• hrv P: 214 nm 1 .••.goga $H: 379 nm , , , o 102o3o time (s) Depth(km) 0.5 , [] Strike "';: -' ale yss ß i frb .• 0.3 0.2 ,.. Figure 4. Observed (solid lines) and best-fitting synthetic (dashed)teleseismicbodywaves(displacement, 0.02-0.05 Hz) of .... i'0" the main shock.Focal sphereswith P and SH nodalsurfacesand Deviation(deg) projectionsof the departingray-pathsare alsoshown.Data are Figure3. (Top) Variance(normalizedby datapower)vs. assumed from stationsof theFederationof theDigital SeismicNetwork. depthof the source.The fault plane solutions(beachbahs), the seismicmomentin unitsof 10]?Nm (firstnumber),andtheperclose-instationsusingseismograms from suitableaftershocks as the empirical Green functionscalibratingthe propagationpaths [e.g.,Mori and Frankel, 1990]. StationHA0 wouldbe a logical (Bottom) Variance vs. deviation from the best-fit estimate of choice for such analysis. Unfortunately, the only aftershock strike,dip, andrakefor a sourceat a 13 km depth. whichtriggeredthis stationshowedvery little coherencewith the mainshockindicating a very different mechanismand source Figure3 (bottom)showstheresolution of thefocalmechanism depth,and thereforecouldnot be used.We usethe aftershockof assuming a northeast-dipping faultplaneanda source depthof 13 June8, 1993 (mb=3.7) whichproducedcoherentrecordsat CaR, km. The bestresolvedparameteris the strikeof the fault plane, 74 km from the epicenter.BecauseCaR was nodal for vertical while the rake of the slip angleis the leastresolved.For a 10% and radial components,the analysiswas carriedout only on the varianceincrease,the boundsare -4ø to +4ø for strike,-7ø to +4ø transversecomponent.The estimatedsourcetime function(Fig. due for dip,and-7øto +9øfor rake,relativeto thebest-fitestimates. 6) hasa durationof 1.8 s. Accountingfor someunderestimate centageof the double-couplecomponentof the momenttensor (second number) estimated for each depth are also shown. Usinglong-period datathesource timefunctioncannotbewell resolved.All we cansayis thatthedurationis shorterthan3 s. TeleseismicBodywave Analysis to constrain the mechanism was probably about 2.0 s. Assuming a circular crack model [Madariaga, 1976] with equalruptureandhealingvelocitiesof 3 kngs(0.8vs)we canestimatethe fault radiusto be about3.0 km. Togetherwith theseismicmomentof 2.5x10•7Nm, thisimplies The resultsof the regionalwave analysisare checkedagainst those obtained from teleseismicbody waves. Typically such analysisis performedusingbroadband P andSH wavesin the3090øepicentraldistancerange[e.g.,Ntib•lek,1984].For theScotts Mills earthquake onlya few goodqualitybroadband bodywaves are available, nevertheless,after bandpassfiltering in the 0.020.05 Hz band (notchin the Earth'sambientnoisespectrum)we were able to obtain good azimuthal distributionof P and SH waveforms to the finite duration of the aftershock, the mainshock duration and seismic moment (Figure4, Table 2). The resultsarein very goodagreementwith the regionalwave analysis.The low-frequencybodywavescould not resolvethe sourcedepth,but fortunatelytwo stationsin Asia (TLY and YSS) producedgoodqualitybroadbandP waveswith distinctP and sP arrivals(pP is nodal).Thesecouldbe usedto determinethe sourcedepthprecisely(Figure 5), in good agreementwith the depthestimatefrom theregionalwaves.The pulse an averageslip on the fault of 22 cm anda stressdropof 40 bar. Summary and Discussion The results of the regional and teleseismicanalysesof the Scotts Mills earthquakeare remarkably consistent(Table 2). They indicatea mechanismwith approximately equalamountsof reverseand fight-lateralstrike-slipcomponents on a fault with a strikeof about300ø anda dip of 50ø.The centroiddepthis 13-15 0.9 tly yss 0.6 widths at these two stations indicate a source duration of ~2 s. time (s) 0.5 Depth(km) SourceDuration From Empirical Green Function and StressDrop Figure 5. Broadbanddisplacementseismogramsof P waves recordedat TLY andYSS (solidlines) showingcleardirectP and The teleseismic estimate of source duration for a medium sP arrivals tightly constrainingthe depth of the mainshock (dashed magnitudeearthquakeis significantlyaffectedby the uncertainty (varianceplot). The best-fittingsyntheticseismograms in the Earth attenuation. A better estimate can be obtained from lines) and the estimatedsourcetime functionare also shown. 