STAGNANT SLABS IN THE UPPER AND LOWER MANTLE TRANSITION REGION
Yoshio Fukao Sri Widiyantoro MasayukiObayashi • EarthquakeResearch Institute Departmentof Geophysics Seismologyand Volcanology Universityof Tokyo and Meteorology ResearchDepartment Tokyo,Japan BandungInstitute of Technology MeteorologicalResearch Bandung,Indonesia Institute Tsukuba,Japan
Abstract. We made a region-by-regionexamination of with present slab subduction appears to be blocked subductedslab images along the circum-Pacificfor some stronglyto turn into predominantlyhorizontal flow in of the recentglobal mantle tomographicmodels, specif- the transitionregion. Recent globaltomographic models ically for two high-resolutionP velocitymodels and two show also a group of lithosphericslabs deeply sinking long-wavelengthS velocity models. We extractedthe throughthe lower mantle, typicallythe presumedFaral- slab imagesthat are most consistentamong different lon slab beneath North and Central America and the models. We found that subducted slabs tend to be sub- presumedIndian (Tethys) slab beneathHimalaya and horizontally deflected or flattened in the upper and the Bay of Bengal. These remnant slabs are not con- lower mantle transitionregion, the depth range of which nectedto the surfaceplates or to the presentlysubduct- correspondsroughly to the Bullen transition region ing slabs and appear to sink independentlyfrom the (400-1000 km). The deflectedor flattenedslabs reside latter. The presenceof thesedeeply sinking slabs implies at different depths,either above or acrossthe 660-km that the pre-Eocene subductionoccurred in much the discontinuityas in Chile Andes, Aleutian, Southern sameway asin the presentday to accumulateslab bodies Kurile, Japan,and Izu-Bonin; slightlybelow the discon- in the transitionregion and that the consequentunstable tinuity as in Northern Kurile, Mariana, and Philippine; downflow occurred extensivelythrough the transition or well below it as in Peru Andes, Java, and Tonga- region in the Eocene epoch to detach many of the Kermadec. There is little indication for most of these surfaceplates from the subductedslabs at depths and slabsto continue "directly" to greater depthswell be- hence to cause the reorganizationof global plate mo- yond the transitionregion. Mantle downflowassociated tion.
1. INTRODUCTION tensen, 1989; Ringwood, 1994; Carlson, 1994; Tackley, 1995a;Ogawa, 1997]. Since the discoveryof the so-called Bullen [1963] called the outermost1000 km of the 660-km discontinuity[Niazi and Anderson,1965; John- Earth's mantle the upper mantle,which was dividedinto son, 1967], however, the boundary between the upper the B and C layers at a depth of about 400 km. The C and lower mantle has been placed at this discontinuity layeris the transitionregion between the B layer (above and the transition zone has been defined as the depth 400 km) and D layer (below1000 km). Figure la shows range between the 410- and 660-km discontinuities.In the depth variations of P and S wave velocities in Figure la we add a typical model having the 410- and Bullen's model. While the gradientsin seismicvelocity 660-km discontinuities,the preliminaryreference Earth and in densityin the B and D layers are steady and model (PREM), at a referencefrequency of 1 s which positivewith respectto increasingdepth, those in the C has the first-order discontinuities at 400- and 670-km layer are anomalouslylarge. These large gradients imply depths[Dziewonski and Anderson, 1981]. The lower half gradualor multiplechanges of composition,or of phase, of the mantle transitionregion in its original definition or of both [Birch, 1952]. Such compositionaland/or structuralchanges in mineral assemblageshould have has sincethen been regarded as the uppermostpart of played a significantrole in convectionprocesses and the lower mantle, where the velocity and densitygradi- evolutionhistories of the mantle [Anderson,1979; Chris- ents remain steadyand normal. This consensushas recently been challenged by Kawakatsuand Niu [1994],who confirmedthe existence Now at Japan Scienceand TechnologyCorporation, of a seismicdiscontinuity beneath Tonga at a depth of Kawaguchi-shi,Japan. 920-950 km to suggestthat the bottom of the transition
Copyright2001 by the AmericanGeophysical Union. Reviewsof Geophysics,39, 3 / August2001 pages291-323 8755-1209/01/1999RG000068515.00 Papernumber 1999RG000068 ß 291 ß 292 ß Fukao et al.' SLABS IN THE MANTLE TRANSITION REGION 39, 3 / REVIEWS OF GEOPHYSICS
(a) Velocity(km/s) (b) RMS Amplitude (%) 4 6 8 10 12 14 0 1 .o 2.0 S ;•-•-...P
500 -
1000 -
500 -
2000 - Bullen
2500 -
3000 0.3 0.6 0.9 Correlation
Figure 1. (a) P and S wavevelocities of the sphericalmodels of Bullen and the preliminaryreference Earth model(PREM) asa functionof depthin the mantle.The depthrange of the Bullentransition region is shaded. (b) A comparisonof the three-dimensional(3-D) seismicmodels constrained by seismicdata alone (SH8M/ WM13) and by seismicand gravitydata (modelA) after Forteand Woodward[1997]. The root-mean-square (RMS) amplitudesof seismicshear velocity heterogeneity described by SH8M/WM13 andmodel A are shown by the dotted and solidcurves, respectively. Units are percentperturbation relative to the sphericalreference velocityat the givendepth. The dashedcurve indicates the globalcross correlation between SH8M/WM13 and model A.
region may be defined by this discontinuityso that it mappingof seismicvelocities). Tanirnoto [1990b] inves- deepensback to its originalplacement by Bullen [1963] tigated correlation of lateral velocity variation among and Gutenberg[1959]. Niu and Kawakatsu[1997] further different layers in the three-dimensional(3-D) model extended their work to show a large depth variation MDLSH of Tanirnoto[1990a] and a combinedmodel of (900-1100 km) of this discontinuity.Its presencehas M84A [Woodhouseand Dziewonski,1984] and L02.56 now been reported in the areas of New Britain [Reve- [Dziewonski,1984]. He found that the correlationsbe- naughand Jordan, 1991], Japan-Kurile-Kamchatka[Pe- tween layersin the upper 1000 km and betweenlayers in tersenet al., 1993], Tonga [Kawakatsuand Niu, 1994], the rest of the mantle are high but that the correlation and Japan-Izu-Boninand Java-Banda-Flores[Niu and betweenthe upper 1000 km and the rest of the mantle is Kawakatsu,1997]. More recently, Vinnik et al. [1998] either low or even negative,as later confirmedby Mon- observedmultiple discontinuities (at 860, 1070,and 1170 tagner [1994]. On the basis of this finding, Tanirnoto km) and the persistentappearance of the 1070-kmdis- [1990b] suggestedplacing the boundary between the continuitybeneath the Sundaarc. These reportsindicate upper and lower mantle at a depth of 1000 km. Wenand that a thickness of several hundred kilometers below the Anderson [1995] found that the correlationsbetween 660-km discontinuityis far more complicated,at least in seismicheterogeneity and graveyardsof subductedslabs the subductionzones, than has ever been thought. [Engebretsonet al., 1992]peak near 1000-kmdepth. They The significanceof separatingthe upper 1000 km of interpreted this result as indicatinga significantbound- the mantle from the rest of the mantle is also indicated ary between800 and 1100 km depth acrosswhich vertical by analysesof previous seismic tomographic models flow of the mantle is stronglyinhibited. (Earth-mantle models obtained by three-dimensional In the present paper we maintain the view that the 39, 3 / REVIEWSOF GEOPHYSICS Fukaoet al.: SLABSIN THE MANTLETRANSITION REGION ß 293 nominalboundary between the upper and lower mantle convectionmodels constrainedby seismictomography lies at 660-km depth.We then refer to the "transition [see alsoButler and Peltier,2000]. These resultsimply region"as the transitionallayer between the upperand that the transitionregion acrossthe 660-km discontinu- lower mantle across the 660-km discontinuityrather ity maybe understoodas a relativelyimpermeable layer than the lower half of the upper mantle.The transition of mantle circulation.Li and Romanowicz[1996] noted regionin thisview maybe understoodas a depthrange in their 3-D model SAW12D that the midmantle is geodynamicallyaffected directly by processesacross the characterizedby the "white" characterof the heteroge- 660-kmdiscontinuity. Assuming that the mantleconvec- neity spectrumand that this "white" heterogeneityis tive circulationis driven by densityheterogeneity in- sandwichedby the degree-2dominant heterogeneity at ferred from global seismictomography models, Forte depthsof 500-800 km and the even strongerdegree-2 and Woodward[1997] carriedout a joint inversionof dominantheterogeneity in the D" layer. On the basisof seismicand gravitydata while simultaneouslyminimiz- thisobservation they suggesteda thermalboundary layer ing verticalflow acrossthe 660-kmdiscontinuity. This nature not only in the D" layer but alsoin the upper and inversionrevealed a family of mantle modelsthat pro- lower mantle transitionzone (500- to 800-km depth). vide as goodfit to the data as the whole mantleflow Downflow of mantle convection occurs mainly by modelsbut generatea stronglyor partiallylayered man- platesubduction. If thisdownflow is stronglyor partially tle flow.Figure lb [Forteand Woodward,1997] compares inhibited at 660-km depth and at other possibledepths a modelof this family (modelA) to the modelinverted in the transitionregion, vertical fluxes of subductedslabs from the seismicdata alone(SH8M/WM13 [Forteet al., mayconvert largely into horizontalfluxes in this region. 1994]).The solidand dottedcurves indicate the shear The aim of the presentpaper is to seekevidence for or velocityperturbations of thesetwo modelsas a function againstthis predictedreorientation of slabfluxes in the of depth,which are obtainedby computingthe root- transitionregion. For this purposewe will investigate mean-square(RMS) of the amplitudesof the model subductedslab images, from region to region alongthe [Ritzwollerand Layely, 1995]. In eitherof the modelsthe circum-Pacific,for some of the recent global mantle RMS velocityperturbation takes its minimumin a depth tomographicmodels and our own regionalmantle tomo- rangeof 1000-2000km. Above this depthrange, how- graphicmodels. It will be shownthat manyof subducted ever, we observea noticeable difference. Model A pos- slabs are deflected or flattened in and near the Bullen sessesa distinctlayer at depthsof 500-1000 km in which transitionregion, in agreementwith the idea that mantle the RMS perturbationis relativelylarge, while SH8M/ downflowis blocked significantlyto turn largely into WM13 has no suchlayer. This relativelyheterogeneous horizontal flow in this region. In the present paper we layer in modelA is dominatedby horizontalflow as a will view the transitionregion as sucha relativelyimper- consequenceof strong inhibition of vertical flow at meable layer of mantle circulationat least in the sub- 670-kmdepth. Similarly, •adek et al. [1997]demon- duction zones. Naturally, the top and bottom of the stratedthat the layeredmantle convectionmodels de- transitionregion in thisview cannot be defineduniquely rived from seismictomographic models can reconcile and only looselycorrespond to thoseof Bullen'stransi- the geoid and the amplitudesof the dynamictopogra- tion region(400 and 1000km, respectively). phiesof the surfaceand the 660-kmdiscontinuity. These modelspossess a pronouncedlow-viscosity zone in a depthrange from 660to 1000km, wherehorizontal flow 2. CHARACTERISTICS OF HETEROGENEITY dominates.The existenceof a low-viscosityzone near or IN THE TRANSITION REGION beneath660-km depth was discussedfirst by Forteet al. [1993]and later by King [1995], Kido and Oadek [1997], Many global tomographicmodels have appeared Oadekand Van den Berg [1998], and others. Cserepes and sincethose of Dziewonski[1984] and Woodhouseand Yuen[1997] found that a low-viscosityzone placedbe- Dziewonski[1984]. They may be grouped into two neath the endothermicphase transition could consider- classes,models from arrival-time data and models from ablyincrease the degreeof layering. digital waveform data. The arrival time of a seismic Imposingplate velocitiesas the surfaceboundary phase(e.g., a P or S wave) essentiallycarries structural condition,and without introducinga low-viscosityzone informationalong its ray path connectingthe sourceand nearor beneath 660-km depth, Oadek and Fleitout [1999] receiver.Arrival times can be measuredeasily and have madea joint inversionof seismicand geoid data, similar beenreported to the InternationalSeismological Centre to the one by Forte and Woodward[1997], to conclude (ISC) for morethan 30 years from many observatories in that the modelsthat best fit the geoid require a strong the world. The amount of available data now exceeds reduction of vertical flow across the 660-km discontinu- several million even for the first arrivals alone. Wave- ity, whichmeans the dominanceof horizontalflow in form data, on the other hand, have become available somedepth range around 660 km. Pari and Peltier [1998] only relativelyrecently and only at limited stations,al- demonstratedthat significantmantle layering is required thoughthe situationis now changingrapidly. In princi- to adequatelyreconcile the observedgeographical dis- ple, one can retrievefrom waveformdata the arrival tribution of the surface heat flow in terms of mantle times and differentialtravel times of many phases(e.g., 294 ß Fukao et al.: SLABS IN THE MANTLE TRANSITION REGION 39, 3 / REVIEWS OF GEOPHYSICS
the S wave and its multiply reflectedwaves and surface but the actual resolution is limited by many factors waves)that carry informationon Earth structurealong including the density and uniformity of path coverages all the relevant ray paths connectingthe source and [Boschiand Dziewonski,1999]. Regarding the actual receiver. For a given source-receiverconfiguration, resolutionof WEPP2, we refer to Tafima et al. [1998], therefore, the waveform data contain far more abundant who showed the result of the checkerboard resolution information than the arrival-time data. On the other test for the top 1000 km of the mantle beneath the hand, a much larger number of source-receiverconfig- westernPacific (see Inoue et al. [1990] for the checker- urations are available from a data set of arrival times. board resolutiontest). Fukao et al. [1992] reported in Recent global modelsbased on arrival-time data include somedetail the resultsof the resolutionanalyses for the Zhou [1996], van der Hilst et al. [1997], Obayashiet al. earlier versionof WEPP2 (WEPP1; see AppendixA), [1997], and Bijwaardet al. [1998] for P waves;Grand et including the checkerboardresolution test, the calcula- al. [1997] and Widiyantoroet al. [1998] for S waves;and tion of resolution kernels for selected blocks within Robertsonand Woodhouse[1995, 1996], Kennett et al. subductedslab images, and the recoverytest for syn- [1998],and Vascoand Johnson [1998] for P and S waves. thetic subductedslab images. The resolution analyses (The S velocitymodel of Grand et al. [1997] has been were made for model P97 by van der Hilst et al. [1997], derivedfrom the arrival times carefullyread by the first Widiyantoro[1997], and van derHilst et al. [1998],includ- author.) The modelsbased on waveformdata are Suet ing the "spike" test [Spakmanet al., 1993] for regions al. [1994], Li and Romanowicz[1996], Masters et al. beneath subductionzones and the recoverytest for syn- [1996], and Resovskyand Ritzwoller[1999] for S waves. thetic subducted slab images. Similar analyseswere Su and Dziewonski[1997] combined the ISC arrival-time made by Widiyantoro[1997] for his regionalmodels. In data and the waveform data to obtain a 3-D model for P the above P velocity models, in general, the well-re- and S waves. solvedregions tend to be more narrowlyconfined to the In this paper we refer mainly to the two P wave vicinity of subductingslabs as the depth of the layer models of van der Hilst et al. [1997] (model P97) and becomesshallower. Suet al. [1994] and Li and Ro- Obayashiet al. [1997] (model WEPP2) in the first class manowicz [1996] made resolution analyses for and the two S wave models of Suet al. [1994] S12WM13 and SAW12D, respectively,using the syn- (S12WM13) andLi and Romanowicz[1996] (SAW12D) thetic degree-ll patterns.Although these analyseswere in the secondclass to examine the