tters 243 (2006) 149–158www.elsevier.com/locate/epsl

Earth and Planetary Science Le

High temperature anomalies oceanward of subducting slabsat the 410-km discontinuity

Masayuki Obayashi ⁎, Hiroko Sugioka, Junko Yoshimitsu, Yoshio Fukao

Institute for Research on Earth Evolution, Japan Agency for Marine-Earth Science and Technology, 2-15 Natsushima, Yokosuka, 237-0061, Japan

Received 2 June 2005; received in revised form 20 December 2005; accepted 27 December 2005Available online 7 February 2006

Editor: R.D. van der Hilst

Abstract

Our P-wave whole mantle tomography revealed a low velocity region oceanward of the Northern Honshu slab of the Pacificplate at depths around the 410-km seismic discontinuity. Resolution tests and scrutiny of the traveltime residuals for the ray pathspassing through the low velocity region indicate that this anomaly is a resolvable feature and not an artifact due to the strong slabanomalies. The existence of the slow anomalies is also supported by the analysis of the P-wave records from the J-array (a large-aperture seismic array in Japan) for a Bonin earthquake. The P arrivals to Northern Honshu (at epicentral distances of 13–20°) arestrongly triplicated because of the 410-km discontinuity. The later arrivals along the retrograde branch, where ray paths passthrough the low velocity region, are anomalously slow. Comparison of the observed and synthetic waveforms indicates not onlyslow anomalies but also depression of the 410-km discontinuity. This depression represents the direct evidence for the low velocityzone of primarily thermal origin. An excess temperature of 200 K and the associated fractional melt of less than 1% can explainboth the results of the tomographic and waveform analyses.© 2006 Elsevier B.V. All rights reserved.

Keywords: mantle; 410-km discontinuity; subducting lithosphere; seismic tomography

1. Introduction

At subduction zones mantle lithosphere descendsinto the Earth's interior along with sediments andoceanic crust, which is collectively called “slab”.Because subducting slabs are colder than the ambientmantle, they are distinct as anomalously high seismicvelocity regions (e.g. [1–3]) and they are also knownas the places of distorting the seismic discontinuityplane at either 410 or 660 km depth. The 410-kmdiscontinuity is generally attributed to the phase

⁎ Corresponding author. Tel.: +81 46 867 9744; fax: +81 46 8679745.

E-mail address: obayashi@jamstec.go.jp (M. Obayashi).

0012-821X/$ - see front matter © 2006 Elsevier B.V. All rights reserved.doi:10.1016/j.epsl.2005.12.032

transition from α-olivine to β-wadsleyite [4–6].Topographic high in the cold slab, which is expectedfrom the positive Clapeyron slope of this transform,has been observed [7].

The mantle wedge, a part of the mantle overlying thesubducting slab, is known in general as a zone of lowseismic velocity and high attenuation [2,8]. Sincesunduction-related activities are largely observedabove the mantle wedge, attention has been mostlydirected to this part (e.g. [9–12]).

On the other hand, little attention has been given tothe opposite side (trench side) of the mantle across theslab and no significant seismic anomaly has beenreported in detail. In this study, we investigate the lowvelocity anomaly oceanward of the Northern Honshu

Fig. 1. P-wave velocity anomalies in the northwest Pacific at depths of(a) 148–190 km, (b) 348–410 km and (c) 410–478 km. Cross-sectionsin Figs. 2 and 6 are along the profiles shown in (b). Seismic records inFig. 5a are from stations indicated in (a). The blue records in Fig. 5a arefrom stations in red. The epicentres are shown in (b) for the eventanalyzed in detail and in (c) for the other events. See Table 1 for theinformation of these events. The map is drawn using the equidistantazimuthal projection with its origin at the epicentre of Event 1. Thelengths of ray path segments in the LVZ410 are defined as those withina green box in (b) and (c).

Fig. 2. Vertical cross-sections of (a) the P-wave velocity anomalies and(b) a result of checkerboard resolution test with unit size5.625°×5.625° along profile K in Fig. 1b. The approximate boundarybetween high and low resolution inferred from the checkerboardresolution test is indicated by an arrow in each of (a) and (b). Black andred triangles in (a) represent the Japan Trench and the “Petit-spot”,respectively.

