The Deep Structure of Continents

VOL. 84, NO. B I3 JOURNAL OF GEOPHYSICAL RESEARCH DECEMBER 10, 1979

DON L. ANDERSON

SeismologicalLaboratory, California Institute of Technology,Pasadena, California 91125

The Lehmann discontinuityat 220-km depth is an important global feature which occursunder both oceansand continents.It is a barrier to penetrationby younglithosphere and marks the baseof seismicity in regionsof continent-continentcollision. The stronglateral variationin uppermantle velocitiesoc- cursmainly abovethis depth. Continentalroots extend no deeperthan about 150-200 km. The basalt- eclogitetransformation and eclogite-harzburgiteseparation may be responsiblefor the geometryof inter- mediatedepth earthquakes.Oceanic and continentalgeotherms converge above about 200 km and be- comeless steep than the meltinggradient at greaterdepth. This impliesa low viscositychannel near 250 km. This would give a decouplingzone of maximumshear beneath continental shields. The Lehmann discontinuitymay be the interfacebetween two distinctgeochemical reservoirs. The velocityjump, and the inferreddensity jump, at 220 km are consistentwith an increasein garnetcontent. The mantle may be garnetlherzolite above and eclogiteimmediately below the Lehmanndiscontinuity. The transitionre- gion may be mainly eclogiteand be the sourceregion for oceanictholeiites.

INTRODUCTION nental values[Jordan and Anderson,1974; Jordan, 1975].Hart et al. [1977] determined an attenuation corrected free oscilla- Continental lithosphereis much thicker than oceanic lithosphere,but the question of how thick a sectionof continent tion averageearth model, QM2, which is comparedwith the translatescoherently during continental drift has not yet been continentalmodel SHR14 of Helmbergerand Engen[1974] in adequatelyaddressed. The bottom of the low-velocity zone is Figure 2. A more direct comparisonis the Pacific-easternU.S. usuallyconsidered to be the bottom of the asthenosphereand curve which is from Cara's [1979] surfacewave study. In this casethe major differencesare above 250 km. it has been presumedthat coherent translation of both oceanic and continental plates takes place above some 200 km. This Okal and Anderson[1975] and Sipkin and Jordan [1976] basicassumption of plate tectonicsis contrary to the idea that studiedthe corereflected shear phase, ScS, with conflictingin- terpretations. Okal and Anderson concluded that all differ- ontinents have roots deeper than 400-500 km [MacDonald, 963; Jordan, 1975, 1979a, b]. ences could be explained in terms of known effects above Surfacewave studieshave shownthat there are large differ- --,180 km, while Sipkin and Jordan concludedthat differences races between oceans and shields above about 220 km [Dor- in velocity must persistto great depths,perhaps extending nan et al., 1960; Brune and Dorman, 1963; Anderson, 1967a; throughout the entire upper mantle. Okal [1977] concluded Kanamori, 1970; Dziewonski, 1971]. Recently, Cara [1979] has that surfacewave data, regionalizedto take accountof the age made a detailed study of regional differencesusing higher of the oceaniclithosphere, are incompatiblewith strong,deep mode surfacewaves. He found strong regional variations be- lateralinhomogeneity anddo notrequire any substantial structure variation below 240 km. tween the Pacific Ocean, western U.S., and eastern U.S. above 250 km and no resolvable difference below this depth. Eng- The ScS phase,of course,averages the velocity throughout the mantle and cannot resolve where the differences occur. land et al. [1978] made a direct comparisonof upper-mantle structure under the North Atlantic and Arctic oceans and the There are more direct ways to isolatethe effect.The velocity old shield of the Russianplatform. Even when maximum dif- structureof the upper mantle has been studiedin many re- ferencesbetween the regions were allowed the P wave data gionsby body wave and surface wave techniques. These stud- could be satisfiedby velocity modelswhich were substantially ies give remarkablyconsistent results when the averagetravel the samebelow 300 km. Cara's modelsare shownin Figure 1. times above 200 or 250 km are calculated.Table 1 presents these results. Shields are the fastest, about 3.5 s faster than Okal and Anderson [1975] used multiple ScS phases to samplethe earth under variousgeological provinces including youngocean. Oceanic models, on the average,are about 1.5-2 oceans and shields. They' concluded that the observations s slowerthan shields,but the differencedecreases with tl•e age of the ocean. were consistent with known differences above about 180 km. We have computed the differences between the $ wave There is therefore good agreement between the body wave and surfacewave data. Jordan [1975], however, proposedthat traveltimes above 200 km for the Rayleigh•ave models and ocean-continentdifferences extend deeper than 400 km and the Jeffreys-Bullenvalues. Figure 3 displaysthe computed ScS residuals for these oceanic and shield models as a func- that the region which translatescoherently in the course of plate tectonicsmay occupy the entire upper 700 km of the tion of crustal age. Measured residualsrelative to Jeffreys- Bullen, are alsoshown. It is clear that the ScStunes can be ex- mantle. This proposalhas reopenedthe questionof the deep structure of continents. plainedby known differences in the upper 200•250 km of the mantle. THE DEEP STRUCTURE OF CONTINENTS Theaverage one-way ScSresidual f•r 'ayerage age'ocean Prior to the recognitionthat attenuationwas important in (70-90 m.y.) is +0.8 s. This can be comparedwith the average interpreting free oscillationperiods, it was thought that aver- residualof +2.0 s for all oceanicdata combined,including the age earth shearvelocities were appreciablyslower than conti- very slow young oceans[Sipkin and Jordan, 1976].This latter value is a straightaverage of all data and ignoresthe variation Copyright¸ 1979 by the Americ.•nGeophysical Union. with age. Paper number 9B0991. 7555 0148-0227/79/009B-0991501.00 7 5 56 ANDERSON:DEEP STRUCTUREOF CONTINENTS