heterogeneitystruc- not particularly focusedon structuresbeneath subduc- ture of the transition region. The two S wave models, tion zones,their resultsin map and cross-sectionalview S12WM13 and SAW12D, are describedhorizontally by yield somehints for the resolutionof the presumedslab sphericalharmonic functions up to degree 12. The grid imagesseen in Plates 1-7. size in the world map (Figure 3) correspondsto the Less extensivelytested in the above tomographic resolutionlength of thesemodels. The verticalvariation studies and in other studies is how stable the images of this heterogeneityis given by Chebyshevpolynomials obtained are against addition or removal of the data. up to degree 13 (from the Moho to the core-mantle Sucha test is important,in particular,when interpreting boundary(CMB)) in S12WM13 and by Legendrepoly- tomographicimages of small amplitudes,because they nomials up to degree 5 (from the Moho to 670-km might criticallydepend on the data additionor removal. depth) and degree 7 (from the 670-km depth to the An example of this effect is seen in Plate 8b, where CMB) in SAW12D. The two P wave models,P97 and WEPP2 is compared with its earlier version, WEPP1 WEPP2, are parameterizedin discreteblocks. The num- [Fukao et al., 1992]. As describedin AppendixA, the ber of blocksin P97 is 180 x 90 x 18 (or 291,600).Their inversionmethod, the model parameterization,and the horizontal size is 2ø x 2ø. The vertical size is depth- relative weights of minimization of model roughnessto dependent,about 100 km in the upper mantle and 200 minimization of data residual are the same between the km in the lower mantle. In WEPP2 the number of basic two models so that the only difference is the amount of blocksis 64 x 32 x 16 (or 32,768),but the total number data. (The data in WEPP2 are about5 timeslarger than is 55,735 after subdivision of the blocks in the western those in WEPP1.) (The crosssection is taken across Pacific (see Appendix A). The horizontal size of basic Japanalong a profile shownby the dotted line in Figure blocks is about 5.6ø x 5.6ø, while the vertical size varies 3.) Both WEPP1 and WEPP2 showthe deflectedslab with depth, from 29 km just below the surfaceto 334 km image in the transition region. In addition, WEPP1 just abovethe CMB. The horizontaland vertical sizesof exhibits some scattered fast anomalies in the lower man- the smallestblocks are one quarter and one half of those tle, many of which have disappearedin model WEPP2. of the basicblocks (see Appendix A). Similar subdivi- This result demonstrates that weak short-wavelength sions have also been made for other subduction zones anomaliesbeneath the stronger,more coherentanoma- along the circum-Pacificby Widiyantoro[1997] to detail lies in the transitionregion have to be interpretedcare- the relevantslab configurations(see AppendixB). His fully. resultswill also be incorporatedin the present study. As is obvious from the results of the resolution anal- The nominal resolutionis givenby the shortestwave- yses,the modelsfrom arrival time data are, in general,of lengthsof model parameterizationas describedabove, much higher resolutionthan thosefrom waveform data. 39, 3 / REVIEWSOF GEOPHYSICS Fukao et al.' SLABSIN THE MANTLE TRANSITION REGION ß 295
WEPP2
410 - 478km 478 - 55 lkm 551 - 629 km
629 -712km 712 - 800km 800 - 893km
fast 2.0% " • • ' ' • '., 20%slow
SAWl 2D
445km 515km 590km
.,
670km 755km 845km
I,
fast 2.0% --• ,,::,J • • •' J :'-1'---• 2.0% slow
Plate 1. Comparisonof the slownessperturbation (_+2%) patternsfor the depth range in the transition regionunder the northwesternPacific [Tajima et al., 1998].(top) High-resolutionP wavetomographic model WEPP2 from the arrival-timeinversion by Obayashiet al. [1997] in which the blockwisevariation of velocity anomaly is slightly smoothed.(bottom) Long-wavelengthSH wave velocity model SAW12D from the waveforminversion by Li and Romanowicz[1996]. (a) Southern Kurile (b) Java
.,
WEP
P97
SAW
S12W
slow fast Plate 2. Vertical mantle sectionsacross (a) the SouthernKurile arc and (b) the Java arc along profiles depictedin Figure 3. Mantle modelsare WEPP2 [Obayashiet al., 1997] (seeAppendix A) and P97 [vander Hilst et al., 1997] for P velocityperturbation and SAW12D [Li and Romanowicz,1996] and S12WM13 [Suet al., 1994]for S velocityperturbation. Wave speedvariations are relativeto the sphericalaverage of the final asphericalmodel for WEPP2, relative to model ak135 [Kennettet al., 1995] for P97 and relative to PREM [Dziewonskiand Anderson,1981] for SAW12D and S12WM13. Blue (red) colorsrepresent fast (slow) wave propagation.The amplitudescale is differentamong the four models(2.0, 1.5, 2.5, and 2.0% for WEPP2, P97, SAW12D, and S12WM13, respectively),as indicatedby the numbersat the sidesof the images.Open circles representhypocenters of earthquakeswithin a band 50 km wide on both sidesof the sectionplane. Two parallel lines indicatethe 410- and 660-km discontinuities. 39, 3 / REVIEWSOF GEOPHYSICS Fukaoet al' SLABSIN THE MANTLE TRANSITION REGION ß 297
(a) JapanA (b) JapanB
WEP
P97 ß 1.5
slow fast
Plate 3. Vertical mantlesections across the Japanarc along(a) profileA and (b) profile B (Figure 3). See Plate 2 for other explanations. 298 ß Fukao et al.: SLABSIN THE MANTLE TRANSITION REGION 39, 3 / REVIEWSOF GEOPHYSICS
(a) Izu-Bonin (b) Mariana
WEP
P97
S12 •
slow fast
Plate 4. Vertical mantlesections across (a) the Izu-Boninarc and (b) the Mariana arc alongprofiles shown in Figure 3. See Plate 2 for other explanations. 39, 3 / REVIEWS OF GEOPHYSICS Fukao et al.: SLABSIN THE MANTLE TRANSITION REGION ß 299
The latter, on the other hand, would be more robust Global againstpossible spurious images arising from, for exam- rms of S anomalies ple, uneven distributionsof sourcesand receivers,be- 0.0 0.4 0.8 1.2 1.6 2.0 I I I cause the waveform data involve many rays with very 0 different ray paths for a given source-receiverdistribu- tion. Hence the long-wavelengthfeatures of models from arrival-time data should be consistent with those 500 from waveformdata. Plate 1 showsan exampleof exam- ining such a consistency[Tajima et al., 1998]. In this plate the map views of the transitionregion under the northwesternPacific are compared at various depths 1000 betweenthe P velocitymodel WEPP2 from arrival-time data and the S velocitymodel SAW12D from waveform data. For both the P and S velocitymodels the strongest fast anomaly appearsbehind the Japaneseislands in a 1500 depth range of 500-600 km. Strongfast anomaliesdis- + WEPP2 appear at depthsbelow 800 km. These imagesare con- + P97 sistentbetween the two very different modelsand can be considered as the first-order robust features of the tran- 2000 SAW12D sition region under the northwestern Pacific. Model ...... S12WM13 WEPP2 indicates the existence of fine structures in these first-orderfeatures, all of which may not be real features. For this reason we always examine the models from 2500 arrival-time data in comparisonwith those from wave- form data. Figure 2 shows the RMS amplitudes of the four 0.0 0.2 0.4 0.6 0.8 1.0 modelsas a functionof depth.Note that the scalein this rms of P anomalies figure is different by a factor of 2 betweenthe P and S wave models.The differenceof this magnitudehas been Figure 2. The RMS value of the asphericalperturbation of reported in severaltomographic studies [Li et al., 1991; the P velocity models WEPP2 and P97 and the degree-12 S Robertson and Woodhouse, 1995, 1996; Kennett et al., velocity models SAW12D and S12WM13 as a function of 1998](see also the P andS velocityperturbation maps of depth in the mantle. Units are percentperturbation relative to Su and Dziewonski[1997]), while the theoreticallyesti- the sphericalreference velocity at the given depth. The refer- mated difference in sensitivityto temperature is about enceis the sphericalaverage of the 3-D model for WEPP2, the 1.7 [Karato, 1993]. A quantitativecomparison of RMS 1-D model ak135 for P97, and the 1-D model PREM for amplitudesis difficult to make betweenthe P and S wave SAW12D and S12WM13. The block models WEPP2 and P97 modelsbecause of the different schemesof parameter- are expanded into spherical harmonics at each depth and truncatedto degree24. In SAW12D, perturbationis not forced ization and regularization.The different predominant to be continuousacross the 670-km discontinuity.The scale is periodsinvolved in the ? and S wavemodels (typically 1 2% for the S velocity anomaly and 1% for the P velocity and 100 s, respectively)may alsobe responsiblefor this anomaly. The depth range of the Bullen transition region is difference[Nakanishi, 1978]. In Figure 2 we expanded, shaded. The upper and lower mantle transition region dis- for the P waveblock models, the lateral heterogeneityof cussedin the text is only in loose correspondencewith this each layer into sphericalharmonics, and the total RMS Bullen transition region. amplitudewith degreesup to 24 is plotted at the mid- depth of each layer. We made suchspherical harmonic expansionand subsequenttruncation for ease of com- Woodward[1997] (see also Figure lb). Ritzwollerand parisonbetween different models.The S velocitymodels Lavely [1995] remarked that the RMS amplitude of SAW12D and S12WM13 have similar RMS amplitudes long-wavelengthheterogeneity reaches a minimum be- at all depths except for the lowermostmantle. On the tween 1000 and 1500 km. In the upper mantle at depths other hand, the P velocitymodels WEPP2 and P97 have above about 400 km, on the other hand, the RMS RMS amplitudesthat are mutually different by a factor amplitude increasesrapidly with decreasingdepth. The of about3 in the uppermostmantle and a factor of about transition region correspondsto a transitional depth 2 in the restof the mantle.Such an amplitudedifference rangebetween this stronglyheterogeneous upper mantle is likely to be due to the difference of regularization at depths above 400 km and the least heterogeneous schemebetween the two models.In Figure 2 these four lower mantle at depthsof 1000-2000 km. In the P wave modelsshow clearly the flat, low amplitudesof hetero- modelsthe top of this transitionregion is marked by an geneity in a depth range from 1000 to 2000 km, as amplitudeminimum at a depth of about 400 km. In the pointed out by Woodwardet al. [1993] and Forte and 3-D model of Li and Romanowicz[1996], the upper and 300 ß Fukao et al.: SLABSIN THE MANTLE TRANSITION REGION 39, 3 / REVIEWSOF GEOPHYSICS
Figure 3. Map illustratingthe locationsof the crosssections of Plates 2-7 and Plates 8a and 8b. The four shadedareas define the regionsof western Pacific for Figures 4-7, South America for Figure 8, Central America for Figure 9, and Tonga for Figure 10 and Plate 8c.
lower mantle transitionzone (500- to 800-km depth) is depths of 1000-2500 km in the western Pacific are dominatedby the degree-2heterogeneity and is distinct smallerthan the global RMS, especiallyfor WEPP2 and from the top 300 km of the mantle dominated by the SAW12D, a contrast opposite to the one in the transi- heterogeneityof degrees 2, 4, 5, and 6 and from the tion region. This contrast between the western Pacific midmantle dominated by the heterogeneityof "white" and the whole globe,however, may be due in part to the spectral nature. effect of regularizationfor poorlyresolved regions in the The characterof the heterogeneityin the transition whole globe. region becomesmore evident by restrictingour interest Figure 5 is another expressionfor the uniquenessof to the westernPacific, where the negativepoloidal com- the transition region under the western Pacific. In this ponent (convergence)of the plate velocityfield is con- figure the correlationof lateral heterogeneityR(z, z') centrated [Lithgow-Bertelloniand Richards, 1998]. The betweentwo depthsz andz' [Tanimoto,1990b; Jordan et westernPacific is also one of the best resolvableregions al., 1993] is plotted for severalfixed valuesof z' at an for whole-mantleP arrival-time tomographybecause of equal interval of 170 km using models WEPP2 and the spatialdistributions of earthquakesand stations.We SAWI 2D in the westernPacific region. In order to make define,following Fukao et al. [1992], the westernPacific a direct comparisonbetween the two models, WEPP2 region as in Figure 3, including the Kurile arc to the was expandedinto degree-12 sphericalharmonics and northeastand the Sumatra arc to the southwest.Figure degree-13 radial Legendre polynomials.The function 4a showsthe RMS amplitudesof the four modelsin the R(z, z') describeshow heterogeneityat depthz' corre- western Pacific region as a function of depth, again lateswith that at depth z and thus givesa measureof the truncatedby degree24 for P wave models.These models continuity of structure in the vertical direction. Natu- indicate clearlya flat, low amplitudeof heterogeneityat rally, R = 1 if z = z'. The range of z in good correlation depthsof 1000-2000 km and a distinctminimum of the with heterogeneity at z' is shaded around z' and is upper mantle heterogeneityat a depth of about 400 km, bounded by the two solid lines. These solid lines show with a relative maximum in between. This is not an different trends acrossthe 970-km depth. The trend for artifact due to the truncation by degree 24, as demon- z' < 970 km indicatesthat the heterogeneitiesat two strated in Figure 4b. In Figure 4b the RMS amplitudes different depths in the transition region are mutually of P velocity anomaly are calculated for the original correlated even acrossthe 660-km discontinuitybut are block models, where the above features are even more anticorrelatedor uncorrelatedif one of the two depthsis emphasized.Thus the transition region is a laterally aboveor belowthe transitionregion. On the other hand, heterogeneousregion distinguishedfrom the uppermost the trend for z' > 970 km indicatesthat the heteroge- mantle. The bottom of this transition region may be neities at two different depths below the transition re- defined by the beginningof the flat, low amplitude of gion are mainly correlatedonly through the radial low- velocity heterogeneityin the middle lower mantle. We pass filter. Figure 5 thus demonstratesthat the long- note in Figures 2 and 4a that the RMS amplitudesat wavelengthheterogeneity in the transitionregion can be 39, 3 / REVIEWSOF GEOPHYSICS Fukao et al' SLABSIN THE MANTLE TRANSITION REGION ß 301 viewed as a verticallycorrelated unit distinctfrom those (a) rmsof S anomalies at shallowerand greater depths. 0.0 0.4 0.8 1.2 1.6 2.0 We note in Figure 4a that the RMS amplitude curve 0 • I of the P wave model of truncatedWEPP2 is very similar to that of the S wave model of SAW12D, exceptfor the amplitudedifference of a factor of 2. Figure 6 plots the 500 RMS amplitudesof fast anomaliesand slow anomalies separatelyfor these two models.The heterogeneityof the transition region in the western Pacific is by far dominatedby the fast anomaly,which takesits maximum 1000 in a depth range of 500-600 km, as we have already remarked in Plate 1. We will show in the next section that the vertical coherenceof long-wavelengthstructure (Figure5) and dominanceof the fastanomaly (Figure 6) 1500 in the transitionregion under the westernPacific are due + WEPP2 mainly to the existenceof deflectedor flattened slabsin [] P97 the transitionregion. 2000 SAW12D
...... S12WM13 3. SLAB IMAGES IN THE WESTERN PACIFIC
2500 Plates2-4 showthe slicesof the mantle alongprofiles shown in Figure 3 acrossSouthern Kurile, Japan-A, Japan-B, Izu-Bonin, Mariana, and Java. All of these slicesare alongthe sameprofiles as thoseof Fukao et al. 0.0 0.2 0.4 0.6 0.8 1.0 [1992]. The modelsshown are the P velocitymodels of rms of P anomalies WEPP2 and P97 in different intensityscales of 2.0 and rms of S anomalies 1.5% and the S velocity models of SAW12D and (b) 0.0 0.4 0.8 1.2 1.6 2.0 S12WM13, again in different scalesof 2.5 and 2.0%, 0 respectively.In graphical representationof the block modelsof P velocity,we have smoothedthese models by giving the velocity perturbation for each block only in f the middle of the block and then conductingregular 500 contouringthrough thesepoints. Each plate superposes the hypocentraldistribution of earthquakeswith magni- tudesgreater than 5.5 in a period 1964-1995 (basedon the ISC catalogues)within a band 50 km wide on both 1000 - sidesof the sectionplane. Followingis a brief descrip- tion of each cross section.