16 N/•B•LEK AND XIA: MOMENT-TENSORANALYSISUSING REGIONALDATA routinemoment-tensor determinations for this regionare now includedin the QuarterlyNetworkReportsof the WRSN compiledby theUniv. of Washington. In specialcircumstances, when the earthquakeoccurswithin 100-150 km from severalstations, higherfrequencydatacanbe includedin the analysisandthe magnitudethresholdis lower. For example,Braunmilleret al. [1994b] were able to applythis techniqueto aftershocks of the KlamathFalls earthquakethat had a magnitudeas low as 3.5. Furtherimprovements in theprecision of theparameter estimates Time (s) and applicationto smallereventswouldrequirea denserstation coverage andbettercalibration of thewavepropagation paths. Figure6. Broadband displacement seismogram (trans.comp.,20 We thankB. Romanowicz, J. Cassidy, H. Benz,and sps,solidline) for the mainshock recordedat COR. The source Acknowledgments. time function(below) wasfoundby deconvolving the mainshock waveform with a waveform from an aftershock(dashedline). A. Frankelwhoprovided datausedforthisstudy.J.Cassidy, R. Crosson, andJ.Deweyprovided helpfulreviews. Thisresearch wassupported by theU.S. Geological Surveygrant1434-93-G-2326. References km and the moment magnitude5.5. Except for the somewhat smaller dip, the parametersfound here are also in excellent Beeson,M.H., K.R. Fecht,S.P.Reedier,andT.L. Toluene,Regional agreementwith thosereportedby Thomaset al. [1993] basedon correlations withintheFrenchman Springsmemberof theColumbia Riverbasaltgroup:New insightsintothemiddleMiocenetectonics of the first-motion analysisof regional short-periodseismograms northwestern Oregon,OregonGeology, 47, 87-96,1985. from the WashingtonRegionalSeismicNetwork(WRSN) (strike = 304ø, dip = 58ø, rake = 138ø;focal depth= 15 km). Usingdata Bonneville,D., A. Del Pino,andD. Kelly, The March25, 1993,Scotts Mills earthquake - effectson structures, EERI SpecialEarthquake from the WRSN and temporary local stationsthey found that aftershocks extended from about 8 to 13 km. A subset of these eventsapparentlylies on a WNW strikingplanewith a dip of 55ø to 60ø to the northeast. A preliminary examination of data recorded by temporary broadband stations deployed in the epicentral area after the mainshock by the Oregon State University (OSU) showsa wide variationof mechanismsof the aftershocks,indicatingthat the aftershockscannotbe interpreted in termsof a singlefault plane. The estimatedstressdrop of 40 bar for the mainshockis average on a world-wide basis, supportingthe view that the rather large damagewas causedprimarily by poor constructionof the pre-1950 structures[Bonnevilleet al., 1993;Deweyet al., 1994], andnot by exceptionalpropertiesof the source. The ScottsMills earthquakeis mostlikely relatedto the Mt. Angel Fault (MAF), whichhasbeenmappedat the surfacewest of the epicenter(Figure1). The strikeof the MAF at its southern mappedendis identicalto thatof themainshock mechanism. The exactrelationship,however,is not certain.Given the estimated dip, centroiddepthandepicenterof themainshock, theprojected fault plane would reachthe surfacesouthwestof the mapped fault, thusif the MAF were the causativefault for the earthquake it wouldhaveto curveupwardasit reachedthe surface.Another alternativeis that the MAF is in fact a zone consistingof several parallelfaultsandthat the earthquake occurredon a stranddis- placeda few km to thesouthfromthemappedfaultplane. The Mt. AngelFaultis a partof the GalesCreek- Mt. Angel structurallineament(GCMAL) extendingacrossthe Willamette Report, 5-8, 1993. Bouchon,M., The completesynthesis of seismiccrustalphasesat regionaldistances, J. Geophys. Res.,87, 1735-1741,1982. Braunmiller,J., T. Dahm,andK.