p and s velocity tomography of the mariana subduction...

RESEARCH ARTICLE10.1002/2014GC005627

P and S velocity tomography of the Mariana subduction systemfrom a combined land-sea seismic deploymentMitchell Barklage1,2, Douglas A. Wiens1, James A. Conder3, Sara Pozgay4, Hajime Shiobara5,and Hiroko Sugioka6

1Department of Earth and Planetary Sciences, Washington University, Saint Louis, Missouri, USA, 2Now at NodalSeismic,LLC, Signal Hill, California, USA, 3Department of Geology, Southern Illinois University, Carbondale, Illinois, USA, 4Greenesk,Cushnie, Alford, UK, 5Earthquake Research Institute, University of Tokyo, Tokyo, Japan, 6JAMSTEC, Yokosuka, Japan

Abstract Seismic imaging provides an opportunity to constrain mantle wedge processes associated withsubduction, volatile transport, arc volcanism, and back-arc spreading. We investigate the seismic velocitystructure of the upper mantle across the Central Mariana subduction system using data from the 2003–2004 Mariana Subduction Factory Imaging Experiment, an 11 month deployment consisting of 20 broad-band seismic stations installed on islands and 58 semibroadband ocean bottom seismographs. We deter-mine the three-dimensional VP and VP/VS structure using over 25,000 local and over 2000 teleseismic arrivaltimes. The mantle wedge is characterized by slow velocity and high VP/VS beneath the fore arc, an inclinedzone of slow velocity underlying the volcanic front, and a strong region of slow velocity beneath the back-arc spreading center. The slow velocities are strongest at depths of 20–30 km in the fore arc, 60–70 kmbeneath the volcanic arc, and 20–30 km beneath the spreading center. The fore-arc slow velocity anomaliesoccur beneath Big Blue seamount and are interpreted as resulting from mantle serpentinization. The depthsof the maximum velocity anomalies beneath the arc and back arc are nearly identical to previous estimatesof the final equilibrium depths of mantle melts from thermobarometry, strongly indicating that the low-velocity zones delineate regions of melt production in the mantle. The arc and back-arc melt productionregions are well separated at shallow depths, but may be connected at depths greater than 80 km.

1. Introduction

The mantle wedge portion of subduction zones, including the fore arc, volcanic arc, and associated back-arcspreading center, is a key component of Earth’s geological system and vital for understanding the processesthat control the dynamic evolution of the mantle. These processes include arc magmatism, the hydration ofthe upper mantle by volatiles from the slab, and the formation of new oceanic lithosphere at back-arcspreading centers. Many questions remain regarding the physical processes in the mantle wedge includingthe mechanics and spatial extent of melt production and transport, the distribution of volatiles given off bythe slab, and the temperature structure of the mantle wedge.

Seismic velocity observations provide important constraints on the structure and dynamics of the mantlewedge. Body wave travel time tomography is considered one of the best tools available to reveal three-dimensional variations in seismic velocity. Many researchers have used body wave tomography to investi-gate the 3-D velocity structure beneath subduction zones and volcanic arcs (see reviews by Zhao [2001],Wiens and Smith [2003], and Wiens et al. [2008]). The Mariana subduction system is a classic example of anintraoceanic subduction zone and an ideal location for evaluating subduction zone processes due to thelack of continental influence on sedimentation and magmatism. In addition, it was a focus site for the MAR-GINS research programs of the U.S. and Japan, and thus many geochemical and petrological aspects arewell characterized [e.g., Stern et al., 2003; Wade et al., 2005; Kelley et al., 2006, 2010; Tatsumi et al., 2008;Shaw et al., 2008; Parman et al., 2011]. Furthermore, the presence of an active back-arc spreading centerallows imaging of spreading center mantle processes.

In this study, we determine the three-dimensional VP and VP/VS seismic velocity structure of the Marianamantle wedge using a combination of local and teleseismic travel times recorded by a temporary array ofland and ocean bottom seismographs. We then discuss the spatial distribution and magnitude of VP and VS

Key Points:� The 3-D velocity structure of the

Mariana subduction system isdetermined� Slow velocities delineate regions of

partial melt in the mantle� Inferred depths of partial melt are

consistent with petrologicalconstraints

Correspondence to:D. A. Wiens,[email protected]

Citation:Barklage, M., D. A. Wiens, J. A. Conder,S. Pozgay, H. Shiobara, and H. Sugioka(2015), P and S velocity tomography ofthe Mariana subduction system from acombined land-sea seismicdeployment, Geochem. Geophys.Geosyst., 16, 681–704, doi:10.1002/2014GC005627.

Received 21 OCT 2014

Accepted 20 JAN 2015

Accepted article online 28 JAN 2015

Published online 11 MAR 2015

BARKLAGE ET AL. VC 2015. American Geophysical Union. All Rights Reserved. 681

Geochemistry, Geophysics, Geosystems

PUBLICATIONS

http://dx.doi.org/10.1002/2014GC005627

http://onlinelibrary.wiley.com/journal/10.1002/(ISSN)1525-2027/

http://publications.agu.org/

anomalies as well as previously determined seismic attenuation anomalies in terms of the thermal structure,hydration, and partial melt distribution within the mantle wedge. We combine our analyses with previouslypublished petrological constraints to show that the slow velocity anomalies beneath the volcanic arc andback-arc spreading center likely delineate the region of melt production in the mantle.

2. Background

2.1. Regional SettingThe Mariana subduction zone occurs where the Jurassic-age Pacific plate subducts into the mantle beneaththe Mariana island arc. The subduction system consists of a highly faulted fore arc including actively ventingserpentinite seamounts [Fryer et al., 1985, 1999], a remnant island arc including the islands of Saipan andGuam, an active volcanic arc, and a back-arc spreading center (for a comprehensive review, see Stern et al.[2003] and references therein). The convergence direction at the trench is oblique with a direction ofN80�W and a rate of 4.5 cm/yr in the region of this study, and the full spreading rate along the back-arcspreading center between the Philippine Sea plate and the Mariana arc is about 2.5 cm/yr at 18�N [Katoet al., 2003]. This rate is consistent with the morphology of the ridge, which is similar to other slow spread-ing centers around the world. The Mariana arc is not experiencing eastward trench roll-back, but rather ismoving slightly westward relative to a hot spot reference frame within an extensional stress regime main-tained by rapid Philippine-Eurasia plate convergence [Stern et al., 2003; Schellart et al., 2007]. The subductedslab, with an age of approximately 180 Ma, becomes nearly vertical below depths of about 250 km andappears to penetrate the 660 km discontinuity [van der Hilst et al., 1991]. A double seismic zone is imagedwithin the slab at depths up to 200 km with a separation of approximately 30 km [Shiobara et al., 2010].

The history of volcanic output and back-arc spreading in the Mariana system has been well sampled by anumber of different geochemical studies. Back-arc basin basalts from the Mariana trough show a water con-centration in the mantle source rock which is considerably higher than the normal range of mid-ocean ridgebasalts, showing that water from the subducted slab plays a role in back-arc melt generation [Stolper and New-man, 1994; Kelley et al., 2006]. Compositions of the arc and back-arc lavas in the Marianas both show a subduc-tion component due to incorporation of volatiles and fluids extracted from the subducted slab [Pearce et al.,2005]. Fluids from the slab that interact with the back-arc basin source are more equilibrated with astheno-spheric mantle than those delivered to the arc source [Pearce and Stern, 2006]. Constraints on spatial patternof melt production and volatile transport provided by tomographic analysis can help to understand thesegeochemical observations and the dynamics of the arc and back-arc magmatic systems.

2.2. Previous Seismic StudiesLarge-scale body wave [van der Hilst et al., 1991; Gorbatov and Kennett, 2003; Miller et al., 2004; Jaxybulatovet al., 2013] and surface wave [Lebedev and Nolet, 2003; Isse et al., 2009] studies show slow velocities in theMariana mantle wedge; however, these studies rely on sparse station coverage scattered throughout thewestern Pacific and do not have sufficient resolution to reveal the fine-scale structure necessary for under-standing wedge processes. Shiobara et al. [2010] use nine ocean bottom seismometers to determine the Pvelocity structure. However, our study is able to obtain much greater resolution and detail in this studyusing data from the 70 stations deployed for the Mariana broadband seismic experiment.

A long seismic refraction line extending from the Mariana fore arc across the West Mariana Ridge at around17�N shows arc and fore-arc Pn velocities of 7.7 km/s and Mariana trough Pn velocities of 7.8–8.0 km/s [Taka-hashi et al., 2007, 2008]. Analysis of a 2-D wide-angle survey in the region surrounding the volcanic arc sug-gests maximum crustal thicknesses of 20–23 km beneath the frontal arc high and Pn velocities of 7.5–7.8 km/s [Calvert et al., 2008]. A passive magnetotelluric survey using ocean bottom electro-magnetometers(OBEM) across the Mariana system in the same region imaged by this study reveals a vertical column ofhigh conductivity beneath the volcanic arc to depths as shallow as 60 km [Matsuno et al., 2010]. However,low conductivity was found to depths of around 100 km beneath the back-arc spreading center, providingno signature of melt production [Matsuno et al., 2010, 2012].