Vs, km/sec TABLE 1. Upper Mantle Vertical Shear Wave Travel Times Above 4.0 4.5 5.0 200 and 250 km 0 Travel Time, s

200 km 250 km Reference ioo Shield 45.2 _+0.3 56.1 +_0.3 (1) • J/'%"EastU.S.- Continent 45.9 +_0.5 57.0 +_0.2 (2) 200 Ocean 46.7 _+0.4 57.7 +_0.3 (3) Minimum 49.2 60.6 (4) 15 m.y. 48.7 59.8 (5)* 70 m.y. 47.1 58.1 (5)* •00 100 m.y. 46.4 57.5 (5)* 150 m.y. 45.8 56.8 (5)* Maximum 46.0 57.2 (4) 400 • D• (1) •4nderson[1967a], •4ndersonand Harkrider [1968], Brune and Fig. 1. Shear velocity versusdepth for Pacific (average age, 90 Dorman [1963], }Vickens[1971], and Massb[1973]. m.y.) and easternU.S. from Cara [1979]. CANSD is Canadian shield (2) Helmbergerand Engen [1974],•4nderson and Julian [1969], Cara from Brune and Dorman [1963]. [19791. ' (3) •4nderson[1967a], Kanamori [19701,Schlue and Knopoff[1976], The residualsfor the oldestocean, > 120 m.y., are scattered, and N. R. Burkhard and D. D. Jackson (unpublished manuscript, 1979) (average ocean). but the mean is -0.7 s. This includesanomalously slow read- (4) Yoshida[1978]; minimum and maximum age groups. ings from the mid-Pacific mountains and the Bermuda Rise. (5) N. R. Burkhard and D. D. Jackson(unpublished manuscript, Therefore the one-way difference in travel times between 1979). shields and old oceans is about 1.3 s. The difference between *SV velocitiesfrom Rayleigh waves.Horizontally traveling SH averageocean and old ocean is about the sameas determined waves(Love waves) give travel times 1.6s (15 m.y.) to 0.5 s (150 m.y.) shorter.ScS waves should be comparedwith Rayleighwave, not Love by Duschenesand Solomon[1977] usingshear waves from Pa- wave,velocities. 'Average' ocean (--,70 m.y.) is about2 s slowerthan cific events. shield.If Love wavesare usedin the comparisonoceans would ap- From the raw ScS data the mean ocean-continent differen- pear to be about 3 s slow. tial time is + 1.5 + 2.8 s, about the same as the upper 250 km alone.Correcting for attenuationreduces the differentialtime sured shear wave velocities in the LVZ are 12-15% lower than slightly. The ScS shield data overlap the shield models,which the high-frequencyvelocities for mineralogiesin assemblages differ from the average earth only above 250 km, but average ranging from pyrolite to eclogite [Anderson,1977]. In addition about I s faster (Figure 3). to the temperatureeffect an additional severalpercent varia- From independent data (Table 1) the travel times above tion is allowed by variations in mineralogy. We take 12% as a 200-250 km in shieldsaverage 1.6 _+0.6 s faster than under plausible variation between shield and ocean mantles,take a average oceans and 0.9-2.0 s faster than 70-100 m.y. old travel time of 57 s above 250 km, and assume that velocities oceans.Therefore the ScS data are in good accord with the are the same at •250 km. This gives3.4 s as a conservativees- surfacewave studies,and it appearsthat all differencescan be timate of possible upper mantle shear wave vertical travel accommodated above 250 km. time variations. This would be the difference between a cold, The ScS data, when correctedfor crust and upper mantle garnet-rich upper mantle and a warm, relaxed, garnet-poor effects above 50 km, suggestthat the mantle under shields may be as much as about 3.4 s fasterthan under old oceans. +5 OC'E•NS At this point, it is instructive to estimate the maximum CONTINENTS plausiblevariations in the upper mantle.The shearvelocity in +z the low-velocity zone (LVZ) in oceansand tectonicregions is •,••yle•ghwaves about 10% lower than subcrustal velocities. This can be ac- c•+3 - \ - counted for by a difference in chemistryor by high-temper- --'+2 - ature stress relaxation mechanisms such as dislocation or grain boundary relaxation [Andersonand Minster, 1979}.Mea- a• -i- I - , \ - Sc \

