JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 99, NO.

Bll, PAGES 22,313-22,329, NOVEMBER 10, 1994

Deep structure of Japan subduction zone as derived

from local, regional, and teleseismic events
Dapeng Zhao
SeismologicalLaboratory, California Institute of Technology,Pasadena

Akira Hasegawa
ObservationCenter for Prediction of Earthquakesand Volcanic Eruptions, Faculty of Science,
Tohoku University, Sendal, Japan

Hiroo Kanamori
SeismologicalLaboratory, California Institute of Technology,Pasadena

Abstract. We have determined a detailed three-dimensionalP wave velocity

structure of the Japan subductionzone to 500-km depth by inverting local,
regional, and teleseismicdata simultaneously. We used 45,318 P wave arrivals
from 1241 shallowand deep earthquakeswhich occurredin and around the Japan
Islands. The arrival times are recordedby the Japan University SeismicNetwork
which coversthe entire Japan Islands denselyand uniformly. We also used 4211
travel time residualsfrom 100 teleseismiceventswhich are read from seismograms
recordedby seismicstationsin northeasternJapan. In comparisonwith the previous
resultsobtainedfrom only local and regionalevents,the presentresult for the area
around the lower plate boundary and the mantle below the plate is determined
more reliably becauseof the addition of 7035 data from 100 teleseismicevents and
41 very deep earthquakes. In the crust and uppermostmantle, low-velocity zones
are clearlyvisiblebeneathactivevolcanoes.In the mantle wedgethe low-velocity
zonesgenerally parallel with the slab and exist continuouslyto a depth of about
200-km, which is consistentwith the petrological,geochemicaland geodynamic
studies. We considerthat the existenceof volcanism-relatedlow-velocity anomalies
in the mantle wedgeis a general seismologicalcharacteristicof subductionzones,
in light of all the availabletomographicresultsfor many subductionzonesin the
world. The Pacific slab beneath Japan is imaged more clearly than in previous
studiesas a high-velocityzone with a thicknessof 80-90 km and a P wave velocity
4-6% higher than the normal mantle. Lowervelocityanomaliesare found in the
mantle below the slab.

Introduction fiected waves at the upper boundary of the subducting

Pacificplate and demonstratedthat the plate boundary
Sincethe pioneeringworkby Wadati[1935],whofirst is a very sharp seismicvelocitydiscontinuity[Okada,
demonstrated the existence of deep earthquakes be- 1971, 1979; Hasegawa et al., 1978; Matsuzawa et al.,
neath the Japan Islands, continuingefforts have been 1986, 1990; Obara and $ato, 1988; Obara,1989]. The
made to study the deep structure of this typical sub- plate boundary estimated from the later arrivals is lo-
duction zone. The lateral variation of seismicvelocitycated just above the Wadati-Benioff seismiczone deter-
and attenuation structures beneath Japan was noted mined from reliable hypocentral locations.
earlier by many investigators[e.g., Katsumata,1960; Many researchershaveusedseismictomographymeth-
Hisamoto,1965a,b; Utsu,1967;Kanamori,1968]. Since odsto investigatethe three-dimensional (3-D) velocity
the advent of plate tectonics, the anomalousstructure and attenuation structures beneath Japan, on both re-
has been associatedwith the subducting Pacific plate gional and local scales. For detailed reviews, see Zhao
[Utsu, 1971], which has highervelocityand lower at- [1991],Zhao and Hasegawa[1993],and Hasegawaet al.
tenuation than the surroundingmantle. Later, many [thisissue].Thesepreviousstudiesconfirmedthe exis-
researchersdetectedand analyzedconvertedand/or re- tence of high-velocity and low-attenuation zonescorre-
spondingto the subductedPacific and Philippine Sea
Copyright 1994 by the American GeophysicalUnion.
plates, and revealed low-velocity and high-attenuation
Paper number 94JB01149. zonesin the crust and mantle wedgebeneath active vol-
0148-0227/94/94JB-01149505.00 canoes.

22,313
22,314 ZHAO ET AL.: DEEP STRUCTURE OF JAPAN SUBDUCTION ZONE

Recently,Zhao et al. [1992a]and ZhaoandHasegawa 145E

[1993]usedan improvedtomographicmethodto study
129E 37E 46N
the deep structure of the Japan subduction zone. They
NSTA: 216
have taken into accountthe effect of the complexshapes
of seismicvelocity discontinuitiessuchas the Moho and
Conrad discontinuitiesand the upper boundary of the
subducting Pacific plate. They used a great number 42N
of both first and later arrivals from shallow and deep
earthquakes, and adopted an efficient 3-D ray tracing JAPAN SEA

technique to calculate travel times and ray paths. As

a result, they determined a detailed 3-D velocity struc-
ture of the Japan subductionzone with a much higher 38N
resolution than the previousstudies.
Zhao et al. [1992a]and Zhao and Hasegawa
[1993]
used local and regional shallow and deep earthquakes,
and determined well the structure of the crust, mantle
wedgeand around the upper boundary of the subduct-
34N
ing Pacific plate. However, their result for the area
around the lower boundary of the slab and the man-
tle below the slab is less reliable because that area is PACIFIC OCEAN
sampledby only the downwarddiving rays from deep 30N
earthquakes which occurred within the slab, and the
rays do not crisscrosswell there.
Figure 1. Locationsof 216 seismicstationsusedin this
In this study, we have extended the tomographic study. Teleseismicdata are recordedby the stations in
method of Zhao et al. [1992a] to make it applica- northeastern
Japan(solidtriangles).Localandregional
ble to simultaneoustomographicinversionsof local, re- earthquake data are recordedby all the stations. Main
gional, and teleseismicevent data. Applying this up- districts in the study area are labeled.
dated method to a large number of high-quality data
recordedby the Japan University SeismicNetwork, we
have improvedthe tomographicimage of the Japan sub- exceptfor Hokkaidoand Kyushu, where there are fewer
duction zone. The present result providesa clearer pic- stations. This new data set is superior, in both station
ture of the subducting Pacific plate, especiallyfor the densityand arrival time accuracybecauseof the higher
area around the lower plate boundary, and the man- sensitivityof seismographs, to the Japan Meteorologi-
tle below the plate, thanks to the addition of data from cal Agencydata set which has been routinely reported
teleseismiceventsand more deepearthquakes.The new to the International Seismological Center and has been
information obtained in this work may help to improve usedso far in the tomographicstudiesin Japan.
our understandingof the dynamic processof subduction Zhaoand Hasegawa
[1993]used50,919arrivaltimes
zones.
from 512 shallow earthquakes and 688 intermediate-
depth and deep earthquakes. The data recorded by
the seismic stations of Tohoku University, which are
Data
situatedin northeasternJapan(trianglesin Figure 1),
are all re-picked directly from the original seismograms.
The data from other stations are collected from JUNEC
In this study, we combined two setsof data to conduct (July 1985- December1986). The data set consists
of
tomographic inversions. One consistsof arrival times many arrivals of SP convertedwavesat the Moho dis-
from shallow and deep earthquakes which occurred in continuity,and PS and SP convertedwavesat the upper
and around the Japan Islands, and the other, travel boundary of the Pacific plate, which were pickedfrom
time residuals from teleseismic events.
seismograms recordedby the seismicnetworkof Tohoku
Local and regional data University(for detailsof the later arrivals,seeZhao et
al. [1992a]).In this study,we added2824 P wavear-
We used local and regional earthquake arrival times rival times from 41 very deep earthquakesto the data
selectedfrom the Japan University Network Earthquake set of Zhao and Hasegawa[1993]. The new data were
Catalog(JUNEC) (July 1985- December1989). The collectedfrom the recent JUNEC issuesfrom January
data set is compiled by the Earthquake ResearchInsti- 1987 to December 1989. All of the 41 events have focal
tute, University of Tokyo, from P and $ wave arrival depths from 300- to 600-km.
times from earthquakes which occurred in and around Figures 2a and 2b show epicentral distributions of
Japan and were observed by the separate seismic net- the 512shallowevents(0-40km) and729 (688plus41)
worksof eight nationaluniversities
in Japan[Tsuboiet intermediate-depthand deep events(40-650 km), re-
al., 1989]. Figure i showsthe geographiclocationsof spectively.We can seefrom Figure 2a that the shallow
the 216 stationswe used. This united seismic network earthquakes are, on the whole, uniformly distributed
densely and uniformly coversthe entire Japan Islands in the Japan Islands. Most of the deep earthquakes
 ZHAO ET AL' DEEP STRUCTURE OF JAPAN SUBDUCTION ZONE 22,315