Previous results from this seismograph deployment show many interesting features related to mantlewedge processes. Rayleigh wave phase velocities indicate shear velocities as low as 3.9 km/s at depths ofabout 60 km with some degree of asymmetry along the arc and back arc as well as shear velocities as low

Geochemistry, Geophysics, Geosystems 10.1002/2014GC005627


as 3.6 km/s at depths of less than 40 km in the fore arc [Pyle et al., 2010]. Body wave attenuation tomogra-phy [Pozgay et al., 2009] shows a �75 km wide columnar-shaped high-attenuation anomaly (QP � 43–60)beneath the spreading center that extends to �100 km depth. A separate weaker high-attenuation region(QP � 56–70) occurs beneath the volcanic arc, and the two anomalies are connected at depths of 75–125 km. High attenuation also occurs beneath serpentinite seamounts in the outer fore-arc region. Shearwave splitting measurements along the Mariana arc show a complex pattern of mantle flow. The results canbe summarized as along-strike fast directions in the fore arc and arc with fast directions rotating to conver-gence parallel near the back-arc spreading center [Pozgay et al., 2007]. P wave receiver functions image sub-ducting oceanic crust at depths of 75–110 km indicating the basalt to eclogite phase transition must occurat a greater depth. They also image a low-velocity zone in the fore-arc mantle at a depth of �40–55 kmwith S wave velocity as low as 3.6 km/s that represents widespread serpentinization of the mantle fore arc[Tibi et al., 2008].

3. Data

We use data from the 2003–2004 Mariana Subduction Factory Imaging Experiment. This 11 month experi-ment consisted of 20 broadband seismic stations deployed on the Mariana Islands and 58 semibroadbandocean bottom seismographs (OBS) deployed across the fore arc, island arc, and back-arc spreading center(Figure 1). The land stations used STS-2 and Guralp CMG-40T sensors and were installed by a team of scien-tists from Washington University. Fifty OBSs used three-component Mark Products L4 sensors with 1 Hz nat-ural period and modified amplifiers to extend long-period performance [Webb et al., 2001]. These OBSswere operated by the OBS instrument facility at Lamont Doherty Earth Observatory. Another eight OBSsequipped with precision measuring devices semibroadband sensors were developed at the University ofTokyo [Shiobara and Kanazawa, 2008]. Of the 58 OBSs deployed, eight were not recovered and an addi-tional 29 stopped recording data after 50 days due to faulty firmware (Figure 1). However, we were able torecover some arrival time data from a total of 70 stations. A list of station locations is given in Pozgay et al.[2007].

We determine the velocity structure by inverting a P and S wave arrival time data set consisting of bothtravel times from local subduction zone earthquakes and arrival times from incoming teleseismic eventsrecorded by the array. The local travel time data set was compiled by manually picking the first P or S arrivalat each station. We used the Antelope software package to detect phases, determine arrival times, and asso-ciate consistent picks to generate an initial earthquake location. Each event was then reviewed by a seis-mologist, and the arrival time picks were adjusted, many picks missed by the automatic procedure wereadded, and the event was relocated using the GENLOC earthquake location program [Pavlis et al., 2004].Where possible, arrivals were picked on the unfiltered data, but the data were generally filtered with anappropriate band-pass filter. P arrivals in the arc and fore arc were typically best observed with a 0.25–0.125 s band-pass filter whereas P arrivals in the higher attenuation back arc were best viewed with a 0.66–0.25 s band-pass filter. S arrivals in the back arc were generally picked using a 2–0.66 s band-pass filter dueto their high attenuation. Each arrival was also assigned a standard deviation typically ranging from 0.1 to0.5 s based on the quality of the pick, and the standard deviation is used in weighting the data for the inver-sion. The local raypaths are entirely contained within the geographic region defined for tomographic inver-sion (Figure 2).

We determined the teleseismic travel times using an adaptive stacking technique [Rawlinson and Kennett,2004]. First, the instrument response was deconvolved to velocity for each trace to maintain coherencybetween waveforms since we use a variety of different types of instruments. Then we determined a pre-dicted arrival time based on the AK135 velocity model [Kennett et al., 1995] to achieve an approximatealignment of traces, which are then stacked to form a reference trace. The reference trace was then com-pared to each station trace and an improvement in alignment was determined based on an L3 measure ofmisfit. The time shift required for improved alignment allowed us to calculate the residuals relative to thestarting model. We found this technique to be a significant improvement over standard cross-correlationtechniques [Van Decar and Crosson, 1990] because it performs better in the different noise environmentspresented in this study between the land and OBS station locations. A standard deviation was assigned toeach teleseismic arrival based on the uncertainty of the estimated residual and ranges from 0.025 to 0.15 s.



Teleseismic P wave data were filtered with a band-pass filter ranging from 3.3 to 1.25 s whereas teleseismicS data were filtered with a band-pass filter ranging from 25 to 4 s. Our choice of data filters and alignmentwindow parameters allowed us to effectively remove the P wave water reverberations recorded at OBS sta-tions from the alignment to avoid bias in the travel time data set (Figure 3) [Blackman et al., 1995].

For our tomographic analysis, we use a total of �850 local events, yielding about 15,000 local P wave arriv-als and about 10,000 S wave arrivals. Each event is required to have a minimum of 20 total arrivals toensure it is well located. We use an additional 58 teleseismic events recorded by the array includingroughly 1500 P wave arrivals and 260 S wave arrivals. These teleseismic raypaths provide informationabout portions of the model space where local raypaths are not present. They also greatly improve raycoverage in the back arc where local arrivals are less frequently observed due to high attenuation in thewedge and the inherent asymmetry of the earthquake distribution. Each teleseismic event is required to

142˚

142˚

144˚

144˚

146˚

146˚

148˚

148˚

14˚ 14˚

16˚ 16˚

18˚ 18˚

20˚ 20˚

0 100 200

West

Mariana

RidgeMariana

Spreading

Center

Maug

Agrihan

Pagan

Alamagan

Guguan

Sarigan

Anatahan

Saipan

Tinian

Rota

Guam

GUMO

Celestial Seamount

−6.0 −4.0 −3.0 −2.5 −2.0 −1.5 −1.0 −0.1

seafloor elevation (km)

scale (km)

A A‘

B B‘

C C‘

Big Blue Seamount

Figure 1. Station configuration and the seafloor bathymetry of the Mariana Arc. Locations of broadband land seismic stations are plottedas blue squares, and long-period ocean bottom seismographs (OBS) are plotted as triangles. OBSs that operated throughout the observa-tion period are plotted in red, and OBSs that recorded only 50 days due to a firmware failure are plotted in yellow. The axis of the back-arcspreading center is traced in red. Also note the presence of Big Blue and Celestial serpentinite seamounts in the fore arc. The horizontalblack lines labeled A-A0 , B-B0 , and C-C0 show the locations of the cross sections presented in the results.



have observations at a minimum of eight stations. The distribution of teleseismic earthquakes used for thisstudy is shown in Figure 4.

4. Methods

We reference all anomalies in this paper to a best fitting 1-D velocity model (Table 1). To solve for the aver-age 1-D model, we invert the local arrival time data using methods similar to the full 3-D tomographic inver-sion, but only solve for a constant velocity at each depth, with a depth increment of 25 km. To stabilize the1-D profile, the inversion result is slightly penalized against an IASP91 velocity model [Kennett and Engdahl,1991], and a first derivative smoothing function is also applied. Results show that average velocities in theupper 200–300 km beneath the Mariana region are slower than the global average (Table 1), consistentwith regional surface wave results [e.g., Pyle et al., 2010].

The starting model for the full 3-D tomographic inversion consists of the best fitting 1-D velocity model forthe Mariana region with a superimposed fast velocity subducting Pacific slab (Figure 5). Previous work hasshown that the insertion of a slab in the starting model improves the ray tracing calculation and eliminatesproblems with nonlinearity in which incorrect raypaths in the first iteration produce biased models. In gen-eral, the inclusion of a slab in the first iteration results in better final models as measured by the resultingtravel time residuals [Zhao et al., 1994]. The a priori slab velocity anomaly used in our model is 90 km thickand is 4% larger than the surrounding mantle for both VP and VS. The slab includes a gradient described bya Gaussian function. The a priori slab anomaly is only used in the first iteration of the inversion to ensurethat the raypaths are reasonable, and tests show that the final structure is relatively insensitive to the detailsand magnitude of the initial slab anomaly. The structure of the superimposed slab used in the startingmodel is shown in Figure 5.

We use the LSQR least squares algorithm [Paige and Saunders, 1982] to jointly invert for the P wave velocityand VP/VS structures. This inverse method is used because it is fast and efficient and can handle large sparsematrices. The model parameters include P slowness and VP/VS values on a 3-D grid with nodes that arespaced 20 km at the center of the box and 50 km along the edges, with depth spacing of 20 km in theupper 200 km and larger depth spacing at greater depths (Figure 2). The velocity model grid extends fromlatitude 16�N, just south of Anatahan, north to latitude 20�N. We account for the curvature of the Earth byreshaping the grid of rectangular elements to a grid of trapezoidal elements and trace the rays through the

−500 −400 −300 −200 −100 0 100 200 300

0

100

200

300

400

500

600

700−500 −400 −300 −200 −100 0 100 200 300

−250

−200

−150

−100

−50

0

50

100

150

200

250

SC VF BB

Agrihan

Alamagan

Anatahan

SC VF BB

km eastkm westkm west km east

dept

h (k

m) km

nor

thkm

sou

thFigure 2. (left) Cross section and (right) map view of the grid used for tomographic inversion. The (0,0) point is located at 18�N, 146�E. The lateral spacing is 20 km at the center of thegrid and 50 km along the perimeter, and the depth spacing is 20 km down to 200 km with larger spacing at greater depth. The black triangle labeled SC denotes the spreading center,VF the volcanic front, and BB the Big Blue Serpentinite Seamount. The location of the spreading center is shown in red in the map view.



trapezoidal mesh. Rays are traced using a two-point bending method [Um and Thurber, 1987]. We do notinvert for the velocity of the crust so we remove that portion of the raypath from the inversion and make acorrection to the observed travel times by assuming a constant 6.0 km/s velocity for the crust for P wavesand 3.5 km/s for S waves. The thickness of the crust is extrapolated from a wide-angle seismic refraction sur-vey conducted just to the south of the main OBS deployment [Takahashi et al., 2007], assuming that crustal

Before Alignment After Alignment

OBS50OBS49OBS48OBS47OBS45OBS44

OBS50OBS49OBS48OBS47OBS45OBS44

reference trace reference trace

0 10 20 30 0 10 20 30time (s) time (s)

Direct P Water Reverberations Direct P Water Reverberations

Figure 3. Example of the teleseismic arrival time determination procedure, showing a teleseismic P wave arrival as recorded by a subset ofthe OBS stations. The water depths for these stations range from 1.9 to 2.6 km depth. The direct P arrival is followed by a strong waterreverberation arriving 2.5–3.5 s later. This figure shows that by carefully selecting the filter and window properties of the adaptive stackingprocedure, we can effectively remove any influence or bias of water reverberations from the travel time measurements.