+050 Average Earth OM2 I" -I •ELDS +025_[7 -2 -

-3 I , \! I04 106 i08 109 -025 Age, yeors Fig. 3. ScS residuals (vemcal bars) as functio,• o! age of lithosphere.Data are from Okal and Anderson[1975], Okal [1978a,b], and -O50 Sipkinand Jordan[1976]. Solid line is calculatedfrom surfacewave 0 I00 200 300 400 500 600 modelswhich differ only above 200 km (N. R. Burkhard and D. D. Depth, km Jackson,unpublished manuscript, 1979). Dashed line indicatesrange Fig. 2. Ocean-continentshear velocity differencesversus depth. of continental residuals [Hales and Roberts, 1970; Poupinet, 1979]. Solid line is from Cara [1979]. Dashed line is attenuation corrected 'Average earth' is free oscillationmodel of Hart et al. [1977] corrected averageearth model QM2 [Hart et al., 1977]minus continentalmodel for attenuation.Region marked shieldsis calculatedfrom body wave SHRI4 [Helmbergerand Engen, 1974]. and surface wave shield models. ANDERSON: DEEP STRUCTURE OF CONTINENTS 7 5 5 7

• ' ' I I • • I ' They should not be comparedunless anisotropy is taken into Rayleigh Wave Phase Velocities. (Shield-Ocean) consideration.An apparent transverseisotropy can be the re- sult of fine layering which will not be detecteddirectly if the +0.2 -- / xx 250-400 seismicwavelengths are longer than the layer thicknesses[An- xxx km - derson,1966]. For example, a thin low-rigidity layer under the •300 •x - oceanic lithosphere, possibly due to melt accumulation by E drainage from the upper mantle, will give an apparent trans- G 0 verseisotropy if not allowed for in the modeling.Variations in the thickness of such a layer could also explain the large

-0.1 spreadin residualsof ScSin older oceanbasins.

THE LEHMANN DISCONTINUITY -0.2 IOO 200 300 The major seismicdiscontinuities in the mantle are near Period, sec 400 and 670 km. These have been interpreted as phase Fig. 4. Rayleigh wave differential phasevelocities (shield minus changes,although the deeperone may involvea composition ocean). Data from Kanamori [1970], Brune and Dorman [1963], Okal [1978a, b], and Cara [1979]. Solid lines showthe effectof distributing changeas well [Anderson,1967b, 1968]. A compositionchange a 3.4-s differencein travel time, as implied by the ScS data, between would be an effective barrier to convectionand explain the 50 km and the depth shown.The dashedline showsthe effectof plac- termination of seismicactivity at -•670 kin. The sharpnessof ing a 5-s differencebetween 250 and 400 kin. The ScS and Rayleigh this discontinuity [Adams, 1971; Whitcomb and Anderson, wave data arc compatible if the differenceis above 200 kin. 1970]together with the large increasein velocityargues for a changein chemistryas well as a changein crystal structure. upper mantle. In this casethe suboceanicupper mantle can be There is another important mantle discontinuityat a depth interpreted as residual or depleted material relative to sub- near 220 kin, the base of the low-velocity zone. shield mantle. A discontinuityat 232-km depth was proposedin 1917 by The near-verticalScS data cannot isolate the depth range Galitzin. The most detailed early studiesindicated the pres- responsiblefor the variation. For this purposewe can use the enceof a discontinuityunder North America and Europe near dispersionof Rayleigh waves.Figure 4 givesthe differencein 215-220 km [Lehmann,1959, 1961, 1967]and we shall hence- phasevelocity between oceans and shieldsas a function of pe- forth refer to it as the Lehmann discontinuity.The early work riod. The data are from Brune and Dorman [1963], Kanamori is summarized in Anderson[1966, 1967a] and Knopoff et al. [1970], Okal [1978a, b], and Cara [1979]. The solid curves [1966].Additional evidencehas accumulatedsince