500 - 3.1. Southern Kurile (Plate 2a) + WEPP2 The deflected slab underneath Southern Kurile has been tomographicallyimaged by van der Hilst et al. [] P97
[1991],Fukao et al. [1992],and van derHilst et al. [1993], 2000 - SAW12D for which support was given by the residual sphere ...... S12WM13 analysisof Ding and Grand [1994]. Tajima and Grand [1995,1998] and Tajima et al. [1998]have shown that the deflected slab imagesbehind the Southern Kurile and 2500 - Japanese arcs are in qualitative agreement with the triplicatedbroadband waveforms of the P raysbottom- ing the transitionzone beneaththese areas. Oreshin et al. [1998] showed that the P arrivals from Kurile-Japan 0.0 0.2 0.4 0.6 0.8 1.0 seismicevents at a seismicnetwork in the Baikal region rms of P anomalies are consistentwith the deflected slab image, which, Figure 4. Same as Figure 2 but the region is limited to the however, could equally be explained by a possible western Pacific as defined in Figure 3. The P velocitymodels 530-km discontinuity.The two recentP velocitymodels are either (a) expandedby the degree-24spherical harmonics WEPP2 and P97 also show the deflected slab image or (b) untruncated. beneathSouthern Kurile. The detail of the slabimage is 302 ß Fukaoet al.: SLABSIN THE MANTLE TRANSITION REGION 39, 3 / REVIEWSOF GEOPHYSICS
WEPP2 SAW 12D
460 630 800 970 1140 1310 !480 460 630 800 970 1140 1310 1480
1 ooo
Figure 5. One-dimensionalprojections of R(z, z') in the westernPacific region (Figure 3) for z' = 460, 630, 800, 970, 1140, 1310, and 1480km at 170-kmintervals, i.e., correlationsof structureat depthz' with the structurespanning the rest of the mantle for modelsWEPP2 and SAW12D. The smallbar attachedto each curverepresents the positionat whichz - z' and R = 1. The area with R valuesgreater than 0.2 is shaded to emphasizethe main part of the curve.This main part is delineatedby the two solid lines,which show differenttrends in two z'- rangesacross 970-km depth. WEPP2 is expandedby the degree-12spherical harmonicsand by the degree-13radial Legendrepolynomials for easeof comparisonwith SAW12D.
rms of S anomalies somewhatdifferent, however:WEPP2 showsa single, 0.0 0.4 0.8 1.2 1.6 2.0 deflectedslab image while P97 has two separatedfast 0 • anomaliesin the samedepth range.Both the two long- wavelengthS velocitymodels, SAW12D and S12WM13, show the horizontal spread of a fast anomalyin the transitionzone, which may well be interpretedas the 500 - long-wavelengthfeature of the deflectedslab. In partic- ular, the qualitative agreementbetween WEPP2 and SAW12D is remarkable if the difference of cutoff wave-
1000 - lengthis taken into account(see alsoPlate 1).
3.2. Japan-Aand -B (Plates3a and 3b) Subductedslabs in the transitionzone beneath Japan WEPP2 .,--, 1500 - have been studied by seismictomography by Zhou -o- fast [1988],Spakrnan et al. [1989],and Karniyaet al. [1988]
O slow usingP wavedata and by Zhou and Clayton[1990] using P andS wavedata. The deflectedslab beneath Japan has 2000 - SAWI2D beenimaged, for instance,by van derHilst et al. [1991], fast Fukao et al. [1992],and van derHilst et al. [1993].Such an imageis clearlyseen in P velocitymodels WEPP2 and slow P97, as shown in Plates 3a and 3b. In Plate 3a, P97 shows 2500 - a fast anomalyextending upward from the leadingedge of the deflectedslab image. This anomaly,not existentin WEPP2, is subduedgreatly on the closelyadjacent cross I ' 0.0 0.2 0.4 0.6 0.8 1.0 section(see Plate 8a) and maynot be a real feature(see rms of P anomalies Figure2b of van derHilst et al. [1998]).Also in Plate 3a, both the P velocitymodels show a weak fast anomaly Figure 6. Separate plots of the RMS valuesof the fast and belowthe deflectedslab image that can be tracedsome- slowanomalies in the westernPacific region (Figure 3) for the how to the bottom of the mantle. modelsWEPP2 (truncatedby degree-24spherical harmonics) This deep fast anomalywas first noticed on a cross and SAW12D. See Figure 2 for other explanations. sectionof P97 alonga profile a little apart from profile 39,3 / REVIEWSOFGEOPHYSICS Fukaoet al' SLABSIN THEMANTLE TRANSITION REGION ß 303
(a) SouthAmerica A (b) SouthAmerica B (c) Aleutian
. •.
Widiyantoro 1.5 1.5 1.0
P97
SAW 12D 1.25
S12
. . slow • .. •...... fast
Plate5. Verticalmantle sections across the (a) ChileAndes, (b) PeruAndes, and (c) Aleutianarcs. The profilesare depicted inFigure 3.See Plate 2for other explanations. InPlate 5c the scale ischanged (1.0, 0.75, 1.25,and 1.0% for Widiyantoro [1997], P97, SAW12D, and S12WM13, respectively) toemphasize the contrast. 304 ß Fukao et al.' SLABS IN THE MANTLE TRANSITION REGION 39 3 / REVIEWS OF GEOPHYSICS
(a) Central America A (b) Central America B
Widiyantoro 1.5
P97 1.5 •I.5
SAWI2D 2.5 2.5
S12
slow fast
Plate 6. Vertical mantle sectionsacross the Middle America Trench along (a) profile A and (b) profile B (Figure 3). See Plate 2 for other explanations. 39, 3 / REVIEWS OF GEOPHYSICS Fukao et al.' SLABS IN THE MANTLE TRANSITION REGION ß 305
(a) TongaA (b) Tonga B
Widiyantoro 1..5 1.5
P97 .
SA 25
S12WM13 2.0
slow fast
Plate 7. Verticalmantle sections across the Tonga-Kermadecarc along (a) the Tongaprofile A and(b) the Kermadecprofile B (Figure3). See Plate 2 for other explanations. (a) JapanA' (b)
WEPP2• '•2.0 WEPP1 2.0
WEPP2 2.0
slow fast
(c)
P97 Widiyantoro,1997
_L
(D
o !
160 170 180 190 160 170 180 190
-1 5% +1.5%
Plate 8. Comparisonof P velocitymodels. (a) Crosssections across the Japanarc alongprofile A whichis closelyadjacent to profileA (Figure3). Modelsare WEPP2 (scale,2.0%) and P97 (scale,1.5%). The cross sectionfor P97 is the sameas that of vander Hilst et al. [1997]and Levi [1997]but with a differentscale. (b) Crosssections across the Ryukyuand Izu-Boninarcs along the profiledepicted by a dashedline in Figure3. Modelsare WEPPI [Fukaoet al., 1992]and its updatedversion, WEPP2 (AppendixA) (scale,2.0%). (c) Anomalypatterns at 550-kmdepth beneath the Tongaregion defined in Figure3 for the globalmodel P97 and the regionalmodel of l/Vidiyantoro[1997] (Appendix B) (scale,1.5%). 39, 3 / REVIEWS OF GEOPHYSICS Fukao et al.: SLABS IN THE MANTLE TRANSITION REGION ß 307
A (Figure 6b of van derHilst et al. [1997] and the second with the fast anomalyin the transitionregion (more than figure of Levi [1997]). Plate 8a reproducesthis cross 1.5%), to regardas evidence for "direct"continuation of section(profile A' in Figure 3) in comparisonwith that the subductedslab from the transitionregion. Assuming for WEPP2. As for profile A, the fast anomaly can be a homogeneityin composition,the P wave velocityper- traced somehow from the top to the bottom of the turbation of more than 1.5% in the transition region mantle. However, for both profiles, there is a marked correspondsto a temperatureperturbation of more than contrast in amplitude between the anomaly continuing 300øC and a density perturbation of more than 0.7%, directly from the Wadati-Benioff zone and the anomaly while the P wavevelocity perturbation of lessthan 0.5% in the rest of the lower mantle. Note that such a contrast in the midmantle correspondsto a temperaturepertur- is difficult to see in the figures of van der Hilst et al. bation of lessthan 170øCand a densityperturbation of [1997] andLevi [1997],where the anomalyis saturatedat less than 0.25% (evaluated at depths of 600 and 1500 an amplitude of 0.5%. The similarly saturated cross km, respectively,based on the figuresof Karato [1993]). sectionshave also been presentedmore recently along Thus the anomalyin the midmantle along profile A or apparentlythe sameprofile and alonga slightlydifferent A' is much weaker than the anomaly in the transition profile by Bijwaard et al. [1998] and Van der Voo et al. region not only in seismicvelocity but also in tempera- [1999a], respectively,to argue for the continuationof ture and density.Moreover, a fast anomalyof more than subductionbeneath Japan all the way downto the CMB, 1.5% changesrather abruptly into a weak anomaly of in supportof the earlier suggestionby van der Hilst et al. lessthan 0.5% along the extensionof the slab image, a [1997]. W. Spakman (personal communication,2000) feature not compatiblewith the experimental expecta- presented the same crosssections in a scale of 2.0%, tion of the gradual depth variations of temperature where the fast anomalyof more than 1.5% in the tran- derivativesof seismicwave velocities [Karato, 1993]. sition region changesrelatively suddenly to the anomaly Thus it would be difficult to regard the above localized, of lessthan 0.5% at greater depths,just as in WEPP2. not sheet-like, weak fast anomaly as the evidence for The detailedconfiguration of the deepweak anomaly "direct" slabcontinuation throughout the mantle. At the only slightly deviated from the reference model would same time we have to note that the relative difference in depend critically on, for example, the addition or re- lateral temperaturevariation betweenthe transitionre- movalof arrival-timedata (see,e.g., Plate 8b), the choice gion and the middle lower mantle tends to be smaller of regularizationscheme (e.g., either norm minimization than indicatedby the relative difference in lateral veloc- or roughnessminimization), and the reference model ity variation. employed(e.g., either a standard1-D Earth model as in The two P velocity modelsin Plate 3b show the very P97 or a laterally averaged version of the final 3-D long, deflected slab image whose eastern and western model as in WEPP2). For example, the deep weak parts lie above and across the 660-km discontinuity, anomaly in P97 is of narrow feature and is apparently respectively.This long deflected slab image splits into continuousthrough the mantle beneath Japan along two, one behind the Izu-Bonin arc and the other behind profile A' but not alongA, which are only 2ø apart from the Ryukyuarc, as it goesfarther to the south(Plates 4a eachother (Plates3a and 8a). This maybe interpretedas and 8b). Both S velocity models SAW12D and a localization of the weak fast anomaly into a narrow S12WM13 showthe horizontal spreadof a fast anomaly column [van der Hilst et al., 1997]. We, however,note in the transition zone, which may be interpreted as the that the deep weak anomaly in WEPP2 is of broader, long-wavelengthimage of the deflected slab. The de- mutually very similar feature along profiles A' and A. flected slab anomalyin the transitionzone has alsobeen Such a difference between P97 and WEPP2 indicates the indicated from the waveform analysisof broadbandP difficulty of resolvingthe detailed configurationof the waves [Tajima and Grand, 1998; Tajima et al., 1998]. weak anomaly only slightly deviated from the back- Plate 3b representsa good exampleshowing the consis- ground reference model. Forte and Woodward[1997] tency between different P models or different S models pointed out that the low amplitude of the midmantle and amongP and S models. heterogeneityimplies that its effect on seismicwaves will be largely obscuredby the much strongerheterogeneity 3.3. Izu-Bonin (Plate 4a) in the upper mantle and that this is the major reasonwhy There are many tomographicstudies for subducted the heterogeneitypatterns at 1000- to 2000-km depths slabsin the transition zone beneath Izu-Bonin [Zhou, are little correlatedamong the modelsinverted from the 1988;Spakman et al., 1989;Kamiya et al., 1988;Zhou and same seismic data with different constraints, as demon- Clayton, 1990; van der Hilst et al., 1991; Fukao et al., strated in Figure lb. The difference between P97 and 1992; van der Hilst et al., 1993]. As in some models in WEPP2 in the detailed configurationof the deep fast thesestudies, the long, deflectedslab has been imagedin anomalyalong profile A' (or A) may be understoodin both P velocity modelsWEPP2 and P97 at depthsnear this context. the 660-km discontinuity.In addition, the flattened slab Whatever the detailed configurationwould be, this image in these new modelsis split into two, one behind deep-seatedfast anomaly seemsto be too weak (less the Ryukyu arc and another behind the Izu-Bonin arc, than 0.5% in P wave velocity perturbation) compared although graphical smoothinghas somewhatobscured 308 ß Fukao et al.: SLABSIN THE MANTLE TRANSITION REGION 39, 3 / REVIEWS OF GEOPHYSICS
sucha feature in this particular crosssection (see also the 660-km discontinuity.The new P velocity models Plate 8b). seem to indicate that the penetrated slab flattens at The plate tectonichistory in the westernPacific [Seno depths around 900 km in a relatively poor resolution and Maruyama, 1984;Hall et al., 1995] seemsto suggest (see also Plate 4b of Bijwaardet al. [1998]). They also that the image of a flattened slabbehind the Ryukyu arc suggestpenetration of the subductedslab below the representsthe remnant of the Pacific slab before the 660-km discontinuityunderneath Philippine. There is no Pacific plate was replaced by the clockwiserotating indicationfor thesepenetrated slabs to descendfarther Philippine Sea plate. According to this interpretation downwardwell beyond the transition region. The long- [Hall et al., 1995],the Izu-Bonin-MarianaTrench at --•25 wavelengthS velocitymodels show the horizontalspread Ma was NW-SE trending and its junction with the of a fast anomaly in the transition region with relative Ryukyu-Nankai-JapanTrench (the TTT junction)was intensificationbeneath Mariana and Philippine in good located near Taiwan, so that the Pacific plate, rather agreementwith the resultsof P wave tomography. than the Philippine Sea plate, was subductedbeneath the Ryukyu-Nankai-JapanTrench. Since --•25 Ma, the 3.5. Java (Plate 2b) PhilippineSea plate with its boundaryat the Izu-Bonin- The penetration of the subductedslab below the Mariana Trench underwent clockwise rotation. This ro- 660-km discontinuityand its subsequentdeflection are tation progressivelymoved the TTT junction to the the two remarkable featuresin the tomographicstudies northeast along the Ryukyu-Nankai Trench. The for thisregion by Fukaoet al. [1992]and Widiyantoroand Ryukyu-NankaiTrench before the passageof the TTT van der Hilst [1996, 1997]. The slab penetration is also junction was a convergenceboundary between the Pa- indicated by Puspitoet al. [1993]. These features have cificand Eurasiaplates, where the formerwas subducted been confirmedby the new P velocity models,WEPP2 beneath the latter. Upon the passageof the TTT junc- and P97. Again, there is little indication for the pene- tion, the Ryukyu-NankaiTrench became a convergence trated slab to extend farther downwardmuch beyond a boundarybetween the PhilippineSea plate and Eurasian depth of 1200 km [see also Bijwaard et al., 1998]. The plate, with the former subductingbeneath the latter. The long-wavelengthS velocity modelsshow the horizontal plate subductedfrom the Ryukyu Trench thus changed spreadof a fast anomalywell below the 660-km discon- from the old, thick, and fast spreadingPacific plate to tinuity, which is in qualitative agreement with the P the young,thin, and relativelyslow spreading Philippine velocityimage of subhorizontaldeflection of the pene- Sea plate. According to this scenario,the already sub- trated slab. ductedPacific slab could be identified in seismictomog- We have seen, in the two recent P velocity models, raphyfrom the presentlysubducting Philippine Sea slab that subductingslabs in the western Pacific tend to with a possiblespatial gap between them. We suggest deflect subhorizontallyin and near the Bullen transition that in Plates 4a and 8b the deflectedslab image behind region, either above or across or below the 660-km the Ryukyu arc representsthe remnant of the already discontinuity.The deflection occurs,in general, in the subducted Pacific slab and that the one behind the forward (continentward)direction of subductionrather Izu-Bonin arc is the presently subductingslab of the than in the backward(oceanward) direction. Beneath Pacificplate (see alsovan der Hilst and Seno[1993] for Mariana, Philippine (Plate 4b), and Northern Kurile the interpretation of the flattened slab image). The (see Plate 1) the resolutionis not good enoughto see remnant Pacific slab lies above the 660-km discontinuity whether the slab simply deflectsor complexlyflattens on the oceanic side but below or across it on the conti- below the 660-km discontinuity.Gorbatov et al. [2000] nental side, in agreementwith the model of Bijwaard et obtained a much sharperimage of the subductedslab al. [1998] (basedon a figure by W. Spakman).In their beneath Northern Kurile by adding a large amount of model the presently subductingslab of the Philippine Russiandata to the ISC data. This slabimage penetrates Sea plate hasbeen imagedmore clearlyas a faster-than- the 660-kin discontinuityand extends down to about averagevelocity anomaly rather than a faster-than-sur- 900-kin depth with considerablecontortion. There is roundingvelocity anomaly as in P97 and WEPP2. Mod- little indication for the penetrated slabsbeneath Mari- els SAW12D and S12WM13 show the long horizontal ana, Philippine,and Northern Kurile to continuefarther spread of a fast anomaly in the transition zone, which downward"directly" to the deeper lower mantle. Thus may be understoodas the combinedeffect of thesetwo slabssubducted beneath the westernPacific are, in gen- flattened slabs. Plate 4a is another good example of eral, deflected or flattened in or near the Bullen transi- showingthe consistencyamong the four models. tion region. Some of them reside above or acrossthe 660-km discontinuityas in Southern Kurile, Japan, and 3.4. Mariana (Plate 4b) Izu-Bonin; slightlybelow the discontinuity,as in Mari- Earlier tomographicstudies in this region include ana, Philippine,and Northern Kurile; or well belowit, as those of Zhou [1988], Spakmanet al. [1989], and Zhou in Java.In a manner consistentwith this depthvariation, and Clayton[1990]. van der Hilst et al. [1991], Fukao et al. the long-wavelengthS velocitymodels showthe lateral [1992], and van der Hilst et al. [1993] have presented spreadsof fast anomalies (wider than the resolution tomographicimages of the subductedslab penetrating length of the models of about 1500 km) at different 39, 3 / REVIEWS OF GEOPHYSICS Fukao et al.' SLABS IN THE MANTLE TRANSITION REGION ß 309 depthsof the transition region and thus render support rms of S fast anomalies for the deflected or flattened slab image. At depths 0.0 0.4 0.8 1.2 1.6 2.0 • I • I • I shallowerthan the transitionregion, the P velocitymod- els,in general,show the narrow,fast anomalyassociated with the Wadati-Benioff zone, which is often sandwiched by broad slow anomalies in the surroundingmantle. 