120o W

60 oW 0

o

60oE

120o E 180 o

W

60 oS

30oS

0o

30o N

60o N

Figure 4. The distribution of teleseismic earthquake epicenters used in this study. The black box denotes the study area.



thickness is consistent along the strike of the arc. Thecrust reaches a maximum thickness of 20 km beneaththe volcanic arc, a minimum thickness of about 8 kmbeneath the back-arc spreading center, and an inter-mediate thickness of 15 km in the fore arc.

We trace the teleseismic raypaths by breaking thepath into two segments: a teleseismic part outside ofthe tomographic model and the local raypath inside.We determine the complete raypath by setting up agrid along the edges of the tomographic model. Wecalculate travel times from the grid nodes to stationsinside the box using the 3-D local velocity model, andwe compute travel times outside the box using theAK135 velocity model [Kennett et al., 1995]. Then weuse Fermat’s Principle of Least Time to determine theentire raypath for a given teleseismic earthquake andrecording station by searching over the grid nodes tofind the two segments that minimize the total traveltime. We account for source-side velocity anomalies

and source location uncertainty by solving for a constant teleseismic travel time anomaly in the inversionfor each teleseismic event and ray type.

We choose an inversion method that simultaneously solves for VP and VP/VS ratios, and does not assumeidentical P and S raypaths [Conder and Wiens, 2006]. We use the discrete form of the P wave travel timewhich can be written as:

t�p5tprep 1

XDgi si (1)

where t* is the travel time, tpre is the predicted travel time through the starting model, Dg 5 2DVP/VP is theP wave slowness perturbation to the starting model, s is the ray length portion associated with that nodedetermined by a weighted distribution of the six surrounding nodes for each ray segment, and i denotesthe particular node. For S waves, the travel time is (2)

t�s 5

ðgpr � ds (2)

where r equals the VP/VS ratio at any point. We use the discrete form of the S wave travel time defined by

t�s 5tpre1X

Dgi ri si1X

giDri si (3)

where Dg and Dr are treated as independent model parameters that affect ts*. This equation effectively cou-ples the VP and VP/VS inversions because the S data simultaneously affect both the g and r parameters. Thepartial derivative matrix and model vector are defined as

G5@dj

@ni

�� @dj

@ri

� �� (4)

and

m5 Dgijc21Dri� �T

(5)

where dj refers to the jth datum, superscript T denotes transpose, and c is a weighting factor to accountfor the disparity in magnitude between r and g. For a perfectly posed problem, the relative magnitudes ofthe partial derivatives is not an issue; however, the minimum length solution and nonideal ray coveragecauses S residuals to map preferentially into g relative to r, resulting in weak calculated VP/VS anomaliesunless the weighting factor c is applied. For the tomography presented here, we setÇ equal to 20 toaccount for the approximate ratio of the magnitudes of the two parameters in the upper mantle [Conderand Wiens, 2006].

Table 1. Average 1-D Velocity Model

Depth (km) Vp (km/s) Vs (km/s)

20 7.663 4.27040 7.684 4.28360 7.710 4.30080 7.738 4.319100 7.767 4.339120 7.807 4.358140 7.875 4.376160 7.949 4.391180 8.020 4.402200 8.091 4.417225 8.194 4.461250 8.330 4.529275 8.456 4.591325 8.720 4.722375 8.903 4.809425 9.410 5.112475 9.578 5.218525 9.746 5.324575 9.914 5.430625 10.082 5.536



For the least squares inverse problem, the existence of outliers can have a strong adverse effect on theresults so it is desirable to eliminate these data from the inversion. However, in a region of large-velocityanomalies high-quality travel times may have large residuals, so eliminating such observations can bias thestructure toward small-velocity anomalies. Therefore, instead of implementing a residual cutoff, we adjustall data weights by a factor, w, determined by

w5 11lexpd2

2r2

� �� 21

(6)

where d is the residual of the datum to the starting model (or previous iteration), l is 0.005, and r is 1 forlocal P waves,

ffiffiffi2p

for local S waves, and 1=ffiffiffi2p

for teleseismic data [Conder and Wiens, 2006]. By using thistechnique, outliers will have a minimal effect on the results, but with additional iterations, the residuals ofhigh-quality data will improve and carry more weight in the inversion.

Smoothing constraints are also included to help stabilize the inversion. Smoothing constraints are placeddirectly into the partial derivative matrix before inverting for tomographic structure. We use a second deriv-ative constraint that depends only on the surrounding nodes [Menke, 1984]. We repeated the tomographicinversion many times using different values of smoothing and compared the results. We seek a balancebetween the reduction of travel time residuals and the smoothness of the velocity model obtained. A valueof 65 is chosen as the best value for our data set and model parameterization. The S wave structure is morepoorly determined than the P wave structure due to fewer arrival times, particularly in the back arc, soapproximately 5 times greater smoothness weighting is necessary for the VP/VS structure than for P wavestructure.

Since both the raypaths and the hypocenters are a function of the velocity structure, the earthquakes arerelocated and the raypaths recalculated between each iteration using the updated 3-D velocity structure.We start by relocating each of the hypocenters using the 1-D Mariana velocity structure including the high-velocity slab. Then the rays are traced through this same structure using the relocated hypocenters. Thestructure is then solved for using the tomographic inversion, and the hypocenters and raypaths are

−250 −200 −150 −100 −50 0 50 100 150 200

0

50

100

150

200

250

300 −5

−4

−3

−2

−1

0

1

2

3

4

5

SC VF BBB B‘

% d

Vp

dept

h (k

m)

Figure 5. The a priori subducting slab velocity structure used in this study for preliminary earthquake location and to facilitate accurateraypaths in the first iteration of the tomographic inversion. The slab is �90 km thick and has a velocity 4% higher than the surroundingmantle, with the shape determined by earthquake locations (white circles). The edge of the anomaly is smoothed with a Gaussian functionbetween the peak anomaly and surrounding mantle.



recalculated using the updated velocity structure. This iteration process continues until further iterations donot significantly decrease the variance. We find that five iterations of the above procedure are sufficient toproduce a model with no significant further decrease in variance.

We relocate hypocenters prior to each tomographic iteration using a weighted least squares technique[Wright and Holt, 1986]. We use both local travel times recorded by this study and outgoing teleseismictravel times reported by the USGS Preliminary Determination of Epicenters (PDE) catalog as data for hypo-center relocation. The PDE travel times provide valuable depth constraints for the hypocenter relocations,but are only available for the largest events. Teleseismic travel time derivatives are found using the AK135velocity model [Kennett et al., 1995], whereas local travel time derivatives are computed using successivelyupdated tomographic velocity models. Data weights are the same as those used for tomographic inversiondescribed above.

We compute checkerboard tests with various cell sizes to assess the ability of our method to recover differ-ent portions of the model space. Synthetic travel times are computed through a checkerboard velocitymodel with various wavelengths. P wave velocities are perturbed between 20.5 and 0.5 km/s from the ini-tial 1-D starting model, and VP/VS ratios are perturbed between 20.15 and 0.15 from the initial ratios withpositive VP/VS anomalies corresponding to cells with negative P wave anomalies. After the synthetic traveltime data set is calculated, travel times are inverted using the same parameters as the actual inversions. Wefind that a cell size of 80 km shows good recovery of both VP and VP/VS anomalies within the triangularregion between the fore-arc and the back-arc spreading center down to about 200 km depth (Figure 6). VP/VS anomalies (and thus Vs structure) are poorly recovered with 60 km cell size beyond the back-arc spread-ing center due to the much reduced raypath coverage, indicating reduced S wave velocity structure resolu-tion there.

SC VF BBB B‘

dept

h (k

m)

Vp/V

s

SC VF BBB B‘

% d

Vp%

dVp

dept

h (k

m)

Vp/V

s

dept

h (k

m)

SC VF BBB B‘

dept

h (k

m)

SC VF BBB B‘

−250 −200 −150 −100 −50 0 50 100 150 200

0

50

100

150

200

250

300

−6

−4

−2

0

2

4

6

−250 −200 −150 −100 −50 0 50 100 150 200

0

50

100

150

200

250

300 1.65

1.7

1.75

1.8

1.85

1.9

1.95

2

−250 −200 −150 −100 −50 0 50 100 150 200

0

50

100

150

200

250

300 1.65

1.7

1.75

1.8

1.85

1.9

1.95

2

−250 −200 −150 −100 −50 0 50 100 150 200

0

50

100

150

200

250

300

−6

−4

−2

0

2

4

6

Figure 6. Resolution tests showing the recovery of a synthetic checkerboard anomaly pattern using the actual raypath coverage of the data for (left) P wave velocities and (right) VP/VS

ratios. Synthetic anomalies of 60.5 km/s were added to the P wave starting model and 60.15 to the VP/VS ratio starting model to generate synthetic travel times for all of the source-receiver pairs used in the inversion (top figures). Bottom figures show the velocity structure recovery when inverting the synthetic arrival times with noise added.