these sum- showthe effectof distributinga 3.4-s differencefrom 50 km to maries. the depth shown. It is clear that the dispersiondata are satis- A. L. Hales et al. (unpublishedmanusvript, 1979), Steinmetz fied if the oceanicdelay is placed above 200 kin. Putting a 5-s et al. [1974], Lukk and Nersesov [1965], and Wiggins and delay between250 and 400 km is clearly not acceptable. Helmberger [1973] have all found evidence for a discontinu- The shorterperiod Rayleigh wave data (20--60 s) including ity between 190 and 230 km from body wave data. Cara higher modescan be satisfiedby a 3.2-s delay between50 and [1979] found high velocity gradients near 220 kin. The in- 250 km under oceans compared to the eastern U.S. [Cara, creasein velocity is the order of 3.5-4.5%. Using the seismic 1979]. Cara's eastern U.S. model is similar to the Canadian equation of state [Anderson,1967c], the associateddensity in- shield model of Brune and Dorman [1963] and gives similar crease is about 3%. travel times above 250 kin. Niazi [1969] demonstratedthat the Lehmann discontinuity Jordan [1975] noticed that the differencebetween oceanic in California-Nevada is a strong reflector and found a depth and continental Love wave phase velocities was much less of 227 +_ 22 kin. than would be predicted from the observeddifferences in ScS Sackset al. [1977] and Jordan and Frazer [1975] found con- travel times if the variations were restrictedto shallow depths, verted phasesfrom a discontinuityat a depth of 200-250 km for example,above 400 kin. He tacitly assumedthat horizon- under both the Canadian and Baltic shields. Reflections from tally and verticallytraveling SH waveshad the samevelocity. a similar depth have been reported from P'P' precursors In a transverselyisotropic media thesevelocities are different, [Adams, 1971; Whitcomb, 1973; Whitcomb and Anderson, but vertically traveling SH has the same velocity as horizon- 1970] for Siberia, western Europe, North Atlantic, Atlantic- tally traveling S V, i.e., ScS times should be compared with Indian Rise, Antarctica, and the Ninety-east Ridge. Evidence Rayleigh wave velocities,or with models based on Rayleigh now existsfor the Lehmann discontinuityin easternand west- wave data, as we have done above. The Love wave and ScS ern U.S., Canadian Shield, Baltic Shield, oceanicridges, nor- observationsare consistentif the oceanicupper mantle in the mal ocean,the Hindu Kush, the Alps, and the African rift. vicinity of the low-velocityzone has a shearwave anisotropy The V•,/Vsratio of recentglobal earth models[Hart et al., of about 5%. This is about the same as required to reconcile 1977] reversestrend at 220 kin. This is indicative of a change Love wave and Rayleigh wave data [Anderson,1966; Anderson in composition,phase, or temperaturegradient. andHarkrider, 1962; Schlue and Knopoff, 1976; N. Ri Burk- There are not yet enough seismicdata to map the variabil- hard and D. D. Jackson,unpublished manuscript, 1979].Thus ity in depthof the '220-km'discontinuity. Most reportedre- this seemsto be a reasonablealternative to deep (>200 kin) flectionsoccur at depthsbetween 190 and 230 kin. Part of this continental roots. variation is due to assumptionsabout the mantle velocity There is therefore good consistencybetween the free oscil- above the reflector. lation, surfacewaves and ScS observationswhen the effectsof The geopotentialpower spectrumyields a depth of 200 km anelasticity and ocean age are taken into account. ScS and for a density discontinuity [Lambeck, 1976;Marsh and Marsh, Love wavesdo not averagethe upper mantle in the same way. 1976]. This givesadditional evidencefor the interface and in- 7 5 5 8 ANDERSON: DEEP STRUCTUREOF CONTINENTS