(a) (b)
129E 14 E
137E 145E 129E 37E
DEP.' 40-650 KM
DEP: 0-40 KM
NEQS - 729 ß
NEQS- 512 44N

40N

36N

32N
40-150
• 150-30
a300-6
,300-65(
NEW
28N
Figure 2. (a) Epicentrallocationsof 512 earthquakeswith focal depthsshallowerthan 40-km.
(b) Epicentrallocationsof 729 earthquakeswith focal depthsfrom 40- to 650-km. Epicenters
are shownin differentsymbolsaccordingto the focal depth. Cross,40-150 km; open triangle,
150-300km; solid triangle and solid circle,300-650km. Solid trianglesdenotedeepeventsused
by Zhao and Hasegawa[1993];solidcirclesdenote41 deepearthquakeswhich are newly added
in this study.
are associated with the subduction of the Pacific plate, cal componentseismogramsat 67 seismicstationsof the
manyof them deeperthan 400-km(Figure2b). Earth- seismic network of Tohoku University, which are situ-
quakesdeeper than 30-km beneath southwesternJapan atedin northeastern
Japan(solidtrianglesin Figure1).
are associated with the subduction of the Philippine The data set includes those at two stations of Hokkaido
Sea plate. The 41 deepearthquakesnewly collectedare University, three stations of Hirosaki University, and
shownby dots in Figure 2b. Note that there are only three stations of University of Tokyo. The teleseismic
a total of 128 deepevents(> 300-km) in the original events selected occurred between November 1988 and
data set of Zhao and Hasegawa[1993](solidtriangles October 1990. Figure 4 showsthe geographiclocations
in Figure 2b) becauseonly a few JUNEC issueswere of the events. We can see that the events selected have
availablewhen they collectedthe data. The addition of a fairly complete azimuthal range except for gaps at
the 41 new deep earthquakesin this work significantly 500-900 and 90ø-1450 from north. All events were at
improvesthe coverageof deepeventsbeneath the Japan epicentral distances between 300 and 900 from north-
Sea, whichplayed an important role in the improvement eastern Japan; thus the effects of complex structures
of resolutionof the tomographicimage, as describedbe- of the core-mantle boundary and the upper mantle are
low. avoided. We only read those P wave arrivals which have
The numbers of P, $ and later arrivals for the total a distinct first motion. Waveform characteristics were
1241earthquakesare 45,318, 8141 and 284, respectively. visually correlated between stations to minimize read-
The accuracyfor most P arrival times is consideredto ing errors and the effectsof attenuation. In total, 4211
be approximately 0.1 s. The accuracyfor some $ ar- P wave arrival times were collected from the 100 tele-
rivals, later arrivals and P arrivals with large epicentral seismic events. The number of arrivals for each event
distances(e.g., > 300-km)is relativelylow, being0.2- rangesfrom 25 to 65. The accuracy of the arrival times
0.3 s. Figure 3 showsthe numberof P wavearrival times is estimated to be 0.1 s for most of the teleseismic data.
versusepicentral distances. Most of the data have epi-
central distances from 50- to 300-km. There are many Method
data with ray paths longer than 900-km. The arrival
times are corrected for the ellipticity of the Earth.
In this study we modified the original tomographic
method of Zhao et al. [1992a]to combineteleseis-
Teleseismic data
mic residualswith local and regionalearthquakearrival
We read first P wave arrivals for 100 teleseismic times in tomographicinversions.This method can deal
events of magnitude M > 5.8, from short-period verti- with a generalvelocity model in which complexvelocity
22,316 ZHAO ET AL.' DEEP STRUCTURE OF JAPAN SUBDUCTION ZONE

rivals. The nonlinear tomographicproblem is solvedby

iteratively conducting linear inversionsuntil a conver-
6000
gent solution is found.
For local and regional earthquakeswhich are located
in the modelingspace,an observed
travel time T.ø.bs
from the fih event at the ith station can be written as
0•4000
bs

2000

I I I I I I
0 300 600 900 1200
DELTA (KM)
Figure 3. Number of P wave arrival times from local
and regional events versusepicentral distances.
discontinuities exist and the velocitychangesin three-
dimensions.The medium under study is divided into
layersby the discontinuities.Three-dimensional grid
nodesare arrangedin every layer. Velocitiesat grid
nodesare taken to be unknownparameters.The veloc-
ity at any point in the model is calculatedby linearly
interpolatingthe velocitiesat the eightgrid nodessur-
roundingthat point. Local and regionalearthquakes
are relocated in the inversionprocess.Arrival times of
later phasescan be used easily in addition to first ar-
N mologicalCentre. Raw travel time residualscan also be
W E
40*
60*
80*
where
T. c.al
- •3 iscalculated
traveltime'
' Tj ' Aj' hj' Tojare
latitude, longitude,focal depth, and origin time of the
1500 jth event; V• is velocityat the kth grid node; A denotes
perturbationof a parameter;and Eij are higherorder
terms of perturbations and error in the observation. We
write the travel time residual as
tij- T,"øbs
-- T. cal
. (2)
Ray paths of seismicwavesfor the local and regional
earthquakesare entirely located in the modelingspace.
We usedthe efficient3-D ray tracing techniqueof Zhao
et al. [1992a]to trace rays betweenhypocentersand
receivers and calculate theoretical travel times T .c.aland
travel time residualstij.
For the teleseismicevents, we use relative travel time
residuals in order to minimize errors introduced by
hypocentralmislocationsand origin times, and the path
effect outside the modeling space. Hypocenterswere
obtained from the Bulletin of the International Seis-
writtenasequation(2), but wecalculatethe theoretical
traveltimes
Wi•
.alin a different
way.Wefirstused
the
iasp911-D Earth model [Kennettand Engdahl,1991]
to determine a teleseismicray betweenthe hypocenter
and receiverand find the intersectionbetweenthe ray
and the bottom plane of the modelingspace.Then we
trace the ray path between the intersectionand receiver
by usingthe 3-D ray tracerof Zhao et al. [1992a].A
relativetravel time residualvii is then calculatedfrom
individualstationresidualsby removingthe meanresid-
ual tj averagedoverall stationsfor eachevent:
- (a)
100 ø
The mean residualis given by

- i "•J
tj = • tij (4)
mj i-•

wheremj is the numberof observedarrivalsfor the jth

event.
These two kinds of residuals form a whole set of data
with a column vector d of dimension N. The entire
Figure 4. Epicentral locationsof the 100 teleseismic
eventsusedin this study. Equidistant-azimuthalprojec- source and medium parameters with a column vector
tion is used. The map center is the center of the seismic in can be arrangedas
networkof TohokuUniversity(38.5ø, 140.5ø).The con- naT -- (A•l, A,•l, Ahl, ATol,...,A•M, A•M,
centric circlescorrespondto epicentraldistancesof 200,
40ø, 60ø, 800 and 100ø. AhM,AToM,AV•, ..., AVK) (5)
 ZHAO ET AL.: DEEP STRUCTURE OF JAPAN SUBDUCTION ZONE 22,317

where K is the number of medium parameters, and M gain an insight into the usefulnessand importanceof
the number of local and regionalevents. the data. First, the relative residualsfor each station
The columnvectorsspecifyingthe data set and the are averagedover all 100 events, resulting in a mean
source and medium parameters can be related as the relative travel time residual
observation equation
i ni
d - Gm + e (6)
ni j--1
wheree is an error vector,and G is an N x (4M + K) where ni is the number of events observed at the ith
matrix, whoseelements consistof the coefficientsof tke
station. These are plotted in Figure 5e. Second,we
derivativesin equation(1).
separatethe eventsinto sourcequadrantsaccordingto
There are severalwaysto solveequation(6). In the the azimuth of the source from the network, also re-
early tomographicstudies,Aki and Lee [1976]and Aki ferred to as the source azimuth or source direction. We
et al. [1977]adopteda dampedleast squaresmethod.
calculate the mean relative residualsfor four quadrants:
They use
NE = (0ø,90ø),SE = (900, 180ø),SW = (180ø,270ø),
(GTG+ e)m- GTd (7) NW = (270ø,360ø). The meanresidualsfor eachsource
quadrant are plotted in Figures 5a-5d.
which
canbeobtained
byminimizing
I d-Gm ]2+ The mean relative residuals, referred to simply as
mTOm, where
O isthedamping
factormatrix.The residuals,averagedfor all 100 events(Figure 5e), pro-
advantageof this method is that the resolutionand co- duce a spatial pattern of early arrivals at stations near
variancematricescan be obtainedtogetherwith the so- the Pacific coast, especially in southeasternTohoku,
lutionvector.The drawbackis that the matrix GIT(] and late arrivalsat stationsalongthe volcanicfront and
needsa large memory storageand its inversionis time the Japan Sea coast. This pattern is generally consis-
consuming, though 131is a relatively sparse matrix. tent with the main structural feature of the Tohoku re-
Thus this approachcan only deal with a relatively small gionasderivedfrom localearthquakestudies[e.g.,Zhao
tomographic problem with numbers of data and un-
knowns less than a few thousand. Note that it is not
138E 142E 138E 142E 138E 142E
uncommon that hundreds of thousands to millions of
data are used to solve tens to hundreds of thousands of
41N
unknownparametersin modern tomographicproblems.
The damped least squares approach has been im-
proved with the introduction of the parameter separa-
39N
tion technique[Pavlis and Booker,1980; Spencerand
Gubbins,1980]. This techniqueexploitsthe ability to
constructan orthogonaltransformationfor each event
that annuls the hypocenter componentswhile trans- 37N
forming but still preserving the velocity part of the
problem. However, the parameter separation itself is
a nontrivial operation and becomesformidable when a
greatnumber(e.g.,thousands)of earthquakes
are used.
Recently, iteration type inversion techniqueshave
beenusedto solvedirectlythe observation
equation(6),
to avoidthe requirementfor the large storagespaceand
theformidable
computation
ofGT 131.
These
techniques
are,e.g.,the algebraicreconstruction
technique(ART)
[Herman,1980],the simultaneous iterative reconstruc-
tion technique(SIRT) [Humphreys and Clayton,1988],
andthe conjugate gradient(CG) method.In thisstudy,
we have useda CG type solver,the LSQR algorithm of
Paige and Saunders[1982],to solvedirectly the large
and sparseobservationequation(6). The LSQR algo-
rithm has been usedby severalresearchers[e.g., No- Figure 5. (a)-(d) Relativetravel time residualsper
let, 1985; $pakman and Nolet, 1988; Lees and Crosson, sourcequadrant, calculatedwith respectto the network
1989;Zhao et al., 1992a]and is confirmedto be a very average.Quadrantsare as follows:(a) northwest,(b)
efficientalgorithmto solvelarge tomographicproblems. northeast,(c) southwest, (d) southeast.(e) Relative
travel time residualsfor all 100 teleseismicevents. (f)
Analyses Locationsof activevolcanoes (solidtriangles)in north-
eastern Japan. Five volcanoeswhich are mentioned in
Teleseismic Residuals the text are labeled. The scale for the size of the residu-
Before conductingtomographicinversions,we first als in Figures5a-5eis shownat bottom. Solidand open
display the observed teleseismicdata in two ways to circlesdenote early and delayed arrivals, respectively.
22,318 ZHAO ET AL.: DEEP STRUCTURE OF JAPAN SUBDUCTION ZONE