5. Results

5.1. Teleseismic TomographyWe first present the results of an inversion using only the P and S wave teleseismic travel time data. Theseresults are useful for comparison to regions where only teleseismic travel time data are available such asmidplate regions and mid-ocean ridges [e.g., Forsyth et al., 1998; Hammond and Toomey, 2003]. Figure 7shows a broad region of slow P wave velocities within the mantle wedge centered under the back-arcspreading center and extending beneath the volcanic arc and into the fore arc. The slow anomalies are con-centrated in the upper 150 km of the mantle with a maximum occurring at a depth of about 40 km,although the depth resolution is poor. The S structure is not as well constrained due to the sparse teleseis-mic S arrival times, particularly in the back arc where S arrivals were often not visible due to high attenua-tion. While these images are useful for determining the bulk characteristics of the mantle wedge, they aredeficient in the fine-scale structure necessary for constraining mantle wedge processes. Including the localtravel time data set greatly increases the resolvability of the velocity structure and provides absolute veloc-ity models that can be easily compared to other regions around the world, rather than relative models.

5.2. VP StructureP wave velocity tomography results from inversion of both local and teleseismic data show a seismically fastslab subducting into the mantle and both slow and fast velocities within the mantle wedge (Figures 8–10).Here we discuss these results, emphasizing the best resolved parts of the image which are generally alongcross section B-B0. The slab is imaged as a continuous feature with P wave velocity anomalies ranging from�2.5 to 7% relative to the average model for the region. Maximum velocity anomalies within the slab occurfrom �80 to 160 km depth. A fast velocity anomaly is also observed in the fore arc just east of the active arcextending from the top of the mantle to the top of the slab. A maximum velocity anomaly of �7% occurs inthis region at 50 km depth, just above the subducting Pacific slab. This feature is labeled on cross section B-B0

as F1.

Slow P wave velocity anomalies are also present in the mantle wedge beneath the back-arc spreading cen-ter, the active arc, and in the fore-arc mantle. Slow P wave velocity anomalies beneath the back-arc

SC VF BBB B‘

% d

Vp

dept

h (k

m)

88

8

8

9

9 9

9

10

10

10

11

1112

−250 −200 −150 −100 −50 0 50 100 150 200

0

50

100

150

200

250

300 −10

−8

−6

−4

−2

0

2

4

6

8

10

Figure 7. P wave velocity anomalies from an inversion using only teleseismic travel time data. Cool colors represent fast seismic velocitiesand warm colors represent slow seismic velocities. White circles represent the earthquake locations. The triangles (from left to right) repre-sent the locations of the back-arc spreading center (SC), the volcanic front (VF), and Big Blue serpentinite seamount (BB).



SC VF BBB B‘

% d

Vp

SC VF BBA A‘

SC VF BBC C‘

dept

h (k

m)

100 150 200500-50-100-150-200-250

100 150 200500-50-100-150-200-250

100 150 200500-50-100-150-200-250

−10

−8

−6

−4

−2

0

2

4

6

8

10

−10

−8

−6

−4

−2

0

2

4

6

8

10

8

8

8

9

9

9

910

10

1011

11

11

0

50

100

150

200

250

88

8

8

99 9

9

10

10

10

11

1112

0

50

100

150

200

250

300

8

8 8

88

8

89

9

910

10

1011

11

0

50

100

150

200

250

300

300

−10

−8

−6

−4

−2

0

2

4

6

8

10

F1F1F2F2 S1S1

S2S2

S3S3

Figure 8. P wave velocities plotted as the percent deviation from the starting model along the three profiles shown in Figure 1. Cool colorsrepresent fast seismic velocities and warm colors represent slow seismic velocities. White circles represent the earthquake locations. Thetriangles (from left to right) represent the locations of the back-arc spreading center (SC), the volcanic front (VF), and Big Blue serpentiniteseamount (BB). The black contour lines trace constant values of the log of the derivative weight sum, providing an estimate of the regionof best resolution. The labels S1–S3 denote slow velocity anomalies and F1–F2 denote fast velocity anomalies discussed in the text.



spreading center (labeled S3) are found in the upper 80–100 km of the mantle wedge. The maximum anom-aly of 26.5% is located at a depth of 20–40 km about 20 km west of the spreading axis. A fast velocityanomaly (labeled F2) is observed at depths shallower than 50 km between the back-arc spreading centerand the active arc, separating the slow anomalies associated with those features.

88

8

8

9

9 9

9

10

10

10

11

1112

−250 −200 −150 −100 −50 0 50 100 150 200

0

50

100

150

200

250

300 7

7.5

8

8.5

SC VF BBB B‘

Vp (k

m/s

)

dept

h (k

m)

SC VF

Vp (k

m/s

)

dept

h (k

m)

−200 −150 −100 −50 0 50

0

50

100

150 7

7.5

8

8.5

Figure 9. Absolute P wave velocities in km/s, along section B-B0 . White circles represent the earthquake locations. The triangles (from leftto right) represent the locations of the back-arc spreading center (SC), the volcanic front (VF), and Big Blue serpentinite seamount (BB). Theblack contour lines trace constant values of the log of the derivative weight sum. The lower figure shows an enlarged view of the uppermantle structure beneath the arc and back arc. The black crosses show the relationship between velocity anomalies and geochemicalresults. The vertical lines show the range of final equilibrium depths for arc and back-arc magmas based on Si and Mg thermobarometry[Kelley et al., 2010]. The horizontal ticks show the depth of the maximum P wave velocity anomaly in the tomographic images.



Slow P wave velocity anomalies are also observed beneath the volcanic arc (labeled S2) to a depth of100 km with a maximum anomaly of 26% occurring at a depth of 60–70 km. The slow velocities persist to agreater depth beneath the active arc than beneath the back arc and appear to be separate and distinct fea-tures with little to no connection in the upper 100 km of the mantle wedge. A third region of slow P wavevelocity (labeled S1) is also present in the outer fore arc beneath the Big Blue serpentinite seamount, wherean anomaly of 25.5% is concentrated in the upper 30 km of the mantle and extends laterally for at least100 km along the strike of the trench (Figures 10 and 12).

5.3. VP/VS Ratio and VS StructureWe determine the S wave structure by dividing the P wave solution by the VP/VS ratio results. Overall, the Swave structure (Figures 11 and 12) is similar to the P wave structure, but with some notable differences.Some of these differences may be due to the lower resolution of the S model resulting from fewer S wavearrival time observations. This is likely the case beneath and to the west of the back-arc spreading center,

8

8

8

9

9

9

9

10

10

10

10

10

11

11

11

11

11

12

12

12

8

8

8

8

888

8

8

9

9

9

9

9

10

10

10

10

11

11

88

8

8

9

9

9

9

99

910

10

10

1011

11

11

8

8

8

9

9

9

9

9

10

10

10

0

10

11

11

1111

12

z=40km z=60km

-5-10 0 5 10

% dVp

z=80km z=100km

−200

−150

−100

−50

0

50

100

−200

−150

−100

−50

0

50

100

−200

−150

−100

−50

0

50

100

−200

−150

−100

−50

0

50

100

−250 −200 −150 −100 −50 0 50 100 150 200 −250 −200 −150 −100 −50 0 50 100 150 200

−250 −200 −150 −100 −50 0 50 100 150 200 −250 −200 −150 −100 −50 0 50 100 150 200

Agrihan

Alamagan

Anatahan

Agrihan

Alamagan

Anatahan

Agrihan

Alamagan

Anatahantrench

A A’

B B’

C C’

Figure 10. Depth slices of the final P wave velocity model plotted as percent deviation from the starting model (Table 1) for four different depths. The black contour lines trace constantvalues of the log of the derivative weight sum. The black triangles in the arc show the location of the arc volcanoes, and the black triangle in the fore arc shows the location of Big BlueSeamount. Black lines in the back arc denote the spreading center segments.



where high attenuation greatly limits the S wave observations. The magnitudes of the slow S wave anoma-lies are �5.5% beneath the back arc, 212% beneath the active arc, and 13.5% in the fore arc with the larg-est anomalies occurring beneath the Big Blue seamount in the Mariana fore arc. In contrast, the P wavestructure shows the greatest velocity anomalies beneath the arc and back arc. The S wave anomaliesbeneath the volcanic arc are greater than S wave anomalies beneath the spreading center, whereas theback-arc spreading center anomalies are greater in the P wave results. This probably results from

8

8

8

8

9

9

9

9

910

10

10

11

11

12

−250 −200 −150 −100 −50 0 50 100 150 200

0

50

100

150

200

250

300

3.8

4

4.2

4.4

4.6

4.8

5

SC VF BBB B‘

dept

h (k

m)

Vs (k

m/s

)

8

8

8

8

9

9

9

9

910

10

10

11

11

12

−250 −200 −150 −100 −50 0 50 100 150 200

0

50

100

150

200

250

300 −20

−15

−10

−5

0

5

10

15

20

SC VF BBB B‘

dept

h (k

m)

Vs (k

m/s

)

Figure 11. (top) Absolute S wave velocity plotted in km/s for the section B-B0 . White circles represent the earthquake locations. The trian-gles (from left to right) represent the locations of the back-arc spreading center (SC), the volcanic front (VF), and Big Blue serpentinite sea-mount (BB). The black contour lines trace constant values of the log of the derivative weight sum. (bottom) S wave velocities plotted aspercent deviation from the starting model for the same profile. Labels are the same as the top figure.



underestimation of the back-arc spreading center anomaly amplitude in the S wave structure due to poorray coverage.