low about this depth [Isacks and Molnar, 1971].Actually, be- 2000- O\,•,c•eFO?•"'"'"'""•/ OI,v,ne//.•.•.tween 200 and 300 km about half the focal mechanisms in- •.••"./•/' ..•WetSohdus dicate down dip compression,and most of the mechanisms below 215 km are compressional.Isacks and Molnar [1971] 1600 ß...... " .... • • •- suggestedthat the slabsencounter stronger or densermaterial • • •-•/ Ohwne • • which resiststheir sinking. •1200 -'• •:• / • • - ....' •.• We proposethat all these observationsare consistentwith • - /?•6'-• Pyroxene•• • -- mechanical barriers near the 670 km and Lehmann discontin- 8oo /?• •- G•otherms_,/ / • rOarnet- uities. A small intrinsic increasein density,due to a change in tl / •o• - chemistry,is a very effectivebrake to penetrativeconvection. For example,a 3% differencein intrinsicdensity can be offset only by a large temperature differential of 1000øC.A similar • 15050 my.my - 400•• - decrease in temperature is required to elevate the olivine- 00 •00 200 300 400 spinelphase boundary to 250 km in the slab. Depth, km There is a relationshipbetween age of subductedplate and Fig. 5. Oceanic and continental pyroxene geothermsfrom Mer- penetrationdepth [Vlaar and Wortzel, 1976],suggesting that cier and Carter [ 1975] and theoretical oceanicgeotherms at 50 and 150 thin lithosphere cannot subduct to great depth. Old litho- m.y. from Schubertet al. [1978]. The critical gradient [Kumazawa and sphereon the other hand is not only colder but may be in- Anderson, 1969] is for constant shear velocity versusdepth in a ho- mogeneousolivine mantle. The critical gradient for garnet is not trinsicallydenser if it growsby freezingeclogite onto its base. much different. The olivine-spinel and garnet + pyroxene to garnet We believe that the seismicitypatterns may be controlled solid solution boundariesare from Akaogi and Akimoto [1977]. by.the mantle discontinuitynear 220 km. There are no important first-order phase changesin the mantle near this depth dicatesthat it is variable in depth. The depth range depends [Ringwoo&1975]. This plus the sharpnessof the discontinuity on the densitycontrast but need only be a few kilometers. suggestthat there is a chemicalchange. An increasein garnet Thus there is a variety of evidenceof supportof an impor- content is the most reasonableway to increasethe density. tant discontinuitynear 220 km. This discontinuityaffects seis- The Lehmann discontinuity may be the boundary between micity and may be a density or mechanical impediment to depleted and fertile lherzolite or betweenperidotite and ec- slab penetration. It marks the depth above which there are 1ogite. large differencesbetween continental shieldsand oceans.Few In regions of continent-continentcollision the distribution earthquakes occur below this depth in continental collision of earthquakesshould define the shapeand depth of the colli- zonesand in regionswhere the subductinglithosphere is less sion zone. The Hindu Kush is characterizedby a seismicity than about 50 m.y. old. pattern terminatingin an active zone at 215 km [Santo, 1969]. A pronouncedminimum in seismicactivity occursat 160 km. SEISMICITY Again, the Lehmann discontinuityappears to mark the lower In most seismicregions, earthquakes do not occur deeper boundaryof the moving plates. than about 250 km. This applies to oceanic,continental, and We proposethat the harzburgiteportion of the slab remains mixed domains. The maximum depths are 200 km in the above -•250 km and only old slabscan penetratedeeper. The South Sandwich arc, Burma, Rumania, the Hellenic arc, and eclogitepart of the lithosphereis denserthan spinelor garnet the Aleutian arc; 250 km in the west Indian arc; and 300 km in peridotite.We suggestthat it is a densitybarrier rather than a the Ryukyu arc and the Hindu Kush. There are large gapsin strength barrier that is responsiblefor the distribution and seismicitybetween ,•250 km and ,•500-650 km in New Zea- stressesof intermediate depth earthquakes.With this model land, New Britain, Mindanao, Sundu, New Hebrides, Kuriles, only the eclogiteportion of the lithospherecan penetrate be- North Chile, Peru, South Tonga, and the Marianas [Isacks low 220 km and the uppermost mantle will be rich in olivine and Molnar, 1971]. In the New Hebrides there is a concentra- and orthopyroxene. The upper mantle below the Lehmann tion of seismicactivity between 190 and 280 km that movesup discontinuityis richer in garnet. to 110 and 150 km in the regionwhere a buoyant ridge is at- The compositionof the mantle between220 and 670 km is temptingto subduct[Chung, 1979]. In the Bonin-Mariana re- the subjectof a separatepaper. The seismicdata are consis- gion there is an increasein activity at 280-340 km to the south tent with eclogite in this region. This may be the source,as and a general decreasein activity with depth down to about well as the sink, of oceanic crust. 230 km. Where earthquakesreach as deep as 400 km, there is TEMPERATURES AND VISCOSITY IN THE MANTLE a pronouncedgap below 150 km. In the Tonga-Kermadecre- gion, seismicactivity decreasesrapidly down to 230 km and, Mercier and Carter [1975] have reanalyzed xenolith data in the Tonga region, picks up again at 400 km. In Peru most and derived the continental and oceanicpyroxene geotherms of the seismicityoccurs above 190-230km, and thereis a pro- shown in Figure 5. They converge above 200-250 kin. The nouncedgap betweenthis depth and 500 km. In Chile the ac- colder oceanic curve is their preferred solution for normal tivity is confined to above 230 km and below 500 km. ocean. The points for 50 and 150 m.y. oceanic mantle are Crosssections of seismicityin theseregions suggest impedi- from a theoretical discussionof Schubertet al. [1978]. There is ments to slab penetration at depthsof about 230 and 600 km. good agreement between the estimatesof temperature using Oceanic lithospherewith buoyant ridgesseems to penetrate geophysicaland petrologicaltechniques, and no evidencefor only to 150 km. deep, >200 kin, differencescorrelated with shields.Solomon Compressionalstresses parallel to the dip of the seismic [1976] alsomade this point. zone are prevalent everywhere that the zone exists below The shearvelocity under shieldsincreases with depth, or is about 300 km, indicating resistanceto downward motion be- constant,to about 100 km in spite of the fact that the shield ANDERSON: DEEP STRUCTURE OF CONTINENTS 7 5 5 9

