GEOPHYSICAL RESEARCH LETTERS, VOL. 28, NO. 1, PAGES 21-24, JANUARY 1,2001 Error spectrum for the global M2 ocean tide R. D. Ray,• R. J.Eanes, 2 G. D. Egbert, 3 andN. K. Pavlis 4 Abstract. The mostaccuratedeterminations of the global The tidal errorspectrum is estimated hereby two comoceantidesarecurrentlybasedon altimetermeasurements plementary methods:by computing differences in existing madeby the Topex/Poseidon satellite.The errorspectrummodelsandby estimating errorsthroughglobalinversecalcorresponding to theM2 tidalsolutionis hereestimated, pri- culationswherereasonably realistica priori errorcovarimarilyby inverse methods andsecondarily by simplediffer- ancesareemployed. Because we areinterested in theeffect encingof severalof the besttidal models.The tidal error of tide modelerrorson the gravityfield as determinedby spectrumis flatterthanthe tidal signalspectrum,and it ex- satellitegeodesy,the erroris expressed in termsof equivaceeds10% of thesignalat sphericalharmonicdegree15 and lent geoidheighterror,aswasrecentlydoneby Wahret al. above. The tide errorsalsoexceedthe anticipatedsensitiv- [1998]. Two possiblesystematic errorsin the conversion to ity of the upcomingGRACEsatellitegravitymissionfor all tidal geoidvariationsare alsoexamined:the commonasdegreesbelow 40, andpossiblybelow 50. sumptions of a constantseawaterdensityand of a spherical earth. In this note we confine ourselves to the dominant 1. Introduction semidiurnal lunar tide M2. Sincethe launchof the Topex/poseidon satellitein 1992 considerable progresshasbeenmadein our knowledgeof the globaloceantides[e.g.,Le Provostet al., 1995]. This progresshas been quantifiedin severalways: by numerouscomparisons with in situtide determinations, primarily with coastaltide gaugesat small islandsand bottompressurerecorderson the sea floor; by variancereductiontests with independentsatellitealtimeterdata; and by comparisonswith tidal estimatesfrom satellitetracking [e.g., An- 2. Formalism dersen et al. , 1995; Shum et al., 1997; Desai et al. , 1997]. (Testsof tidal currentvelocitiesfrom someof the new tide modelshavealsobeenmadeby Dushawet al. [ 1997] andby others,althoughtidalcurrentsarenotthetopicof thepresent paper.) Anotherapproachto understanding accuraciesof ocean tide modelsis to determinetheir errorspectra--thatis, their erroramplitudes asa functionof the degreen of a spherical harmonicexpansion.In additionto characterizing modelaccuracies,reliableerrorspectrahavemanyapplications[e.g., Tsaoussiand Koblinsky,1994]. One application(prompting the presentstudy)involvesestimatingthe effectof aliased tide modelerrorson monthlydeterminations of the Earth's gravityfield as anticipatedto occurduringthe upcoming GRACEsatellitemission. The first step of this work is to determinethe errorspectrumof the bestavailablemodels. We employthe sphericalharmonicnormalization commonly usedin physicalgeodesy[Heiskanenand Mortiz, 1967],wheretheLegendre functions Pnm(cos O)satisfy .;_• [Pnm(•t)] 2dtt - 2(2 - Sm,0). 1 The contributionof the oceantidesto the geoidundulation N is expressedas •SN(O, qo, t) - a • • Pnm(cos0)[•SCnmcOsmqo+•SSnm n= 1 m=0 (1) wherea is the earth'smeanradiusand (O,9) are spherical polarcoordinates. Thetime-varying Stokescoefficients are computed fromthetidalelevation •(O,q0, t) by [cf Wahret al., 1998] •Snm(t) Me(2n+ 1)f/ •Pnm sinmq0 {•Cnm(t) }--a2pw(l +k•n) {cosmq0 }dg•(2) wherePwisthemeandensity of theocean, Me isthemassof theearth,andk•nisa loading Lovenumber of degree n which accounts for thegeoidperturbation fromoceantidalloading of the solid earth. Define the Stokescoefficients•Cnmand •Snmas the root overa complete tidalcycleT (12.42hours)of 1Laboratory forTerrestrial Physics, NASA GoddardSpaceFlight meansquares thetime-varying coefficients. Thenthetidalcontribution to Center,Greenbelt,Maryland. 