150 M. Obayashi et al. / Earth and Planetary Science Letters 243 (2006) 149–158

slab of the Pacific plate at depths around the 410-kmseismic discontinuity and the associated anomaly intopography of this discontinuity. The results will beinterpreted in terms of whether the observed velocityanomaly is of thermal or chemical origin.

2. Tomographic image

We obtained a three-dimensional P-wave velocitymodel for the whole mantle by inverting about 7.4

million first arrivals reported to the InternationalSeismological Centre. The mantle was discretized intoblocks with variable sizes [2,13]. We calculated the totallength of ray path segments within each block and if thislength was large enough, the block was subdivided intofiner blocks. The largest and smallest blocks are11.25°×11.25° and 1.40625°×1.40625° in size, respec-tively, with about 160,000 blocks in total for the wholemantle.

Fig. 1 shows the P-wave velocity anomalies nearJapan in three different layers. The fast slabs subductedfrom the Kurile, Japan, Izu-Bonin trenches are clearlyimaged (Fig. 1a). The mantle wedge regions behindthese trenches are slow at depths down to more than 200km, with an exception in Northern Honshu where slowanomalies are limited to depths down to about 100 km.Another remarkable low velocity zone is observed onthe Pacific side of the subducting slabs around 410 kmdepth (Fig. 1b and c) (referred to as LVZ410 hereafter).The amplitudes of the low velocity anomalies arecomparable to those of the subducting slabs.

151M. Obayashi et al. / Earth and Planetary Science Letters 243 (2006) 149–158

Fig. 2a shows the vertical cross-section along profileK (Fig. 1b) across the Northern Honshu arc. Thesubducting slab is sandwiched by two slow anomalies;a shallower one on the continental side (mantle wedge)and a deeper one on the ocean side. Fig. 2b shows theresult of a checkerboard resolution test along the sameprofile as for Fig. 2a, where the oceanward end of highresolution is marked by an arrow. Clearly, the LVZ410is a resolvable feature. We made two tests to check ifthe slow anomaly is an artifact of the presence of a fastslab. The first test is to see how well a synthetic slabmodel can be reconstructed with little spurious imagesusing the same earthquake-station pairs as in the realtomography [2]. We constructed a synthetic slab modelwith a positive anomaly of 5%, a value 2–3 times as

Fig. 3. Result of synthetic test. Upper profiles show model slabs to which a fafrom the synthetic traveltime data calculated for the input model. Note that theside (1.5% slow).

high as the anomaly in the real tomography. Noperturbation was added elsewhere. We calculated thetraveltime residuals for this synthetic slab model. Wethen relocated the earthquakes and ran tomography.Fig. 3 shows the result of the synthetic test incomparison with the input model. Note that velocityanomaly scale is very different in the positive side(5%) and the negative side (1.5%). The fast slabs arereconstructed well with some smearing at theirmargins. Although spurious slow anomalies areobserved in a region corresponding to the LVZ410,their spatial extent is more narrowly limited and theiramplitudes are no more than 0.5%, much smaller thanthose in the real tomography. The second test is toscrutinize the traveltime residuals for the ray paths

st anomaly of 5% is given. Lower profiles show the slabs reconstructedscale is different between the positive side (5.0% fast) and the negative

152 M. Obayashi et al. / Earth and Planetary Science Letters 243 (2006) 149–158

passing through the LVZ410. Fig. 4 shows thefrequency distributions of the traveltime residuals ofthe ray paths through the LVZ410 (defined in depthfrom 348 to 478 km and in area by green lines in Fig.1b and c). The residual data are grouped according tothe length of the relevant ray path segment within theLVZ410. Each panel shows the frequency distributionof the residuals in each group. The averages of theresiduals are more positive for the ray path segmentslonger in the LVZ410. This and the result of thesynthetic test indicate that the LVZ410 is not an aliasedimage due to the surrounding fast anomalies. Althoughthe slow anomaly apparently extends further ocean-ward beyond the eastern end of high resolution, itremains to be answered whether it is localized in a

Fig. 4. Frequency distributions of the traveltime residuals of the ray paths presiduals are binned according to the length of ray path segment within theresiduals. The range of ray path segment lengths is shown at the top left. Numshown.

close proximity of the subducting slab or extendsfurther oceanward.