et al., 1990a,b, 1992a,b]. Largedelayedarrivalsexist Tohoku. This was confirmedby conductinga numer-
for stationsin volcanicareas,e.g., Akita-Komagatake, ical simulation with a model having the geometry of
Iwaki-San,Chokai-San,and Azuma-San(seeFigure5f the Pacificplate as that shownin Figure 6 [Zhao and
for locationsof the volcanoes).Althoughstationsin Hasegawa,1994].
the Nasudake volcanic area show early arrivals, they These observations demonstrate that the teleseismic
are much later than thoseof the adjacentstations. data, independentof the local and regional data, con-
The pattern and magnitude of the residualsfor the tain a wealth of important information on the deep
events from the NW, NE and SE quadrants are all structure of the Japan Islands.
similar(seeFigures5a, 5b, 5d). The arrivalsat sta- FollowingZhaoet al. [1992a]and ZhaoandHasegawa
tions along the volcanicfront and the Japan Sea coast [1993],we conductedtwo kindsof inversions.The first
are later than those on the Pacific coast. This is be- is free inversions,in which we did not introduce any a
cause the rays to stations in central and western To- priori information into the starting model; the second
hoku travel longerin the slow mantle wedge. The early is constrained inversions,in which the Pacific plate is
arrivals in northern Tohoku and southern Hokkaido for taken into account in the model.
the eventsin the NE quadrant are probably causedby
Free Inversion
the rays traveling in the Pacific plate subductingbe-
neath the Kuriles. The pattern and magnitudeof the The major aim of the free inversionis to know if and
residualsfor eventsin the SW quadrant(Figure5c) are how well we can image the subductingPacificplate by
very different from thosefor the other quadrants. The
stations in southern Tohoku and northern Kanto show
using the present method and data set. The starting
1-D velocity model for the free inversion is the same as
very early arrivals, which are consideredto be caused
that usedby Zhao and Hasegawa[1993]. The medium
by the changeof the geometryof the Pacificplate be- under study is divided into three layersby the Conrad
neath Kanto. We can seefrom Figure 6 that the Pacific and the Moho discontinuities. The three layers corre-
plate south of Kanto dips toward the southwestand spond to the upper crust, lower crust and upper man-
has a larger dip angle(,-, 400 - 45ø) than that north tle, respectively. The upper crust and lower crust are
of Kanto (,-, 30ø). The incidentangleof the teleseis- two constant-velocitylayers with a P wave velocity of
mic raysis generallysmall(within ,-, 300of the vertical 6.0 and 6.7 km/s, respectively.The Conradand Moho
direction). Thus, for eventsin the southwestern
quad- discontinuity are curved surfaceswith two-dimensional
rant, becauseof the coincidenceof the ray direction geometriesas shown in Figure 7. This crustal model is
and the dip direction of the steeperpart of the Pacific determinedby inverting a large number of arrival times
plate, the rays arriving at stations in northern Kanto from shallowearthquakes[Zhao et al., 1992b],and is
and southernTohokutravel longerin the high-velocity confirmed to represent well the seismicvelocity struc-
Pacificplate than the raysat the northernareas,result- ture of the crustin the JapanIslands[Zhaoet al., 1992b;
ing in earlier arrivals in northern Kanto and southern Zhao and Hasegawa,1993]. In the upper crust, lower
crust and upper mantle, we individually set 1, i and
132E 136E 140E •44E
16 layersof grid meshes.The separationbetweengrid
44N nodes is 25-35 km in the horizontal direction and 15-35
km in the vertical direction. Figure 8 showsthe con-
figuration of the grid mesh in a horizontal plane. The
depth and initial velocity of each mesh layer are the
sameas thoseusedby Zhao and Hasegawa
[1993](see
Table i in their paper).
40N Applying our techniqueto all the local, regionaland
teleseismicdata and adopting the initial velocity model
as describedabove,we performeda tomographicinver-
sion. The teleseismic residuals are corrected for the un-
50KM dulations of the Conrad and Moho discontinuities. Be-
causeno plate boundary exists in the model of the free
, 36N inversion, we did not use the 248 arrival times of con-
verted wavesat the plate boundary. We simultaneously
solved the inverse problem for 1241 x 4 hypocentral
parameters and 13,749 velocity perturbations for grid
nodeswith hit counts(numberof rayspassingthrough
eachgrid) largerthan 10. More than 99%of thosegrid
32N nodeshave hit countslarger than 20. The root-mean-
square(rms) travel time residualis 0.521 s beforethe
Figure 6. Depth distributionof the upperboundary inversion and reduced to 0.375 s after the inversion. The
of the subductingPacificplate in andaroundtt,e Japan variance reduction amounts to 48%. Two vertical cross
Islands[afterZhaoandHasegawa, 1993].Locationsof sectionsof P wavevelocityperturbationsin percentde-
the crosssectionsin Figure 9 and Plate i are shown. terminedby the free inversionare shownin Figures9a
 ZHAO ET AL.' DEEP STRUCTURE OF JAPAN SUBDUCTION ZONE 22,319

N • N
Conrad Depth

UNIT-KM

•o

Moho Depth

(b ) UNIT'KM
(a)

•igure 7. Depthdistribution of (a) theConradand(b) theMohodiscontinuities [afterZhaoet

al., 1992b].Solidlinesdenotedepthcontours,dashedlinesthe standarderrorsof the estimated
depths.The outerdashedline encloses
the regionwherethe standarderroris lessthan 2-km,
and the inner one less than 1-km.

and 9b. Locations of the cross sections are shown in tion). This improvement
of the slabimageis easyto
Figure 6. understand. The teleseismicrays propagatebasically
For comparison,we alsoconductedan inversionwith in a vertical direction. They crisscross
with the hori-
the samemethod and starting model but only usedthe zontallypropagatingraysfrom the deepeventsin the
data from the 1241local and regionalevents.The result slab,resultingin goodcoverage
of raysaroundthe slab
is shownin Figures 9c and 9d along the same profiles and the mantle below the slab. We can see in Figure
as Figures9a and 9b. We can seethat the imagesfor 9 that a continuoushigh-velocityzoneis clearlyvisible
the crust,mantle wedgeand the slab in Figures9c and from 50 to 400-500 km in the two cross sections. The
9d are very similar to thosein Figures9a and 9b. The slablike zone has a thickness of 70-100 km and a P wave
differencesare in areasaround the lower plate boundary velocity2-6% higherthan the normal mantle.
and the mantle belowthe plate. Note that by just using The improvedslabimagein Figure9, however,is still
the local and regional data, we have achieveda much not very satisfactory.The upper and lowerboundaries
better image of the high-velocityPacificslab than that of the high-velocity zoneshowa sawtoothshape.Rel-
by ZhaoandHasegawa [1993](this canbe seenby com- atively low velocityanomaliesare visiblein someareas
paring Figure 8b in their paper with the presentFigure within the high-velocityzonecorresponding to the Pa-
9c). The major reasonfor this differenceis considered cific plate. The high-velocityzone is even brokenby
to be the 41 deep earthquakeswe added in the present the low-velocityanomaliesin severalplaces.Obviously,
data set. Zhao and Hasegawa[1993]usedtoo few deep the imagesin Figure9 do not showthe realformof the
eventsto image the slab well with the free inversion. slabbut just its obscurepicturebecauseof the imperfect
With the addition of the teleseismicdata, although resolutionand parameterizationof the free inversions.
the tomographicimagesfor the crust,mantle wedgeand Constrained Inversion
the slab have changedlittle, the imagesare determined
more reliably, especiallyfor the area around the lower There are at least three reasons which motivated us
plate boundaryand the mantlebelowthe plate (seethe to conduct the constrained inversions in which we in-
resolution analysesas describedin the following sec- troducedthe high-velocityPacificplateinto the model.
 22,320 ZHAO ET AL.' DEEP STRUCTURE OF JAPAN SUBDUCTION ZONE

129E 133E 137E 141E 145E

45N

eeleleeee•eetleeeleleeee•eeel

41N

37N

eeeeeleee eeeee*e

,1i e9• ee,ii ee, ,11

•e 9E.e ,11
e e 9 e e '11
e ee e
33N
* "* ,• e t ,• e e e ,• e e • • e e e ß ,• e9 "•e t e e e • e e e e e e e
4k E e e

Figure 8. Configurationof the grid net in a horizontalplane.