-P. Bonjer,Sourcemechanism of the 1992Roermond earthquake fromsurface waveinversion of regional data,Geophys. J. Int., 116, 663-672,1994a. Braunmiller, J., J. Nfib•lek,B. Leitner,andT. Qamar,The 1993Klamath Falls, Oregon, earthquakesequence:Sourcemechanismsfrom regionaldata,Geophys. Res.Lett.,in press,1994b. Dewey, J.W.,B.G.Reagor, D.Johnson, G.L.Choy, andF.Baldwin, The ScottsMills, Oregon,earthquakeof March 25, 1993: Intensities, strong-motion data,and teleseismicdata,U.S. GeologicalSurvey, Open-FileReport94-163, 1994. Giardini, D., E. Boschi,and B. Palombo,Moment tensorinversionfrom Mednetdata(2) Regional earthquakes of theMediterranean, Geophys. Res. Lett., 20, 273-276, 1993. Jacobsen, R.S.,Map of Oregonseismicity, 1841-1986,OregonDept.of Geol.andMin. Ind. GeologicalMap SeriesGMS-49,1986. Madariaga,R., Dynamicsof anexpanding circularfault,Bull.Seis.Soc. Am., 66, 639-666, 1976. Madin, I.P., G.R. Priest,M.A. Mabey, S. Malone, T.S. Yelin, andD. Meier, March25, 1993 ScottsMills earthquake - westernOregon's wake-upcall, OregonGeology,55, 51-57, 1993. Mori, J., and A. Frankel, Sourceparametersof small eventsassociated with the 1986 NorthPalmSprings,CA, earthquake determined from empiricalGreenfunctions,Bull. Seis.Soc.Am., 80, 278-295, 1990. Nfib•lek, J., Determinationof earthquakesourceparametersfrom inversionof bodywaves,Ph.D.thesis,Mass.Inst.of Tech.,1984. Ritsema,J., andT. Lay, Rapidsourcemechanism determination of large (Mw>5)earthquakes in thewesternUnitedStates,Geophys. Res.Lett., 20, 1611-1614, 1993. Valley [Beesonet al., 1985]. The MAF structurehas been Romanowicz, B., D. Dreger,M. Pasyanos, R. Urhammer, Monitoringof mappedusingwaterwell andindustryseismicreflectiondataby strainreleasein centralandnorthernCaliforniausingbroadband data, Werner et al. [1992]. The offsetsof sedimentaryhorizonsindi- Geophys.Res.Lett., 20, 1643-1646, 1993. catea postMioceneage.In 1990 a seriesof smallearthquakes Thomas,G.C., R.S. Crosson,S. Dewberry,et al., The 25 March 1993 (mb< 2.5) occurredsouthof Woodbumat thenorthwest endof ScottsMills, OregonEarthquake: Aftershockanalysisfromcombined permanentand temporary digital stations(abstract),EOS, Trans. the mappedfault (Figure 1). The eventsoccurredat a depthof Arner.Geophys.Un., 74, 201, 1993. about15 km andhad a strike-slipmechanismconsistent with the orientationof the MAF there,indicatingactivedeformationand seismic potentialof thefault[Werneret al., 1992].At present we do not have sufficient data to estimate the likelihood of an earth- quakeinvolvingthe entireGCMAL, but giventhelengthof the Trthu, A., I. Asudeh,T. Brocker,J. Luetgert,W. Mooney,J. Nfib•lek, and Y. Nakamura, Crustal architectureof the Cascadiaforearc, Science,in press,1994. Werner,K., J. Nfib•lek,R. Yeats,andS. Malone,TheMountAngelfault: Implicationsof seismic-reflection data andthe Woodbum,Oregon, earthquake sequence of Aug. 1990,OregonGeol.,54, 112-117,1992. structure(about 150 km) and the seismogenicthicknessof the crust(15 km) an earthquake of magnitude7 is conceivable. In conclusion, ourstudyshowsthata relativelysimpleanalysis JohnN•b•lek, Collegeof Oceanic andAtmospheric Sciences, Oregon of datafrom sparseregionalbroadband stationscanyield quite StateUniversity,Ocean.Admin.Bldg.104,Corvallis,OR 97331-5503. preciseestimatesof sourceparameters.In the northwestern GanyuanXia, now at the Dept. of GeologicalSciencesandInstitute United Statesand western Canada, where there are 30 broadband for Geophysics,Universityof Texasat Austin,Austin,TX 78713. seismicstationsoperatingat thistime,thistechnique canbe routinelyappliedto eventsof magnitude greaterthan4. The OSU (ReceivedJuly14, 1994;acceptedSeptember19, 1994.)