3. Seismic array analysis

The existence of the LVZ410 can be seen in theprevious tomographic studies (e.g. [2,13,14]). Howeverit has been paid little attention and has been seldomargued due in part to lack of other evidence. We herepresent independent evidence. Fig. 5a shows P-waveseismograms from the J-array [15] stations in northeastJapan (open triangles in Fig. 1a) for the 1998 Boninearthquake (Table 1). The epicentre is indicated in Fig.1b. Based on the observed P-wave arrivals we drew atraveltime curve as shown by broken lines. The solid

assing through the LVZ410 (defined by the green box in Fig. 1). TheLVZ410. Each panel shows the frequency distribution of the binnedber of the data, average and standard deviation of the residuals are also

Fig. 5. Observed and synthetic vertical seismograms. (a) J-array records of Event 1 at the northeast Japan stations (Fig. 1a). Blue waveforms are fromstations indicated by red triangles (Fig. 1a). The predicted traveltime curve of IASP91 is indicated by solid lines with the triplication branch labels.The observed traveltime curve is visually drawn as broken grey lines. (b) Synthetic waveforms calculated for M250T. The reference traveltime curveand the observed traveltime curve, the same as in (a), are also drawn. (For interpretation of the references to colour in this figure legend, the reader isreferred to the web version of this article.)

153M. Obayashi et al. / Earth and Planetary Science Letters 243 (2006) 149–158

lines indicate the reference traveltime curve of IASP91[16], which is very different from the observedtraveltime curve. At epicentral distances 11–17°, the

Table 1Earthquakes and networks used in this study

Eventno.

Date Latitude Longitude Depth Magnitude Network(km)

1 1, Jan.1998

23.93N 141.94E 100.5 6.3 Mb J-array

2 14, Oct.2000

23.56N 141.83E 137.5 5.5 Mw J-array

3 6, Oct.2002

46.03N 149.65E 153.9 4.9 Mb Hi-net

observed arrivals along the AB branch, which representP-waves passing through the upper mantle with theturning points in a depth range 100–300 km, aresignificantly early. The BC branch beyond 13° desig-nates P-waves bouncing on the 410-km discontinuity.The observed arrivals along this branch are anomalouslylate. The observed arrivals along the CD branch, whichrepresent P-waves refracted at the 410-km discontinuity,are also late, although the anomaly becomes small withincreasing distance.

A marked contrast between the early arrivals alongthe AB branch and the late arrivals along the BC and CDbranches implies the coexistence of fast anomaly in theuppermost mantle and slow anomaly at depths near the

154 M. Obayashi et al. / Earth and Planetary Science Letters 243 (2006) 149–158

410-km discontinuity, a feature consistent with thecontrasted tomographic images as discussed before. Fig.6a shows the ray paths with an epicentral distance of 15°superposed onto the tomographic cross-section alongprofile L (Fig. 1b). The AB ray traverses all the waythrough the fast slab in the uppermost mantle, thus

Fig. 6. Vertical cross-sections of the P-wave velocity anomalies. Thetriplicated ray paths for model M250T are shown in green for the ABand CD branches and in blue for the BC branch. (a) Cross-sectionalong profile L (Fig. 1b) with the ray paths at an epicentral distance of15°. (b) Cross-section along profile M with the ray paths at anepicentral distance of 20°. (c) Cross-section along profile N with theray paths at an epicentral distance of 14.6°. (For interpretation of thereferences to colour in this figure legend, the reader is referred to theweb version of this article.)

causing anomalously early arrivals along this branch.The BC and CD rays pass through the LVZ410, resultingin the anomalously late arrivals along these branches.The situation becomes different at an epicentral distanceof 20° (Fig. 6b). The AB phase no longer remains in thefast slab but grazes through the LVZ410 so that itsadvance at shallower depths tends to be cancelled out bythe delay at greater depths. The CD phase begins to travelfurther below the LVZ410 so that its delay becomes lessdistinct as compared to the delay at shorter distances. TheBC phase, on the other hand, keeps its deepest point atthe 410-km discontinuity so that its delay remains to bevery large.

