JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 95, NO. B5, PAGES 6787-6798, MAY 10, 1990 Three-Dimensional Structure of the Upper Mantle Beneath the Atlantic Ocean Inferred From Long-Period Rayleigh Waves 2. Inversion A. MOCQUET1 Laboratoire de Gdophysiq•te Interne, Institut de Gdologie, Rennes, France B. ROMANOWICZ Laboratoire de Sismologie, Institut de Physique du Globe, Paris, France The three-dimensional shear wave velocity structure of the upper mantle beneath the Atlantic Ocean is investigated by inversion of fundamental mode of Rayleigh wave group and phase velocities. The results are reliable down to 300 km depth and improve upon the image of upper mantle structure given by global tomographic models. The inversion of two-dimensional lateral distributions of group and phase velocities, regionalized without a priori constraints, shows that shear wave velocities are correlated with lithospheric age beneath the North Atlantic. Our data set fails to quantify the correlation between shear velocities and lithospheric age, because of the small lateral extent of young regions(<22 Ma) and of strong transversegradients. In contrast with similar studies in the Indian and Pacific oceans,slow velocities associatedwith the location of the Mid-Atlantic Ridge are restricted to the lithosphere. Therefore the Atlantic mid-oceanic ridge behaves as a shallow structure, which is weakly related to deep upwelling asthenospheric flows. At depths greater than 200 km, the central part of the Atlantic is characterized by the highest velocity anomaly. This anomaly can be interpreted in different ways. Part of this anomaly mightcorrespond to a negative (VsH-- Vsv)/Vsv ratioimplicitely included in theisotropic inversion. We also tentatively interpret this anomaly as an important contributor to a shallow degree two pattern of shear wave velocity distribution for the Atlantic area. 1. INTRODUCTION In global tomographicmodels[e.g.,Woodho•seand Dziewonski, 1984; Natal et al., 1986; Taniraoto, 1986], the Atlantic upper mantle appearsrelatively homogeneous.Mocquet et al. [1989](hereafterreferredaspaper 1) haveshown that a spatial resolution of 1500 km, at least, was neededin order to detect fine lateral heterogeneitiesin the Atlantic. In paper 1, 86 paths for group velocity and 114 paths for locities and lithosphericageheld for group velocitiesbut not for phase velocities. However, our data set did not allow us to quantify precisely the correlations between surface wave velocitiesand lithosphericage. In fact, similar studiesin the Pacific [e.g., Forsyth,1977; Nishiraura and Forsyth,1988] have shown that the main contrast between shear wave ve- locitiesoccursfor regionsyoungerand older than 22 Ma. In a slow spreading ocean such as the Atlantic, the former is about 500 km wide, and lateral velocity gradients on each phasevelocity,wereregionalized usingMontagner's[1986] side of the Mid-Atlantic Ridge are high. Therefore paths method of regionalizationwithout geographicala priori con- obliqueor perpendicularto the ridge are hardly sensitive,in straints. Each path was corrected for surficial layer ef- the framework of the first-order optical approximation, to fects before regionalization. Three subregionswith different this heterogeneity. Only paths parallel to the Mid-Atlantic group and phasevelocitydistributionswere defined(Figure Ridge axis manage to determine preciselythe upper mantle 1). North of 35øN (regionA), we foundno correlationbe- structurebeneaththe ridge[e.g., Jacobyand Girardin,1980]. tween group or phasevelocitiesand lithosphericage. South of 35øN (regionB), phasevelocitieswereslightlylowerbeneath young(< 22 Ma) oceanicregionsthan beneathold ocean basins, but this velocity versus age relation was not observedfor group velocities. In contrast, in the South Atlantic (region C), this correlationbetweensurfacewave ve- 1Temporarily at Department of EarthSciences, Nagoya University, Nagoya, Japan. Therefore we will not attempt to construct shear wave velocity versusage models,and we will only mutually compare the three main regionsA, B, and C. After briefly presenting the inversion method, which has already been extensively detailed by Montagner and Joberr [1981] and Natal et al. [1986], we will discussthe resolution at depth, and velocity versus depth profiles will be constructedfor the different large subregionsdefined in paper 1. Such a block regionalization will be useful to compare our results with previous studies, particularly with the body wave analysisof Grand and Helmberger[1984b]in region B. Finally, we will interpret the lateral velocity varia- Copyright 1990 by the American GeophysicalUnion. tions of shear wave velocities and discuss the differences in Paper number 89JB02839. the group and phase velocity distributionsbetween regions 0148-0227]90! 