2Center for SpaceResearch, University of Texas,Austin,Texas. thevarianceof thegeoid,takenovertheentireglobe,is 3Collegeof OceanicandAtmospheric Sciences, OregonState University,Corvallis,Oregon. 1 //f•N2dtd•--a2E(•2nmq-•n 4Raytheon ITSS, Greenbelt, Maryland. 4r•T We takethe amplitudespectrumto be the quantity Copyright 2001 bytheAmericanGeophysical Union. 1/2 Papernumber2000GL011674. (3) •Nnm-a(•2nmq-•2nm) 0094-8276/01/2000GL011674505.00 21 22 RAY ET AL.' ERROR SPECTRUM The degreeamplitudespectrumis [Wahret al., 1998] •Nn -- a n --2 (Z aXn •Cnmq- rn \m-O 3. Tidal error spectra (4) FOR THE M2 TIDE tions(LTE) and prior statisticalassumptions aboutthe errorsin theseequations[Egbertet al., 1994]. Posteriorerrors canthusbe calculatedto characterize uncertaintyin the estimatedfieldsarisingfrom incompletealtimetercoverageand errorsin thedata.We usea Monte Carloapproach[Dushaw et al., 1997], which we summarize here. A series of ran- dom realizationsof the LTE forcingerror (which accounts in the equations,bathymetryerrors,and Figure 1 showsthe spectrumof the oceantide computed for approximations numerical grid truncation) are generatedusingtheprior cofrom threerecentglobalmodelsof the M2 oceantide. (The variance derived in Egbert et al. [ 1994]. For eachrealization figure strictly showsthe tide elevationspectrum,not the i the perturbed LTE are solved numericallyto yield synthetic geoidspectrum;the two are relatedby the factorpreceding tidal elevations •i, i = 1,..., N. (Fortheresults reported here the integralin (2).) The threemodelsare all deducedfrom we use N -20.) These elevation fields are each sampled Topex/Poseidon altimetryusingdifferentmethodsof analysis, including(for TPXO.4) formal data assimilationinto a with thespatialandtemporalpatternof thealtimeter,random of noiseandnon-tidaloceanograhydrodynamicmodel. All are improvements to earliertide dataerrors(representative modelsthat Shumet al. [1997] found were amongthe most phy in the T/P data) are added,and the syntheticdata are accurateof thoseavailablefor the entireglobalocean.Solu- inverted(usingthe sameprocedureappliedto real T/P data tion GOT99.2is describedby Ray (1999); CSR4.ois similar togenerate TPXO.4) forestimates •i. Thedifferences •i- •i in designbut differentin somedetails,includingthe method are representative of errorsin the assimilationestimatesof usedfor tidal analysis.AssimilationsolutionTPXO.4 is an tidal elevations.The error fields are found to be largestin updateto thatdescribedby Egbertet al. [1994] andEgbert thepolarseas(beyondthe satellitedatacoverage)andin the [1997].BothCSR4.O andGOT99.2 default toapurelyhydro- shallowseassurroundingIndonesia.The root meansquare dynamicsolution[Le Provostet al., 1994] in polarregions amplitudesof sphericalharmoniccoefficientsof theN error abovethe Topex inclinationof 66ø, while TPXO.4 extends fieldsareusedto estimatethe tidal errorspectrum. polewardsvia its own linearizedhydrodynamicmodel. The Parametersof the prior covariancecanbe adjustedsothat threespectraareclearlyin closeagreement,althoughcareful the amplitudeand