The blue seismograms in Fig. 5a were recorded atstations above the junction of the Northern Honshu slaband the Izu-Bonin slab (red triangles in Fig. 1a). Ascompared to the records further to the east, the positivetraveltime anomalies of the BC and CD phases are smallwhile the negative anomalies of the AB phase are almostthe same. These observations are consistent with thetomographic image of the LVZ410: the BC and CD raystoward these stations pass through the western rim of theLVZ410 where the anomaly is less extensive, while theAB rays remain in the fast slabs (see profile N in Fig. 1band the corresponding cross-section in Fig. 6c).

The negative and positive traveltime anomaliesassociated with the fast slab and the LVZ410 are alsoobserved for other earthquakes listed in Table 1. Fig. 7aand b shows the J-array seismograms for Event 2 and theHi-net [17,18] seismograms for Event 3, respectively.The similar advances of the AB arrivals and the similardelays of the BC and CD arrivals along the mutuallyreversed profiles (for Events 1 and 2 versus Event 3)warrant the following one-dimensional analysis.

We constructed a one-dimensional P-wave velocitymodel that can explain the observed traveltime anomaliesof Event 1. We computed synthetic seismograms for trialmodels modified from IASP91 using the propagator matrixmethod [19] and compared them with the observedseismograms. In a trial and error process, we found thatboth a velocity reduction near the 410-kmdiscontinuity anda depression of this discontinuity relative to IASP91 arenecessary to explain the observation. Only the velocityreduction near the 410-km discontinuity makes it difficultto match simultaneously the traveltime and slowness of theBC branch to the observed ones. Only the discontinuitydepression places the end point C of the BC and CDbranches at a distance much greater than suggested by theobservation.

A low velocity zone atop the 410-km discontinuitywas detected in eastern China using multiple-ScSreverberation [20] and in northwestern United States

Fig. 7. Vertical component seismograms. (a) J-array records for Event 2 at stations with epicentral distances of 11–22°. (b) Hi-net records for Event 3at stations with epicentral distances of 10–22°. The predicted traveltime curves for IASP91 are indicated by solid lines.

155M. Obayashi et al. / Earth and Planetary Science Letters 243 (2006) 149–158

using refracted S and scattered S-waves [21]. Adepression of the discontinuity was also implied in thelatter study. These observations have been interpreted asevidence for silicate melt atop 410 km.

4. Discussions

Although the 410-km discontinuity is generallyattributed to the phase transition of (Mg,Fe)2SiO4, itsexact depth and seismic velocity change should beaffected by various factors such as temperature [5,6,22],Mg/Fe ratio [4–6,23], aluminium content [24] andwatercontent [25,26]. Increase of either Fe or H2O contentproduces slow anomaly, while it elevates the 410-kmdiscontinuity. Increase of Al content, on the other hand,depresses the discontinuity but produces fast anomaly.

Among the above four factors, only high temperatureanomaly causes slow anomaly and depression of thediscontinuity simultaneously. We estimated the temper-ature anomaly by assuming that the traveltime anomaliesof the BC and CD branches are controlled only bythermal anomaly, not by chemical anomaly. A constanthigh temperature anomaly of dTwas introduced at depthsfrom 350 km down to the discontinuity. Above 350 kmthe anomaly was assumed to decrease linearly with depthto vanish at 290 km. We estimated P velocity reductiondVp and discontinuity depression dh relative to IASP91corresponding to temperature anomaly dT, using coef-ficient values of dVp /dT=−5.3 E−04 km s−1 K−1 [22]and dh / dT=0.13 km K−1 [6]. P velocity at depths belowthe displaced 410-km discontinuity was unchanged. Thefast anomaly corresponding to the subducting slab was

156 M. Obayashi et al. / Earth and Planetary Science Letters 243 (2006) 149–158

introduced in a depth range 35–150 km. For models withdifferent values of dT, we calculated traveltimes of theBC branch arrivals. The root mean square (rms) value oftheir residuals relative to the observed traveltimes isplotted as a function of dT in Fig. 8. The rms residualtakes its minimum when dT=250 K. The rms of thismodel is about 85% smaller than the rms of modelIASP91. Thin line in Fig. 9 shows this model (M250T).The synthetic seismograms for this model (Fig. 5b) are inqualitative match with the observed seismograms.Although model CD arrivals are slightly earlier thanthe observed arrivals, this could be improved by a littlevelocity reduction immediately below the 410-kmdiscontinuity. The model contains 1.5% slow anomaly

Fig. 8. Rms residuals of traveltimes of the BC branch for variousmodels relative to IASP91. The residuals are calculated as a function oftemperature anomaly dT atop the 410-km discontinuity. Solid circlesare for models having only temperature anomaly. In (a) solid lines arefor models having both temperature anomaly and Fe anomaly. In (b)solid lines are for models having both temperature anomaly andvelocity reduction due to partial melting. See the text for detail.