89JB-02839$05.00 B and C. 6787 6788 MOCQUETAND ROMANOWICZ: STRUCTURE OF ATLANTICUPPERMANTLE,2 I ! I 60'N -- 30'N -- 30'5 60'5 -- 90øW 60'W 30'W 0 30'E Fig. 1. The four subregionsin the Atlantic area for which mean velocity models have been calculated[Mocqaetet al., 1989]. 2. METHOD matricesCd and Cp, respectively.The errorson the data are assumedto be uncorrelated so that Cd is diagonal. In In the isotropiccase,shearvelocitiesV$, compressional order to obtain a smoothly varying model with respectto velocitiesVp, and densityp determinesurfacewave veloci- depth,eachmatrix elementCp0is definedusinga correla- ties. The mostinfiuentparameteris V$, and the Rayleigh tion lengthLco, [Nata.[et al., 1986]: wave fundamentalmode doesnot contain enoughinformation to invert simultaneously for those three parameters. Nata.[et al. [1986]broughtforth a priori informationin or- -(hi - h2) 2] = der to retrieve theseparameters simultaneouslyin the transverselyisotropiccase,using the group and phasevelocities of Rayleigh and Love waves. They found that the two best 2Lco,(hl)Lco,(h2) (1) resolved parameters areVs andtheratio(Vs, -Vsv )/Vsv, which definesthe polarization anisotropyof shear wavesin the transverselyisotropic case. Nevertheless,they found a strong trade-off between these two parameters in the shallower upper mantle, down to 160 km. Furthermore, the trade-offbetweendensityand Vsv is higherthan the density resolution,and Vp cannotbe resolvedeither [Nata.[et al., 1986]. A more accuratedeterminationof the density requireshigher modedata [Cara et al., 1984]. In our data set, only the Rayleigh wavefundamentalmode is available. Therefore we restricted our inversion to the distribution wherehi and h2 are depths,andao(h) is the a priori error on V$ at depth h. In this paper, the subscriptzero alwaysrefers to the starting model. The problem is assumedto be linear, so that the relation d=g(p) (2) which links the model p to the data d is written of shear wave velocitiesin the isotropiccase. Doing so, we keep in mind that the match of our velocity maps to the data could be improvedby taking into accountanisotropy [Reganand Anderson,1984]. The inversionat depth of group and phasevelocity distributions is the secondstep of a two-step procedure im- plementedby Montagnerand Joberr[1981], after the algorithm of least squaresinversiondevelopedby Tarantola and with G,j : (Ogi/Opj) (4) where the indice i refersto the velocitymeasurementat Valeire[1982].The first step(paper1) wasthe regionaliza- period Ti and the indicej refersto the investigateddepth. tion without geographicala priori constraintsof individual path measurementsat constant period. The errors on the data d (groupand phasevelocities)and the modelparame- ters p (Vs in our case)are takeninto accountby meansof Defining pe the best estimator of model p, the relation ap MOCQUET AND ROMANOWICZ: STRUCTURE OF ATLANTIC UPPER MANTLE,2 also holds in the linear case, and the inverse problem is solved by TABLE 1. Crustal Parameters of the Starting Model Depth, krn where P0 refers to the starting model, and the resolution matrix R is givenby [Tarantola and Valette,1982] n- %oc*(c + c%oc*)-c 6789 (7) In (6) and (7), the upperscriptT denotesmatrix transpose. The final error on parameters is related to the resolution by [e.g.,Montagnerand Joberr,1981;MontagnerandNatal, 1988] p, g crn -3 gp, gs, krn/s km/s 0.0 1.02 1.45 0.00 5.7 1.02 1.45 0.00 5.7 2.00 1.65 1.00 6.0 2.00 1.65 1.00 6.0 2.60 5.79 3.19 7.4 2.60 5.79 3.19 7.4 2.75 6.00 3.47 7.5 2.75 6.00 3.47 7.5 2.90 6.79 3.89 12.1 2.90 6.79 3.89 12.1 3.38 8.10 4.47 %( hl' h3)-- [•(hl, h2)- R(A1,h2)]' %0(h2,h3) 0 where•(hl, h2) ----1 if h1 ----h2and•(hl, h2) ----0 if not. Equation(8) meansthat if a parameteris unresolved,the a posterJorierror is equal to the a priori error. We computedthe partial derivativesof group and phase velocitiesfor the starting model using Wiggins'[1972] algorithm and usedsimultaneouslythe distributionsof group and phasevelocitiesobtainedin paper I for a periodrange from 50 to 175 s. The inversionwas performedat each 5ø by 5ø cellof the regionalized maps.The a posteriorierrors on group and phase velocitiesafter regionalizationat con- 200 •00 stant period were used as errors