spatialstructureof the randomrealizaexaminationrevealsminor differences.The largestvariation tions•i aregrossly consistent (in amplitude andspatialstrucin the spectrais a functionof degreen, notorderm. The cor- ture) with the T/P data, but it mustbe admittedthat signifirespondingdegreeamplitudespectrumis shownas the top cantuncertaintiesremainaboutthe "true" prior covariance. We have thus performedcalculationsfor severaldifferent curvein Figure2 (basedon TPXO.4). The degreeamplitudespectraof the tide model errors, (consistent)prior covariancesto test the sensitivityof our to poorlyconstrained parameters.Curves1 and basedon differentapproaches to estimatingthe errorsin the conclusions coefficients•SC,mand•SS, m,are shownasthe coloredcurves 2 in Figure 2 give tidal error amplitudespectracalculated labeled1-4 in Figure2. ']['hey stemfromtwocomplementary usingdecorrelationlength scalesof 300 and 600 km, remethods. spectively,for the LTE forcingerrors. The two curvesare in goodagreementfor degreeslessthan 10. For higherde3.1 Error estimates from inverse methods greesthe shorterdynamicalerrordecorrelationlengthscale The assimilationsolutionTPXO.4 canbe viewedasan op- results(quitesensibly)in somewhatlargererrors. But diftimal interpolation or smoothingof the T/P altimeterdata, ferencesbetweenthe two casesare not so great;both cross usinga covariance determinedby the Laplacetidal equa- the GRACEsensitivitycurveneardegree50. O GOT99.2 • -," TPXO.4a • Figure1. Spherical harmonicspectra of threerecentglobaloceantidemodelsasa functionof degreen andorderm. Vertical axisrepresents spectralamplitudein seasurfaceheight. RAY ET AL.: ERROR SPECTRUM FOR THE M2 TIDE 23 derestimate the real errors since identical methods of anal- ysis were used, and indeedcurve 4 falls below curves 1-3 everywhereexceptfor the very lowestdegreeswhereit rises abovecurves1 and 2. Exceptfor thesevery low degrees, curves3 and4 areconsistently belowcurves1 and2. 10ø E 3.3 Other systematicerrors ß •. 10'•- . E o 10 =,. . 10• 0 ., 10 20 30 40 50 60 degree Figure 2. Degreeamplitudegeoidspectraof the globalM2 tide, variousestimatesof the present-dayM2 errors,and the anticipatedsensitivityfrom 3 monthsof data from the GRACE satellitegravity mission. Error curvesare shown in color and are numberedas follows: (1) inverseestimate corresponding to TPXO.4 assuminga 300-km decorrelation lengthfor the dynamicalerrorcovariance;(2) sameexcept with 600-km decorrelationscale; (3) estimatefrom the differencein modelsTPXO.4 and CSR4.o;(4) estimatefrom the differencein modelsCSR4.oandCSR3.0;(5) errorcausedby the neglectof earthflattening;(6) errorcausedby useof a constantdensityfor seawater,with the largererroremploy- Curves5 and6 in Figure2 exploretheeffectof twopossible systematic errorsin theestimates of thetidalgeoidvariations. For this purpose,both assumethat thereis no error at all in thetidalelevations • (whicharetakenfrommodel TPXO.4).Curve5 showsthe errorcausedby the neglectin Eq. (2) of the earth'sflattening.Specifically, it is the spectral differencebetweena harmonicexpansion basedupona sphericalearth surface(radius 6371 km) and one basedon a "serrated"ellipsoidalsurface[Jekeli, 1981]. The error is seento peakarounddegrees5 to 7. Curve 6 (actuallycomprisingtwo curves)showsthe error causedby the commonassumptionof constantseawater density,whichallowedPwin Eq. (2) to be outsidethedouble integral.Theerrorin suchassumption hasbeencomputed by