Fig. 9. P-wave velocity profiles of the reference model IASP91 (thickline), model M250T with an excess temperature of 250 K (thin line),model M200I with an excess temperature of 200 K and an Fe increaseof 3.0 mol% (broken line), and model M200M with an excesstemperature of 200 K and a velocity reduction of 1.0% due to partialmelting (dotted line).

at 410 km depth and a discontinuity depression of 32.5km relative to IASP91.

The 1.5% slow anomaly is comparable to thetomographically imaged slow anomaly (Fig. 1). Sincethe latter is somehow a blurred image of the real anomaly(see Fig. 2b), the real intensity of the slow anomaly islikely to be greater than indicated by model M250T.Hence, we searched for a model having slow anomaly ofgreater than 1.5% by taking additional factors intoaccount. We consider two cases of (1) inserting an Fe-rich layer above the discontinuity and (2) inserting apartially molten layer above the discontinuity.

In case (1), we placed the Fe-rich layer so that itcoincides with the temperature anomaly zone. As theeffects of Fe-addition we used coefficient values of−0.25%/mol% [23] and 1.8 km/mol% [6] forvelocity increase and discontinuity elevation, respec-tively. We increased the Fe-content up to 5 mol% at0.5 mol% interval and the temperature anomaly up to250 K with 50 K interval, and for each pair of theFe and temperature anomalies we calculated thetraveltimes of the BC branch. Fig. 8a plots the rmstraveltime residuals as a function of temperatureanomaly for several Fe-rich models. The rms residualcurve for the 3 mol% Fe-rich model takes itsminimum at dT=200 K, where the residual value

157M. Obayashi et al. / Earth and Planetary Science Letters 243 (2006) 149–158

is comparable to that of M250T. In this model(referred to as M200I) the slow anomaly at 410 kmdepth is 2.0% and the discontinuity depression is 22km relative to IASP91 (broken line in Fig. 9).

In case (2), we placed the 50 km-thick partiallymolten layer above the discontinuity. Such a layer islikely to exist, as will be discussed later, if the transitionzone between the 410- and 660-km discontinuities ishydrated and if hot upwelling rises through this hydratedtransition zone [27]. We assumed that the effect ofpartial melting is to reduce seismic velocities with littlechange in discontinuity depth. We imposed a constantvelocity reduction of up to 2.0% in the partially moltenlayer. Fig. 8b plots the rms residuals as a function oftemperature anomaly for several partial melt models.The rms residual curve for the melt model with 1.0%velocity reduction takes its minimum at dT=200 K,where the rms value is as small as that of M250T. A P-wave velocity reduction of 1.0% corresponds to a meltfraction of 0.3–1.0%, the uncertainty being dueprimarily to unknown partial melt geometry [28]. Thetotal decrease of P-wave velocity due to high temper-ature and partial melt is 2.2% at 410 km depth and thediscontinuity depression is 26 km relative to IASP91.The dotted line in Fig. 9 shows the Vp model in this case(M200M).

Either the Fe-rich model or the partial melting model,together with a temperature anomaly of 200 K, canproduce a slower anomaly of about 2.0% relative to the1.5% anomaly obtained tomographically. Following is ascenario based on the partial melting model, whichappears to be more compatible with the required hightemperature anomaly. The excess temperature of 200 Kis at a level of the minimum estimate for the hotanomalies beneath the Hawaiian swell (200–300 K)[29], yet it would be high enough for the olivine-dominant upper mantle to exceed its wet solidus.Although water can be little stored in the olivine-dominant upper mantle, a mechanism of supplyingwater from below has been suggested [27,30]. Becauseof a large sustainability of water in wadsleyite, thewadsleyite-dominant transition zone is likely to containa large amount of water. When hot upwelling risesthrough the hydrated transition zone, it can absorb waterby reaction with wet wadsleyite. The consequenthydrated hot upwelling would release water by phasetransition from wet wadsleyite to olivine at the 410-kmdiscontinuity [23] and hence partial melting wouldoccur atop the discontinuity. This hydrous melt isheavier than the ambient mantle [31,32] and is able toexist stably to form a low velocity zone in some depthrange above the discontinuity.