500 Since these fast and slow anomalies tend to be canceled out throughlow-pass filtering, it may not be surprising that the long-wavelengthS velocitymodels do not show a slab-relatedfast anomalyat suchdepths. lOOO Figure 7 plots the RMS amplitude of fast anomalies in the western Pacific as a function of depth for the original(not truncatedto degree24) P modelsWEPP2 and P97 and the long-wavelengthS models SAW12D •.• 15oo + WEPP2 and S12WM13. As noted before, all four models consis- tently indicatea relativemaximum of RMS amplitudeat [] P97 depthsof 500-600 km and a relativeminimum at depths 2000 SAW12D around 400 km. Thus the fast anomalyin the transition ...... S12WMI3 region is well identified from one in the uppermost mantle. This anomalouslayer is alsowell distinguished from the least heterogeneouspart of the mantle at 2500 depthsfrom 1000 to 2000 km. Taking into accountthe cross-sectionalimages in Plates 2-4, we suggestthat deflectedor flattened slabsare responsiblefor the dom- 0.0 0.2 0.4 0.6 0.8 1.0 inance of the fast anomalyin the transitionregion. rms of P fast anomalies
Figure 7. The RMS value of the fast anomalyin the western 4. SLAB IMAGES IN OTHER SUBDUCTION ZONES Pacific region (Figure 3) for the (untruncated) P velocity ALONG THE CIRCUM-PACIFIC modelsWEPP2 and P97 and the S velocitymodels SAW12D and S12WM13. See Figure 2 for other explanations. We will review in this sectionslab images in other subduction zones along the circum-Pacific,counter- clockwise from South America, Central America, Aleu- the 660-kmdiscontinuity (profile A) or belowit (profile tian, and Tonga. In what follows we replace WEPP2, B) and extendsfarther continentward,as remarked by whosetarget regionwas the westernPacific, with models Engdahl et al. [1995]. Although Engdahl et al. [1995] of Widiyantoro[1997], who did regional + whole mantle suggesteda possibilitythat the anomalyis smearedhor- tomographyfor severalselected subduction zones (see izontally in the EW direction, the images obtained by AppendixB). The rectanglesin Figure 3 indicatesome Widiyantoro[1997] using a more extensivedata set are of theseselected regions. Each region extendsvertically lessaffected by suchsmearing, as the resultsof his spike to a depth of 1600 km. In what followswe take the cross testsdemonstrate. The deflectedpart of the slab is well sectionsalong the sameprofiles as adopted by Widiyantoro reproducedin P97, althoughthe amplitudeof the anom- [1997].The data set usedby Widiyantoro[1997] is essen- aly is somewhatsmaller. There is no indicationof down- tiallythe sameas that usedby vander Hilst et al. [1997],yet ward continuationfor thesedeflected slabs beyond their the resolutionin a target regionis significantlydifferent, leadingedges. The deflectedslab image across•he Peru owingto the nature of differentmodel parameterization Andes(profile B; seealso Plate 4d of Btjwaard" et al. and numberof ray paths(see Appendix B). We take the [1998]) is well supportedby the long-wavelengthS ve- amplitudescale to be 1.5% for the model of Widiyantoro locitymodels that showthe predominantfast anomalyin [1997]as for P97 rather than 2.0% as for WEPP2. the relevantdepth range. Along the Chile Andes (profile A), on the other hand,the S velocitymodels do not show 4.1. South America-A and -B (Plates 5a and 5b) clearlythe fast anomalycorresponding to the deflected In an earlier work, Engdahl et al. [1995] produced slab image in the P velocitymodels. In South America, high-resolutionP wave tomographic images for the in general,the coincidenceamong the four modelsis not downgoingslab beneathwestern South America. Along as good as in the westernPacific. profilesA and B, taken closeto crosssections 3d and 3e Figure 8 plots the RMS amplitudeof fast anomalies of Engdahlet al. [1995], the regional + whole mantle in South America as a function of depth for the above model of Widiyantoro[1997] has resolved clearly the two (untruncated)P modelsand two long-wavelengthS strongfast anomalyassociated with the inclinedsubduct- models.An additionalplot is made for (untruncated) ing slab.This fast anomalydeflects at depthseither near WEPP2 whosetarget region was in the westernPacific. 310 ß Fukao et al' SLABS IN THE MANTLE TRANSITION REGION 39, 3 / REVIEWS OF GEOPHYSICS
rms of S fast anomalies part of the Caribbean region. Profile A (Plate 6a) 0.0 0.4 0.8 1.2 1.6 2.0 traverses the Middle America Trench and Lesser Anti- 0 • I lles with the Caribbean Sea in between. In the regional model of Widiyantoro[1997] the slab portion now sub- ductingfrom the Middle America Trench is imaged as a faster-than-surroundingvelocity rather than as a faster- 500 - than-averagevelocity, whereas such a feature is hardly seenin the whole mantle model P97. Both the P velocity models show a remarkable fast anomalythat dips down
1000 - at a moderate angle in a depth range correspondingto
.- .- the Bullen transition region and at a steeper angle at .- greater depths to about 1500 km. The plate tectonic history [e.g., Gordon and Jurdy, 500 - 1986] seemsto suggestthat this fast anomalyrepresents + WEPP2 the remnant slabof the Farallon or South Farallon plate -' Widiyantoro before its breakup into the Cocosand Nazca plates at 25 Ma. The faster-than-surroundingvelocity along the Wa- 2000 - [] P97 dati-Benioffzone at shallowerdepths is then interpreted SAW12D as the subductedslab of the Cocosplate. The Farallon
...... S 12WM13 slabimage is apparentlytruncated upward at a depth of 500-600 km. Although this apparenttruncation may be 2500 - due to the lack of resolution in the upper mantle, a direct continuation of the Farallon slab to the Cocos slab is unlikely from the plate tectonicpoint of view, accord- 0.0 0.2 0.4 0.6 0.8 1.0 ing to which the Cocosplate was born by fragmentation rms of P fast anomalies of the Farallon plate upon subduction of the ridge- transformfault system[Menard, 1978]. It would be more Figure 8. The RMS value of the fast anomaly in South reasonable to assume that the Farallon slab is discon- America (Figure 3) for the (untruncated)P velocitymodels nected from the Cocosslab by the subductedridge. We Widiyantoro[1997], P97, and WEPP2 and the S velocitymodels interpret the observedgap between the tomographic SAW12D and S12WM13. See Figure 2 for other explanations. imagesof the deep Farallon slab and the shallowCocos slab as a slab window formed by the ridge subduction [Thorkelson,1996]. The RMS of the three models of Widiyantoro[1997], The regional model of Widiyantoro[1997] in Plate 6a SAW12D, and WEPP2 showsa strong fast anomaly in showsa high-velocityzone extendingwestward farther the transition region well distinguishedfrom the fast deep beyond the leading edge of the Wadati-Benioff anomalyin the uppermostmantle (Figure 8). Model P97 zone from the Puerto Rico Trench and then deflecting exhibits a similar fast anomaly in the transition region above the 660-km discontinuity.Such a deflected slab but in the absence of a fast anomaly in the poorly image confirmsthe earlier observationby van der Hilst resolved uppermost mantle. The fast anomaly in the and Spakman [1989] and van der Hilst and Engdahl transition region in these models commonly possesses [1991].The deflectedpart of this slabis alsoindicated in two peaks,one abovethe 660-km discontinuityand the the whole mantle model of P97. Both long-wavelengthS other below it in a depth range of 900-1000 km. Among velocity models SAW12D and S12WM13 exhibit a hor- the two, the peak at the greater depth is higher than the izontal spread of a fast anomaly at depths below the one at the shallowerdepth. Plate 5 indicatesthat these 660-km discontinuityunder the Caribbean Sea, which two peaks correspondto the deflected slab images at may be due in part to the slabsunder the CaribbeanSea, depthsabove or acrossthe 660-km discontinuityand well but also due, to a considerableextent, to a smearing below it, respectively. effect of the image of the extensivelydeflected slab underneathSouth America (Plate 5b). 4.2. Central America-A and -B (Plates 6a and 6b) Profile B (Plate 6b) crossesthe Middle America The tomographicstudies in this region include Grand Trench and the Bahamasand is the sameprofile as that [1994], using S data, and van der Hilst and Spakman of van der Hilst et al. [1997]. In the P velocitymodels, as [1989] and van der Hilst and Engdahl [1991], using P noticed by van der Hilst et al. [1997] and Grand et al. data. They showedthe existenceof a fast anomaly pre- [1997], the mogt remarkable feature is the deep fast sumablyassociated with the past subductionat the Mid- anomalythat startsat depthsbelow 500 km, dipsdown at dle America Trench. The P wave models of van der Hilst a moderate angle in the transition region and at a and Spakman [1989] and van der Hilst and Engdahl steeper angle at greater depths, and reachesthe lower- [1991] revealed a subductedslab beneath the eastern most mantle. This down-dipping fast anomaly is also 39, 3 / REVIEWS OF GEOPHYSICS Fukao et al.' SLABSIN THE MANTLE TRANSITION REGION ß 311
seenin P velocitymodel WEPP2. This anomalymay be rms of S fast anomalies compared to an inclined zone of faster-than-average 0.0 0.4 0.8 1.2 1.6 2.0 velocityin the S velocity model SAW12D and two con- 0 • t oo centric fast anomaliesaligned along the relevant dip in S12WM13. Comparisonof Plates 6a and 6b indicates that the leading edge of this anomaly deepensnorth- 500 - ward. In both crosssections the slab image in the tran- sitionregion dips at a lower anglethan the slabimage in the deeper lower mantle. This deep anomaly continues
northwardto the deep anomalybeneath North America, 1000 - which altogether has been interpreted as imaging the Farallon graveyard of the slabs [Fukao et al., 1994; Grand, 1994; van der Hilst et al., 1997; Grand et al., 1997; Bijwaardet al., 1998] (see Engebretsonet al. [1992] for 1500- the conceptof lithosphericgraveyards). -•- WEPP2 Figure 9 plots the RMS amplitude of fast anomalies + Widiyantoro in Central America as a function of depth for the two + P97 (untruncated)P models and two long-wavelengthS 2000 - modelsin addition to (untruncated)WEPP2. The four SAW12D models,Widiyantoro, P97, SAW12D, and WEPP2, show ...... S12WM13
the relative maximumof a fast anomalyin the transition 2500 - regionwith the two peaksnear its top and bottom. This .. -. feature is similar to but less distinct than one in South -: :
America (Figure 8). The RMS amplitude of the fast ' i lø i ' i ' i ' anomalyin Central America decreasesmore slowlywith 0.0 0.2 0.4 0.6 0.8 1.0 increasingdepth from 1000 to 1500 km and remains rms of P fast anomalies larger at depthsof 1500-2000 km than in South Amer- ica. This feature may be diagnosticof the subductedslab Figure 9. The RMS value of the fast anomaly in Central extendingdeep beyond the transitionregion. America (Figure 3) for the (untruncated)P velocity models Widiyantoro[1997], P97, and WEPP2 and the S velocitymodels 4.3. Aleutian (Plate 5c) SAW12D and S12WM13. See Figure 2 for other explanations. Earlier tomographicstudies for upper mantle struc- ture beneath parts of the Aleutian islandshave been conductedby Engdahl and Gubbins [1987] and Abers and the down-dippingAleutian slab as the slabwindow [1994].Engdahl and Gubbins[1987] observedthat the createdby subductionof the ridgebetween the Kura and subducted slab beneath the central Aleutian islands ex- Pacificplates [Thorkelson,1996]. tendswell below the deepestseismicity to reach a depth of about 400 km. In Plate 5c the anomalyis, in general, 4.4. Tonga-Aand -B (Plates7a and 7b) weak, due in part to the lack of resolution.The Aleutian Previoustomographic investigations of the subducted section should be regarded as less reliable than other lithospherebelow the Tonga Trench usingP wave trav- sections. To emphasize the anomaly patterns, we el-time datawere carriedout by Zhou [1990]and van der changedthe amplitude scalearbitrarily for each of the Hilst [1995]. They observedthat the slabbeneath Tonga crosssections (such a changewas made only along the and Kermadec penetrates into the uppermost lower Aleutian profile). The regional model of Widiyantoro mantle.Zhao et al. [1997] made a detailedtomographic [1997] showsa clear image of the deflectedslab above studyof the Tonga slab,but the depth extentwas insuf- the 660-km discontinuity,as well as the image of the ficient to discussits ultimate fate. In Plate 7, profile A subductingslab from the deep-seatrench. Although the traverses,from east to west, the Tonga Trench, the Fiji other three modelsdo not showsuch a clear image of the islands,and the Vanuatu Trench. This is the sameprofile deflected slab, they commonly exhibit a horizontal as given by van der Hilst [1995]. Both the P velocity spread of weak anomaly at the correspondingdepths. models show the subductedslab image dipping down More recently, Gorbatov et al. [2000] added a large from the Tonga Trench and then deflectingsubhorizon- amount of Russian data to the ISC data in their travel- tally near the 660-km discontinuity.There is a kink on time tomographyto obtain a much sharperimage of the the deflected slab image, to the east of which it lies horizontal slab that spreadswidely beneath the Bering abovethe 660-km discontinuityand to the west of which Sea and is clearly detachedfrom the presentlysubduct- it lies below the discontinuity,in good agreementwith ing Aleutian slab, in strong support of the image ob- the result of van der Hilst [1995] (see also Plate 4c of tained by Widiyantoro[1997]. Gorbatovet al. [2000] in- Bijwaard et al. [1998]). The long, subhorizontallyde- terpretedthe significantgap betweenthe horizontalslab flectedslab imaged by the P wave tomographyis consis- 312 ß Fukao et al' SLABS IN THE MANTLE TRANSITION REGION 39, 3 / REVIEWS OF GEOPHYSICS
rms of S fast anomalies [Walcott,1987]. The strongeranomaly below the 660-km 0.0 0.4 0.8 1.2 1.6 2.0 discontinuitymay correspondin time to the period of I back-arcspreading (35-25 Ma) during which the Ker- madecTrench retreated rapidly [Walcott,1987] so that slab deflection might have occurred more extensively [van der Hilst and Seno, 1993;van der Hilst, 1995]. This 500 strongeranomaly seen in the P velocity models is re- markablyconsistent with one in the long-wavelengthS velocitymodels. The depth and horizontalextent of the lOOO fast anomalyare different betweenprofiles A and B in a very consistentmanner for all four models. Figure 10 plotsthe RMS amplitudeof fast anomalies in Tonga (Figure 3) as a functionof depth for the two 500 (untruncated)P models and two long-wavelengthS -•- WEPP2 modelsin addition to (untruncated)WEPP2. All five -' Widiyantoro models show clearly the relative maximum of the fast anomalyin the transitionregion, althoughthe detail is 2000 [] P97 different from model to model. Plate 7 indicates that this SAW12D relative maximum is due to the deflected slab in the
...... S12WM13 transitionregion. Note also that the lateral heterogene-
2500 ity is leastin a depth range from 1500 to 2000 km in all the models.