on the data. The choice of the a priori error on parameters is important because it definesthe domain of variation allowed for the final V$ model. The amplitude of the velocity anomaliesis greater at shallowdepth. Forinstance,in the Vsv modelof Natal et al. [1986],maximum valuesof 0.6 km/s are reachedlocally at a depth of 50 km. Nevertheless,the maximum values foundin this latter model are more generallyequalor lower than 0.3 km/s. Thus the a priori error on parametershas beenfixedto 0.3 km/s downto the low-velocityzone(LVZ), and to 0.25 km/s below this depth. These values correspond to a maximum relative variation of 6% with respect to M1066A [Gilbertand Dziewonski,1975]Vsv model. 500km 0 The correlationlengthLcor (equation(1)) wasfixed to 100 km down to 88 km depth. Below this depth, Lcor increasedlinearly to reach a value of 200 km at a depth of 400 km. For the period range used in this study, surface waves are weakly sensitive to the upper mantle discontinuities of a spherically symmetric Earth. Therefore we chose km M1066Awith a modifiedoceaniccrust(Table1) asstarting model. This choiceis supportedby the result of Grand and km Helmberger[1984b],who did not observediscontinuitiesin the upper 400 km of the mantle beneaththe Atlantic using Fig. 2. Resolutioncurvesat selecteddepths. The target depths SH body wave data. areindicatedby an arrow;O'p,corresponding a posteriori errorin krn/s (equation(8)). A priorierrorsare 0.3 and0.25 km/s above 3. MEAN SHEAR VELOCITIES UPPER IN THE ATLANTIC and below 190 km depth, respectively. MANTLE Before describingthe results and discussingthem, it is necessary to investigatethe resolutionat depth. In Figure 2, we plotted for different depths typical resolutioncurvesobtained using equation(7). These curvesdo not vary significantly from one geographicalpoint to another. Down to 300 km, the maximum of the curves coincides with the target depth, except at 190 km depth. Below300 km depth, the peaks widen and the resolution is poor. Therefore we will only be interested in the results above 300 km. 6790 MOCQUET AND ROMANOWICZ:STRUCTUREOF ATLANTIC UPPER MANTLE,2 The resolution kernels are 100 km wide at midheight, at a depth of 150 km. This width enlargesto 200 km at a I I I I I i [ I ] ! i I i i i depth of 300 kin. These values are close to those obtained by Weldnet[1974]in the Atlanticand by Nata! et al. [1986] on a global scale. They are better by an order of magnitude than thoseobtainedby Tanimoto[1986].Furthermore, in this latter study, the target depth and the peak of the resolution kernel coincidedonly at a depth of 200 kin; in lOO the work of Nata! et al. [198½;], the coincidence wasreached at depthsgreaterthan 150 kin. Thereforeour study will help to detail the shallow structure of the upper mantle. At 190 km (Figure 2), the target depth and the peak of the resolution kernel are separated by an amount of 30 km. This discrepancymight be associatedwith the presenceof the LVZ in this depth range, and with the high-velocitygradient it involvesvertically. If such gradients are too sharp to be modeled by the inversion method used, our continous approachwill tend to smooththem. The differenceof 30 km betweenthe target depth and the resolutionpeak can hence be consideredas the uncertainty upon the location at depth of the LVZ 200 eA OB OC ß CAR in our model. We will now present mean velocity models for a tectonic province: the Caribbean, and three oceanic regions 300 definedin paper 1 (Figure 1). Continentaland tectonicareas bordering the Atlantic ocean were unavoidable in our i study(paper 1). Nevertheless,apart from the Caribbean where the path coveragewas good, they were not sufficiently resolved to attempt a modelization of their mean shear velocity structure. Specific continental studies for shear waves are, among others, those of Hadio•tche and Joberr[1988]for Africa, Panza et ai. [1980]for Europe, Osagie [1986] for South America, and Cara [1979], Grand and Helmberger[1984a],and Grand [1987]for North America. Shear velocity versus depth profiles for each of the four selectedregionsare shownin Figure 3, and the corresponding values and standard deviationsof the mean are listed in Table 2. Above the LVZ (down to 150 km) shear velocitiesare identical in the North (regionsA and B) and i 4.2 i 4.4 4.6 4.8 Vs , KM.S-I Fig. 3. Mean V$ profilesbeneaththe fourlargesubregions defined in Figure 1. Error bars are standard deviations of the mean. All values are listed in Table 2. Solid curve, Weldher's [1974] model fortheAtlantic