On the tomographic cross-section in Fig. 2a thelow velocity anomaly near 400 km depth apparentlycontinues weakly down to the mid mantle where theanomaly becomes again intense, suggesting that thehot upwelling originates in the mid mantle. Theseafloor above the hot upwelling, on the other hand,has been suggested to be a site of submarinevolcanic activity [33] (named as Petit-spot; see thered triangle in Fig. 2a) which produced alkali pillowbasalts with 40Ar–39Ar ages of about 6 Ma that werecarried westward with the moving Pacific plate andare now sampled from the toe of the oceanwardslope of the northern Japan Trench [33]. In factalkali basalts with ages less than 1 Ma werediscovered by the 2004 expedition for the Petit-spotat the expected site in the eastern flank of the outerrise of the Japan Trench [34]. We speculate that thePetit-spot activity is the ultimate surface expressionof the hot upwelling.

Acknowledgements

We are grateful to the members of the J-array projectfor use of their data. We also thank National ResearchInstitute for Earth Science and Disaster Prevention foruse of the Hi-net (high-sensitivity seismograph network)data.

References

[1] R. van der Hilst, R. Engdahl, W. Spakman, G. Nolet,Tomographic imaging of subducted lithospere below northwestPacific island arcs, Nature 353 (1991) 37–42.

[2] Y. Fukao, M. Obayashi, H. Inoue, M. Nenbai, Subducting slabsstagnant in the mantle transition zone, J. Geophys. Res. 97 (1992)4809–4822.

[3] Y. Fukao, S. Widiyantro, M. Obayashi, Stagnant slabs in theupper and lower mantle transition region, Rev. Geophys. 39(2001) 291–323.

[4] E. Ito, E. Takahashi, Postspinel transformations in thesystem Mg2SiO4–Fe2SiO4 and some geophysical implica-tions, J. Geophys. Res. 94 (1989) 10637–10646.

[5] T. Katsura, E. Ito, The system Mg2SiO4–Fe2SiO4 at highpressures and temperatures: precise determination of stabilities ofolivine, modified spinel and spinel, J. Geophys. Res. 94 (1989)15663–15670.

[6] T. Katsura, et al., Olivine–wadsleyite transition in the system(Mg, Fe)2SiO4, J. Geophys. Res. 109 (2004), doi:10.1029/2003JB002438.

[7] J.E. Vidale, H.M. Benz, Upper-mantle seismic discontinuitiesand the thermal structure of subduction zones, Nature 356 (1992)678–683.

[8] M. Barazangi, B. Isacks, Lateral variations of seismic-waveattenuation in the upper mantle above the inclined earthquakezone of the Tonga Island Arc: deep anomaly in the upper mantle,J. Geophys. Res. 76 (1971) 8493–8515.

158 M. Obayashi et al. / Earth and Planetary Science Letters 243 (2006) 149–158

[9] D. Zhao, A. Hasegawa, H. Kanamori, Deep structure of Japansubduction zone as derived from local, regional, and teleseismicevents, J. Geophys. Res. 99 (1994) 22313–22329.

[10] A.T. Hsui, M.N. Toksöz, The evolution of thermal structuresbeneath a subduction zone, Tectonophysics 60 (1979) 43–60.

[11] H. Iwamori, Transportation of H2O and melting in subductionzones, Earth Planet. Sci. Lett. 160 (1998) 65–80.

[12] P.E. Keken, The structure and dynamics of the mantle wedge,Earth Planet. Sci. Lett. 215 (2003) 323–338.

[13] H. Bijwaard, W. Spakman, E.R. Engdahl, Closing the gapbetween regional and global travel time tomography, J. Geophys.Res. 103 (1998) 30055–30078.

[14] D. Zhao, Global tomographic images of mantle plumes andsubducting slabs: insight into deep Earth dynamics, Phys. EarthPlanet. Inter. 146 (2004) 3–34.