0.0 0.2 0.4 0.6 0.8 1.0 5. STAGNANT SLABS AND DEEPLY SINKING rms of P fast anomalies SLABS
Figure 10. The RMS value of the fast anomalyin the Tonga The tomographicmodels we have examinedindicate region (Figure 3) for the (untruncated)P velocity models that subductingslabs along the circum-Pacifictend to be Widiyantoro[1997], P97, and WEPP2 andthe S velocitymodels deflected or flattened in and near the Bullen transition SAW12D and S12WM13. See Figure 2 for other explanations. region with little evidencefor their "direct" downward continuationsto the deeperpart of the lower mantle.An exceptionto this general tendencyis Central America, tent with the horizontalspread of a fast anomalyin the where the Farallon slab directly continuesto the lower- transitionregion in the long-wavelengthS velocitymod- most mantle with its possibletailing edge in the transi- els. In the regional model of Widiyantoro[1997] a fast tion region(Plate 6b). We have suggestedin the previ- anomalybranches off from the flattened slabimage in a oussection that this Farallon slabis likely detachedfrom directionupward to the Vanuatu Trench. This anomaly the presentlysubducting Cocos slab with the subducted trends subparallel to the NW-SE trending Vanuatu ridge in between.Farther to the north, at latitudesfrom Trench, as indicatedin Plate 8c, and may imply a rem- 20ø to 30ø, where no subductionoccurs presently, the nant of the slab detached from this trench. Note that Farallon slab image can be traced upward to 500-km such a remnant slab image well delineates the low, depth but not up to 200 km [Bijwaardet al., 1998, Plate scattered deep seismiczone west of the main Tonga 1]. The Farallonslabs beneath North America havebeen Wadati-Benioffzone [Okal and Kirby, 1998]. imagedin the regionalS wave model of Van der Lee and Profile B is almost perpendicularto the Kermadec Nolet [1997] and in the globalP and S wave modelsas Trench. The inclined seismiczone is imaged only as a describedby Grand et al. [1997]. The regional S wave higher-than-surroundingvelocity anomaly rather than as model of Van der Lee and Nolet [1997] is intended to a higher-than-averagevelocity in the two P wavemodels. resolvethe Farallon slab in the upper mantle, while the The most remarkablefeature in the P velocitymodels is global S wave model of Grand et al. [1997] probably a strongfast anomalybelow the 660-km discontinuityin better resolves it in the lower mantle because of better the forward direction of the Wadati-Benioff zone. Al- path coveragethere. If we combinethe slab imagesin thoughthis stronganomaly does not smoothlycontinue thesetwo S wave modelsalong a latitude line at roughly upward to the Wadati-Benioff zone anomalyin the two 30øN, we obtain an image of the Farallon slab that P wave models shown in Plate 7b, they connect each reachesthe lowermostmantle but is upwardtruncated at other more continuouslyin the model of Bijwaardet al. a depth around 350 km. Van der Lee and Nolet [1997] [1998] (basedon a figure by W. Spakman).The plate interpretedthis truncationas a consequenceof breakage tectonichistory also suggestsa continuoussubduction of of the subductedslab in the Oligoceneepoch (38-24 the Pacificplate at the KermadecTrench since--•40 Ma Ma). Farther to the north, along the latitudinalline at 39, 3 / REVIEWSOF GEOPHYSICS Fukao et al.: SLABSIN THE MANTLE TRANSITION REGION ß 313
40ø, l/an derLee and Nolet [1997] identifiedthe horizon- the model of Bijwaardet al. [1998] showsthe deep fast tally orientedfragment of the subductedFarallon plate, anomalyat depthsbelow 1500 km beneathSiberia which without upward continuationabove the 410-km discon- Van der Voo et al. [1999a] interpreted as the Mongol- tinuity. Okhotskgraveyard of the slabs.Kaneshima and Helffrich Besides Central America, Japan has often been [1998, 1999] detected a dipping low-velocitylayer at quoted as one of the regions suggestingthat tectonic depthsof 1400-1600 km which they interpreted as the platesdescend "continuously" through the mantle [e.g., fossil oceanic crust associated with ancient subduction of Levi, 1997]. Such suggestionsare based mainly on the the Indonesianslab long before the Eoceneplate reor- crosssection of P97 acrossprofile A' in Figure 3. We ganization.Thus almost all the reported slab images haveargued, however, that the deep fast anomalyin this deep in the lower mantle do not continueupward to the crosssection (Plate 8a) is too weak, comparedwith the surfaceplates or connectdirectly to the presentlysub- fast anomalyin the transition region, to regard as evi- ductingslabs. dence for "direct" continuation of the subducted slab Figure 11 summarizesschematically the slab config- from the transition region. urations discussedabove. This is a highly caricatured Thus the region-by-regionexamination of the slab illustration.For example,although the Izu-Bonin slabin configurationin the existingtomographic models indi- this cartoonis deflectedjust abovethe 660-km disconti- catesthat circum-Pacificsubducting slabs tend to deflect nuity,the actualslab image (Plate 4a) is more complex, or fiatten in and near the Bullen transitionregion with- whichis not entirelyabove the discontinuitybut slightly out direct continuationto much greater depthsin the acrossit. Although the slab configurationsin Figure 11 lower mantle. Such a feature can be seen beneath South are groupedinto three types,forward deflection (deflec- America, Puerto Rico, Aleutian, Kurile-Japan-Izu-Bo- tion towardthe subductiondirection), backward deflec- nin-Mariana, Ryukyu-Philippine,Java, and Tonga-Ker- tion (deflectionin the oppositedirection), and complex madec. Besidesthese typical oceanicsubduction zones, flattening(e.g., flatteningof a slabin both the forward continentalsubduction of the Indo-Australian plate is and backwarddirections), this groupingis basedsolely occurringin Himalaya and the Bay of Bengal.Bijwaard on the apparent configurationof a tomographic slab et al. [1998] and Van der Voo et al. [1999b] presented image,whose reliability is very different from region to crosssections showing subhorizontal deflection of the region. For the deflected or flattened slabs below subductedslab above the 660-km discontinuitybeneath 660-km depth, it is, in general, difficult to define Himalaya and Hindu Kush, respectively.It may be con- uniquelytheir bottom depthsbecause of the gradationof cluded that the Pacificplate, the Nazca plate, the Phil- anomaly amplitude. The bottom depth given in Figure ippine Sea plate, and the Indo-Australian plate now 11 is an averageof rather subjectivemeasurements of activelysubducting are deflectedsubhorizontally or in slabbottom depths for the four modelsin Plates2-7. For other casesare flattened to lose their slab shapesin a example,we measuredthe bottom depths of the Java depth range correspondingto the Bullen transition re- slab in Plate 2b to be 1300, 1245, 960, and 1075 km for gion. modelsWEPP2, P97, SAW12D, and S12WM13, respec- On the other hand, we have already seen that the tively, with an average of 1145 km, and hence the slab Farallon slab beneath Central and North America ex- bottom is placed at 1200-km depth in Figure 11. We tends deep in the lower mantle but does not continue should also point out in Figure 11 that where the reso- upwardto the surfaceplate or to the presentlysubduct- lution is not good enough, either slab continuationor ing slab.The recent tomographicmodels also showthe detachment is assumedbased on seismicityor plate EW trendingfast anomalyin the lower mantle beneath tectonic history. For example, as discussedbefore, we Himalaya and the Bay of Bengal that has been inter- assumedthat the Cocosslab subducting beneath Central pretedas the remnantIndian (Tethys)slab [Fukao et al., America is separatedfrom the deeply sinkingFarallon 1994; Grand, 1994; van der Hilst et al., 1997; Grand et al., slab.Similarly, the youngPhilippine Sea slabsubducted 1997; Van der Voo et al., 1999b].This slabimage can be beneaththe Ryukyuarc hasbeen assumedto be distinct traced upwardto a depth of 800 km beneathHimalaya from the remnant deep Pacific slab. The Pacific slab [Grandet al., 1997, Figure 2; Bijwaardet al. [1998, Plate subducted from the Kermadec Trench was assumed to 1] but is disconnectedboth vertically and horizontally be continuousfrom the surface to depths below the from the presentlysubducting slab image of the Indo- 660-kmdiscontinuity, based mainly on the plate tectonic Australian plate that is deflected above the 660-km history.The crosssection in a more recent tomographic discontinuity[Bijwaard et al., 1998,Figure 11c].(Van der model of Bijwaardet al. [1998] showsa somewhatmore Voo et al. [1999b] suggestedthat the upper and lower continuousslab image alongthe sameprofile than those slabimages are connectedto eachothers at a "pinpoint" in Plate 7b (basedon a figure by W. Spakman).The in the vicinity of Hindu Kush.) Farther to the east, horizontalslab widely spreadingbeneath the Bering Sea beneaththe Bay of Bengal,the Indian slabimage can be was assumedto be the remnant slab of the Kula plate seenat depthsbelow 1050 km [Grandet al., 1997,Figure followingthe interpretationby Gorbatovet al. [2000]. 2] but not at a depthof 900 km [vander Hilst et al., 1997, On the basisof Figure 11 we may classifysubducted Figure l c]. Besidethese two well-knownremnant slabs, slabsinto two categories.Slabs in the first categoryare 314 ß Fukao et al.: SLABS IN THE MANTLE TRANSITION REGION 39 3 / REVIEWS OF GEOPHYSICS
depth (km)
Pa Pa Pa Pa Pa P P Pa Ph Ph Nz Nz Co
200
400
600
800
1000
1200
1400
• forwardslab deflection Pa:Pacific slab In- Au: Indo- Australian slab • backwardslabdeflection Ph:PhilOpine slab In' Indian (Tethys) slab •complexslab flattering Nz:Nazca slab Fa: Farallon s lab Co' Cocos s lab Ku: Kula s lab
Figure 11. Schematicillustration of slab configurations.This cartoon summarizesthe tomographicimages of stagnantslabs and deeplysinking slabs but involvesmany simplificationsand assumptionsas cautionedin the text. The slab images under North America, Himalaya, and Bay of Bengal are based on the figures presentedby van der Hilst et al. [1997], Grand et al. [1997], and Bijwaardet al. [1998] (see text for details).
connectedto the surfaceplates and tend to be deflected Jordan, 1997;Bunge et al., 1998] (see Lay [1994] for a or flattened in a depth range correspondingto the reviewof earlierwork). Many of the argumentsin favor Bullen transitionregion. The bottom depthsdiffer con- of slab continuitythrough the mantle have been based siderablyamong different subductionzones. Most of the on a very limited number of crosssections of a tomo- presentlysubducting slabs are in this category.We call graphic model. Such cross sectionshave often been these slabs "stagnant slabs" when it is necessaryto presented in a saturated scale so that one could not emphasizetheir geodynamicsignificance, as will be ex- distinguisha strong anomaly in the transition region plained in the next section[Fukao et al., 1992]. Slabsin from a weak anomaly in the deep lower mantle. The the secondcategory are disconnectedfrom the surface purposeof this paper is to make a comprehensivereview plates and are now deeply sinking through the lower of the cross-sectionalimages of subductedslabs along mantle. Their tailing edgesmay still be within the tran- the circum-Pacificfor severaltomographic models. For sitionregion, as in the casesof the Farallon and Tethys this purposewe do not impose a cutoff amplitude for graveyards.It may be difficult to classifythe remnant illustratingtomograms in an attempt to extract the slab Pacificslab behind the Ryukyu arc and the remnant Kula imagesthat are consistentamong different models.We slabbehind the Aleutian arc into the abovetwo catego- have not attemptedto searchfor "best" profilesfor the ries. present purpose but simply have employed the same profiles as in our previousstudies [Fukao et al., 1992; Widiyantoro,1997]. 6. DISCUSSION Our systematicsurvey has demonstrateda remark- able tendencyfor subductedslabs to flatten in the tran- There have been considerable debates about the con- sition region. Support for the flattened slabs in the tinuity of subductedslabs through the mantle or, in a transitionregion comesfrom observationsof precursors more general term, about the stratificationof mantle to SS waves [Shearerand Masters,1992; Shearer,1993; convection[e.g., van der Hilst et al., 1997; Pusterand Flanagan and Shearer, 1998]. Flanagan and Shearer 39, 3 / REVIEWS OF GEOPHYSICS Fukao et al.: SLABS IN THE MANTLE TRANSITION REGION ß 315
[1998] reported that the largestdepressions (about 20 "direct" slab continuationto greater depthsbeyond the km) in the 660-kmdiscontinuity occur beneath the Japan transitionregion. This implies a relativelyimpermeable and Kurile slabs,west of the Izu-Bonin slab, west of the nature of the transitionregion, through which vertical Tonga Trench [see also Roth and Wiens, 1999], and motion of subducted slabs tends to be inhibited and under South America, whereasmore moderate depres- thereby converted into predominantlyhorizontal mo- sions(about 10 km) are seenbelow the Philippineand tion. Sucha nature of the transitionregion may not be Java Trenches.These depressionsindicate the areas of fully explainedby the experimentallyexpected endother- lower-than-averagetemperatures, which are correlated mic reactionat the 660-km depth and a viscosityincrease remarkablywell with the areasof flattened slab images acrossit alone. Some additional features may be re- [Flanaganand Shearer,1998] (seea remarkby Vinniket quired,such as significant reduction of viscosityby which al. [1996]about the arealextent of the depression).Such verticalflow couldeasily be changedinto horizontalflow a correlation is almost absent for the 410-km disconti- [•7adeket al., 1997;Cserepes and Yuen, 1997; •7adek and nuity [Flanaganand Shearer,1998, 1999]. This difference Van den Berg, 1998]. between the two discontinuitiessuggests that the tem- We have learned that whatever the detailed mecha- perature anomaliesassociated with the subductedslabs nismwould be, subductedslabs entering into the transi- are not confined to the inclined part of the Wadati- tion regiontend to be oncestagnant and then to spread Benioff zone but extendhorizontally along the flattened subhorizontallyin this region.We examinethe possible part of the slab imagesat depths around the 660-km relevanceof this nature of the transition region to the discontinuity. subductionhistory of plates.The plate tectonichistory The 660-km discontinuityis a major endothermic experienceda major reorganizationof global plate mo- phase boundaryof the mantle [Ringwoodand Irifune, tion from Mesozoicto Cenozoicpatterns that occurred 1988; Ito and Takahashi,1989; Bina, 1991;Akaogi and primarily during the Eocene epoch (53.5-37.5 Ma) Ito, 1993; Bina and Helffrich, 1994; Ruble and Brearley, [Rona and Richardson,1978]. The reorganizationin- 1994;Stixrude, 1997]. The good correlationbetween the volved fragmentation of the Farallon plate into two 660-km depressionand the tomographicimages of the plates (North and South Farallon plates) at about 48 flattened slabis at least in part interpretedby the endo- Ma, upon which its convergencerate against South thermic phase reaction [Takenakaet al., 1999]. There America decreasedby a factor of about 2 [Lithgow- have been many numericalexperiments of thermal con- Bertelloniand Richards,1998]. This fragmentationwas vectionthat take into accountthis phase reaction [Chris- roughlysynchronous with the disappearanceof the Kula tensen, 1989; Macbetel and Weber, 1991; Peltlet and Sol- plate from the Earth's surface,which was followed by heim, 1992; Tackley et al., 1993; Honda et al., 1993; the northwardsubduction of the Pacificplate [Lithgow- Solheimand Peltlet, 1994b].According to thesestudies, Bertelloniand Richards,1998]. The plate reorganization cold downwellingfluid tends to stagnateabove the en- alsoinvolved reorientation of motionof the Pacificplate dothermicphase boundary to spreadalong it, a feature with the large N-S componentto the large E-W compo- that appearsto be qualitativelyconsistent with the hor- nent at 43 Ma, as marked by the Hawaiian-Emperor izontal spreadsof subductedslabs in the transition re- bend [Gordonand Jurdy,1986]. Since43 Ma, structures gion. There is, however,a noticeabledifference in the inferredto havebeen other than subductionzones (e.g., induced disturbancebetween the computer-simulated transformfaults) during the late Mesozoic along the thermal convectionand the mantle flow pattern inferred western margin of the South Pacific and North Pacific from seismictomography. In the computer-simulated becamesubduction zones including the Philippine,Bo- convection,cold downwellingmaterials once pile up on nin, Marianas, Yap, Palau, and Tonga trench-arc sys- the endothermicphase boundary to eventuallyflush into tems [Rona and Richardson,1978]. The rate of relative greater depths [e.g., Brunet and Macbetel, 1998]. This motion betweenthe India and Eurasiaplates decreased situation makes the vertical correlation minimum be- by continentalcollision by a factor of approximately2 tween two depthsacross the endothermicphase bound- fromabout 10 to 5 cmyr -• through53 to 40 Ma [Rona ary with or without a viscositycontrast across it [Puster andRichardson, 1978]. At aboutthe time of the collision, andJordan, 1994, 1997] (see also discussion by Masters et and possiblyas a direct result of it, the Indian and al. [1996]).In contrast,the verticalcorrelation of seismic Australian plateswere fused to form the modern Indo- disturbancesin the westernPacific (Figure 4) showsno Australianplate [Scoteseet al., 1988].Australia began to such minimum acrossthe 660-km discontinuity,but is drift northwardby relativelyrapid spreadingalong the highthroughout the transitionregion, while it is very low Australia-Antarcticplate boundary[Cande and Mutter, between the transition region and the deeper mantle. 1982]. This event was synchronouswith the reactivation The high correlationthrough the transitionregion is a of subductionbelow the Sundaarc [Audley-Charleset al., long-wavelengthexpression for the horizontalspread of 1988; Scoteseet al., 1988]. In the Caribbean region a subducted slabs either above or across or below the pattern of N-S compressionand subductionthat had 660-km discontinuity,rather than only aboveit. The low prevailedsince about 110 Ma at the northern and south- correlationbetween the transitionregion and the deeper ern boundariesof the Caribbeanplate changedto pre- mantle may be diagnosticof the rare occurrenceof dominant E-W strike slip motion from 54 to 38 Ma 316 ß Fukao et al.: SLABSIN THE MANTLE TRANSITION REGION 39, 3 / REVIEWS OF GEOPHYSICS
PresentDay Depth72.50 (km) densityheterogeneity due to the presumed subducted slabsas a functionof depth [seealso Deparis et al., 1995; Wenand Anderson, 1995]. The model of Lithgow-Bertel- loni and Richards [1998] is reproducedin Figure 12, showing a dramatic change in heterogeneitypattern acrossa depth interval between 940 and 1380 km, al- Depth 362.50 (km) thoughthese depthsare thosebased on many assump- tions and give just rough measures.The heterogeneity pattern at 940 km is due largelyto subductedslabs after the plate reorganization,which is dominated by long, continuousbands of slabanomalies essentially along the presentsubduction zones. The pattern at a depthof 1380 __ Depth 652.50 (kin) km is due mostlyto subductedslabs in the epochof the plate reorganization,which is characterizedby a global scatter of slab anomaly patches.Among these patches the relatively pronouncedtwo are the anomaly of the Farallon graveyardbeneath North and Central America and the anomalyof the Tethysgraveyard beneath India. Depth 942.50 (km) The slabsin the firstcategory (slabs connected to surface plates and now deflectedin the transitionregion) are thosewhich, accordingto Figure 12, startedor restarted subductionafter the Eocene plate reorganization.The slabsin the secondcategory (slabs disconnected from surfaceplates and now deeplysinking through the lower Depth 1377.50 (km) mantle) are thosewhich startedsubduction before the reorganization and were disconnectedfrom surface plates during the reorganizationor after it. As we have reviewed,the presentlysubducting slabs are largely those after the Eocene plate reorganization, which tend to be stagnantto spread subhorizontallyin Depth 2102.50 (km) the transitionregion. We suggestthat this is the quasi- stationarymode of subductionin a sensethat subduction before the Eocene reorganizationhad occurredalso in this mode. The pre-Eocene deep subductionin this mode had accumulated slab bodies in the transition region,which might haveeventually fallen into the deep Depth 2827.50 (km) lower mantle by the inducedRayleigh-Tayler type grav- itational instability. Once the instability of this type occurredin someregion, it might have propagatedrap- idly to the whole subductionzones. We suggestthat the Eocene plate reorganization is a surface manifestation of the worldwide occurrenceof this dynamicinstability that led to the globalfall of accumulatedslab bodies into the deep lower mantle. We argue that this global fall of Figure 12. Mantle density heterogeneitymodel based on subductionhistory [Lithgow-Bertelloniand Richards, 1998]. subductedslabs changed the slab pull forcesand hence Contouredregions represent areas of high densitywith respect the internal balanceof torque acting on surfaceplates to the surroundingmantle, in other words, the location of the [Solomon and Sleep, 1974; Forsyth and Uyeda, 1975; subductedslabs at that depth. In the top panel the plate Solomonet al., 1975; Chappleand Tullis, 1977], hence boundariesand the continentaloutlines in their presentposi- leading to the reorganization of global plate motion. tion have been superposedfor reference. Upon thisglobal change many (but not all) of the deeply subducted slabs were detached from their shallower part, either at the Earth's surface or at intermediate [Rona and Richardson,1978; see alsoRoss and Scotese, depths.Many of the plates left on the Earth's surface 1988]. began to move differently from the detached deeply The aboveseries of the Eocene reorganizationevents subducted slabs. The former became the slabs in the first is well reflected in the density heterogeneitymodel of category,and the latter representthe slabsin the second the mantle by Lithgow-Bertelloniand Richards [1998], category.An exampleof the latter is the Indian (Tethys) who mapped, based on the plate tectonic history, the slab. Upon the Eocene event this slab was probably 39 3 / REVIEWS OF GEOPHYSICS Fukao et al.: SLABS IN THE MANTLE TRANSITION REGION ß 31 7
Pre-Eocene Eocene Post-Eocene Present .• PA EU
NW Pacific P.•.•__AEUP•A • EUPA -)
•-• FA • P•[ NA
N. America
FA N•Z NZ• SA N•Z S. America
ß
FA: Farallon, PA: Pacific, NZ: Nazca, EU: Eurasia, NA: N. America, SA: S. America
Figure 13. Cartoon speculativelyillustrating what might have happenedacross the Eocene epochof global plate reorganizationin the northwestPacific, North America, and SouthAmerica. The shadedzones represent the upper and lower mantle transitionregion. The arrowsindicate the convergencerate of an oceanicplate againstits counterpartcontinental plate. In the Eocene epoch the NS-dominantmotion of the Pacificplate changedinto the EW-dominantmotion. The Farallon plate was broken up into the northern Farallon plate and the southernFarallon (Nazca) plate. The convergencerate of the Nazca plate againstSouth America becameabout half of that of the precedingFarallon plate. The convergencerate of the Farallon plate against North America did not change significantlyin the Eocence epoch, but the convergencestopped in the post-Eocenceby collisionof the ridge system.In this cartoonthe pre-EocenePacific slab under the northwest Pacific and the pre-Eocene Farallon slab under South America are suggestedto have already sunk to the lowermost mantle.
detachedfrom the surfaceplate (Indo-Australianplate), curred in the northwestern Pacific, South America, and the convergencerate of which became only half of that North America. of the precedingIndian plate. The detachedslab contin- The abovescenario implies that the transitionregion, ued to sink through the transitionregion, and its tailing in general, strongly inhibits mantle downflow but can edge is now at depths of 800-1000 km (Figure 11), becomepermeable intermittently to causeglobal change although the actual descendinghistory is likely to be in plate motion. Stagnancyand intermittencyare the two more complex[Van der Voo et al., 1999b].The Farallon closelylinked phenomenathat may characterizeflow in slab in North and South America, on the other hand, the mantle [Christensenand Yuen, 1984, 1985;Machetel remainedattached to the surfaceplate (Farallon plate) and Weber, 1991; Peltier and Solheim, 1992; Bercovici et through the Eocene epoch,where the convergencerate al., 1993; Honda et al., 1993; Tackleyet al., 1993, 1994; changed little [Lithgow-Bertelloniand Richards, 1998]. Steinbach and Yuen, 1994; Solheim and Peltier, 1994a, The Farallon slab was later detached from the surface 1994b;Machetel et al., 1995; Tackly, 1995b;Brunet and plates (Pacificand Cocosplates) by the ridge subduc- Machetel, 1998]. For recent discussionsabout the layer- tion. The detached slab continued to sink through the ing of the mantle flow and its relevance to seismic transitionregion, but its tailing edge is still in the upper tomography,geoid, topography,and heat flux anoma- mantle. This sinkingslab dips at a shallowerangle in the lies, we refer to Le Stunff and Ricard [1995, 1997], transitionregion than in the deep lower mantle (Plates Thoravalet al. [1995], Forte and Woodward[1997], Tho- 6a and 6b), a feature consistentwith the idea of relative raval and Richards [1997], Wen and Anderson [1997], impermeabilityof the transitionregion. Figure 13 is a •adek et al. [1997],Pari andPeltier [1995, 1998], and schematic summary illustrating what might have oc- Christensen[1998]. 318 ß Fukao et al.: SLABS IN THE MANTLE TRANSITION REGION 39, 3 / REVIEWS OF GEOPHYSICS
7. SUMMARY APPENDIX A: WEPP2, A TOMOGRAPHIC MODEL FOR THE WESTERN PACIFIC We have reviewed the tomographicimages of sub- ducted slabsshowing, in general, subhorizontaldeflec- Model WEPP2 [Obayashiet al., 1997] is an update of tion in the upper and lower mantle transition region the P velocitytomographic model for the westernPacific acrossthe 660-km discontinuity.The review resultsmay by Fukao et al. [1992] (abbreviatedas WEPP1). Their be summarized as follows. main differenceis the volume of the data. Obayashiet al. 1. In the mantle of areasof plate convergence,seis- [1997] selected 12,000 earthquakeswith magnitudes mic heterogeneitytakes a relative maximumin a depth greater than 5 from the bulletins of the International range correspondingto the Bullen transition region, SeismologicalCentre for a period of 28 yearsfrom 1964 distinctfrom the strongheterogeneity in the uppermost to 1991 in an effort to obtain an epicentralcoverage as uniform as possible.This selectionresulted in a data set mantle and minimal heterogeneityat depths of 1000- of about 2 million P first arrival times, which is 5 times 2000 km. The heterogeneityin the transition region is aslarge asone usedfor WEPP1. Thesetravel timeswere dominated by fast anomalies due to flattened slabs correctedfor ellipticity [Dziewonskiand Gilbert, 1976] above or acrossor below the 660-km discontinuity. and elevation of the stations.Obayashi et al. [1997] 2. For the Pacificplate, slabflattening occurs either parameterized,as in the work by Fukao et al. [1992],the above and acrossthe 660-km discontinuity(Southern whole mantle structureby blocks,with finer onesin the Kurile, Japan to Izu-Bonin, and Aleutian) or slightly western Pacific. The mantle is divided into 32, 64, and 16 below the discontinuity(Northern Kurile and Mariana) basic blocks in latitude, longitude, and radius, respec- or above,across, and below it (Tonga-Kermadec).For tively. The horizontalcell size is 5.625ø x 5.625ø, except the Nazca plate, deflection occurs either above and for the polar region, where Obayashiet al. [1997] em- acrossthe discontinuity(Chile Andes) or well below it ployeda somewhatlarger cell size.The radial divisionis (Peru Andes). For the Indo-Australianplate, deflec- the same as that of Inoue et al. [1990], where the layer tion occurseither well below the 660-km discontinuity thicknessvaries from 29 km just belowthe surfaceto 334 (Java) or above it (Himalaya). There is little indica- km just above the CMB. Beneath the western Pacific, tion for these slabs to continue "directly" to the Obayashiet al. [1997] gradually subdividedthe basic deeper lower mantle. The tendency of subhorizontal blocksinto smallerblocks in the sameway asFukao et al. deflection or flattening of subducted slabs indicates [1992].The sizeof the smallestblock is one quarter of a that the transition region is understoodas a relatively basic block in latitude and longitude and one half in impermeable layer against slab penetration. We do depth. The total number of blocksis 51,383. not necessarilymean by "the relatively impermeable A travel-time residual is causedby the sourcemislo- layer" a layer with distinct physicalproperties but a cation and mantle heterogeneityso that its relation can be written as depth range in which subducted slabs tend to be flattened by some mechanisms. 3. There are a group of lithospheric slabs now deeply sinkingthrough the lower mantle. They include the remnant slab of the Farallon plate beneath North where and Central America and that of the Indian (Tethys) B matrix of hypocenter-relatedpartial derivatives; plate beneath Himalaya and the Bay of Bengal. These G matrix of P slowness-relatedpartial derivatives; deeply sinking slabsare not connectedto the surface 8h vector of hypocenterand origin time plates or to the presently subductingslabs. The Far- perturbations; allon slab in the transition region dips at a shallower Ill vector of P slownessperturbations; angle than its downward continuation, in support of vector of P travel-time residuals. the idea of relative impermeability of the transition region. Equationsare solvedfor 8h and m iteratively(not si- 4. The aboveresults suggest that mantle downflow multaneously).The systemof equationsfor model pa- is relatively stronglyinhibited in the transition region rametersm is solvedwith the conjugategradient (CG) and is in large part changedinto horizontalflow there. method by imposing the first-order smoothnesscon- However, inhibition of downflow does not seem to be a straint to suppressartificial fluctuation of the solution persistentfeature: Tomographicevidence and plate tec- [Inoueet al., 1990]: tonic history suggestthat extensiveslab penetration mighthave occurred through the transitionregion in the Eocene epoch of the plate reorganization.Many (but not all) of the subductedslabs were detachedfrom the The relativeweight between the data equationin the surfaceplates and beganto sinkindependently upon the top row and the smoothnessequation in the bottom row Eocene plate reorganization. was taken to be the same as that of Fukao et al. [1992]. 39, 3 / REVIEWSOF GEOPHYSICS Fukao et al.: SLABSIN THE MANTLE TRANSITION REGION ß 319