chemical change at the Lehmann discontinuity would make an even more pronouncedminimum viscositychannel at this io• • ' UpperMantleo-: I00 borsViscosity - depth, and would make this region the most likely sourcefor mantle diapirs,the precursorsof basalticvolcanism. •io•3 I- • - SUMMARY AND DISCUSSION There has been to date no seismicstudy which has detected '•io••\ •Shield - resolvable ocean-shield differences in velocity below about 250 km. The large observedvariation in ScS times can be ac- '•o1021• Ocean counted for by changesabove this level which also corre- spondsto a seismicand seismicitydiscontinuity. Deeper vari- •qio2ø• ations also exist but there is no evidencethat they are rigidly iOI9 / i I i I I I • I coupledto shallowerplate motions.The study of deep lateral 0 I00 200 300 400 variationsby combiningseismic data of differenttypes is com- Depth, km plicated by the necessityof allowing for anelasticity,ani- Fig. 6. Viscosityversus depth for old ocean(cold oceangeotherm sotropyand lithosphericaging. in Figure 5) and shields.Dislocation climb in olivine is assumed.The The most detailed recent studies are England et al. [1978] viscosityvaries as the squareof the assumedstress. for body waves and Cara [1979] for surfacewaves, including highermodes. In thesestudies the geometryand analysistech- geothermis steeperthan the critical gradient for a low-veloc- niques were particularly favorable for detectingsuch differ- ity zone down to about 200 km. This requiresthat the miner- encesif they exist. Known variationsabove -•250 km are con- alogy and/or compositionchange with depth. This suggests sistent with observed ScS times. that the garnet content increases,in agreementwith the pe- The geothermsunder young oceansjoin the 1300øC adia- trology of kimberlite pipes [Boyd and McCallister, 1976; Jor- bat at relatively shallowdepths, -• 150 km. Temperaturesun- dan, 1978] der older oceans and shields converge above about 200 km The stable continentalupper mantle shear velocitiesabove and join the 1300øCadiabat near 220 km. This will be a mini- --•250 km are higher than oceanic shear velocitieseven when mum viscositychannel and the recoupling zone for continen- corrected for the difference in temperature. The difference, tal plates. Continental collisionearthquakes will be mainly •8%, can be due to partial melting or to dislocationrelaxation above this zone. The Lehmann discontinuityis probably due [Andersonand Minster, 1979], or it may indicate a difference to a change in chemistry and representsa density barrier to in chemistry. slab penetration.The eclogiticportion of the oceanic litho- The adiabatic gradient is less than the critical gradient. sphere,however, can penetratethis barrier. Separationof ec- Therefore below some 200 km the shear velocity should in- logite and residualharzburgite may be responsiblefor inter- crease with depth and the K/t• ratio should reverse, as ob- mediate depth earthquakesand the steepeningof the seismic served. This is also the depth at which the geothermsare zone. closestto the melting point of mantle minerals. This is the condition for minimum viscosity. Acknowledgments. I thank Tom Jordan, Frank Richter, Hiroo Kanamori, and Raymond Jeanloz for stimulating discussions,and The temperature structure between shields and ocean ba- Thomas Jordan, Michel Cara, David Jackson, and Geoffrey Davies sinsleads to substantialdifferences in viscosity.The viscosity for copiesof preprints.This researchwas supportedby NSF grant •/ depends on stress, temperature, and pressure. For dis- EAR 77-14675. Contribution 3212, Division of Geological and Plan- location climb [Nabarro, 1967], etary Sciences,California Instituteof Technology,Pasadena. •1= •rkTG•/Dbo• REFERENCES where G, D, b, and o are, the shearmodulus, diffusivity, Bur- ger's vector, and stress,respectively, and k T has the usual Adams, R. D., Reflections from discontinuities beneath Antarctica, meaning.The diffusivityD is a strongfunction of temperature Bull. Seismol. Soc. Amer., 61, 1441-1451, 1971. Akaogi, M., and S. Akimoto, Pyroxene-garnetsolid solution equi- and presure, libria in the systemsMg4Si4Oi2-Mg3A12Si3Oi2 and Fe4Si4Oi2- D -- Do exp (-gT,,,/T) Fe3AI2Si30•2 at high pressures and temperatures, Phys. Earth Planet. Interiors, 15, 90-106, 1977. Where T,• is the pressuredependent liquidus temperatureof Anderson, D. L., Recent evidence concerningthe structureand com- the major phase. For the upper mantle it is usually assumed positionof the earth'smantle, in Physicsand Chemistryof the Earth, vol. 6, edited by L. Ahrens, F. Press,and K. Runcorn, Pergamon, that olivine controlsthe rheology.From the geothermsof Fig- New York, 1966. ure 5 and constantsfrom Ashby and Verrall [1978] and Goetz Anderson, D. L., Latest information from seismic observations, in The [1978], we calculate the viscosity profiles of Figure 6. The Earth's Mantle, edited by T. F. Gaskell, Academic, New York, shield and ocean values are the same below 200 km and ex- 1967a. Anderson, D. L., Phase changesin the upper mantle, Science,157, hibit a minimum at about 230 km where the geothermsjoin 1165-1173, 1967b. ' the 1300øC adiabat. Below this depth the viscosityincreases Anderson,D. L., A seismicequation of state, Geophys.J. Roy. Astron. becausethe adiabat divergesfrom the melting curve. Note Soc., 13, 9-30, 1967c. that above 150km, shieldsare at least 1«orders of magnitude Anderson, D. L., Chemical inhomogeneity in the mantle, Earth more viscous than oceans. The mantle will be most fluidlike Planet. Sci. Lett., 5, 89-94, 1968. Anderson, D. L., Composition of the mantle and core, in Annual Re- near 230 km, and this is the most likely horizon for a shear viewof Earth and PlanetarySciences, vol. 5, edited by F. A. Donath, boundary layer betweenthe continentalplate and the under- Annual Reviews, Palo Alto, Calif., 1977. lying mantle. A thermal boundary layer associatedwith a Anderson,D. L., and D. Harkrider,The effectof anisotropyon con- 7560 ANDERSON:DEEP STRUCTUREOF CONTINENTS