[15] J-Array Group, The J-Array program: system and present status,J. Geomagn. Geoelectr. 45 (1993) 1265–1274.

[16] B.L.N. Kenett, E.R. Engdahl, Travel times or global earthquakelocation and phase identification, Geophys. J. Int. 105 (1991)429–465.

[17] Y. Okada, K. Kasahara, S. Hori, K. Obara, S. Sekiguchi, H.Fujiwara, A. Yamamoto, Recent progress of seismic observationnetworks in Japan—Hi-net, F-net, K-NET and KiK-net, EarthPlanets Space 56 (2004) xv–xviii.

[18] K. Obara, K. Kasahara, S. Hori, Y. Okada, A densely distributedhigh-sensitivity seismograph network in Japan: hi-net byNational Research Institute for Earth Science and DisasterPrevention, Rev. Sci. Instrum. 76 (2005) 021301, doi:10.1063/1.1854197.

[19] R. Wang, A simple orthonormalization method for stable andefficient computation of Green's functions, Bull. Seismol. Soc.Am. 89 (1999) 741–773.

[20] J. Revenaugh, S.A. Sipkin, Seismic evidence for silicate meltatop the 410-km mantle discontinuity, Nature 369 (1994)474–476.

[21] T.A. Song, D.V. Helmberger, S.P. Grand, Low-velcoity zone atopthe 410-km seismic discontiuity in the northwestern UnitedStates, Nature 427 (2004) 530–533.

[22] J. Ita, L. Stixrude, Petrology, elasticity, and comparison of themantle transition zone, J. Geophys. Res. 97 (1992) 6849–6866.

[23] S.V. Sinogeikin, T. Katsura, J.D. Bass, Sound velocities andelastic properties of Fe-bearing wadsleyite and ringwoodite,J. Geophys. Res. 103 (1998) 20819–20825.

[24] D.J. Weidner, R. Wang, Phase transformations: implications formantle structure, in: S. Karato, et al., (Eds.), Earth's Deep InteriorMineral Physics and Tomography From the Atomic to the GlobalScale, Geophys. Monogr. Ser., vol. 117, AGU, Washington D.C.,2000, pp. 215–235.

[25] H. Yusa, T. Inoue, Compressibility of hydrous wadsleyite (β-phase) in Mg2SiO4 by high pressure X-ray diffraction, Geophys.Res. Lett. 24 (1997) 1831–1834.

[26] J. Chen, T. Inoue, H. Yurimoto, D.J. Weidner, Effect of water onolivine–wadsleyite phase boundary in the (Mg,Fe) 2SiO4 system,Geophys. Res. Lett. 29 (2002), doi:10.1029/2001GL014429.

[27] E. Ohtani, K. Litasov, T. Hosoya, T. Kubo, T. Kondo, Watertransport into the deep mantle and formation of hydroustransition zone, Phys. Earth Planet. Inter. 143 (2004) 255–269.

[28] W.C. Hammond, E.D. Humphreys, Upper mantle seismic wavevelocity: effects of realistic partial melt geometry, J. Geophys.Res. 105 (2000) 10975–10986.

[29] N.H. Sleep, Hotspots and mantle plumes: some phenomenology,J. Geophys. Res. 95 (1990) 6715–6736.

[30] D. Bercovici, S. Karato, Whole-mantle convection and thetransition zone water filter, Nature 425 (2003) 39–44.

[31] K.N. Matsukage, Z. Jing, S. Karato, Density of hydrous silicatemelt at the conditions of Earth's deep upper mantle, Nature 438(2005) 488–491.

[32] T. Sakamaki, A. Suzuki, E. Ohtani, Stability of hydrous melt atthe base of the Earth's upper mantle, Nature 439 (2006)192–194.

[33] N. Hirano, K. Kawamura, M. Hattori, K. Saito, Y. Ogawa, A newtype of intra-plate volcanism: young alkali-basalts discoveredfrom the subducting Pacific Plate, northern Pacific Japan Trench,Geophys. Res. Lett. 28 (2001) 2719–2722.

[34] N. Hirano, et al., New type volcanism along fractures on theflexural Cretaceous Pacific Plate, Eos Trans. AGU 86 (52) (2005)(Fall Meet. Suppl. Abstract V51A-1461).