By the nature of the systemof equations,the final model resolutionsof large-scalestructures such as subducted dependslittle upon the initial model, althoughthe con- and deflected slabsare good, except for the Scotia re- vergencerate dependsupon it. Startingwith WEPP1 as gion, where data coverageis poor. The resolutiongen- the initial model, changesof the hypocentersand the erally degradesin mantle regions away from the slabs model became negligibleafter two iterations. due to decreasingsampling by seismicrays. The resolu- Plate 8b is a comparisonof the crosssections along tions of structuresin the upper mantle, in particular the same profile (dotted line in Figure 3) acrossthe underneathback arc regions,are improvedby the inclu- northern end of the Ryukyu arc between WEPP1 and sion ofpP andpwP phases.For further explanationson WEPP2. WEPP2 showsup two separate,flattened slab the detailed implementationof P wave tomographyin images,one behind the Izu-Bonin arc and another be- each region being investigated,readers may refer to hind the Ryukyu arc, while in WEPP1 these two are Widiyantoro[1997]. indistinguishableand are blurred into a single,flattened Plate 8c is a comparison of the map view of the slab image. WEPP1 showsscattered fast anomaliesin heterogeneityat a depth of 550 km in the Tonga-Ker- the lower mantle, many of which have disappearedin the madecregion betweenP97 and Widiyantoro[1997]. The new model WEPP2. The model parametersof WEPP2 regional model of Widiyantoro[1997] showsa zone of are availableat http://ohpdmc.eri.u-tokyo.ac.jp/. fast anomaly subparallelto the plate boundary with a hairpin curve at the northern end of the Tonga arc and with a sharpbend at the southeasternend of the Vanu- APPENDIX B: REGIONAL TOMOGRAPHIC MODELS atu arc. It is remarkable that deep seismicityoccurs all FOR THE CIRCUM-PACIFIC BY WIDIYANTORO along this fast anomalyzone rather than only along the [1997] Wadati-Benioff zone to the west of the Tonga Trench [Okal and Kirby, 1998]. Suchdetails are not resolvedin FollowingFukao et al. [1992],Widiyantoro and van der the globalmodel (P97) of van der Hilst et al. [1997]. Hilst [1996, 1997] successfullyimaged the lithospheric slabstructure beneath the Indonesianregion by employ- ing a combinationof a high-resolutionregional and a ACKNOWLEDGMENTS. We thank Hitoshi Kawakatsu, low-resolutionglobal inversion. Widiyantoro [1997] then Craig Bina, Barbara Romanowicz,and Lev Vinnik for their constructive comments. We also thank Al Plueddemann, John applied a similar techniqueto investigatemantle struc- ture beneath subduction zones around the Pacific di- VanDecar, Suzan Van der Lee, and Wim Spakmanfor their thorough reviews,which were very helpful in improving the vided into 10 different study areas includingthe Indo- manuscript.Wim Spakmankindly sent us numerousfigures of nesianregion. Widiyantoro[1997] discretizedthe entire slicesthrough the model of Bijwaard et al. [1998] that he mantle by means of local basisfunctions, i.e., a uniform preparedfor us. This work was supportedby the Grant-in-Aid grid of constantvelocity cells of 5ø x 5ø with 16 layers for Creative Basic Research to the Ocean Hemisphere net- down to the base of the mantle, but in the studyregions work Project (OHP), by a JSPSpostdoctoral fellowship to S. he used a finer grid of 1ø x 1ø with 19 layers down to Widiyantoro and by a COE researchfellowship to M. Oba- 1600 km to allow resolution of relatively small scale yashi. feature. JamesSmith was the Editor responsiblefor this paper. He thanks John VanDecar and Susan van der Lee for technical In these regional studies,Widiyantoro [1997] used a reviewsand Albert J. Plueddemannfor the cross-disciplinary global arrival-time data set that has been carefully re- review. processedby Engdahlet al. [1998]. He used not only P phases,but also the surface-reflecteddepth phasespP and pwP in order to get better samplingin the upper mantle. He employed summaryrays [Widiyantoroand REFERENCES van der Hilst, 1996, 1997], but if both the source and Abers,G. A., Three-dimensionalinversion of regionalP and S receiverwere locatedwithin the studyregions, individual arrival times in the East Aleutians and sources of subduc- ray paths were used to optimize the sampling. The tion zone gravity highs,J. Geophys.Res., 99, 4395-4412, datum, i.e., the residualtime assignedto the summary 1994. ray, was the median of all data consideredfor that Akaogi, M., and E. Ito, Refinementof enthalpymeasurement summary ray. of MgSiO3 perovskiteand negativepressure-temperature For inversions,Widiyantoro [1997] made use of the slopesfor perovskiteforming reactions, Geophys. Res. Lett., 20, 1839-1842, 1993. iterative LSQR method of Paige and Saunders[1982], Anderson, D. L., Chemical stratification of the mantle, J. which is available at http://www'netlib'ørg/linalg/lsqr' Geophys.Res., 84, 6297-6298, 1979. This is a conjugategradient technique that was first Audley-Charles,M. G., P. D. Ballatyne, and R. Hall, Mesozo- appliedto seismictomography by Nolet [1985]. He then ic-Cenozoicrift-drift sequenceof Asian fragmentsfrom assessedthe reliability of the resultingimages by con- Gondwanaland,Tectonophysics, 155, 317-330, 1988. Bercovici,D., G. Schubert,and P. J. Tackley,On the penetra- ducting resolution tests. He carried out checkerboard tion of the 670-km phase by mantle downflows,Geophys. tests as well as hypotheticaltesting using slab models. Res. Lett., 20, 2599-2602, 1993. The results of these tests suggestthat, in general, the Bijwaard,H., W. Spakman,and E. R. Engdahl,Closing the gap 320 ß Fukao et al.: SLABSIN THE MANTLE TRANSITION REGION 39, 3 / REVIEWSOF GEOPHYSICS
between regional and global travel time tomography,J. of lateral heterogeneityin P velocityup to degreeand order Geophys.Res., 103, 30,055-30,078, 1998. 6, J. Geophys.Res., 89, 5929-5952, 1984. Bina, C. R., Mantle discontinuities,U.S. Natl. Rep. Int. Union Dziewonski,A.M., and D. L. Anderson,Preliminary reference Geod. Geophys.1987-1990, Rev. Geophys.,29, 783-793, Earth model, Phys.Earth Planet.Inter., 25, 297-356, 1981. 1991. Dziewonski,A.M., and F. Gilbert, The effectof smallaspheri- Bina, C. R., and G. Helffrich, Phase transition Clapeyron cal perturbationson travel times and re-examinationof the slopesand transitionzone seismic discontinuity topography, correctionsfor ellipticity, Geophys.J. R. Astron. Soc., 44, J. Geophys.Res., 99, 15,853-15,860, 1994. 7-16, 1976. Birch, F., Elasticityand constitutionof the Earth's interior, J. Engdahl, E. R., and D. Gubbins, Simultaneoustravel time Geophys.Res., 57, 227-286, 1952. inversion for earthquake location and subductionzone Boschi,L., andA.M. Dziewonski,High- and low-resolution structurein the central Aleutian islands,J. Geophys.Res., images of the Earth's mantle: Implications of different 92, 13,855-13,862, 1987. approachesto tomographicmodeling, J. Geophys.Res., 104, Engdahl,E. R., R. D. van der Hilst, and J. Berrocal,Imaging 25,567-25,594, 1999. of subductedlithosphere beneath South America, Geophys. Brunet, D., and P. Machetel, Large-scaletectonic features Res. Lett., 22, 2317-2320, 1995. inducedby mantle avalancheswith phase,temperature, and Engdahl, E. R., R. D. van der Hilst, and R. Buland, Global pressurelateral variationsof viscosity,J. Geophys.Res., 103, teleseismicearthquake relocation with improved travel 4929-4945, 1998. times and proceduresfor depth determination,Bull. Seis- Bullen, B., Introductionto the Theoryof Seismology,381 pp., mol. Soc. Am., 88, 722-743, 1998. CambridgeUniv. Press,New York, 1963. Engebretson,D.C., K. P. Kelley, H. J. Cashman,and M. A. Bunge, H.-P., M. A. Richards, C. Lithgow-Bertelloni,J. R. Richards, 180 million years of subduction,GSA Today,2, Baumgardner, S. P. Grand, and B. Romanowicz, Time 93-95, 1992. scalesand heterogeneitystructure in geodynamicEarth Flanagan,M.P., and P.M. Shearer,Global mappingof topog- models, Science,280, 91-95, 1998. raphyon transitionzone velocity discontinuities by stacking Butler, S. L., and W. R. Peltier, On scalingrelations in time- SS precursors,J. Geophys.Res., 103, 2673-2692, 1998. dependentmantle convectionand the heat transfer con- Flanagan,M.P., and P.M. Shearer,A map of topographyon straint on layering,J. Geophys.Res., 105, 3175-3208, 2000. the 410-km discontinuityfrom PP precursors,Geophys. Res. 0•adek,O., and L. Fleitout,A globalgeoid model with imposed Lett., 26, 549-552, 1999. plate velocitiesand partial layering,J. Geophys.Res., 104, Forsyth,D. W., and S. Uyeda, On the relative importanceof 29,055-29,075, 1999. drivingforces of plate motions,Geophys. J. R. Astron.Soc., 0•adek,O., andA. P. Van denBerg, Radial profiles of tem- 43, 163-200, 1975. perature and viscosityin the Earth's mantle inferred from Forte, A.M., and R. L. Woodward, Seismic-geodynamiccon- the geoid and lateral seismicstructure, Earth Planet. Sci. straints on three-dimensional structure, vertical flow, and Lett., 164, 607-615, 1998. heat transfer in the mantle, J. Geophys.Res., 102, 17,981- 0•adek,O., H. Cizkova,and D. A. Yuen,Can long-wavelength 17,994, 1997. dynamicalsignatures be compatiblewith layered mantle Forte, A.M., A.M. Dziewonski, and R. L. Woodward, As- convection?,Geophys. Res. Lett., 24, 2091-2094, 1997. pherical structureof the mantle, tectonic plate motions, Cande, S.C., and J. C. Mutter, A revised identification of the nonhydrostaticgeoid, and topographyof the core-mantle oldestsea-floor spreading anomalies between Australia and boundary,in Dynamicsof theEarth's Deep Interior and Earth Antarctica, Earth Planet. Sci. Lett., 58, 151-160, 1982. Rotation,Geophys. Monogr. Set., vol. 72, edited by J. L. Le Carlson, R. W., Mechanisms of Earth differentiation: Conse- Mou•l, D. E. Smylie, and T. Hering, pp. 135-136, AGU, quencesfor the chemical structure of the mantle, Rev. Washington,D.C., 1993. Geophys.,32, 337-361, 1994. Forte, A.M., R. L. Woodward, and A.M. Dziewonski, Joint Chapple,W. M., and T. E. Tullis, Evaluationof the forcesthat inversionsof seismicand geodynamicdata for models of drive plates,J. Geophys.Res., 82, 1967-1984, 1977. three-dimensionalmantle heterogeneity,J. Geophys.Res., Christensen,U. R., Mantle rheology,constitution, and convec- 99, 21,857-21,877, 1994. tion, in Mantle Convection,Plate Tectonicsand Global Dy- Fukao, Y., M. Obayashi,H. Inoue, and M. Nenbai, Subducting namics,edited by W. R. Peltier, pp. 595-655, Gordon and slabsstagnant in the mantle transition zone, J. Geophys. Breach, Newark, N.J., 1989. Res., 97, 4809-4822, 1992. Christensen,U. R., Dynamic phaseboundary topography by Fukao,Y., S. Maruyama,M. Obayashi,and H. Inoue, Geologic latent heat effects, Earth Planet. Sci. Lett., 154, 295-306, implicationof the whole mantle P-wave tomography,J. 1998. Geol. Soc.Jpn., 100, 4-23, 1994. Christensen, U. R., and D. A. Yuen, The interaction of a Gorbatov, A., S. Widiyantoro, Y. Fukao, and E. Gordeev, subductionslab with a chemical or phase boundary, J. Signature of remnant slabs in the North Pacific from P- Geophys.Res., 89, 4389-4402, 1984. -•__ wave tomography,Geophys. J. Int., 142, 27-36, 2000. Christensen,U. R., and D. A. Yuen, Layered convection Gordon, R. G., and D. M. Jurdy, Cenozoicglobal plate mo- inducedby phase transitions,J. Geophys.Res., 90, 10,291- tions,J. Geophys.Res., 91, 12,389-12,406, 1986. 10,300, 1985. Grand, S. P., Mantle shear structure beneath the Americas and Cserepes,L., and D. A. Yuen, Dynamical consequencesof surroundingoceans, J. Geophys.Res., 99, 11,591-11,621, midmantleviscosity stratification on mantle flowswith an 1994. endothermicphase transition, Geophys. Res. Lett., 24, 181- Grand, S. P., R. D. van der Hilst, and S. Widiyantoro,Global 184, 1997. seismictomography: A snapshotof convectionin the Earth, Deparis, V., H. Legros,and Y. Ricard, Mass anomaliesdue to GSA Today, 7, 1-7, 1997. subductedslabs and simulationsof plate motion since200 Gutenberg,B., Physicsof the Earth's Interior, 240 pp., Aca- My, Phys.Earth Planet.Inter., 89, 271-280, 1995. demic, San Diego, Calif., 1959. Ding, X. Y., and S. P. Grand, Seismicstructure of the deep Hall, R., J. R. Ali, C. D. Anderson,and S. J. Baker, Origin and Kurile subduction zones, J. Geophys. Res., 99, motion historyof the PhilippineSea Plate, Tectonophysics, 23,767-23,786, 1994. 251, 229-250, 1995. Dziewonski,A.M., Mapping the lower mantle:Determination Honda, S., D. A. Yuen, S. Balachander, and D. Reuteler, 39, 3 / REVIEWSOF GEOPHYSICS Fukao et al.: SLABSIN THE MANTLE TRANSITION REGION ß 321