tinentaland oceanicsurface wave dispersion (abstract), J. Geophys. Lehmann,I., Velocitiesof longitudinalwaves in the upperpart of the Res., 67, 1627, 1962. earth's mantle, Ann. Geophys.,15, 93-118, 1959. Anderson,D. L., and D. Harkrider, Universaldispersion tables, II, Lehmann, I., S and the structureof the upper mantle, Geophys.J. Variationalparameters for amplitudes,phase velocity, and group Roy. Astron. Soc., 4, 124-138, 1961. velocity for first four Love modes for an oceanic and continental Lehmann, I., Low-velocitylayers, in The Earth'sMantle, edited by earth model, Bull. Seismol.Soc. Amer., 58, 1407-1499, 1968. T. F. Gaskell, pp. 41-62, Academic,New York, 1967. Anderson,D. L., and B. R. Julian,Shear velocities and elastic param- Lukk, A. A., and I. L. Nersesov,Structure of the upper mantle as etersof the mantle, J. Geophys.Res., 74, 3281-3286, 1969. shownby observationsof earthquakesof intermediatefocal depth, Anderson,D. L., and J. B. Minster, Seismicvelocity, attenuation, and Dokl. Akad. Nauk. SSSR Earth Sci. Sect., 162, 14-16, 1965. theologyof the uppermantle, Phys. Earth Planet. Interiors, in press, MacDonald,G. J. F., Thedeep structure of Continents,Rev. Geophys., 1979. 1, 587-665, 1963. Ashby, M. F., and R. A. Verrall, Micromechanismof flow and frac- Marsh, B. D., and J. G. Marsh, On globalgravity anomaliesand two- ture, and their relevanceto the rheologyof the uppermantle, Phil. scalemantle convection,J. Geophys.Res., 81, 5267-5280, 1976. Trans. Roy. Soc. London, Ser. A, 288, 59-95, 1978. Mass6, R. P., Shear velocity distributionbeneath the Canadian shield, Boyd,F. R., and R. H. McCallister,Densities of fertileand sterile gar- J. Geophys.Res., 78, 6943-6950, 1973. net peridotites,Geophys. Res. Lett., 3, 309-512, 1976. Mercier, J.-C., and N. L. Carter, Pyroxenegeotherms, J. Geophys. Brune, J., and J. Dorman, Seismic waves and earth structure in the Res., 80, 3349-3362, 1975. Canadian shield, Bull. Seismol.Soc. Amer., 53, 167-210, 1963. Nabarro, F. R. N., Steady-statediffusional creep, Phil. Mag., 16, 231- Cara,_M., Lateral variationsof S-velocityin the upper mantlefrom 237, 1967. higher Rayleigh modes,Geophys. J. Ro)4Astron. Soc., 57, 649-670, Niazi, M., Use of sourcearrays in studiesof regionalstructure, Bull. 1979. Seismol. Soc. Amer., 59, 1631, 1969. Chung,W.-Y., Vari•fi0n ofseismic source parameters andstress drop Okal, E. A., The effect of intrinsic oceanic upper-mantle hetero- within a descendingslab as revealedfrom body-wavepulse-width geneityon regionalizationof long-periodRayleigh-wave phase ve- and amplitudeanalysis, Ph.D. thesis,Calif. Inst. of Technol.,Pasa- locities, Geophys.J. Roy. Astron. Soc., 49, 457, 1977. dena, 1979. Okal, E. A., Observedvery long period Rayleigh-wavephase veloci- Dorman,J., M. Ewing,and J. Oliver,Study of shear-velocitydistribu- ties acrossthe Canadian shield, Geophys.J. Roy. Astron. Soc., 53, tion in the uppermantle by mantleRayleigh waves, Bull. Seismol. 663-668, 1978a. Soc. Amer., 50, 87-115, 1960. Okal, E. A., Seismicinvestigations of upper mantle lateral hetero- Duschenes,J. D., and S.C. Solomon,Shear wave travel time residuals geneity, Ph.D. thesis,Calif. Inst. of Technol., Pasadena,1978b. fromoceanic earthquakes and the evolution of oceaniclithosphere, Okal, E. A., and D. L. Anderson,A studyof lateral inhomogeneities J. Geophys.Res., 82, 1985-2000, 1977. in the upper mantle by multiple ScStravel-time residuals, Geophys. Dziewonski,A.M., Upper mantlemodels from "purepath" dis- Res. Lett., 2, 313-316, 1975. persiondata, J. Geophys.Res., 76, 2587-2601, 1971. 'Poupinet,G., Relation entre le tempsde parcoursvertical des ondes England,P. C., B. L. N. Kennett, and M. H. Worthington,A com- sismigueset l'age de la lithospherecontinentale, Bull. Soc. Geol. parison of the upper mantle structure beneath Eurasia and the France,in press,1979. North Atlantic and Arctic Oceans,Geophys. J. Roy. Astron.Soc., Ringwood, A. E., Compositionand Petrologyof the Earth's Mantle, 54, 575-585, 1978. 