The most prominent feature of the VP/VS structure (Figure 13) is the region of extremely high VP/VS ratios inthe outer fore arc, with VP/VS ratios of 1.9 or higher extending from the top of the mantle to a depth of60 km. High VP/VS ratios of 1.85 are also present beneath the active arc at depths of 40–80 km. This regionforms a zone of high VP/VS at about 70 km depth that extends from east of the back-arc spreading center tothe subducting Pacific slab. High VP/VS is also present at in the slab in the depth range 70–120 km, coinci-dent with the double seismic zone. A very low VP/VS ratio is found at shallow depth between the arc andback-arc spreading center, in the region of high VP and VS. Low VP/VS ratios also occur in the inner fore arcand in the shallow (<70 km) and deep (>120 km) parts of the slab.

5.4. Variation Along StrikeThere is significant variation along the strike of the arc in both the P and S wave velocity structures. Thismay be partly due to variations in resolution along strike, with maximum resolution obtained in the center

8

8

8

8

8

8

8

9

9

9

9

10

10

10

11

8

8

8

99

9

9

10

10 10

11

8

8

8

8

9

9

9

9

9

10

10

10

11

11

11

8

8

8

9

9

9

9

10

10

10

10

10

11

11

11

11

11

12

-15 -10 -5 0 5 10 15 20-20

% dVs

z=40km z=60km

z=80km z=100km

−200

−150

−100

−50

0

50

100

−200

−150

−100

−50

0

50

100

−200

−150

−100

−50

0

50

100

−200

−150

−100

−50

0

50

100

−250 −200 −150 −100 −50 0 50 100 150 200 −250 −200 −150 −100 −50 0 50 100 150 200

−250 −200 −150 −100 −50 0 50 100 150 200 −250 −200 −150 −100 −50 0 50 100 150 200

Agrihan

Alamagan

Anatahan

Agrihan

Alamagan

Anatahan

Agrihan

Alamagan

Anatahan

Agrihan

Alamagan

Anatahan

Figure 12. Depth slices of the final S wave velocity model plotted as percent deviation from the starting model for four different depths. The black contour lines trace constant values ofthe log of the derivative weight sum.



SC VF BBB B‘

SC VF BBA A‘

SC VF BBC C‘

dept

h (k

m)

100 150 200500-50-100-150-200-250

100 150 200500-50-100-150-200-250

100 150 200500-50-100-150-200-250

Vp/V

s

8

8

8

8

8

9

9

9

910

10

11

0

50

100

150

200

250

300 1.65

1.7

1.75

1.8

1.85

1.9

1.95

2

88

9

9

9

9

10

10

10

11

11

0

50

100

150

200

250

300 1.65

1.7

1.75

1.8

1.85

1.9

1.95

2

8

8

8

89

9

9

9

910

10

10

11

11

12

0

50

100

150

200

250

300 1.65

1.7

1.75

1.8

1.85

1.9

1.95

2

Figure 13. VP/VS ratios. White circles represent the earthquake locations. The triangles (from left to right) represent the locations of theback-arc spreading center (SC), the volcanic front (VF), and Big Blue serpentinite seamount (BB). The black contour lines trace constant val-ues of the log of the derivative weight sum. The color bar is unitless with warm colors representing high VP/VS ratios and cool colors repre-senting low VP/VS ratios.



of the image (profile B-B0) along the line of ocean bottom seismographs, but there are nonetheless someresolvable differences. The slow velocities in the fore arc have a greater magnitude toward the north of thestudy area. Mantle P wave velocities in the fore arc along section A-A0 are as slow as 7.1–7.2 km/s whereas Pwave velocities along C-C0 range from 7.4 to 7.5 km/s. This correlates well with the presence of serpentinitemud volcanoes on the seafloor extending north of latitude 18�N. At 40 km depth, the largest amplitudelow-velocity anomalies in the fore arc occur just to the west of Big Blue serpentinite seamount (Figures 10and 12).

The distribution of the arc and back-arc anomalies also vary along strike. A large-velocity transition atdepths of 60–100 km from high velocities in the fore arc to low velocities in the arc and back arc occursalong the entire length of the arc (Figures 10 and 12). However, the slow velocity anomalies beneath thevolcanic arc vary in magnitude along strike and appear to be strongest beneath active volcanoes. For exam-ple, at 80 km depth, the slowest velocities are concentrated beneath the volcanic islands at Pagan andGuguan. This suggests the arc volcanoes are fed from specific low-velocity regions in the upper mantle, asobserved in higher-resolution images in NE Japan [Nakajima et al., 2001; Wang and Zhao, 2005].

6. Discussion

6.1. Comparison Between Seismic Velocity and Attenuation StructureProfiles through the 3-D velocity structure obtained in this study show a striking similarity to the 2-D attenu-ation structure obtained previously using data from the same deployment [Pozgay et al., 2009] (Figure 14).Since the P wave data set is larger and better resolved for both velocity and attenuation models, we usethose results to make some direct comparisons between the two models. A slice from the 3-D velocitymodel along latitude 18�N is used for comparison to the 2-D attenuation model at the same location. Theimages show low velocity and high attenuation at similar locations and depths beneath the volcanic arcand the back-arc spreading center (Figure 14). Exceptions to the correlation are in the fore arc, where thelargest-velocity anomalies are at shallow depth beneath Big Blue seamount whereas the attenuation imagesshow a high-attenuation anomaly deeper (50–100 km deep) in the slab, and in the far back arc to the westof the spreading center where there is low resolution in both images due to poor raypath coverage.

The similarity of the velocity and attenuation images is generally expected; in most regions seismic attenua-tion and velocity anomalies show a strong inverse correlation such that slow velocity anomalies correspondto regions of high attenuation [Roth and Wiens, 2000; Artemieva et al., 2004; Dalton et al., 2009]. Laboratorystudies indicate that high temperature [Jackson et al., 2002; Faul and Jackson, 2005; Sundberg and Cooper,2010; McCarthy et al., 2011], melt [Faul et al., 2004; McCarthy and Takei, 2011], or water [Karato, 2003; Aizawaet al., 2008] all may produce strong attenuation and also shear modulus relaxation (and thus velocity reduc-tion) in the seismic band. However, variations in the magnitude of velocity and attenuation anomalies canhave important implications for interpretation [e.g., Shito et al., 2006; Wiens et al., 2008]. In the following sec-tions, we interpret the observed velocity anomalies from this paper along with the attenuation structurefrom Pozgay et al. [2009] in terms of the geodynamic processes operating in the Mariana subductionsystem.

6.2. Interpretation of Incoming Plate and Slab AnomaliesThe incoming Pacific plate and downgoing slab generally show fast velocities, low VP/VS ratios, and lowattenuation as expected because of their colder temperatures relative to the average mantle. Slow S wavevelocities and high VP/VS ratios in the subducting slab at depths of 70–120 km, in the region of the doubleseismic zone, represent an exception to this general observation (Figures 10 and 12). The location of theseanomalies corresponds to a strong high-attenuation anomaly in the attenuation structure [Pozgay et al.,2009]. These reduced velocities in the upper part of the slab are potentially caused by water released bydehydration of the crust and upper mantle of slab. The VS anomaly is located in the region where wide-spread dehydration of the slab is expected due to increased pressure and temperature [e.g., Hacker et al.,2003] and which serves as the source of water for the volcanic arc. The presence of fluids could cause highattenuation either through the presence of a free fluid or by incorporation into the crystal structure of man-tle minerals [Karato, 2003] and corresponding shear modulus relaxation at seismic frequencies. Similar fea-tures are observed within other subducting slabs, for example, beneath Nicaragua and Costa Rica [Syracuseet al., 2008] and within the Japan slab [Nakajima et al., 2009].



6.3. Fore-Arc Structure and SerpentinizationA strong low-velocity anomaly occurs at depths less than 50 km in the outer fore arc beneath just to thewest of Big Blue serpentinite seamount, with S velocities of �3.8 km/s and high VP/VS of greater than 1.9.This region does not show a corresponding high-attenuation anomaly (Figure 14), with moderate QP valuesof about 150 [Pozgay et al., 2009]. The Moho in the outer Mariana fore arc is located at depths of 10–15 km[Takahashi et al., 2007], and the shallow thrust zone at the top of the slab is at depths of 30–50 km just tothe west of Big Blue seamount [Emry et al., 2011]. This indicates that the slow velocity anomaly is localized

dept

h (k

m)

dept

h (k

m)

Vp (k

m/s

)10

00/Q

p

SC VF BBB B‘

SC VF BBB B‘

−250 −200 −150 −100 −50 0 50 100 150 200

0

50

100

150

200

250

300

7.2

7.4

7.6

7.8

8

8.2

8.4

250 300 350 400 450 500 550 600 650

0

50

100

150

200

250

300 0

5

10

15

20

Figure 14. Cross sections comparing (top) the P velocity structure from this study with the P attenuation (1/Qp) structure [Pozgay et al.,2009]. Labels are the same as Figure 8.



in the mantle fore arc immediately above the subducting slab interface. Cold temperatures expected in themantle fore arc [Conder et al., 2002; van Keken, 2003; Wada et al., 2011] and the absence of an attenuationanomaly argue against a thermal origin and suggest a compositional effect such as the formation ofhydrous minerals. An alternative interpretation is that the high fore-arc attenuation anomaly at a depth of50–100 km (Figure 14, bottom) is not well localized in depth and may possibly correspond to the slowvelocity anomaly at shallower depths in the outer fore arc. This would be compatible with Nakajima [2014],who found a region of low velocity and high attenuation in the Kanto area, and interpreted both anomaliesas the result of serpentinized upper mantle.