Three-dimensional instabilities of mantle convection with velocitymodel of the mantle,Philos. Trans. R. Soc.London, multiple phase transitions,Science, 259, 1308-1311, 1993. Ser. A, 354, 1385-1411, 1996. Inoue, H., Y. Fukao, K. Tanabe, and Y. Ogata, Whole mantle Menard, H. W., Fragmentationof the Farallon plate by pivot- P-wave travel time tomography,Phys. Earth Planet.Inter., ing subduction,J. Geol., 86, 99-110, 1978. 59, 294-328, 1990. Montagner, J.-P., Can seismologytell us anythingabout con- Ito, E., and E. Takahashi, Postspineltransformations in the vection in the mantle?,Rev. Geophys.,32, 115-137, 1994. systemMg2SiO4-Fe2SiO 4 and some geophysicalimplica- Nakanishi, I., Regional differencesin the phasevelocity and tions,J. Geophys.Res., 94, 10,637-10,646, 1989. the qualityfactor Q of mantle Rayleighwaves, Science, 200, Johnson,L. R., Array measurementsof P velocities in the 1379-1381, 1978. uppermantle, J. Geophys.Res., 72, 6309-6325,'1967. Niazi, M., and D. L. Anderson, Upper mantle structure of Jordan,T. H., P. Puster,G. A. Glatzmaier, and P. J. Tackley, western North America from apparent velocities of P Comparisonsbetween seismic Earth structuresand mantle waves,J. Geophys.Res., 70, 4633-4640, 1965. flow models based on radial correlation functions, Science, Niu, F., and H. Kawakatsu,Depth variation of the midmantle 261, 1427-1431, 1993. seismic discontinuity,Geophys. Res. Lett., 24, 429-432, Kamiya, S., T. Miyatake, and K. Hirahara, How deep can we 1997. see the high-velocityanomalies beneath the Japan islands?, Nolet, G., Solving or resolvinginadequate and noisy tomo- Geophys.Res. Lett., 15, 828- 831, 1988. graphicsystem, J. Cornput.Phys., 61,463-482, 1985. Kaneshima, S., and G. Helffrich, Detection of lower mantle Obayashi,M., T. Sakurai,and Y. Fukao, Comparisonof recent scatterersof the Mariana subductionzones using short- tomographicmodels (abstract), in InternationalSymposium period array data, J. Geophys.Res., 103, 4825-4838, 1998. on New Images of the Earth's b•terior ThroughLong-Term Kaneshima,S., and G. Helffrich, Dipping low-velocitylayer in Ocean-Floor Observations,edited by K. Suyehiro, p. 29, the mid-lower mantle: Evidencefor geochemicalheteroge- Earthquake Res. Inst., Univ. of Tokyo, Tokyo, 1997. neity, Science,283, 1888-1891, 1999. Ogawa, M., A bifurcation in the coupled magmatism-mantle Karato, S., Importance of anelasticityin the interpretationof convectionsystem and its implicationsfor the evolutionof seismictomography, Geophys. Res. Lett., 20, 1623-1626, the Earth's upper mantle, Phys.Earth Planet. Inter., 102, 1993. 259-276, 1997. Kawakatsu, H., and F. Niu, Seismic evidence for a 920-km Okal, E. A., and S. H. Kirby, Deep earthquakesbeneath the discontinuityin the mantle, Nature, 371,301-305, 1994. Fiji Basin,SW Pacific:Earth's most intensedeep seismicity Kennett, B. L. N., E. R. Engdahl, and R. Buland, Constraints in stagnant slabs,Phys. Earth Planet. Inter., 109, 25-63, on seismic velocities in the Earth from travel times, Geo- 1998. phys.J. Int., 122, 108-124, 1995. Oreshin, S., L. Vinnik, A. Treussov, and R. Kind, Subducted Kennett, B. L. N., S. Widiyantoro, and R. D. van der Hilst, lithosphereor 530 km discontinuity?,Geophys. Res. Lett., Joint seismictomography for bulk-soundand shear wave 25, 1091-1094, 1998. speedin the Earth's mantle,J. Geophys.Res., 103, 12,469- Paige, C. C., and M. A. Saunders,LSQR: An algorithm for 12,493, 1998. sparse linear equations and sparse least squares, Trans. Kido, M., and O. (2adek,Inferences of viscosityfrom the Math. Software,8, 43-71, 1982. oceanicgeoid: Indication of a low viscosityzone below the Pari, G., and W. R. Peltier, The heat flow constraint on mantle 660-km discontinuity,Earth Planet Sci. Lett., 151, 125-137, tomography-basedconvection models: Toward a geody- 1997. namically self-consistentinference of mantle viscosity,J. King, S. D., Radial modelsof mantle viscosity:Results from a Geophys.Res., 100, 12,731-12,752, 1995. geneticalgorithm, Geophys. J. Int., 122, 725-734, 1995. Pari, G., and W. R. Peltier, Global surface heat flux anomalies Lay, T., The fate of descendingslabs, ,4nnu. Rev. Earth Planet. from seismic tomography-basedmodels of mantle flow: Sci., 22, 33-61, 1994. Implicationsfor mantle convection,J. Geophys.Res., 103, Le Stunif, Y., and Y. Ricard, Topographyand geoid due to 23,743-23,780, 1998. lithosphericmass anomalies, Geophys. J. Int., 122, 982-990, Peltier, W. R., and L. P. Solheim,Mantle phasetransitions and 1995. layeredchaotic convection, Geophys. Res. Lett., 19, 321-324, Le Stunff, Y., and Y. Ricard, Partial advectionof equidensity 1992. surfaces:A solutionfor the dynamictopography problem?, Petersen, N., J. Gossler, R. Kind, K. Stammler, and L. Vinnik, J. Geophys.Res., 102, 24,655-24,667, 1997. Precursors to SS and structure of the transition zone of the Levi, B. G., Earth's upper mantle:How low can it flow?,Phys. northwesternPacific, Geophys. Res. Lett., 20, 281-284, 1993. Today,50, 17-20, Aug. 1997. Puspito,N. T., Y. Yamanaka,T. Miyatake, K. Shimazaki,and Li, X.-D., and B. Romanowicz,Global mantle shear-velocity K. Hirahara, Three-dimensionalP-wave velocity structure model usingnonlinear asymptoticcoupling theory, J. Geo- beneath the Indonesianregion, Tectonophysics,220, 175- phys.Res., 101, 22,245-22,272, 1996. 192, 1993. Li, X.-D., D. Giardini, and J. H. Woodhouse, Large-scale Puster, P., and T. H. Jordan, Stochasticanalysis of mantle three-dimensionaleven-degree structure of Earth from convectionexperiments using two-point correlation func- splittingof long-periodnormal mode,J. Geophys.Res., 96, tions, Geophys.Res. Lett., 21,305-308, 1994. 551-557, 1991. Puster, P., and T. H. Jordan, How stratified is mantle convec- Lithgow-Bertelloni,C., and M. A. Richards,The dynamicsof tion?, J. Geophys.Res., 102, 7625-7646, 1997. Cenozoicand Mesozoicplate motions,Rev. Geophys.,36, Resovsky,J. S., and M. H. Ritzwoller,A degree8 mantle shear 27-78, 1998. velocity model from normal mode observationsbelow 3 Machetel,P., and P. Weber, Intermittent layeredconvection in mHz, J. Geophys.Res., 104, 993-1014, 1999. a model mantle with an endothermicphase changeat 670 Revenaugh,J., and T. H. Jordan, Mantle layering from ScS km, Nature, 350, 55-57, 1991. reverberations,2, The transitionzone, J. Geophys.Res., 96, Machetel, P., C. Thoraval, and D. Brunet, Spectral and geo- 19,763-19,780, 1991. physicalconsequences of 3-D sphericalmantle convection Ringwood, A. E., Role of the transition zone and 660 km with an endothermicphase change at the 670 km disconti- discontinuityin mantle dynamics,Phys. Earth Planet.Inter., nuity,Phys. Earth Planet.Inter., 88, 43-51, 1995. 86, 5-24, 1994. Masters, G., S. Johnson, G. Laske, and H. Bolton, A shear Ringwood, A. E., and T. Irifune, The nature of the 650-km 322 ß Fukao et al.: SLABS IN THE MANTLE TRANSITION REGION 39, 3 / REVIEWS OF GEOPHYSICS
discontinuity:Implications for mantle dynamics,Nature, zone, U.S.Natl. Rep.Int. UnionGeod. Geophys. 1991-1994, 331, 131-136, 1988. Rev. Geophys.,33, 275-282, 1995a. Ritzwoller,M. H., and E. M. Lavely,Three-dimensional seis- Tackley,P. J., On the penetrationof an endothermicphase mic modelsof the Earth's mantle,Rev. Geophys.,33, 1-66, transitionby upwellingand downwelling,J. Geophys.Res., 1995. 100, 15,477-15,488, 1995b. Robertson, G. S., and J. H. Woodhouse, Evidence for the Tackley, P. J., G. A. Stevenson,G. A. Glatzmaier, and G. proportionalityof P and S heterogeneityin the mantle, Schubert,Effects of an endothermicphase transition at 670 Geophys.J. Int., 123, 85-116, 1995. km in a sphericalmodel of convectionin the Earth'smantle, Robertson, G. S., and J. H. Woodhouse, Ratio of relative S to Nature, 361,699-704, 1993. P velocityheterogeneity in the lower mantle,J. Geophys. Tackley, P. J., G. A. Stevenson,G. A. Glatzmaier,and G. Res., 101, 20,041-20,052, 1996. Schubert,Effects of multiplephase transitions at 670 km in Rona, P. A., and E. S. Richardson,Early Cenozoicglobal plate a three-dimensionalspherical model of convectionin the reorganization,Earth Planet.Sci. Lett., 40, 1-11, 1978. Earth's mantle,J. Geophys.Res., 99, 15,877-15,901,1994. Ross,M. I., and C. R. Scotese,A hierarchicaltectonic model of Tajima, F., and S. P. Grand, Evidenceof high-velocityanom- the Gulf of Mexico and Caribbeanregion, Tectonophysics, alies in the transition zone associated with southern Kurile 155, 139-168, 1988. subductionzone, Geophys.Res. Lett., 22, 3139-3142, 1995. Roth, E.G., and D. A. Wiens, Depressionof the 660-km Tajima, F., and S. P. Grand, Variation of transitionzone high discontinuitybeneath the Tonga slab determined from anomaliesand depressionof the 660-km discontinuityas- near-verticalScS reverberations,Geophys. Res. Lett., 26, sociated with subduction zones from the southern Kuriles 1223-1226, 1999. to Izu-Bonin, J. Geophys.Res., 103, 15,015-15,036,1998. Rubie, D.C., and A. J. Brearley, Phasetransition between [3 Tajima, F., Y. Fukao,M. Obayashi,and T. Sakurai,Evaluation and •/(MgFe)2SiO4 in the Earth's mantle:Mechanism and of slab imagesin the northwesternPacific, Earth Planets rheologicalimplications, Science, 264, 1445-1448,1994. Space,50, 953-964, 1998. Scotese,C. R., L. M. Gahagan,and R. L. Larson,Plate tectonic Takenaka, S., H. Sanshadokoro,and S. Yoshioka, Velocity reconstructions of the Cretaceous and Cenozoic ocean ba- anomaliesand spatialdistributions of physicalproperties in sins,Tectonophysics, 155, 27-48, 1988. horizontallylying slabsbeneath the northwesternPacific Seno, T., and S. Maruyama, Paleogeographicreconstruction region,Phys. Earth Planet.Inter., 112, 137-157, 1999. and origin of the PhilippineSea, Tectonophysics,102, 53- Tanimoto, T., Long wavelength S-wave velocity structure 84, 1984. throughout the mantle, J. Geophys.Int., 100, 327-336, Shearer,P.M., Global mappingof upper mantle reflectors 1990a. from long-periodSS precursors,Geophys. J. Int., 115, 878- Tanimoto, T., Predominanceof large-scaleheterogeneity and 904, 1993. the shiftof velocityanomalies between the upperand lower Shearer,P.M., and T. G. Masters,Global mappingof topog- mantle, J. Phys.Earth, 38, 493-509, 1990b. raphy on the 660-km discontinuity,Nature, 355, 791-795, Thoraval, C., and M. A. Richards, The geoid constraintin 1992. global geodynamics:Viscosity structure, mantle heteroge- Solheim,L. P., and W. R. Peltier, Avalancheeffects in phase neitymodels and boundary conditions, Geophys. J. Int., 131, transition modulated thermal convection: A model of 1-8, 1997. Earth's mantle,J. Geophys.Res., 99, 6997-7018, 1994a. Thoraval, C., P. Machetel, and A. Cazanave,Locally layered Solheim,L. P., and W. R. Peltier, Phaseboundary deflection at convectioninferred from dynamicmodels of the Earth's 660 km depth and episodicallylayered isochemical convec- mantle, Nature, 375, 777-780, 1995. tion in the mantle, J. Geophys.Res., 99, 15,861-15,875, Thorkelson,D. J., Subductionof divergingplates and the 1994b. principlesof slabwindow formation, Tectonophysics, 255, Solomon,S.C., and N.H. Sleep,Some simple physical models 47-63, 1996. for absoluteplate motions,J. Geophys.Res., 79, 2557-2567, van der Hilst, R. D., Complexmorphology of subductedlitho- 1974. spherein the mantlebeneath the Tongatrench, Nature, 374, Solomon,S.C., N.H. Sleep,and R. M. Richardson,On the 154-157, 1995. forcesdriving plate tectonics:Inference from absoluteplate van der Hilst, R. D., and E. R. Engdahl, On ISC PP andpP velocitiesand intraplatestress, Geophys. J. R. Astron.Soc., data and their use in delay-timetomography of the Carib- 42, 769-801, 1975. bean region,Geophys. J. Int., 106, 169-188, 1991. Spakman,W., S. Stein, R. D. van der Hilst, and R. Wortel, van der Hilst, R. D., and T. Seno, Effects of relative plate Resolutionexperiments for NW Pacific subductionzone motion on the deep structureand penetrationdepth of tomography,Geophys. Res. Lett., 16, 1097-1100,1989. slabs below the Izu-Bonin and Mariana island arcs, Earth Spakman,W., S. van der Lee, and R. D. van der Hilst, Travel- Planet. Sci. Lett., 120, 395-407, 1993. time tomographyof the European-Mediterraneanmantle van der Hilst, R. D., and W. Spakman,Importance of the down to 1400 km, Phys.Earth Planet.Inter., 79, 3-74, 1993. referencemodel in the linearizedtomography and images Steinbach,V., and D. A. Yuen, Effects of depth-dependent of subductionbelow the Caribbeanplate, Geophys.Res. propertieson the thermal anomaliesproduced in flush Lett., 16, 1093-1096, 1989. instabilitiesfrom phasetransitions, Phys. Earth Planet.In- van der Hilst, R. D., E. R. Engdahl, W. Spakman,and G. ter., 86, 165-184, 1994. Nolet, Tomographicimaging of subductedlithosphere be- Stixrude,L., Structureand sharpnessof phasetransitions and low northwest Pacific island arcs, Nature, 353, 37-43, 1991. mantle discontinuities,J. Geophys.Res., 102, 14,835-14,852, van der Hilst, R. D., E. R. Engdahl,and W. Spakman,Tomo- 1997. graphicinversion of P andpP data for asphericalmantle Su, W., and A.M. Dziewonski, Simultaneousinversion for 3-D structure below the northwest Pacific region, Geophys. variationsin shear and bulk velocityin the mantle, Phys. J. Int., 115, 264-302, 1993. Earth Planet. Inter., 100, 135-156, 1997. van der Hilst, R. D., S. Widiyantoro, and E. R. Engdahl, Su, W., R. L. Woodward, and A.M. Dziewonski,Degree 12 Evidencefor deepmantle circulation from globaltomogra- model of shear velocity heter0geneityin the mantle, J. phy,Nature, 386, 578-584, 1997. Geophys.Res., 99, 6945-6980, 1994. van der Hilst, R. D., S. Widiyantoro,K. C. Creager,and T. J. Tackley, P. J., Mantle dynamics:Influence of the transition McSweeney,Deep subductionand asphericalvariations in 39, 3 / REVIEWSOF GEOPHYSICS Fukao et al.: SLABSIN THE MANTLE TRANSITION REGION ß 323
P-wave speed at the base of Earth's mantle, The Core- Extending shear-wavetomography for the lower mantle Mantle BoundaryRegion, Geodyn. Set., vol. 28, edited by E. usingS and SKS arrival-timedata, Earth PlanetsSpace, 50, Gurnis et al., pp. 5-20, AGU, Washington,D.C., 1998. 999-1012, 1998. Van der Lee, S., and G. Nolet, Seismicimage of the subducted Woodhouse,J. H., and A.M. Dziewonski,Mapping the upper trailing fragmentsof the Farallon plate, Nature, 386, 266- mantle: Three-dimensionalmodeling of Earth structureby 269, 1997. inversionof seismicwaveforms, J. Geophys.Res., 89, 5953- Van der Voo, R., W. Spakman,and H. Bijwaard, Mesozoic 5986, 1984. subducted slabsunder Siberia, Nature, 397, 246-249, 1999a. Woodward, R. L., A.M. Forte, W.-J. Su, and A.M. Dziewon- Van der Voo, R., W. Spakman,and H. Bijwaard, Tethyan ski, Constraintson the large-scalestructure of the Earth's subducted slabs under India, Earth Planet. Sci. Lett., 171, mantle, in Evolution of the Earth and Planets, Geophys. 7-20, 1999b. Monogr. Set, vol. 74, edited by E. Takahashi, R. Jeanloz, Vasco, D. W., and L. R. Johnson, Whole earth structure and D. Rubie, pp. 89-109, AGU, Washington,D.C., 1993. estimatedfrom seismicarrival times,J. Geophys.Res., 103, Zhao, D., Y. Xu, D. A. Wiens, L. Dorman, J. Hiderbrand, and 2633-2671, 1998. S. Webb, Depth extentof the Lau back-arcspreading center Vinnik, L., G. Kosarev, and N. Petersen, Mantle transition and its relation to subductionprocesses, Science, 278, 254- zone beneath Eurasia, Geophys.Res. Lett., 23, 1485-1488, 257, 1997. 1996. Zhou, H., How well can we resolvethe deep seismicslab with Vinnik, L., F. Niu, and H. Kawakatsu, Broadband converted seismictomography?, Geophys. Res. Lett., 15, 1425-1428, phasesfrom midmantlediscontinuities, Earth PlanetsSpace, 1988. 50, 987-997, 1998. Zhou, H.-W., Mapping of P-wave slab anomaliesbeneath the Walcott, R. I., Geodetic strain and the deformationhistory of Tonga, Kermadec and New Hebrides arcs, Phys. Earth the North Island of New Zealand duringthe late Cainozoic, Planet. Inter., 61,199-229, 1990. Philos. Trans. R. Soc. London, Set. A, 321,163-181, 1987. Zhou, H.-W., A high-resolutionP wavemodel for the top 1200 Wen, L., and D. L. Anderson, The fate of slabs inferred from km of the mantle, J. Geophys.Res., 101, 27,791-27,810, seismictomography and 130 million years of subduction, 1996. Earth Planet. Sci. Lett., 133, 185-198, 1995. Zhou, H., and R. W. Clayton,P and S travel time inversions Wen, L., and D. L. Anderson,Layered mantle convection:A for subductingslab under the island arcs of the northwest model for geoid and topography,Earth Planet. Sci. Lett., Pacific,J. Geophys.Res., 95, 6829-6851, 1990. 146, 131-143, 1997. Widiyantoro, S., Studiesof seismictomography on regional and global scale,Ph.D. thesis,Aust. Natl. Univ., Canberra, Y. Fukao,Earthquake Research Institute, University of To- A.C.T., Australia, 1997. kyo, Tokyo,Japan ([email protected]) Widiyantoro, S., and R. D. van der Hilst, Structure and evo- M. Obayashi,Japan Science and TechnologyCorporation, lution of lithosphericslab beneath the Sunda arc, Indone- sia, Science,271, 1566-1570, 1996. 4-1-8 Honcho,Kawaguchi-shi 332-0012, Japan.(mohbayas@ Widiyantoro, S., and R. D. van der Hilst, Mantle structure mri-jma.go.jp) beneath Indonesia inferred from high-resolutiontomo- S. Widiyantoro,Department of Geophysicsand Meteorol- graphicimaging, Geophys. J. Int., 130, 167-182, 1997. ogy, BandungInstitute of Technology,J1. Ganesha,Bandung Widiyantoro, S., B. L. N. Kennett, and R. D. van der Hilst, 40132, Indonesia.([email protected])