First, by detectingand analyzingconvertedand re- geometriesof the Conrad and the Moho are the same
flected waves, many previous researchershave demon- as thoseshownin Figure 7. We take the depth distri-
strated the existenceof the upper boundaryof the Pa- bution of the upper boundaryof the subductingPacific
cificplate and foundthat the boundaryis a very sharp plate as that shown in Figure 6, which is constructed
seismicvelocitydiscontinuitywith a complexgeometry. by ZhaoandHasegawa[1993]by referringto the studies
Forreviews, seeZhao[1991],ZhaoandHasegawa [1993], using later arrivals and accurate hypocentral locations
and Hascgawaet al. [thisissue].Second,withouttak- of deepearthquakes.In the four layers,we individually
ing into accountthe plate boundaryin the model,we set 1, 1, 16 and 16 layersof grid meshes,which are the
cannot calculate accurately ray paths and travel times sameas thoseof Zhao and Hasegawa [1993](seeTable
for long rays from deep events, and thereforethe final i in their paper). The configuration of the grid meshis
tomographic image, especiallythose near the disconti- shownin Figure 8.
nuities, will be distorted[Zhao, 1991; Kokctsu,1991; From the resultof the free inversion(Figure9), we
Zhao and Hasegawa,1993].Third, later phasessuchas can see that the subductingPacific plate has a thick-
the reflectedand convertedwavesat the plate boundary, nessof 70-100km and a P wavevelocity2-6% higher
which are frequentlyobservedin Japan, cannotbe used than the normal mantle. Therefore in the constrained
in the tomographicinversionsif the plate boundaryis inversion,we assignthe initial thicknessand velocityof
not introducedinto the model. Later phasescontain the plate to be 85-km and 4% high, respectively.Ap-
very important information on the interior structure of plying our method to all the local, regionaland tele-
the Earth, especiallythe structure closeto the discon- seismicdata and adopting the initial model as shown
tinuities. When later phasesare used,hypocenterscan in Figure 10, we performed a constrainedinversion.In
be locatedmore accuratelyand ray coveragecan be this inversion we used all the later arrivals in addition
improvedbecausetheir trajectoriesare quite different to first arrivals. We solvedthe problemsimultaneously
from those of first arrivals.
for 1241x 4 hypocentralparametersand 16,061velocity
FollowingZhao andHascgawa [1993],we adopta ve- perturbationsfor grid nodeswith hit countslargerthan
locity model as shownin Figure 10 in the constrained 10. The rms travel time residual is 0.421 s before the
inversions. The medium under study is divided into inversion and reduced to 0.353 s after the inversion.
four layersby the Conrad, the Moho and the upper Two vertical crosssectionsof P wave velocity per-
boundaryof the subductingPacificplate. The four lay- turbations determinedby the constrainedinversionare
ers correspondto the upper crust, lower crust, mantle shownin Plates la and lb. A map view of the tomo-
wedgeand the mantle belowthe plate boundary. The graphic image at four depth levelsis shownin Plate 2.
 ZHAO ET AL.: DEEP STRUCTURE OF JAPAN SUBDUCTION ZONE 22,321

Honshu pacific A
A Japan
Sea .... .•,,
ß
ß ß eOO.**eOe-
Honshu
Pacifica ß
+++****...e..
Ocean/H• Japan Sea
........... oAo
..... t ..... Ocean
rt
......... -•__•t-_
Moho .................. ••==_oooooo ..... ++
:4t-+;+++..
(a)
......
.... ee .....
....oo..+.....o....oooo.....o..++-t--•++.+
eeoc .... eOOOOOOOOOe-*+
....
( C) Moho ...............................
.**eOOeo*..eeeee.*eOOOOeeeeeO-* ;;•-j:$::: ....
............. ee ...... s$s$ ......... s$+++* ....
:::::::::::::::::::::::::::::::::::::::::::
........... *** .... .o.o..***+++......++++....
_ ........... •eee--.*+***.****++* ..... ***.. - lOO
......... ee•ee...*++ .... **++++**''**+++++''
............ e•eee--.*+++*---*++++++*
................ **..*++++**+++++++** .... ..... ....
.... ** ....... ++++++*++-t--t--t-+++++++++* ..... **++* .... ::::;;:::::::;;;;;;•;•++ .................
+++++++++++++. .....
............... *+++++++++_'{-H-•'.++++++++* ..... +*** ..... ........... ...**+-•+++++++++++++o.+++.o.
..... . ..... .***++++++++++-H.++++****o**...
.... ....e..**++++++++++++-t-f-++ ............

- ...........
..........
...... ::::::::l::i:;::::11:•+
::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
.+**+++++.
.+++++****......+***
...... .++....+*.leee
.... ...lee
=========================
.....
.........
...eel...**
lice-*+*
....
....
...........
..........
........ **oo.+++***..**++++++++-H-o.•
o+++++o
..... -200•
......... .++++++........++++. ........ ee ......... eOOee.*++ .... .......... ..+++..::::;;•-+++++ ++++-H-+. ........
......
............ .+++++++..**...+++. ....... eeee ......... e•ee.++* ............ o.+++++..o.o+++++++++++++.
............ .+++++++ ........ ++ ....... eeeee .... ee.-.e==ee-*++ .... ............ **+++++..***+++++****+++..
.... ..... ..++..++++++ ..... ......ee..e•eeee .... e•e..eOOe..*+*.e-. .......... o****+++++.o**+++** .........
..... ***....o..++++++.***... .......
...........++..+•+++.
.... .........+++ +++. .........
......... eeeeeeOOeee
eeeOOOOOe ..... .... eeee.e*OOe..**
eee.eeOOe..*+*.ee. ..... ......
......
..***o.+++++++++.o**
+.o....+++++++++ .............
.........
- ......+++....+++++++
.+++++...+++++.. ......... eeooOOOOOOe..... .... eeOOoeeOOOe.-** ............
....
.... ..++++++++++..
........
..........
eeeeeoooooo
eeeeoooooe ......
oeoooeeoooo
eeooeeeeooe .........
-
..... oo++++++++++++++.
+++++++o.
300 •
.... ..+++++++++. ............ eeeoooooe--..*..eeooe.eeooe ......... ...... o++++++++++oo.oo .......
....... ..++++++++.. .......... eoeeeooooe--*** .... eooeeeeeoe ......... .......... ++++++++++**o
.......... +++++++++
........ ++++++++.. .......... eeeeee•ee-.**. ..... eeeeee .............
...... ..++++++.. ............ •eeeeee=e--*++* ...... eeeee ....... ß ..... ....... .o++++++++.ooo+** .....
....... **++++++******* ....
...... +.+++++** ........... •eeeeeeeee--**++* ...... eeeee ............. ......... ..+++ ............
...... .....++. ............ eeeeeeeee•..**+* ...... eeee• ..............
.......... ,,,+,,,,+ .......
...... ....**** ...... ß .... coooeec ...... **** .... eeeeeee ..............
_ ß ..... .o*****o .........
- ::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
......... ... .... eeoo..oeeoOee ......... * ...... coocOlo..ecole ........
4OO
........... +. .... eoOee-eeeeeOe ......... ** ...... eeoeOOeooeeoe .........

.eeeee© ...... eeeee .........................

::::::::::::::::: -6% 0% 6%
coOOil ..... eeOOee ............................ oleoeeoc ..... ::::::::::: ...... O o o o ..... + + +
eellee .... oleOel ............ * ................. eellee .......
5OO

........... ß ß o_o•.•,•x•- ....... +. ßu- +•++ ß•,.•u*u{• 0 ß ß +,+++. ß

Moho :::::::::::::::::::::::::::: ....... :: .... ::::: ............. Moho'::::':::::!!!!!!:tt:•:?

øøø'øe:::::::•i!ii-:!ii:::•i:;•;':
_(b) .....................................................
..............
..............
ooo0...* ..... ++++++++•:•-+. .oo.'
oo ..... + .... +lii i iil i ,, ' "' ......
_ ß (d) ..........................................................
ß ....... ß ß ß ß ß ß ooo ß ß ß ß ß ß ß ß. ß +++++++++++ ß ßß ßß ß ß ß ß _
loo
....... oOOOOoe ß .... *.I'+++*+1 i i i i1 Ii i ,+* ...... ::::::: ;•;;::::::;::::; ;tt•-•T+++.::::::::::
.......... oOOOOo
ß...... ++++++•m m• m:++++++
.... '{,•+'+ .......... oooooo ß ..... ß ++++++-•++ ß ..........
......... ß ooooo. ....... +++++:;;;;. ......... ........... oOooeoe ...... ,i,+ + + +-,I--{--I--{--I--• + + + ............
......... eoOOeeeeee.- ß '* * + + + +-,I-t-I.•-•+ .1..... ++* ........ .......... ß ,,o,e,,o ..... + ++•H-•++++ + ..............
......... + .......... .e.....+++++++++++. .............. A
.......... o ............. +++++++**+++++
...................... +++++++

200•
.................. ß .++.+++++.. o... ........ ++..oo .......
............. .. + +.++++++++++..oo... ...... .++. ß ß ß ....... ............. .+++++++++++++++. .... ****o.o***o .....
..... ß ....... ++++++++++o......oo .......... .. ........... ..... ß ..... o.++++++++++++++++ .....................
- ..... ...... ++++++++++.. ....... oo .... ß ß ................. - .....ee .... ++++++++++..+++++.. ....................
......... ee ß ß *.1- +++-I--t--I--1-•++ ,m....... eOe .... me ......... .e..o++++•++-H-++o*o.+++o. ................ ....
............. +++++'t-•+++''****ø ......................
.................
...':: 22::;:: 2;+,+++* ++:1:{:1:{:2:+
:,*+.+_m_•+.T.:
i::::: .': ' .':::.:::
ß ß .......