normal basins; dashed curve, V$H Atlantic modelof Grand and Helrnberõer[1984b]. depth, Grand's model for the Caribbean becomes simi- lar to model ATL [Grand and Helmberger,1984b]. Similarly, in the 200-300 km depth range (Figure 3), shear veSouthAtlantic (regionC), closeto Weldher's[19741model, locities for regions CAR. are in good agreementwith this whereasthe Caribbean (CAR) shear velocitiesare signif- model,correctedfor attenuationeffects[Kanamoriand Anicantly slower, of the order of 0.2 km/s. Below 250 km, derson,1977]. our V$ valuesare higherthan Weldher's[1974]model. We In contrast, though model ATL correspondsto a sampling are more confident in our models than in Weidner's below from regionB, the two modelsdiffer significantly,exceptat a that depth, becauselonger periods have been used in our depthof 200 km (Figure 3). Grandand Helmberger [1984b] study. Below the LVZ, Caribbean shear velocities become similar to those of regions A and C. The similarity of the Caribbean seismicstructure with those of regionsA and C below 150 km depth indicates that the tectonic character of the Caribbean disappearsbelow that depth. Grand [1987] constructeda V$H tomographicmodelof the upper and lower mantle beneath North America, the northwestern At- constrained the upper 200 km of the mantle by adjusting the thicknessand the velocity of an homogeneous lid to the travel times and waveformsof the SH motion at regional distances(10ø-16ø). A 100-km-thickhigh-velocitylid gave the best fit to their SH body wave data. Our approachis different becausewe use Rayleigh waves and we assumea smoothly varying velocity with depth. The uncertainty on model ATL is estimated at 0.04 km]s [Grand and Helmberger,1984b]. Using the a priori errors previouslymen- lantic and the Caribbean, using S and SS body waves. In his model [Grand, 1987, Figure 17], the Caribbeandisplays tioned, the a posteriorierror on our model(equations(7) lowvelocities (-3%) downto the 140-235km depthrange. and (8)) increases regularlyfrom 0.06 km/s abovethe LVZ In fact, studiesof the upper mantle using body wavesand surface waves are complementary. The resolvingpower of surfacewavesis best above200 km for the period range used in our study, whereasSH body waveshavemore resolution at depths greater than 200 kin. Taking into account the respectivevertical resolvingpowersof body wavesand surface waves down to 300 kin, we consider that Grand's model and ours are mutually consistent. Between 235 and 320 km to 0.2 km/s at 318 km depth (Figure 2). Thereforethe difference between model ATL and our model above the LVZ is too large to be entirely explained by the different techniques used, or by the respectiveerrors of the models. At greater depth, model ATL lies at the edge of the domain of variation allowed for region B. Polarization anisotropy is implicitely included in the isotropic models, and we can tentatively explain their differencesin terms of anisotropy. MOCQUET AND ROMANOWICZ:STRUCTUREOF ATLANTIC UPPER MANTLE,2 6791 TABLE 2. Mean Shear Wave Velocities in the Four RegionsDefined in Figure 1 A B C CAR 45 4.59 0.11 4.56 0.19 4.62 0.07 4.45 6:2 4.56 0.09 4.53 0.16 4.58 0.07 4.38 0.03 0.03 88 4.49 0.10 4.48 0.]4 4.5:2 0.07 4.:29 0.0:2 113 4.41 0.11 4.40 0.10 4.43 0.07 4.27 0.01 139 4.34 0.09 4.34 0.08 4.34 0.08 4.27 0.01 165 4.32 0.07 4.36 0.09 4.31 0.07 4.31 0.01 190 4.35 0.04 4.43 0.07 4.34 0.04 4.37 0.01 216 4.39 0.03 4.52 0.08 4.38 0.04 4.41 0.01 242 4.45 0.02 4.63 0.10 4.44 0.06 4.46 0.01 267 4.52 0.04 4.72 0.12 4.50 0.08 4.50 0.01 293 4.60 0.07 4.79 0.14 4.57 0.13 4.55 0.01 318 4.66 0.07 4.83 0.11 4.65 0.08 4.61 0.01 Velocitiesin km/s. The discrepancybetweenour modelfor regionB and model At depths greater than 100 km, regionsB and C display ATL wouldcorrespond to a (Vsn-Vsv )/Vsv ratioofabout different lateral shear wave velocity variations. In region B, +5% abovethe LVZ and of-5% below,valueswhich are consistentwith the result of Nard! et al. [1986,Figure 30] in the North Atlantic. Nevertheless, the comparison between two isotropicmodelsonly providesan upper bound on polar- the best visualcorrelationbetweenlithosphericage and Vs is observedat 139 km and disappearsdeeper. In region C, such a correlation is not observed but, in contrast, the ve- locity anomaliesstrike along an east-westdirection. In paization anisotropy[Reganand Anderson,1984; Montagner per 1, we presentedthe azimuthal path coverageof our data and Anderson,1989]. Hencethe inclusionof Love wavesin set and showed that the east-west direction is not