Serpentinization of the fore-arc mantle is often invoked to explain low seismic velocities and high VP/VS

ratios [Graeber and Asch, 1999; Kamiya and Kobayashi, 2000]. Since serpentine has a much lower shear mod-ulus relative to the bulk modulus, large VP/VS ratios are a diagnostic feature of serpentinized peridotite[Christensen, 1996]. P velocity anomalies of 24 to 6% and VP/VS ratios ranging from 1.9 to 1.95 indicate thatthe degree of serpentinization of mantle material could be relatively high, with as much as 30–60% serpen-tinized material [Hacker et al., 2003; Christensen, 2004; Reynard et al., 2007] assuming a chemically boundwater content of 4–6% [Carlson and Miller, 2003]. This interpretation is reinforced by the presence of activemud volcanoes containing serpentinized mantle peridotite in the Mariana fore arc [Fryer et al., 1999]. A simi-lar level of serpentinization has also been postulated for the Cascadia margin of North America [Bostocket al., 2002], and a somewhat lower degree of serpentinization of 15–25% has been estimated for the CostaRica fore arc [DeShon and Schwartz, 2004].

Another interesting feature of the fore arc is the extremely high velocity in the inner fore arc at depths ofabout 50 km just to the east of the volcanic front. This region shows some of the highest velocities withinthe entire mantle wedge, with VP of 8.1–8.2 km/s and VS of 4.7–4.8 km/s. The region is bounded to the westby a sharp velocity decrease, with changes as large as 15%, immediately beneath the volcanic arc. The highfore-arc velocities suggest cold temperatures and a lack of significant hydrous minerals. We interpret thisarea as representing the ‘‘cold nose’’ of the mantle wedge, caused by the decoupling of the downgoingplate from the fore arc and the stagnation of fore-arc material, with a sharp transition to the much warmerarc section of the wedge [Wada et al., 2011]. The highest velocities in the inner fore arc are observed atdepths of about 40 km (Figure 9, inset), with somewhat lower velocities at depths of 60 km and greater,suggesting some hydration of the lowermost part of the inner fore arc by water from the slab. This is con-sistent with a velocity reduction around 50 km depth in the inner Mariana fore arc imaged using seismicreceiver functions [Tibi et al., 2008].

6.4. Magma Production Region Beneath the Mariana Volcanic ArcWe image slow velocity anomalies centered beneath the active arc extending from the top of the mantle toa depth of 100 km. Such low-velocity anomalies have been observed at similar depths and locationsbeneath volcanic arcs worldwide [e.g., Zhao, 2001; Nakajima et al., 2001; Reyners et al., 2006; Syracuse et al.,2008]. The strongest anomaly occurs at a depth of around 60 km and shows a minimum VP of 7.3 km/s anda minimum VS of 3.9 km/s, for a VP/VS ratio of 1.87. This is similar to average values of 7.4 km/s, 4.0 km/s,and 1.85 observed in a variety of volcanic arc tomographic studies in other arcs around the world [Wienset al., 2008]. This anomaly corresponds well with a strong anomaly in the attenuation image, with a maxi-mum P wave attenuation of about 0.016 (QP 5 56) (Figure 14).

There is strong evidence that the reduced seismic velocity beneath the arc results from the presence of par-tial melt and that the anomaly images the source region of arc magma production in the mantle. Using theSi and Mg thermobarometer developed by Lee et al. [2009], Kelley et al. [2010] find that the final equilibriumdepths for Mariana arc melts is 34–87 km, essentially identical to the depths of the strong velocity andattenuation anomalies. In addition, the large magnitude of the velocity and attenuation anomalies is impos-sible to produce with without the effect of partial melt [Wiens et al., 2008; Pozgay et al., 2009].

Seismic and electromagnetic (EM) methods are sensitive to different material properties and therefore pro-vide complementary information on the structure of back-arc basins. Seismic observations are generallymore sensitive to temperature variations whereas EM methods are generally more sensitive to water con-tent in olivine. Both methods are sensitive to melt content [Wiens et al., 2006a]. A magnetotelluric (MT) sur-vey across the Mariana arc by Matsuno et al. [2010] images a zone of highly conductive material above theslab beneath the arc, extending upward to a depth of about 60 km. This result is consistent with the



presence of slab-derived water in the mantle above the slab as well as aqueous melts at the depths indi-cated by the seismic imaging.

6.5. Structure of the Mariana Back-Arc Spreading CenterWe image slow P wave velocities beneath the back-arc spreading center of 7.1–7.2 km/s and S wave veloc-ities of 4.0–4.1 km/s. A very strong attenuation anomaly is also imaged beneath the back-arc spreading cen-ter at depths of less than 70 km, with a maximum P attenuation of about 0.023 (Q 5 43) [Pozgay et al.,2009]. The maximum P wave anomalies are at a depth of 20–30 km and are located just west of the spread-ing axis. This depth is similar to results from Si and Mg thermobarometry of back-arc spreading centerbasalts, showing a final equilibrium depth of 21–37 km for the back-arc spreading center melts [Kelley et al.,2010]. This suggests that the back-arc anomaly delineates the melt production region beneath the spread-ing center, where basaltic melts are formed by decompression melting. The slight offset of the maximumanomaly to the west of the spreading center may result from upwelling mantle brought in from the west bythe subduction zone mantle flow pattern driven by the downgoing slab [e.g., Conder et al., 2002], consistentwith shear wave splitting results that suggest EW flow to the west of the spreading center [Pozgay et al.,2007].

Both the attenuation result and the P wave image show the back-arc anomaly is somewhat stronger thanthe anomaly beneath the volcanic front, whereas the S velocity images show the anomaly beneath the arcto be �0.2 km/s slower than the anomaly observed beneath the back arc. S wave velocity anomalies in theback arc are more poorly resolved than P wave anomalies due to many fewer arrival times resulting fromthe longer raypaths and high attenuation, so that the resolution of the S wave structure in this region is sig-nificantly poorer than the P structure (Figure 6). So it is likely that the magnitude of the back-arc anomaly isunderestimated in the S wave results. Both P and S images show lower resolution to the west of the back-arc spreading center (Figure 6), so although the eastern side of the back-arc spreading center anomaly iswell resolved in the P wave image, the extent of the back-arc spreading center anomaly to the west of thespreading center is not well resolved. Therefore, it is difficult to confirm from the velocity image whetherthe back-arc anomaly is a narrow vertical feature as suggested by the attenuation image (Figure 14).

Experimental observations predict that partial melt will greatly increase attenuation and significantly reduceseismic velocities [Faul et al., 2004; McCarthy and Takei, 2011], providing an obvious interpretation for thearc and back-arc anomalies. However, quantification of melt porosity in the mantle by seismic observationsremains difficult [e.g., Abers et al., 2014]. There is considerable uncertainty about the effect of melt on seis-mic velocities and different studies have reached different conclusions about the effect of melt on seismicvelocities [Faul et al., 2004; Hammond and Humphreys, 2000; Takei, 2002; McCarthy and Takei, 2011]. Resultsindicating high attenuation for samples containing only a small melt fraction (<1%) [Faul et al., 2004; McCar-thy and Takei, 2011] suggest that even small melt fractions may cause large dissipation and modulus relaxa-tion, producing large attenuation and velocity anomalies.

Laboratory studies suggest that an increase in water content of upper mantle minerals is expected toreduce the seismic velocity of the material [Karato, 2003; Aizawa et al., 2008]. The mantle source of theMariana back-arc basin basalts contains 0.17 wt % water which is high relative to most other back-arc basinsaround the world [Kelley et al., 2006; Wiens et al., 2006b]. This amount of water is expected to reduce theshear velocity by about 1.5% [Karato, 2003] so there may be a contribution to the observed velocities fromwater content. However, the expected velocity reduction due to mantle water is small relative to theobserved velocity anomalies, which are on the order of 8%. Also, the largest-velocity anomalies are foundimmediately beneath the back-arc spreading center, whereas the concentration of water in the mantledecreases with distance from the slab [Kelley et al., 2006]. Therefore, we suggest that water in the mantleplays a minor role in affecting the seismic velocities in the back arc and that the velocity anomalies beneaththe back-arc spreading center result from the presence of partial melt.

The Mariana MT survey imaged an entirely resistive region beneath the back-arc spreading center with athickness of 80–100 km [Matsuno et al., 2010] in the same region where we observe a strong low-velocityanomaly. The resistive region beneath the back arc can be explained by dry mantle with typical upper man-tle temperatures implying that water plays an insignificant role in the conductivity structure of the back arc.A further modeling study of the MT results shows that suggests that the largest melt region that was notdirectly imaged by the MT data, but that is compatible with the observations, has a 3-D shape on a ridge-



segment scale focused toward the spreading center and 0.1–1% interconnected silicate melt [Matsunoet al., 2012]. Thus, the EM data are compatible with a very small melt fraction focused beneath inflated seg-ments of the Mariana spreading center.