.................
ß ...... e. ............
.............
++***+'1-•++*
+++...+++,+++
...............
................
ß .............
. .........
........ o o ... ++++.+++++.+... ........................ . ........ ............ .++.++++++++++.. ........................
....... . ++++++++++++++. .... o o ß o ..... ß ........................ ....... o++++++++++++++++.++o. .......................
............... ++.+++.+.++. ß ....... .. ..... .. ........................ ............... **++++.++++**++++.....
........... $, +., $ + .1.+-t..t..t. + +. ,m...... eOeOOe, ,, cool ............. ee ........... .. .... +++++++++.....o.....

- ...:: 2..: ....:22,:.2.++++:

2:: +.•.:•1:++: .... oo.,i::
................................. ..........ooooe ß ß eooo ß ................... - - ====================================== 300 •L
........ ***++++++.e.++++++ ......... eeoeeeeeeooooee ........ eee .... ee ............... ++++*+***+++++* ...........
....... e++++++++++**++++**oee ...... eeoe--eeeOOOOee ........ eee...eeee ........... +*++++++***++++++ ............
....... ***+++++++++****+++*.*ee ...... eeOe..eoOOOOeee ........ ee ..... eee .......... **e+.++++++***+++++ .............
...... i•l .... :::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
,'..oo+
.... llli''''•J& .:.11 ............... oooo...oo ...............................
::::::::::::::::::::::::::::::::::::::
-++++-*.*+++- ...................
.... ++o. ,.+
i ''* ++.H-•_++++++++.... ............... eeOOOOeoeeeeee ........ ' ........... eeeeee ...++++++++**++*.ee*ee ..............
'**+!' .... " .... +++.eooeo-*++** ........ eOOOOOOeeeeeee ........... ..... ****e. .**++ ++ ++
+-00000.-*+*e* ........
++q-•4+***
..............................................................,.•+++++•
..++ .......
•:...........................................................
• . ++
• •i•
eoooo.
+ +++•+
+...............................................................
..................... oo
4OO
+ .... eeOOee .... eOOOOOOOOOOOOeeo.... ß ...... ß ........ ee- +
li•+.+!!i:::::::m:: ..................................................... ml,•m ....................

5OO

Figure 9. (a andb) Verticalcrosssections of fractionalP wavevelocityperturbations

(in
percent)obtained by•hefreeinversion
(see•ex•forde•ails).Local,regional
and•eleseismic
da•a
areall usedin •heinversion.
(c andd) SameasFigures9a and9b bu• only•he da•afromlocal
andregionaleventsareusedin the inversion.Crossandcircledenotehighandlowvelocities,
respectively.
The perturbation scaleis shownin Figure9c. The landareaandactivevolcanoes
areshownat the top of eachfigureby the boldhorizontallinesandsolidtriangles,respectively.
Locationsof the crosssectionsare shownin Figure 6.

For comparison,we alsoconducteda constrainedinver- i are, as a whole, similar •o •hosein Figure 9. The
sion usinglocal and regionalda•a only. The resuk is imagesof •he crus•,mantlewedgeandslabin Pla•eslc
shown in Pla•es lc and ld along •he same profiles as and ld are very similar to •hosein Pla•es la and lb.
Pla•es la and lb. We can see •ha• •he imagesin Pla•e However,•he resuk for •he area aroundand below•he
SURFACE
lower boundaryof •he slab in Pla•es lc and ld is less
reliable. Wi•h •he addition of •he •eleseismicda•a, •he
CONRAD
................. s•ruc•ure of •he mantle around and below •he slab is de-
HOHO termined much more reliably because•he ray coverage
•here is dramatically improved.
We also conducted a number of constrained inversions
by changing
•he initial •hickness
and initial velocityof
•he Pacific slab. We found •ha• akhough somevaria-
tionsoccurin •he amplkudesof velockyanomalies,•he
patternof •he •omographic imageasshownin Pla•es1
and 2 exhibits almos• no changewhen •he initial slab
•hicknessis in •he range of 10-120km and •he initial

Figure 10. Schematicillustrationof the seismicve-

locitymodeladoptedfor the constrained inversion(see
text for details). Dots denotegrid nodesof 3-D grid
nets individually set in the four layers. Hatched area
denotesthe subductingPacific plate.
22,322 ZHAO ET AL.: DEEP STRUCTURE OF JAPAN SUBDUCTION ZONE

(• ) H,t,aaa (• ) H,T,
aaa
O 0 0 0 0 0 0 0 0 0
0 0 0 0 0
0 0 0 0 0 ,• Oq

ß o
0

•X• .

ß
 ZHAO ET AL.: DEEP STRUCTURE OF JAPAN SUBDUCTION ZONE 22 323

(a) (b) (c) (d)

-6%

0%

6%

10km ' 24kin ;' 40kin ' 65km

Plate 2. FractionalP wavevelocityperturbations(in percent)at four depthsdeterminedby
the constrained
inversion(seetext): (a) 10-km,(b) 24-km,(c) 40-km,(d) 65-km. Red and blue
denotelow and high velocities,respectively.The perturbation scaleis shownon the left of the
figure. Solid trianglesdenoteactive volcanoesin the Japan Islands.

slab velocity is 1-7%. These tests indicate that the fi-
nal tomographicimage is well constrainedby the whole
data set from local, regionaland teleseismicevents,and
does not depend on the initial model very much.
The final rms residual of the constrained inversion
(0.353s) is smallerthan that of the freeinversion(0.375
s). To know if this rms residualreductionis statisti-
callysignificantor not, we usedthe F test [Draperand
Smith,1966].A detailedexplanationof the application
of the F test to tomographicmodels can be found in
the workby Zhao and tlasegawa[1993].Throughsome
simple calculations,we found that the decreasein the
rms residual for the constrainedinversionis surely sig-
nificant from a statistical point of view. Therefore we
prefer the result obtained by the constrainedinversion
and take it as our final result.
Resolution and Result
We have used the method of checkerboard resolution
test to assessthe adequacyof the ray coverageand to
evaluate the resolution. For details of the method, see
Humphreys and Clayton[1988],Inoue et al. [1990],and
Zhao and tlasegawa[1993]. Figure 11 showsthe result
of a checkerboardresolutiontest with a grid separation
of 33-km for the velocity model of the constrainedin-

Plate 1. (a andb) Verticalcrosssections offractionalP wavevelocityperturbations (in percent)

obtainedby the constrained inversion(seetext for details). Local,regionaland teleseismicdata
are all usedin the inversion.(c and d) Sameas Plates la and lb but only the data from local
and regional events are used in the inversion. Blue and red denote high and low velocities,
respectively.The perturbation range is shownin Plate lc. The land area and active volcanoes
are shownat the top of eachfigure by the bold horizontallines and solid triangles,respectively.
Open circlesdenoteearthquakeswhich are usedin this study and occurredwithin a 80-km width
along the profiles. Locationsof the crosssectionsare shownin Figure 6.
22,324 ZHAO ET AL.- DEEP STRUCTURE OF JAPAN SUBDUCTION ZONE

130E 144E 130E 144E 13 0E 144E

10 40 KM ::.' 120 KM 44N
ß

..o

..o

oeo, ß

0o.

38N

,32N
..

270 KM -"' 390 450 • 0.

.0 ......
4N
ß

,.oo.,......
...
.e.......

e.o.
ß
.... :•:..
........oo?o
..

.0oo.oo,..eo. ©ooo.OOo,O..©..
.e.

.::::;.O.o;:.. ...........
.eoeoO

O*.Oeoe.•O. ,..ooooo.
..o

3BN
.e

--*ooo•DOO0 o.o,o

.
.:;•:o.:•
...0....¾;• .:;..o•:.
,.o.oo.00

eo ''ø'ø'oO.OOO.O
' eO ..0.. eo
oooeo.o.

o.o;::$:'
eoeo,ooOooo o0o0o0oo00
32N
000 0 o ß ß ß ß 000

-3% 0% 3%
Figure 11. Resultof a checkerboard
resolution
test (seetext) for sixtypicallayers.The depth
of eachlayer is shownat the upperleftcornerof the map. The grid separationin this test is
33-km. Solidand open circlesdenotehigh and low velocities,respectively.The scaleis shownat
the bottom of the figure.

version. The resolutionis generally high in the central western and eastern Japan. The velocity under western
part of the study area, which is expected. It is gener- Japan except Kyushu is generally high and relatively
ally good for the layers shallower than 250-km. In the homogeneous, in contrastwith an averagelowervelocity
deeper areas correspondingto the subducting Pacific and a significantinhomogeneitybeneatheasternJapan,
plate, the resolutionis generallygood and the pattern where many active volcanoesexist. This pattern is con-
is, on the whole,correctlyreconstructed.At the edge sistentwith the distributionof the horizontalgradientof
and deepestpart of the study area, the pattern of the the gravity anomaly in Japan determinedby Kono and
checkerboard is not correctlyreconstructed.Figure 12 Furuse[1989](seealsoFigure16 in ZhaoandHasegawa
shows the result of another checkerboard resolution test [1993]).In Hokkaidoand easternHonshu,low-velocity
with a gridseparationof 50-km. We canseein Figure12 zones are distributed in and around active volcanoes.
that the resolutionis highand the checkerboardpattern In addition to the volcanic areas, low-velocityanoma-
is correctlyreconstructedfor most of the study area. lies in the upper crust are also presentnear Urakawa in
From the two resolutiontests, we can say that the Hokkaido and around Tokyo Bay in the Kanto District
tomographicimageobtainedin the presentstudy has a (seeFigureI for the locationsof the two areas).This
spatialresolutionof about 33-km for the upper250-km is consistentwith previous studies by Miyamachi and
depths and 33-50 km for the deeper areas. Note that Moriya [1984],Horie and Aki [1982],and Ishida and
the resolution, as a whole, becomesbetter than that Hasemi[1988].The low-velocityzonesin the two areas
of the resultdeterminedby Zhao andHasegawa [1993] correspondto the thick Quaternary sedimentary layers
for the there.
(seeFigures12 and13in theirpaper),especially
area around the lower plate boundary and the mantle Low-velocityanomaliesamountingto 6% are contin-
below the plate. uously distributed to a depth of about 200-km in the
The structureof the crust and the shallowerportion mantle wedge along the volcanic front from Hokkaido
of the upper mantle is very similar to that by Zhao and to Kanto-Tokai. The low-velocity zonesdip toward the
Hasegawa[1993](seePlates I and 2). Largeregional west and are generally parallel with the high-velocity
variationsexist in the crustalvelocitystructurebetween Pacificplate (see Plate 1). In Kyushu,low-velocity
 ZHAO ET AL.: DEEP STRUCTURE OF JAPAN SUBDUCTION ZONE 22,325