overs amthe inversionis neededto test theseproposedhigh valuesof pled in region C. Such large east-westtrending anomalies polarization anisotropy. This will increase the difficulty of are in good agreementwith the observationsof Honda and the inversion because, as discussedin paper 1, continental Tanimoto[1987]in this depth range. Below 200 km depth, the most striking feature is a strong regionsare unavoidablein such a study. Continental marginsare characterizedby strongtransversevelocitygradients high-velocity anomaly beneath the central part of the Atwhich induce surfacewave scatteringor, equivalently,mode lantic, which mainly contributesto the high mean veloccoupling[e.g., $nieder, 1987]. Thereforethe joint inversion ity of regionB (Figure 3). In the Pacific [Montagnerand of Love and Rayleigh waves will require the use of newly Joberr,1981; Wielandtand Knopoff,1982]andIndian oceans developedformalisms [$nieder and Romanowicz,1988; Ro- [Montagner and Joberr, 1988], slow shear wave velocities rnanowiczand $nieder, 1988] which describe,for instance, associatedwith mid-oceanic ridges can be followed down to depths greater than 350 km. These deep slow velochigher-modecontaminationby short-periodLove waves. 4. LATERAL VARIATIONS OF SHEAR VELOCITIES The regionalizedgroup and phase velocitiesobtained in paper I were inverted at depth in each 5ø by 5ø cell, using the method describedin section2. The resultinglateral variations of shear wave velocities in the Atlantic area are presentedat selecteddepths in Figure 4. Montagner and Joberr[1988] showedthat one of the advantagesof Montagher's[1986]method is that the a posterioritrade-offbetween parameters vanishesif a suitable correlation length is used in the regionalization. For instance,we found no trade-off between the North and South Atlantic, using a correlation length of 1500 km (paper 1). The use of this correlation length is equivalentto a spatialfiltering and determinesthe smoothnessof the final model. On the other hand, assuming smooth lateral velocity variations, together with the firstorder optical approximation,doesnot permit to modelsharp lateral velocity gradients. These are the reasonswhy, at a depth of 88 km (Figure 4), the Mid-Atlantic Ridge only displays two local slow velocity anomalies,in regionsB and C. The absenceof slow velocities in other regions along the ridge (in regionA, for instance)doesnot mean that they do not exist but only that our data set and the method of regionalizationusedare not adequateto model them. ity anomaliesare usually interpreted in terms of large-scale upwelling asthenosphericflows beneath mid-oceanicridges. In contrast, in our model, the Mid-Atlantic Ridge displays slow velocities only in the lithospheric depth range. Below the LVZ, the distribution of negative anomaliescannot straightforwardlybe interpreted in terms of asthenospheric upwelling flows beneath the Mid-Atlantic Ridge. If such flows exist, their lateral extent is too small to be detected by our data set and our method of investigation, so that their influence on the distribution of shear wave velocities is maskedby the global asthenosphericstructure. In M84C [Woodho•seand Dziewonski,1984], the Atlantic area is a major contributorto the high-velocityanomalyof the global degree two pattern of shear wave velocity distribution, for depthsgreater than 450 km. Romanowiczet ai. [1987] suggested several depths for the global degree two pattern of shear wave velocity distribution, one of them concentrated in the depth range 300-500 km. In our study, the highestvelocity anomaly, in the depth range 200-300 km, might correspondto the top of this latter heterogeneity. Test o! the Model The final Vsv modelwastestedby calculatingthe correspondinggroup and phasevelocitydistributions.The misfit 6792 MOCQUET ANDP•OMANOWICZ: STRUCTURE OFATLANTIC UPPER MANTI,E,2 60 ø W 30øE .... ::'•:' ";:i'-- .:.::½':-•i!::.; • '•-•}!:':":",:::.'27•.-! .......... ': .' ' ....................... ..... :............ .:....::..:•:.:•:::. --•:.:-::::•.--}• .•..:?.:• .......•.•.•:. ..• ................ :::--:.,. :... •,-............. ::•,•:.• •... • ;•.:.....-.: ...... .:..... ..:.•:..?. :..,:.;::•.-.::...•:...;•. ?..-..:--:::•::.:.•:--.•::-•.:...... ... ..... . .--%::: .'..":..========================================= ••. ':: • .:':.:;:'"-'":'•:..... :;;:::...:.:• . '•:': ................... •.: ::-....:::: •• •: . :. -.::• .. . •- :.... ....:::;.......... ?.;•..• ....ß :..:.• , , ::': .•: •:•. i•:•;:•;:•.: .......... .... :.•:?'"'