6.6. Relationship Between Arc and Back-Arc AnomaliesOne of the central questions to understanding subduction systems is the transfer of material from the slabto the arc and back-arc spreading center largely because the pathways of migration for this process remainpoorly understood. Arc and back-arc magmas are quite distinct, but there is geochemical evidence of trans-fer of water and fluid mobile elements between the arc and back arc [Kelley et al., 2006, 2010; Pearce et al.,2005; Pearce and Stern, 2006]. The velocity tomography in this study and the attenuation tomographyresults from Pozgay et al. [2009] image a slow velocity, high VP/VS, high-attenuation region at 75–100 kmdepth that connects the arc and back-arc anomalies. We propose that the slow velocity, high-attenuationregion between the back-arc and volcanic arc anomalies is comprised of mantle material that is hot enoughto permit the transport of small quantities of fluids and aqueous melts from the subarc region where water-rich arc magmas are generated to the back-arc spreading center melting column. The low velocity and highattenuation in this channel may indicate the presence of a very small fraction of water-rich melt, as meltwould lower the velocity and increase attenuation as observed. A tenuous but persistent interconnection ofthe volcanic arc and the back-arc spreading center fluids and melts at these depths is likely responsible forthe elevated water contents and the observed geochemical slab signature in the back-arc basin basalts[Pearce and Stern, 2006].

Another interesting feature is the extremely fast velocities and low VP/VS ratios observed at depths down toabout 40 km between the volcanic arc and the back-arc spreading center (Figure 9). This region contains thefastest velocities within the mantle wedge of about 5.0 km/s for S and 8.5 km/s for P, and a very low Vp/Vs ratioof about 1.7. These velocities are fast even for typical oceanic lithosphere of a similar age (<6 Ma). A very lowattenuation region is also found in the same area between the volcanic arc and the back arc (Figure 14). Weinterpret this region to be cold refractory lithosphere which may be unusually thick or depleted due to proc-esses involved with the initial rifting of the Mariana back arc. This region represents a stagnant portion of themantle wedge and acts as barrier for exchange of materials between the arc and back arc at shallow depths.

7. Conclusions

Three-dimensional images of the seismic velocity structure of the mantle for the Mariana subduction zoneilluminate subduction zone processes. The tomographic results help constrain the subduction water cyclefrom release into the fore arc and wedge to the formation of aqueous melts beneath the volcanic arc andback-arc spreading center. A layer of low P and S velocities and high VP/VS in the outer Mariana fore arc indi-cate that a large amount of water is transported into the fore-arc mantle with as much as 30–60% serpenti-nization of mantle minerals. A low-velocity, high VP/VS region is found in the slab between 70 and 120 kmdepth suggesting that dehydration of the slab exists to at least this depth. These fluids are then transferredto the mantle wedge where they facilitate the formation of aqueous melts beneath the volcanic arc. In addi-tion, we image a separate low-velocity anomaly that represents the melt production region for the back-arcspreading center. These images combined with laboratory experiments and MT results are consistent withthe presence of small amounts of partial melt beneath the back-arc spreading center.

Our results show that the source region for arc melts form at a greater depth than for back-arc spreadingcenter, corroborating prior petrological and geochemical work [Kelley et al., 2010]. Arc melts are dominatedby those that form due to the presence of fluids within the hot core of the mantle wedge, whereas back-arcmelts form within a shallow zone of decompression melting beneath the spreading center. Furthermore,these two regions appear to be connected at 75–100 km depth which is a channel for fluids and fluid-mobile element transfer from the slab, to the arc, to the back arc and can explain the high water contentsfound in arc and back-arc lavas.

ReferencesAbers, G. A., K. M. Fischer, G. Hirth, D. A. Wiens, T. Plank, B. K. Holtzman, C. McCarthy, and E. Gazel (2014), Reconciling mantle attenuation-

temperature relationships from seismology, petrology, and laboratory measurements, Geochem. Geophys. Geosyst., 15, 3521–3542, doi:10.1002/2014GC005444.

AcknowledgmentsWe thank numerous people forassistance with deploying andrecovering the seismographs,particularly P. Shore, S. Webb,A. Sauter, B. Shiro, P. Jonke,J. Camacho, J. Kaipat, H. Iwamoto, andR. Chong, as well as the captains andthe crews of the R/V Kaiyo, the R/VWecoma, and the Super Emerald. Landseismic instrumentation was providedby the PASSCAL program of theIncorporated Research Institutions inSeismology (IRIS); the Lamont OceanBottom Seismograph Facility and theUniversity of Tokyo provided oceanbottom seismographs. The data usedfor this study are available through theIRIS Data Management Center: http://www.iris.edu/ds/nodes/dmc/data/Dataset: MARIANAS 2003–2004.Network Code: XO. The paper wasfinished while D. Wiens was supportedas a visiting professor by theEarthquake Research Institute,University of Tokyo. This research wassupported by the MARGINS programunder National Science Foundationgrants OCE-0001938 and EAR-0549056.



http://www.iris.edu/ds/nodes/dmc/data/Dataset




Aizawa, Y., A. Barnhoorn, U. H. Faul, J. D. Fitz Gerald, I. Jackson, and I. Kov�acs (2008), Seismic properties of Anita Bay dunite: An exploratorystudy of the influence of water, J. Petrol., 49, 841–855.

Artemieva, I. M., M. Billien, J.-J. Leveque, and W. D. Mooney (2004), Shear wave velocity, seismic attenuation, and thermal structure of thecontinental upper mantle, Geophys. J. Int., 157, 607–628.

Blackman, D. K., J. A. Orcutt, and D. W. Forsyth (1995), Recording teleseismic earthquakes using ocean-bottom seismographs at mid-oceanridges, Bull. Seismol. Soc. Am., 85(6), 1648–1664.

Bostock, M. G., R. D. Hyndman, S. Rondenay, and S. M. Peacock (2002), An inverted continental Moho and serpentinization of the forearcmantle, Nature, 417, 536–538.

Carlson, R. L., and D. J. Miller (2003), Mantle wedge water contents estimated from seismic velocities in partially serpentinized peridotites,Geophys. Res. Lett., 30(5), 1250, doi:10.1029/2002GL016600.

Calvert, A. J., S. L. Klemperer, N. Takahashi, and B. C. Kerr (2008), Three-dimensional crustal structure of the Mariana island arc from seismictomography, J. Geophys. Res., 113, B0140, doi:10.1029/2007JB004939.

Christensen, N. I. (1996), Poisson’s ratio and crustal seismology, J. Geophys. Res., 101(B2), 3139–3156.Christensen, N. I. (2004), Serpentinites, peridotites, and seismology, Int. Geol. Rev., 46, 795–816, doi:10.2747/0020–6814.46.9.795.Conder, J. A., D. A. Wiens, and J. Morris (2002), On the decompression melting structure at volcanic arcs and back-arc spreading centers,

Geophys. Res. Lett., 29(15), doi:10.1029/2002GL015390.Conder, J. A., and D. A. Wiens (2006), Seismic structure beneath the Tonga arc and Lau back-arc basin determined from joint Vp, Vp/Vs

tomography, Geochem. Geophys. Geosyst., 7, Q03018, doi:10.1029/2005GC001113.Dalton, C. A., G. Ekstrom, and A. M. Dziewonski (2009), Global seismological shear velocity and attenuation: A comparison with experimen-

tal observations, Earth Planet. Sci. Lett., 284, 65–75.DeShon, H. R., and S. Y. Schwartz (2004), Evidence for serpentinization of the forearc mantle wedge along the Nicoya Peninsula, Costa Rica,

Geophys. Res. Lett., 31, L21611, doi:10.1029/2004GL021179.Emry, E. L., D. A. Wiens, and D. Garcia-Castellanos (2014), Faulting within the Pacific plate at the Mariana Trench: Implications for plate inter-

face coupling and subduction of hydrous minerals, J. Geophys. Res. Solid Earth, 119, 3076–3095, doi:10.1002/2013JB010718.Faul, U. H., and I. Jackson (2005), The seismological signature of temperature and grain size variations in the upper mantle, Earth Planet.

Sci. Lett., 234, 119–134.Faul, U. H., J. D. Fitzgerald, and I. Jackson (2004), Shear wave attenuation and dispersion in melt-bearing olivine crystals: 2. Microstructural

interpretation and seismological implications, J. Geophys. Res., 109, B06202, doi:10.1029/2003JB002407.Forsyth, D. W., et al. (1998), Imaging the deep structure beneath a mid-ocean ridge: The MELT experiment, Science, 280, 1215–1218.Fryer, P., E. L. Ambos, and D. M. Hussong (1985), Origin and emplacement of Mariana forearc seamounts, Geology, 13, 774–777.Fryer, P., C. G. Wheat, and M. J. Mottl (1999), Mariana blueschist mud volcanism: Implications for conditions within the subduction zone,

Geology, 27, 103–106.Gorbatov, A., and B. L. N. Kennett (2003), Joint bulk-sound and shear tomography for Western Pacific subduction zones, Earth Planet. Sci.

Lett., 210, 527–543.Graeber, F., and G. Asch (1999), Three-dimensional models of P wave velocity and P-to-S velocity ratio in the southern central Andes by

simultaneous inversion of local earthquake data, J. Geophys. Res., 104(B9), 20,237–20,256.Hacker, B. R., G. A. Abers, and S. M. Peacock (2003), Subduction factory: 1. Theoretical mineralogy, densities, seismic wave speeds, and H2O

contents, J. Geophys. Res., 108(B1), 2029, doi:10.1029/2001JB001127.Hammond, W. C., and E. D. Humphreys (2000), Upper mantle seismic wave velocity: Effects of realistic partial melt geometries, J. Geophys.