130E 144E 130E 144E 130E 144E

ß 44N
ß ,e

12KM ß
100KM ß 200KM o.
,,

ø• ßO
ß O• O• ß eO•
ß
O• ß tO• * col(
ß ß O• ß ß 004 ß ß ß oOoo

ß
eee• i i
38N
ß ß ß

ß ß 0
eø ß

coo. eo .... ß.... ßßß oeOe'O ß

32N

eeeoeoeO 44N
300KM"
ß
I,eO
O
ß 400KM. ß. .. o 500KM ...o
.oo o.
ß . .•00 ß ...eeoO0, •e..oOOO.o.Oo 0
eO ©eO©o-O• ©•eo oOooOO.0o
eeOlO• 'eO'lOeO ''0oOOIotoIooo,
ß ee.ooeOee 'tleo..lOlOeO.ol
ß eOOeOlO• tll-olotOtolllO
O0.OOOeO eo-e•oo.oo..
-.OoOOOo oOOOOOeO•O OeoOo.o•Oooee..
eeeeOlO• ß 'e-eOtOeOt eOO-ooOOoOOOO
.... '*-oI0t0©• ©''*tolOlOl• 38N
ß oOoO0• ß ee*Oto• ol-lO C•
'coOl ee•e, OeOoOe
eeOeO ß '-e.0o.' 0-0-o
-oo0•
•-o..o ß .Oo.e.o • OOe
• e.e, •t010' o.OeOI

eo oeOO•oe eoOo * ooe

32N
OOOo o o . ß cOOl

-3% 0% 3%

Figure 12. Resultof a checkerboard resolutiontest (seetext) for six typicallayers.The depth
of each layer is shownat the upperleft corner of the map. The grid separationin this test is
50-km. Solid and open circlesdenotehigh and low velocities,respectively.The scaleis shownat
the bottom of the figure.

zonesare also presentbeneath active volcanoesand ex- Plate 1 showsthat the upper boundary of the slab
tend to the upper mantle (seealsoFigure 9b of Zhao bends upwardsbelow about 200-km depth, while this
andHasegawa [1993]). feature doesnot showup in the result of the free inver-
The subducting Pacific plate has P wave velocities sions(Figure 9). The geometryof the plate boundary
4-6% higherthan the surroundingmantle. In the to- we adoptedhere(Figure6) is mainly determinedfrom
mographicimage, we can seethat the thicknessof the the distribution of deep earthquakes.We do not think
plate changesslightly at different depths, but in the that the seismicityalone can define this small bending
rangeof 80-90 km (seePlate 1). Within the Pacific curvaturechange,and thereforethis feature is consid-
plate the P wave velocity has variationsof about 2%. ered to be meaningless.We have conducteda number
In a previousstudyof tomographicimagesin northeast- of inversionsby taking into accountthe small curvature
ernJapan,Zhaoet al. [1992a]foundin a mapviewthat changeof the plate boundary,and foundthat this small
most earthquakesin the lower plane of the double seis- curvaturechangehas almost no effect on the result we
obtained.
mic zone occur in relatively high-velocityareas of the
In the mantle below the slab, we can seesomevelocity
plate for areas shallower than 200-km. Note that the
double seismiczone exists from 60- to 200-km depth in
anomalies which have spatial extents of about 100-km
or larger and P wavevelocities2-3% lowerthan that of
northeasternJapan; the two seismicplanesmerge to a
the normal mantle. Taking into accountthe resolution
singleplanebelow200-kmdepth [Hasegawaet al., this
issue].For detailsof the correlationbetweenthe lower scalein that area (33-50 km), we considerthat these
lower-velocityanomaliesare real features, but not the
plane seismicityand velocityvariationswithin the slab, artifacts of the inversion due to the lack of resolution.
seeFigure18 of Zhaoet al. [1992a].This featureis also We will discuss these anomalies in more detail in the
found in the presentresultfor the northeasternJapan.
followingsection.
The seismicvelocitystructureof the Japan subduction
zone determinedby this study is generallyconsistent Discussion
with the temperaturemodels[e.g., Toksozet al., 1971; So far, two kinds of travel time tomographic stud-
Furukawa,1993]. ies have been actively conducted,one using data from
22,326 ZHAO ET AL.: DEEP STRUCTURE
local and regional earthquakes,originatedby Aki and
Lee [1976],the other usingteleseismic
residuals,orig- canoesin the crust and mantle wedge. We consider
inated by Aki et al. [1977]. The two methodshave that this feature (volcanism-related
inherent advantagesand drawbacks.The local and re- liesin the crustand mantlewedge)is a commonseis-
gional tomographycan determinethe shallowstructure
of an area to a depth abovewhich earthquakesexist or
the divingrays (like Pn) propagate,but cannotdeter- flow model, Furukawa[1993]suggested
mine the deeper structure. In contrast, teleseismicto-
mography can determine the deep structure of an area
(the modelingdepthdependson the apertureof the ar-
ray used),but usuallycannotdeterminewell the shal-
low structure simply becauseteleseismicrays basically
travel in a vertical direction and do not crisscross well
near the surface. In this study, we conducted tomo-
graphicinversionsby usingthe data from local,regional
and teleseismiceventssimultaneously,which preserves
the advantagesof the two separateapproachesand over-
comestheir drawbacks. The successful applicationsof
the joint inversionmethod in this study demonstrate
the power of this new approach.
In this study, we used the teleseismicdata collected
only from the stations in northeasternJapan because
we could only accessthe seismograms recordedby the
seismic network of Tohoku University. If teleseismic
data from all the JUNEC stationsin the entire Japan
Islandswill be availablein the future, we will be able to
extend the modelingspaceto the lowermantle because
the aperture of the array is greatly enlarged.Thus we
may determine the structure around the 650-km dis-
continuity and get some information on the slab pene-
tration in the mantle beneath Japan. With teleseismic
data from stationsin southwesternJapan, we may also
determinethe structureof the PhilippineSeaplate well.
The seismicity associatedwith the subduction of the
young Philippine Sea plate exists at depths shallower
than ~120-km in southwesternJapan. One interesting
and important questionis how deep the Philippine Sea
plate penetrates beneath this area. Does it just reach
the depth of the deepest seismicity or penetrate to a
greater depth aseismically?A joint tomographicinver-
sion of local and teleseismicdata as adopted in this
study may help to solvethis problem. This kind of ap-
plication has beenmade for the Alaskasubductionzone,
and a promisingresultobtained[Zhaoet al., 1993].
We found that in the Japan subductionzone,promi-
nent low-velocity anomaliesexist in the crust and man-
tle wedge above the subducting Pacific plate, and the
low-velocity anomalies seem to be associatedwith the
volcanism in the islandarc [Hasegawa et al., 1991;Zhao
et al., 1992a]. Similar structureshave been found in
many other subductionzonesin the world, e.g., Aleu-
tians[Engdahland Gubbins, 1987],Alaska[Zhaoet al.,
1992c,1993],Cascadia[Michaelsonand Weaver,1986;
Rasmussenand Humphreys,1988; Zhao et al., 1992d],
Caribbean[van der Hilst and Engdahl,1991],Tonga,
Kermadec and Hebrides [Zhou, 1990], New Zealand
[Kuge and Satake, 1987; $atake and Hashida, 1989],
Indonesia[Puspitoe! al., 1993],Mediterranean[$pak-
man et al., 1993], and Taiwan [Roeckeret al., 1987].
Although the resolution scale is different for each of
OF JAPAN SUBDUCTION ZONE
the results, all of them show low velocities beneath vol-
low-velocityanoma-
mologicalcharacteristicof subductionzones. Through
numerical calculations with a two-dimensional viscous
that the tem-
perature structures and stress fields in the crust and
mantle wedgeare all similar under variousisland arcs,
and independent of the subduction parameters such as
the angle and speed of slab subduction, the temper-
ature structures of the slab and back-arc region, and
the thicknessof the overlyinglithosphere.This seemsa
good explanation of the seismologicalobservations.
Our tomographic result shows that the prominent
low-velocity zonesin the mantle wedgeextend contin-
uously from the Moho discontinuityto a depth of ap-
proximately 200-km beneath Japan. This is consistent
with petrologicaland geochemicalstudies.Geochemists
havesuggestedthat a key factor of magmatismin island
arcsis the aqueousfluids dehydratedfrom the subduct-
ing slab[Boettcher,1977;$akuyamaandNesbitt,1986].
From the resultsof high-temperatureand high-pressure
experimentsunderwet conditions,Tatsumi[1989]sug-
gestedthat amphibolite, chlorite, phlogopite,talc, and
serpentine are major hydrous minerals in the mantle
wedge. The temperature at the upper surfaceof the
slabis about 800øCat depthsgreaterthan 100-km[Fu-
rukawa,1993]. Under thesetemperatureand pressure
conditions, these hydrous minerals, which are carried
deeper by the downward induced flow in the mantle
wedge,will break down at depths shallowerthan 200-
km, whichis the stabilitylimit of phlogopite[Tatsumi,
1989]. The terminationof dehydrationin the subduct-
ing oceaniccrust is essentiallydefinedby the stability
limit of hornblende,which is about 100-km. The tem-
perature of the upper surfaceof the slab will changea
little with the age of the slab. Thus the major dehy-
dration reactionsare consideredto take place at depths
shallowerthan 200-km in the mantle wedge. Aqueous
fluids released from these hydrous minerals will move
upward becauseof buoyancy;the migration mechanism
may be the buoyancy-drivenfracturing for magma mi-
grationunderislandarcs[Furukawa,1993].
The existenceof lower-velocityanomaliesin the man-
tle below the sub ducted Pacific slab in and around
Japan has been suggestedby previous tomographic
studies,e.g., Zhouand Clayton[1990],van der Hilst et
al. [1991],and Fukaoet al. [1992],whousedthe data
reported in the Bulletin of the International Seismolog-
ical Centre(ISC) and determinedtomographicimages
with a spatial resolutionscaleof 100-kmor larger. The
subslab low-velocity anomalies show different images,
which might be causedby variationsin the study area,
data set, analysismethod, and resolutionscalein each
of the studies. The tomographicimage beneath Japan
determinedby the presentstudy hasa muchhigherreso-
lution than that of all the previousstudiesusingthe ISC
data set and therefore is believed to be more reliable.
It is unclear how the subslab lower-velocity anomalies
 ZHAO ET AL.: DEEP STRUCTURE OF JAPAN SUBDUCTION ZONE 22,327