618 pp., McGraw Hill, New York, 1975. Goetze,C., The mechanismsof creepin olivine,Phil. Trans. Roy. Soc. Sacks,I. S., J. A. Snoke, and E. S. Husebye,Lithosphere thickness be- London, Ser. A, 288, 99-119, 1978. neath the Baltic shield, CarnegieInst. Yearb., 76, 805-822, 1977. _Hales,A. L., andJ. L. Roberts,The traveltimes of S andScS, Bull. Santo, T., Regional study on the characteristicseismicity of the world, Seismol.Soc. Amer., 60, 461-489, 1970. I, Hindo Kush region, Bull. Earthquake Res. Inst., 47, 1035-1048, 1969. Hart, R. S., D. L. Anderson, and H. Kanamori, The effect of attenuation on grossearth models,J. Geophys.Res., 82, 1647-1654,1977. Schule,J. W., and L. Knopoff, Shearwave anisotropyin the upper. Helmberger,D. V., andG. R. Engen,Upper mantle shear structure, J. mantle of the Pacific basin, Geophys.Res. Lett., 3, 359-362, 1976. Geophys.Res., 79, 4017-4028, 1974. Schubert, G., D. A. Yuen, C. Froidevaux, L. Fleitout, and M. Isacks,B., and P. Molnar,Distribution of stressesin the descending Souriau,Mantle circulationwith partial shallowreturn flow: Effects lithospherefrom a global surveyof focal mechanismsolutions of on stressesin oceanicplates and topographyof the sea floor, J. mantleearthquakes, Rev. Geoœhys. Space Phys., 9, 103,1971. Geophys.Res., 83, 745-758, 1978. Sipkin, S. A., and T. H. Jordan, Lateral heterogeneityof the upper Jordan,T. H., The continentaltectosphere, Rev. Geophys.Space mantle determined from the travel times of multiple ScS, J. Phys., 13, 1-2, 1975. Geophys.Res., 81, 6307-6320, 1976. Jordan,T. H., Compositionand developmentof the continentaltectosphere,Nature, 274, 544-548, 1978. Solomon,S.C., Geophysicalconstraints on radial and lateral temperature variationsin the upper mantle, Amer. Mineral., 61, 788-803, Jordan,T. H., Mineralogies,densities, and seismic velocities of garnet 1976. lherzolitesand their geophysicalimplications, in The Mantle Sample:Inclusions in Kimberlitesand Other Volcanics,edited by Steinmetz,L., A. Hirn, and G. Perrier,Reflexions sismigues a la base de l'asthenosphere,Ann. Geophys.,30, 173-180, 1974. F. R. Boydand H. O. A. Meyer,pp. 1-14, AGU, Washington, D.C., 1979a. Vlaar, N.J., and M. J. R. Wortzel, Lithosphericaging, instability and Jordan,T. H., The deepstructure of the continents,Sci. Amer., 240, subduction,Tectonophysics, 32, 331-351, 1976. 92-107, 1979b. Whitcomb, J. H., A study of the velocitystructure of the earth by use Jordan, T. H., and D. L. Anderson, Earth structurefrom free oscilla- of core phases, Ph.D. thesis, Calif. Inst. of Technol., Pasadena, 1973. tions and travel times,Geophys. J. Roy. Astron.Soc., 36, 411-459, 1974. Whitcomb, J. H., and D. L. Anderson, Reflections of P'P' seismic Jordan,T. H., and N. Frazer, Crustal and upper mantle structure waves from discontinuitiesin the mantle, J. Geophys.Res., 75, 5714-5728, 1970. from Sp phases,J. Geophys.Res., 80, 1504-1518, 1975. Kanamori, H., Velocity and Q of mantle waves,Phys. Earth Planet. Wickens,A. J., Variations in lithosphericthickness in Canada, Can. J. Interiors, 2, 259-275, 1970. Earth Sci., 8, 1154-1162, 1971. Knopoff, L., S. Mueller, and W. L. Pilant, Structure of the crust and Wiggins,R. A., and D. V. Helmberger,Upper mantle structureof the upper mantle in the Alps from the phasevelocity of Rayleigh westernUnited States,J. Geophys.Res., 78, 1870-1880, 1973. waves,Bull. Seismol.Soc. Amer., 56, 1009-1044, 1966. Yoshida,M., Group velocitydistributions of Rayleighwaves and two Kumazawa,M., and O. L. Anderson,Elastic moduli pressure deriva- upper mantle models in the Pacific Ocean, Bull. Earthquake Res. tivesand temporalderivatives of single-crystalolivine and single- Inst. Tokyo Univ., 53, 319-338, 1978. crystalforsteritc, J. Geophys.Res., 74, 5961-5972, 1969. (Received March 28, 1979; Lambeck,K., Lateral densityanomalies in the upper mantle, J. revisedMay 25, 1979; Geophys.Res., 81, 6333-6340, 1976. acceptedJuly 2, 1979.)