• ]•.................. .::.::.;:::.::;::.?•: %:. ...... :......... :•::::•";.' ß . •' •' '":: •,'•.- ß• "' ..... :•--•i. ...... :?:......:....::......:.:-:: ..::•.-..• ß.......... :•;•:::•:.:.•;•; •"• • .... . ':,f•;.',..:.:•':•*:' "'":'::: •:.•;•;:•?:•::•:;':•':-•:•:•::2•.: ........ :.."::::'"':4::::::::: :' ::' ß '":•'.•:":•::::.•:•:.½:.• ................................. ::•:."::•.,• l•t ..... •;... .............:?.:i•: ...... :::•m:.-•.•::.:::•.: ..•.::::•:.:;:;'•.;• .. • • ..... . t:.:-:• :::.•:.:?-•:.:....:::•.I-' • ---<.:.W- • : •?•::....::..:? i•, . '" '{• . c<.:..%..:. •:. :•'•< .:•. ......::.... •...:-•i:::•::• •-• .•-- ":..;:::........... ':::•: ........... .:... ß.•::,• ...•' , -...... ....... 242Km 139"K rn......................... . 'VSv Fig.4. Lateralvariations ofshearwavevelocities at selected depths. Percentages areanomalous velocities withrespect to M1066A[GilbertandDziewonski, 1975]. between synthetized andobserved velocities should attempt wavevelocityanomalyat $$ km depth beneaththe Brazilto matchthe observational error. In Figures5 and 6, the ian shield. As pointedout in paper 1, the lack of data consynthetized velocitydistributions arecompared with the re- cerning the thicknessof the crust made accurate surficial gionalized mapsobtainedin paper1. Figures7 and8 show layercorrections impossible in this region,anda thickconthe relation between data residuals and observational er- tinentalcrustmightaccount for thislow-velocity anomaly. rorsfor phaseand groupvelocities,respectively.In what Therefore,evenin the ideal caseof a perfectfit between follows,we will only discussthe resultsbeneath oceanicar- synthetized and observed groupand phasevelocities, this eas, because of the lack of resolution in continental areas. anomalyhas no meaningconcerning the structureof the For instance,our modeldisplaysa strongnegativeshear- upper mantle. 73S 102S lllllllllll!11111111111111111 !I • / • Ji i i • iIIIII1111111-• i • • • • • -.' i :: • • •1111• 171S • • - - ::111111111111111! ! :• SYNTHETIZED REGIoNALIZED I [ J J•l I ! :: •-•.11111111111HIIl[:-":'...'.:.::• Fig. 5. Comparison between synthetized andobserved phase velocities at threeselected periods. Anomalous velocities arein percentwith respectto M1066A[GilbertandDziewonski, 1975]. 70S 100S 175S SYNTHETIZED REGIoNALIZED Fig. 6. Same as Figure 5 for group velocities. 73S i 171S _ lli !•"g _ .... crD Fig. 7. Comparison between a posterJori phase velodtyresiduals andobservational erroreYD.Thescale for eY D is in percent withrespect to M10OOA [G//Mr/auclDzieumuski, 1975]. 70S lOOS 175s 13 RES Fig. 8. Same as Figure 7 for group velocities. MOCQUET AND ROMANOWICZ: STRUCTURE OF ATLANTIC UPPER MANTLE,2 Our model tends to overestimate the relation between group and phasevelocitiesand lithosphericage (Figures 5 and 6). In the North Atlantic, the increaseof short-period (T=73 s) phasevelocitiestoward old oceanbasinsis synthetizedwith an accuracyof 2% (Figure5). The synthetized groupvelocitiesalsodisplaysucha relation with lithospheric age, though it was not observedin the regionalized maps (Figure 6). This remark alsoholdsat intermediateperiod (T=100 s). In paper 1, we observedin the North Atlantic a velocity-age correlation for phase velocities but not for group velocities. In the South Atlantic, this correlationwas observedfor group velocitiesbut not for phasevelocities. In this latter region, the lack of denselydistributed refraction 6797 errors and data residuals, together with the associatedun- certainty on the location at depth of the LVZ (Figure 2), indicate that the structure of the LVZ must be more com- plicated. Furthermore, all these studies,including ours, deal with isotropic models, and it has been recognizedthat the inclusion of anisotropy in the inversion at depth of surface wave data substantially reduces the importance of the LVZ [e.g., Dziewonski and Anderson, 1981; Regan and Anderson, 1984]. 