are formed. One possibility is that they are associated some of their stations. We thank K. Aid and D. L. Ander-
with small-scale mantle convections. The subduction of son for fruitful discussions.H. Zhou, C. Finn, K. Suyehiro,
the oceanic plate probably generatesa secondaryme- and two anonymous referees critically read the manuscript
chanicallyinduced convectionin the overlying mantle and provided us with helpful comments, which improved the
manuscript. This work was partially supported by a Na-
wedge[e.g., Mckenzie, 1969; Sleepand Toksoz,1973; tional Science Foundation grant, EAR-92-04748. The first
Toksozand Bird, 1977]. The subductingslab is not author (D.Z.) has beensupportedby a fellowshipfrom the
a simple planar plate, but has a complex geometry in Southern California Earthquake Center and the Texaco Re-
both strike and dip directions. As a result, a similar search Fellowship from the California Institute of Technol-
mechanicallyinduced convectionmay be alsogenerated ogy. Division of Geological and Planetary Sciences,Califor-
in the mantle below the slab by the subduction of the nia Institute of Technology,Contribution No. 5300.
oceanicplate, which results in the lateral variation of
densityand/or temperature,and thereforecausesthe References
lateral heterogeneitiesof seismicvelocity. Another pos-
sibility is that the lower-velocity anomalies below the Aid, K., and W. H. K. Lee, Determination of three dimen-
slab showthe image of mantle plumesarisingfrom the sional velocity anomalies under a seismicarray using first
deep interior of the Earth, e.g., from the 650-km dis- P arrival times from local earthquakes, 1, A homogeneous
continuityand/or the core-mantleboundary.To clarify initial model, J. Geophys.Res., 81, 4381-4399, 1976.
this problem,further high-resolution
tomographicstud- Aid, K., A. Christoffersson, and E. S. Husebye, Determi-
nation of the three-dimensional seismic structure of the
iesare neededin this regionand other subductionzones.
lithosphere, J. Geophys.Res., 8œ,277-296, 1977.
Geodynamicstudiesin both theoretical and experimen- Boettcher, A. L., The role of amphiboles and water in
tal aspectswill also be helpful. circum-Pacific volcanism, in High-Pressure Research: Ap-
plications in Geophysics,edited by M. H. Manghnani and
Conclusions S. Akimoto, pp. 107-126, Academic, San Diego, Calif.,
1977.
Draper, N. R., and H. Smith, Applied RegressionAnalysis,
We simultaneouslyused a large number of high accu- John Wiley, New York, 1966.
racy data from local, regional and teleseismiceventsto Engdahl, E. R., and D. Clubbins, Simultaneous travel time
determine a detailed three-dimensional P wave velocity inversion for earthquake location and subduction zone
structure of the Japan subduction zone to a depth of structure in the central Aleutian Islands, J. Geophys.Res.,
9œ, 13,855-13,862, 1987.
500-km. The subductingPacific plate is clearly imaged
Fukao, Y., M. Obayashi, H. Inoue, and M. Nenbai, Subduct-
by a free inversion without any a priori constraint on ing slabs stagnant in the mantle transition zone, J. Geo-
the slab in the starting model. However, constrained phys. Res., 97, 4809-4822, 1992.
inversionsare found to give a result which explains Furukawa, Y., Magmatic processesunder arcs and formation
the whole data set better. Compared with previous of the volcanic front, J. Geophys. Res., 98, 8309-8319,
1993.
results obtained from only local and regional events,
the presentresult has a much higher resolutionfor the Hasegawa, A., N. Umino, and A. Takagi, Double-planed
deep seismiczone and upper-mantle structure in the north-
area around the lower boundary of the Pacific plate
eastern Japan arc, Geophys. J. R. Astron. Soc., 5•, 281-
and the mantle below the plate, thanks to the addi- 296, 1978.
tion of 7035 data from 100 teleseismic events and 41 Hasegawa, A., D. Zhao, S. Hori, A. Yamamoto, and S. Ho-
very deep earthquakes. Prominent low-velocity anoma- riuchi, Deep structure of the northeastern Japan arc and
lies exist in the crust and upper mantle beneath active its relationship to seismic and volcanic activity, Nature,
volcanoes.In the mantle wedgethe low-velocityanoma- 35œ, 683-689, 1991.
lies dip toward the west and extend continuouslyto.a Hasegawa,A., S. Horiuchi, and N. Umino, Seismicstructure
of the northeastern Japan convergentmargin: A synthe-
depth of about 200-km, which is consistentwith petro- sis, J. Geophys.Res., this issue.
logical, geochemicaland geodynamicstudies. We be- Herman, G. T., Image Reconstruction from Projections:
lieve that the existence of the volcanism-related mantle
The Fundamentals of Computerized Tomography,316 pp.,
wedgelow-velocityanomaliesis a commonseismological Academic, San Diego, Calif., 1980.
feature of subductionzones,in light of all the available Hisamoto, S., On the anomaly of travel time of S waves
tomographicresultsfor many subductionregionsin the observedin easternJapan, 1 (in Japanese),J. Seismol.
Soc. Jpn., 18, 142-153, 1965a.
world. The subducting Pacific plate beneath Japan is Hisamoto, S., On the anomaly of travel time of S waves
imaged well as a high-velocity zone with a thicknessof observedin eastern Japan, 2 (in Japanese),J. Seismol.
80-90 km and a P wave velocity 4-6% higher than the Soc. Jpn., 18, 195-203, 1965b.
normal mantle. Lower-velocity anomalies are visible in Horie, A., and K. Aid, Three-dimensionalvelocity structure
the mantle below the slab. beneath the Kanto District, Japan, J. Phys. Earth, 30,
255-281, 1982.
Humphreys, E., and R. W. Clayton, Adaptation of back
Acknowledgments. We used arrival times reported in projection tomography to seismic travel time problems,
the Japan UniversityNetwork EarthquakeCatalogpublished J. Geophys. Res., 93, 1073-1085, 1988.
by the Earthquake ResearchInstitute, University of Tokyo. Inoue, H., Y. Fukao, K. Tanabe, and Y. Ogata, Whole man-
We thank the staff members of the seismic networks of tle P wave travel time tomography, Phys. Earth Planet.
Hokkaido University, Hirosaki University, and University of Inter., 59, 294-328, 1990.
Tokyo for allowing us to use the teleseismicdata recordedat Ishida, M., and A. H. Hasemi, The three-dimensional fine
22,328 ZHAO ET AL.: DEEP STRUCTURE OF JAPAN SUBDUCTION ZONE