includingPwwithinthespatialintegration whereit thenrepresentsthe averagedensitywithin the watercolumnat any location.This densityhasbeencomputedfrom the Levims et al. [ 1994] climatologicaldensities.The two errorcurves correspond toassuming Pwiseithera constant 1025kgm-3 or1035kgm-3. Thelatterismorerepresentative ofthetrue ingdensity 1025kgm-3 andthesmaller themorerealistic globalmeanseawaterdensity. The formeris morerepre1035kgm-3. sentativeof surfacewaters,andit inducesthe largererrors, actuallycrossing severalof theothercurvesbelowdegree6. 3.2 Error estimates from model differences A moresimplisticapproachto estimatingthe error in the tidal coefficientsis by computingthedifferencesamongour three adoptedmodels. Since all are basedon the samealtimeterdata, this approachlikely underestimates the errors 4. Discussion Curves1-3 in Figure2 represent ourbestestimates of the errorspectrum for presentmodelsof the globalM2 ocean somewhat, but it is still a useful check on the inverse results. Figure 3 showsthe maximum elevationdifferencebetween TPXO.4and CSR4.o,computedat every point of the global ocean. Throughoutmost of the open ocean where Topex/Poseidon data are of greatestbenefit,the differences betweenthese two models are small, generally less than 2 cm. In coastalregionswhere the tidal wavelengthsare shorterandthetideis lesseasilymappedby relativelywidely spacedsatelliteobservations, modeldifferencesare significantlylarger.Someof thesedifferencesare alsoin regions of extremelylargetidal amplitudeswhereevena smallpercentageerrorcanproducelargediscrepancies. As expected, modeldifferencesarealsolargein polarlatitudesoutsidethe satellitelimits; resultsin theseregionsrely completelyon hydrodynamicmodelingand (possibly)a sparseset of tide gaugemeasurements. Figure 3 certainlysuggests locations wherefuturemodelingeffortsshouldbe directed. Curve3 in Figure'2showsthedegreeamplitudespectrum computedfrom Figure 3. A similar resultis obtainedfrom differences involvingtheGOT99.2model.Curve4 is a spectrum computedfrom the differencebetweenCSR4.oand its earlierversionCSR3.0.This curveis especiallyproneto un- 30'$ 80'S 60'E 120'E 0 180' i 2 4 120'W 6 8 10 $O'W 20 o' oll Figure3. Maximumseasurfaceheightdifference between theCSR4.oandTPXO. 4 modelsof theM2 globaloceantide, in cm. Note the color scale is nonlinear. The maximum difference(in HudsonStrait) exceeds4 meters.Somebasin boundary errors(notin TPXO.4)arediscussed by Smithand Andersen[ 1997]. 24 RAY ET AL.: ERRORSPECTRUMFORTHE M2 TIDE tide. Curve 4 is, as expected,an underestimate.The er- Desai,S. D., J. M. Wahr,andY. Chao,Erroranalysis of empirical oceantidemodels estimated fromTopex/Poseidon altimetry, J. ror spectrumis considerably flatterthanthesignalspectrum. Geophys.Res.,102, 25157-25172, 1997. Abovedegree15 or so,the errorexceeds10% of the signal. B. D. Cornuelle, Becausecurve 5 everywherefalls below curves1-4 we Dushaw,B. D., G. D. Egbert,P. F. Worcester, B. M. Howe,andK. Metzger,A Topex/Poseidon globaltidal concludethat errorscausedby neglectingthe earth'sflatmodel(TPXO.2)andbarotropic tidalcurrents determined from teningare smallerthan the errorsinducedby incomplete long-range acoustic transmissions, Prog.Oceanogr., 40, 337367, 1997. knowledgeof the tidal elevations •. Until furthersignifiEgbert, G. D., Tidaldatainversion:interpolation andinference, cantimprovements in ( areforthcoming, thespherical earth Prog. Oceanogr.,40, 53-80, 1997. approximationappearsjustified (althoughthe "serrated"elEgbert, G. D., A. F.Bennett, andM. G. G. Foreman, Topex/Poseilipsoidapproximationis nearlyas simpleto implementas dontidesestimated usinga globalinverse method, J. Geophys. the sphericalapproximation).Similarly,errorsinducedby Res., 99, 24821-24852, 1994. W. A. andH. Moritz,PhysicalGeodesy, Freeman and usinga constantseawaterdensityin (2) can be ignoredif Heiskanen, Co., San Francisco, 1967. one adoptsa sufficientlyrepresentative densitynear 1035 Jekeli, C., The downwardcontinuationto the Earth's surfaceof kgm-3;a density of 1025kgm-3 induces errors exceeding curves 1 and 2 when n < 5. truncated spherical andellipsoidal harmonic series ofthegravity andheightanomalies, Rep.323, Dept. Geodetic Science, Ohio In the contextof the upcomingGRACEgravityfield mis- State Univ., 1981. C., M. L. Genco,F. Lyard,P.Vincent,andP.Canceil, sion,weobserve thattheexpected GRACE sensitivity curve Le Provost, Spectroscopy of theworldocean tidesfroma finiteelement hyin Figure2 crosses abovethetidalsignalcurveonlyat dedrodynamic model,J. Geophys. Res.,99, 24777-24797,1994. grees above 56.Thisimplies thatocean-tide forcemodeling LeProvost, C., A. F.Bennett, andD. E. Cartwright, Oceantidesfor onthesatellite mustemploymuchhigherdegree spherical andfromTopex/Poseidon, Science, 267,639-642, 1995. harmonic expansions thanis traditionally donein satellite Levitus,S., R. Burgett,andT. Boyer,WormOceanAtlas1994, U.S. Dept.Commerce, Washington, 1994. Ray, R. D., A global ocean tide model fromTopex/Poseidon alMoreimportantly, weobserve thatpresent-day tidemodel timetry: GOT99.2, NASATech.Memo.209478,58pp.,1999. errorsexceed theexpected GRACEsensitivity at all spheri- Shum,C. K., et al.,Accuracy assessments of recentglobalocean calharmonic degrees belowabout40 or50. It is important tidemodels,J. Geophys. Res.,102,25173-25194,1997. geodesy. to acknowledge, however,the temporalmismatch in these Smith, A. J. E. and O. B. Andersen,Errors in recentoceantide curves.TheGRACEsensitivity corresponds to a long-term models:possible originandcause,Prog.Oceanogr., 40, 325(3-month)estimate of thegravityfield;thetidalerrorcurves 336, 1997. Tsaoussi, L. S. andC. J. Koblinsky,An errorcovariance modelfor correspondto mean errors over 12.4 hours. To someextent theeffectof thetideerrorswill bemitigated by temporal seasurface topography andvelocityderivedfromTopex/Poseidonaltimetry,J. Geophys. Res.,99, 24669-24683, 1994. averaging. To what extent is still unclear. The tide errors Wahr,J.,M. Molenaar, andF.Bryan,Timevariability of theEarth's willcertainly aliasintolonger, non-tidal periods (depending gravityfield:hydrological andoceanic effectsandtheirpossible detection usingGRACE,J. Geophys. Res.,103, 30205-30229, onthesatellitesamplingcharacteristics). Whethertheerrors 1998. couldseriously hindertheestimation of thegravityfieldat thelevelof theinstrument precision orhinder thegeophys- R. J. Eanes,CSR, Univ. Texas,3925 W. BrakerLane, Austin, icalinterpretation of thegravityfieldtemporal variations is TX 78759-5321 anobjectof ongoingstudy. G. D. Egbert,COAS, OregonStateUniversity,Corvallis,OR 97331-5503 References N. K. Pavlis, RaytheonITSS, 7701 GreenbeltRd., Greenbelt, MD 20770 R. D. Ray, NASA/GSFC, Code 926, Greenbelt,MD 20771; Andersen, O.B.,P.L. Woodworth, andR.A. Flather, Intercompar-richard.ray@gsfc.nasa.gov isonof recentoceantidemodels, J. Geophys. Res.,100,2526125282, 1995. (ReceivedApril 9, 2000; accepted October4, 2000.)