5. CONCLUSION Considering the previously discussedlimitations of our model, the following conclusionscan be drawn: (1) the profiles had led us to fix crustal parameters to standard oceanic values. Therefore the correlation between group velocities and lithospheric age might still be explained by surficial layer effects, such as lateral variations in crustal slower velocities beneath the Mid-Atlantic Ridge than beneath old ocean basins. The largest velocity contrast is 6% velocity data appear to be too highly sensitive,in a nonlin- deep slow-velocity anomalies below the location of the mid- North Atlantic lithosphere displays classical structures: with an uncertaintyof 2%. (2) In contrastwith otheroceans, thickness.More generally,short-period(T < 100 s) group the Atlantic asthenosphericstructure is not characterizedby ear way (paper 1), to surficiallayer effectsto provideuseful oceanicridge, but the central regionof the Atlantic displays information concerning the structure of the upper mantle. a high-velocity(+6% 4- 3%) anomaly in the depth range In contrast, the better fit between synthetized and observed phase velocities in the North Atlantic permits interpretation of the velocity-age correlation in this region in terms of lithosphericcooling and plate thickeningwith age [e.g., Nishim•tra and Forsyth,1988]. The amplitude of the velocity anomaliesand the associatedtransversegradients will have to be better constrained by incorporating amplitude data in the inversion. The broadening of the resolution curves below 200 km depth (Figure 2) implies a greater uncertaintyon the location at depth of the V$ v anomalieswith increasingdepth. This uncertainty results in a shift in period between synthetized and observed phase velocity distributions for peri- 200-300 kin. Part of this anomaly might be explained by a negative polarization anisotropy implicitely included in the isotropic model. We also tentatively interpret this anomaly as a contributor to a shallow global degree two pattern of shear wave velocity distribution for the Atlantic area. Acknowledgements. This work was conductedunder grant GEOSCOPE ASP 1987, sponsored by INSU of CNRS. We are thankful to J.P. Montagner, who provided us with his programs. His help at the early stage of this work was greatly appreciated. Constructive reviews from the editors, M. Cara, and an anonymous reviewer have been helpful to improve this paper. This is IPG contribution nø 1081. odsgreater than 100 s (Figure 5). This is particularly clear for the Central Atlantic high-velocityanomaly,whoselargest amplitude(6%) was observedover a large area at a period of 102 s, and which is better synthetized at a period of 171 s. Therefore,as can alsobe seenin Figure 2 (h > 190 km), our model only providesan averagedestimate of the anomaliesfor depths greater than 200 kin. As statedpreviously,the misfit a! betweensynthetized REFERENCES Cara, M., Lateral variations of S velocity in the upper mantle kom higher Rayleigh modes, Geophys. J. R. Astron. Soc., 57, 649-670, 1979. Cara, M., J.J. L•v•que, and V. Maupin, Density-versus-depth models from multimode surface waves, Geophys. Res. Left., 11, 633-636, 1984. and observedgroup and phasevelocity distributions should attempt to match the observational error a,t. This is Dziewonski, A.M., and D.L. Anderson, Preliminary reference Earth model, Phys. Earth Planet. Inter., •5, 297-356, 1981. achieved for all periods in the western part of the North Forsyth, D.W., The evolution of the upper mantle beneath midAtlantic, becausethis region displayedthe best path cover- age (seeFigure 1 in paper 1). The misfit increasebeneath young regionsof the North Atlantic reflectsour difficulty to model velocity versuslithospheric age relations. The small erroron phasevelocitydistributions (ad < 1%, Figure7) appears to be underestimated. An error of 2% is more realistic in view of the group velocity distributions, for which a! canbe smallerthanad (Figure8). The largestdifferencebetweena! and ad is obtainedfor intermediateperiodgroupand phasevelocities(T - 100 s) which are mostly sensitiveto the V$ structure of the LVZ ocean ridges, Tectonophysics,38 (1-2), 89-118, 1977. Gilbert, F., and A.M. Dziewonski, An application of normal mode theory to the retrievM of structural parameters and source mechanisms from seismic spectra, Philos. Trans. R. Soc. London Set. A, 278, 187-269, 1975. Grand, S.P., Tomographic inversion for shear velocity beneath the North American plate, J. Geophys. Res., 92, 14,065-14,090, 1987. Grand, S.P., and D.V. Helmberger, Upper mantle shear structure of North America, Geophys. J. R. A stroh. Soc., 76, 399-438, 1984a. (see Figure 13 in paper 1). In this depth range (Figure 4), shearwave velocitiesare homogeneously distributed Grand, S.P., and D.V. Helmberger, Upper mantle shear structure beneath the northwest Atlantic Ocean, J. Geophys. Res., 89, throughout the Atlantic Ocean. Similarly, we noted previ- ouslythat the Vsv modelsfor regionsA, B, C, and CAR 11,465-11,475, 1984b. coincidedwithin a pronouncedLVZ (Figure 3). Though Hadiouche, O., and N. Joberr, Geographical distribution of this result is consistent with previous works in this region [e.g., Weidner, 1974], the large misfit betweenobservational surface-wave velocities and 3-D upper mantle structure in Africa, Oeophys. J., 95, 87-110, 1988. 