velocity structure and hypocentral distribution of earth- Rasmussen, J., and E. Humphreys, Tomographic image of
quakes beneath the Kanto-Tokai District, Japan, J. Geo- the Juan de Fuca plate beneath Washington and western
phys. Res., 93, 2076-2094, 1988. Oregon using teleseismicP-wave travel times, Geophys.
Kanamori, H., Travel time to Japanesestationsfrom Long- Res. Lett., 15, 141%1420, 1988.
shot and their geophysicalimplications, Bull. Earthquake Roecker,S. W., Y. H. Yeh, and Y. B. Tsai, Three-dimensional
Res. Inst. Univ. Tokyo, •6, 841-859, 1968. P and S wave velocity structure beneath Taiwan: Deep
Katsumata, M., The effect of seismiczone upon the trans- structure beneath an arc-continent collision, J. Geophys.
missionof seismicwaves(in Japanese),Q. J. Seisrnol.,25, Res., 92, 10,547-10,570, 1987.
89-95, 1960. Sakuyama, M., and R. W. Nesbitt, Geochemistry of the
Kennett, B. L. N., and E. R. Engdahl, Traveltimesfor global Quaternary volcanic rocks of the northeast Japan arc, J.
earthquake location and phase identification, Geophys. J. Volcanol. Geotherrn. Res., œ9,413-450, 1986.
Int., 105, 429-465, 1991. Satake, K., and T. Hashida, Three-dimensionalattenuation
Kono, Y., and N. Furuse, 1:1 Million Scale Gravity Anomaly structure beneath North Land, New Zealand, Tectono-
Map in and Around the Japanese Islands, University of physics, 159, 181-194, 1989.
Tokyo Press, Tokyo, 1989. Sleep, N., and M. N. Toksoz, Evolution of marginal basins,
Koketsu, K., Three-dimensional ray tracing in a subduction Nature, œ33,548-550, 1973.
zone(in Japanesewith Englishabstract),J. Seisrnol.Soc. Spakman, W., and G. Nolet, Imaging algorithms, accuracy
Jpn., 44, 165-176, 1991. and resolutionin delay time tomography,in Mathematical
Kuge, K., and K. Satake, Lateral segmentationwithin the Geophysics,edited by N.J. Vlaar, G. Nolet, M. J. R.
subducting lithosphere: Three-dimensionalstructure be- Wortel, and S. A. P. L. Cloetingh,pp. 155-187,D. Reidel,
neath the North Island, New Zealand, Tectonophysics, Norwell, Mass., 1988.
139, 223-237, 1987. Spakman, W., S. van der Lee, and R. van der Hilst, Travel
Lees, J. M., and R. S. Crosson, Tomographic inversion for time tomography of the European-Mediterranean mantle
three-dimensional velocity structure at Mount St. He]ens down to 1400 km, Phys. Earth Planet. Inter., 79, 3-74,
using earthquake data, J. Geophys.Res., 94, 5716-5729, 1993.
1989. Spencer, C., and D. Gubbins, Travel-time inversion for si-
Matsuzawa, T., N. Umino, A. Hasegawa,and A. Takagi, Up- multaneousearthquake location and velocity structure de-
per mantle velocity structure estimated from PS-converted termination in laterally varying media, Geophys. J. R.
wave beneath the northeastern Japan arc, Geophys. J. R. Astron. Soc., 63, 95-116, 1980.
Astron. Soc., 86, 767-787, 1986. Tatsumi, Y., Migration of fluid phasesand genesisof basalt
Matsuzawa, T., T. Kono, A. Hasegawa,and A. Takagi, Sub- magmasin subductionzones,J. Geophys.Res., 9•, 4697-
ducting plate boundary beneath the northeastern Japan 4707, 1989.
arc estimated from SP converted waves, Tectonophysics, Toksoz, M. N., and P. Bird, Formation and evolution of
181, 123-133, 1990. marginal basinsand continental plateaus, in Island Arcs,
Mckenzie, D. E., Speculations on the consequencesand Deep Sea Trenches,and Back-Arc Basins,Geophys. Mo-
causes of plate motions, Geophys. J. R. Astron. Soc., nogr. Set., Vol. 1, edited by M. Talwani and W. C. Pit-
18, 1-32, 1969. man, pp. 379-393,AGU, Washington,D.C., 1977.
Michaelson, C. A., and C. S. Weaver, Upper mantle struc- Toksoz, M. N., J. W. Minear, and B. R. Julian, Temper-
ture from teleseismicP wave arrivals in Washington and ature field and geophysicaleffectsof a downgoingslab,
Northern Oregon, J. Geophys.Res., 91, 2077-2094, 1986. J. Geophys.Res., 76, 1113-1138, 1971.
Miyamachi, H., and T. Moriya, Velocity structure beneath Tsuboi, S., K. Koketsu, K. Takano, T. Miyatake, K. Abe,
the Hidaka Mountains in Hokkaido, Japan, J. Phys. Earth, and Y. Hagiwara, Hypocenterdeterminationprocedureof
$œ, 13-42, 1984. JapanUniversityNetworkEarthquakeCatalog(in Japan-
Nolet, G., Solving or resolvinginadequate and noisy tomo- ese),J. Seisrnol.Soc. Jpn., •œ,277-284,1989.
graphic systems, J. Cornput. Phys., 61, 463-482, 1985. Utsu, T., Anomalies in seismicwave velocity and attenua-
Obara, K., Regional extent of the S wave reflector beneath tion associatedwith a deep earthquake zone, 1, J. Fac.
the Kanto district, Japan, Geophys. Res. œett., 16, 839- Sci. HokkaidoUniv. Set. 7 (Geophys.),3, 1-25, 1967.
842, 1989. Utsu, T., Seismologicalevidencefor the anomalousstruc-
Obara, K., and H. Sato, Existence of an S wave reflector ture of island arcs with special reference to the Japanese
near the upper plane of the double seismic zone beneath region, Rev. Geophys.,9, 839-890, 1971.
the southern Kanto district, Japan, J. Geophys.Res., 93, van der Hilst, R., and E. R. Engdahl, On ISC PP and
15,037-15,045, 1988. pP data and their use in delay-time tomography of the
Okada, H., Forerunners of ScS waves from nearby deep Caribbean region, Geophys. J. Int., 106, 169-188, 1991.
earthquakesand upper mantlestructurein Hokkaido(in van der Hilst, R., E. R. Engdahl, W. Spakman, and G. No-
Japanese),J. Seisrnol.Soc. Jpn., œJ,228-239,1971. let, Tomographicimagingof subductedlithospherebelow
Okada, H., New evidence of the discontinuousstructure of northwest Pacific island arcs, Nature, 353, 37-43, 1991.
the descendinglithosphereas revealedby ScSpphase, J. Wadati, K., On the activity of deep-focusearthquakesin
Phys. Earth, œ7, Suppl., S53-S64, 1979. the Japan Islands and neighbourhood,Geophys.Mag., 8,
Paige, C. C., and M. A. Saunders,LSQR: An algorithmfor 305-326, 1935.
sparse linear equations and sparse least squares, Assoc. Zhao, D., A tomographicstudy of seismicvelocity structure
Cornput. Mach. Trans. Math. Software, 8, 43-71, 1982. in the Japan Islands, Ph.D. thesis,301 pp., Tohoku Univ.,
Pavlis, G. L., and J. R. Booker, The mixed discrete con- Sendai, Japan, 1991.
tinuous inverse problem: Application to the simultane- Zhao, D., and A. Hasegawa,P wave tomographicimaging
ous determination of earthquake hypocentersand velocity of the crust and upper mantle beneath the Japan Islands,
structure, J. Geophys.Res., 85, 4801-4810, 1980. J. Geophys. Res., 98, 4333-4353, 1993.
Puspito, N. T., Y. Yamanaka, T. Miyatake, K. Shimazaki, Zhao, D., and A. Hasegawa,Teleseismicevidencefor lateral
and K. Hirahara, Three-dimensionalP-wavevelocitystruc- heterogeneitiesin the northeastern Japan arc, Tectono-
ture beneath the Indonesian region, Tectonophysics,œœ0, physics,in press, 1994.
175-192, 1993. Zhao, D., S. Horiuchi, and A. Hasegawa,3-D seismicvelocity
 ZHAO ET AL.: DEEP STRUCTURE
structure of the crust and uppermost mantle in the north-
eastern Japan arc, Tectonophysics,181, 135-149, 1990a.
Zhao, D., S. Horiuchi, and A. Hasegawa, P-wave velocity
of the uppermost mantle off the Japan Sea coast, Tohoku
Geophys. J., 33, 177-183, 1990b.
Zhao, D., A. Hasegawa,and S. Horiuchi, Tomographicimag-
ing of P and S wave velocity structure beneath northeast-
ern Japan, J. Geophys.Res., 97, 19,909-19,928, 1992a.
Zhao, D., S. Horiuchi, and A. Hasegawa, Seismic velocity
structure of the crust beneath the Japan Islands, Tectono-
physics, œ1œ,289-301, 1992b.
Zhao, D., D. Christensen, and H. Pulpan, Tomographic
imaging of the Alaska subductionzone, Eos Trans. A G U,
73, 199, 1992c.
Zhao, D., G. Rogers, and K. Wang, Tomographicimaging
of Cascadia subduction zone in and around Vancouver
Island, Eos Trans. AGU, 73, 199, 1992d.
Zhao, D., K. Aki, and N. Biswas, How deep does slab pen-
etrate beneath Alaska? Eos Trans. A GU, 74{,312, 1993.
OF JAPAN SUBDUCTION ZONE 22,329
Zhou, H., Mapping of P-wave slab anomalies beneath the
Tonga, Kermadec and New Hebrides arcs, Phys. Earth
Planet. Inter., 61, 199-229, 1990.
Zhou, H., and R. W. Clayton, P and S wave travel time
inversionsfor subducting slab under the island arcs of the
northwest Pacific, J. Geophys.Res., 95, 6829-6851, 1990.
A. Hasegawa,ObservationCenter for Prediction of Earth-
quakes and Volcanic Eruptions, Faculty of Science,Tohoku
University, Senda.i980, Japan.
H. Kanamori and D. Zhao, SeismologicalLaboratory 252-
21, California Institute of Technology,Pasadena, CA 91125,
USA.
(ReceivedJune 28, 1993;revisedApril 25, 1994;
acceptedApril 28, 1994.)