6798 MOCQUET ANDROMANOWICZ: STRUCTURE OFATLANTICUPPERMANTLE,2 Honda, S., and T. Tanimoto, Regional three-D heterogeneities Romanowicz, B., and R. Snieder, A new formalism for the effect by the wave form inversion-Application to the Atlantic area, of lateral heterogeneity on normal modes and surface waves, Geophys. J. R. Astron. $oc., 91, 737-753, 1987. 2, General anisotropic perturbation, Geophys. J., 93, 91-99, 1988. Jacoby, W.R., and N. Girardin, The evolution of the lithosphere at the southeast flank of Reykjanes ridge from surface wave Romanowicz,B., G. Roult, and T. Kohl, The upper mantle dedata, J. Geophys., 47, 271-277, 1980. Kanamori, H., and D.L. Anderson, Importance of physical dispersionin surfacewave and free oscillationproblems: Review, Rev. Geophys., 15, 105-112, 1977. Mocquet, A., B. Romanowicz, and J.P. Montagner, Threedimensional structure of the upper mantle beneath the Atlantic Oceaninferred from long-periodRayleigh waves,1, Group and phase velocity distributions, J. Geophys.Res., 94, 7449-7468, 1989. Montagner,J.P., Regionalthree dimensionalstructuresusinglong period surface waves, Ann. Geophys., ,l, 283-294, 1986. Montagner, J.P., and D.L. Anderson,Petrologicalconstraintson seimic anisotropy, Phys. Earth Planet. Inter., 54, 82-105, 1989. Montagner, J.P., and N. Joberr, Investigation of upper mantle structure under young regions of the southeastPacific using long-period Rayleigh waves, Phys. Earth Planet. Inter., 27, 206-222, 1981. gree two pattern: Constraints from GEOSCOPE fundamen- tal spheroidal mode eigenfrequencyand attenuation measurements, Geophys. Res. Left., 14, 1219-1222, 1987. Snieder,R., Surfacewave scatteringtheory: With applicationto forward and inverseproblemsin seismology,Ph.D. thesis,Univ. of Utrecht, Netherlands, 1987. Snieder, R., and B. Romanowicz, A new formalism for the effect of lateral heterogeneityon normal modesand surfacewaves,1, Isotropicperturbations,perturbationsof interfacesand gravitational perturbations, Geophys. J., 92, 207-222, 1988. Tanimoto, T., The backus-Gilbert approach to the three-D structure in the upper mantle, II, SH and SV velocity, Geophys. J. R. Astron. Soc., 84, 49-69, 1986. Tarantola, A., and B. Valette, Generalized non linear inverse problems solved using the least squares criterion, Rev. Geophys., 20, 219-232, 1982. Weidner, D.J., Rayleigh wave phase velocities in the Atlantic Ocean, Geophys. J. R. Astron. Soc., 36, 105-119, 1974. Montagner, J.P., and N. Joberr, Vectorial tomography,II, Appli- Wielandt, E., and L. Knopoff, Dispersionof very-longperiod P•yleigh waves along the East Pacific Rise: Evidence for S cation to the Indian Ocean, Geophys. J., 9,1,309-344, 1988. wave velocity anomaliesto 450 km depth, J. Geophys. Res., Montagner, J.P., and H.C. Nataf, Vectorial tomography,I, The87, 8631-8641, 1982. ory, Geophys. J., 94, 295-308, 1988. Nataf, H.C., I. Nakanishi, and D.L. Anderson, Measurementsof Wiggins, R.A.,The general linear inverse problem: Implication mantle wavevelocitiesand inversionfor lateral heterogeneities, of surfacewavesand free oscillationsfor Earth structure, Rev. Geophys., 10, 251-285, 1972. 3, Inversion, J. Geophys.Res., 91, 7261-7307, 1986. Woodhouse, J.H., and A.M. Dziewonski,Mapping the upper manNishimura, C.E., and D.W. Forsyth, Rayleigh wave phaseveloctle: Three-dimensional modeling of Earth structure by inverities in the Pacific with implicationsfor azimuthal anisotropy sion of seismicwaveforms,J. Geophys. Res., 89, 5953-5986, and lateral heterogeneities, Geophys. J., 94, 479-501, 1988. 1984. Osagie, E.O., Anelasticity of the crust and upper mantle of South America from the inversion of observed surface wave attenua- tion, Geophys. J. R. Astron. Soc., 86, 1-17, 1986. Panza, G.F., St. Mueller, and G. Calcagnile,The grossfeatures of the lithosphere-asthenosphere system in Europe from seismic surfacewavesand body waves,P•re Appl. Geophys.,118, 1209-1213, 1980. Regan, J., and D.L. Anderson, Anisotropicmodelsof the upper mantle, Phys. Earth Planet. Inter., 35, 227-263, 1984. A. Mocquet, Department of Earth Sciences,NagoyaUniversity, Chikusa, Nagoya 464, Japan. B. Romanowicz,Laboratoire de Sismologie,Institut de Physique du Globe, 4 place Jussieu, 75252 Paris, Cedex 05, France. (Received December 15, 1988; revised August 22, 1989; accepted September 6, 1989.)