Res., 105(B5), 10,975–10,986.Hammond, W. C., and D. R. Toomey (2003), Seismic velocity anisotropy and heterogeneity beneath the mantle electromagnetic and

tomography experiment (MELT) region of the East Pacific Rise from analysis of P and S body waves, J. Geophys. Res., 108(B4), 2176, doi:10.1029/2002JB001789.

Isse, T., et al. (2009), Seismic structure of the upper mantle beneath the Philippine Sea from seafloor and land observation: Implications formantle convection and magma genesis in the Izu-Bonin-Mariana subduction zone, Earth Planet. Sci. Lett., 278, 107–119.

Jackson, I., J. D. Fitzgerald, U. H. Faul, and B. H. Tan (2002), Grain-size-sensitive seismic wave attenuation in polycrystalline olivine, J. Geo-phys. Res., 107(B12), 2360, doi:10.1029/2001JB001225.

Jaxybulatov, K., I. Koulakov, and N. L. Dobretsov (2013), Segmentation of the Izu-Bonin and Mariana plates based on the analysis of theBenioff seismicity distribution and regional tomography results, Solid Earth, 4, 1–15, doi:10.5194/se-4-1-2013.

Kamiya, S., and Y. Kobayashi (2000), Seismological evidence for the existence of serpentinized wedge mantle, Geophys. Res. Lett., 27(6),819–822, doi:10.1029/1999GL011080.

Karato, S.-I. (2003), Mapping water content in the upper mantle, in Inside the Subduction Factory, Geophys. Monogr. Ser., vol. 138, edited byJ. M. Eiler, pp. 135–152, AGU, Washington, D. C.

Kato, T., J. Beavan, T. Matsushima, Y. Kotake, J. T. Camacho, and S. Nakao (2003), Geodetic evidence of backarc spreading in the Marianatrough, Geophys. Res. Lett., 30(12), 1625, doi:10.1029/2002GL016757.

Kelley, K. A., T. Plank, T. L. Grove, E. M. Stolper, S. Newman, and E. Hauri (2006), Mantle melting as a function of water content beneathback-arc basins, J. Geophys. Res., 111, B09208, doi:10.1029/2005JB003732.

Kelley, K. A., T. Plank, S. Newman, E. Stolper, T. L. Grove, S. Parman, and E. Hauri (2010), Mantle melting as a function of water contentbeneath the Mariana arc, J. Petrol., 51, 1711–1738, doi:10.1093/petrology/egq036.

Kennett, B. L. N., and E. R. Engdahl (1991), Travel times for global earthquake location and phase identification, Geophys. J. Int., 105(2), 429–465.

Kennett, B. L. N., E. R. Engdahl, and R. Buland (1995), Constraints on seismic velocities in the Earth from travel times, Geophys. J. Int., 122(1),108–124.

Lebedev, S., and G. Nolet (2003), Upper mantle beneath Southeast Asia from S velocity tomography, J. Geophys. Res., 108(B1), 2048, doi:10.1029/2000JB000073.

Lee, C.-T. A., P. Luffi, T. Plank, H. Dalton, and W. P. Leeman (2009), Constraints on the depths and temperatures of basaltic magma genera-tion on Earth and other terrestrial planets using new thermobarometers, Earth Planet. Sci. Lett., 279, 20–33, doi:10.1016/j.epsl.2008.12.020.

Matsuno, T., et al. (2010), Upper mantle electrical resistivity structure beneath the central Mariana subduction system, Geochem. Geophys.Geosyst., 11, Q09003, doi:10.1029/2010GC003101.

Matsuno, T., R. L. Evans, N. Seama, and A. D. Chave (2012), Electromagnetic constraints on a melt region beneath the central Mariana back-arc spreading ridge, Geochem. Geophys. Geosyst., 13, Q10017, doi:10.1029/2012GC004326.



http://dx.doi.org/10.1029/2002GL016600

http://dx.doi.org/10.1029/2007JB004939

http://dx.doi.org/10.2747/0020-6814.46.9.795

http://dx.doi.org/10.1029/2002GL015390

http://dx.doi.org/10.1029/2005GC001113

http://dx.doi.org/10.1029/2004GL021179

http://dx.doi.org/10.1002/2013JB010718

http://dx.doi.org/10.1029/2003JB002407

http://dx.doi.org/10.1029/2001JB001127

http://dx.doi.org/10.1029/2002JB001789

http://dx.doi.org/10.1029/2001JB001225

http://dx.doi.org/10.5194/se-4-1-2013

http://dx.doi.org/10.1029/1999GL011080

http://dx.doi.org/10.1029/2002GL016757

http://dx.doi.org/10.1029/2005JB003732

http://dx.doi.org/10.1093/petrology/egq036

http://dx.doi.org/10.1029/2000JB000073

http://dx.doi.org/10.1016/j.epsl.2008.12.020

http://dx.doi.org/10.1016/j.epsl.2008.12.020

http://dx.doi.org/10.1029/2010GC003101

http://dx.doi.org/10.1029/2012GC004326

McCarthy, C., and Y. Takei (2011), Anelasticity and viscosity of partially molten rock analogue: Toward seismic detection of small quantitiesof melt, Geophys. Res. Lett., 38, L18306, doi:10.1029/2011GL048776.

McCarthy, C., Y. Takei, and T. Hiraga (2011), Experimental study of attenuation and dispersion over a broad frequency range: 2. The univer-sal scaling of polycrystalline materials, J. Geophys. Res., 116, B09207, doi:10.1029/2011JB008384.

Menke, W. (1984), Geophysical Data Analysis: Discrete Inverse Theory, 260 pp., Academic, Orlando, Fla.Miller, M. S., B. L. N. Kennett, and G. S. Lister (2004), Imaging changes in morphology, geometry, and physical properties of the subducting

Pacific plate along the Izu-Bonin-Mariana arc, Earth Planet. Sci. Lett., 224(3–4), 363–370.Nakajima, J. (2014), Seismic attenuation beneath Kanto, Japan: Evidence for high attenuation in the serpentinized subducting mantle, Earth

Planets Space, 66, 1–11.Nakajima, J., T. Matsuzawa, A. Hasegawa, and D. Zhao (2001), Three-dimensional structure of Vp, Vs, and Vp/Vs beneath northeastern

Japan: Implications for arc magmatism and fluids, J. Geophys. Res., 106(B10), 21,843–21,857.Nakajima, J., Y. Tsuji, A. Hasegawa, S. Kita, T. Okada, and T. Matsuzawa (2009), Tomographic imaging of hydrated crust and mantle in the

subducting Pacific slab beneath Hokkaido, Japan: Evidence for dehydration embrittlement as a cause of intraslab earthquakes, Gond-wana Res., 16, 470–481.

Paige, C. C., and M. A. Saunders (1982), LSQR: An algorithm for sparse linear equations and sparse least squares, ACM Trans. Math. Software,8(1), 43–71.

Parman, S. W., T. L. Grove, K. A. Kelley, and T. Plank (2011), Along-arc variations in the pre-eruptive H2O contents of Mariana arc magmasinferred from fractionation paths, J. Petrol., 52(2), 257–278.

Pavlis, G. L., F. Vernon, D. Harvey, and D. Quinlan (2004), The generalized earthquake-location (GENLOC) package: An earthquake-locationlibrary, Comput. Geosci., 30(9–10), 1079–1091.

Pearce, J. A., and R. J. Stern (2006), Origin of back-arc basin magmas: Trace element and isotope perspectives, in Back-Arc Spreading Sys-tems: Interactions Among Physical, Chemical, Biological, and Geological Processes, Geophys. Monogr. Ser., vol. 166, edited by D. M. Christieet al., pp. 63–86, AGU, Washington, D. C.

Pearce, J. A., R. J. Stern, S. H. Bloomer, and P. Fryer (2005), Geochemical mapping of the Mariana arc-basin system: Implications for thenature and distribution of subduction components, Geochem. Geophys. Geosyst., 6, Q07006, doi:10.1029/2004GC000895.

Pozgay, S. H., D. A. Wiens, J. A. Conder, H. Shiobara, and H. Sugioka (2007), Complex mantle flow in the Mariana subduction system: Evi-dence from shear wave splitting, Geophys. J. Int., 170, 371–386.

Pozgay, S. H., D. A. Wiens, J. A. Conder, H. Shiobara, and H. Sugioka (2009), Seismic attenuation tomography of the Mariana subduction sys-tem: Implications for thermal structure, volatile distribution, and slow spreading dynamics, Geochem. Geophys. Geosyst., 10, Q04X05, doi:10.1029/2008GC002313.

Pyle, M. L., D. A. Wiens, D. Weeraratne, P. J. Shore, H. Shiobara, and H. Sugioka (2010), Shear velocity structure of the Mariana mantle wedgefrom Rayleigh wave phase velocities, J. Geophys. Res., 115, B11304, doi:10.1029/2009JB00697.

Rawlinson, N., and B. L. N. Kennett (2004), Rapid estimation of relative and absolute delay times across a network by adaptive stacking,Geophys. J. Int., 157, 332–340.

Reynard, B., N. Hilairet, E. Balan, and M. Lazzeri (2007), Elasticity of serpentines and extensive serpentinization in subduction zones, Geo-phys. Res. Lett., 34, L13307, doi:10.1029/2007GL030176.

Reyners, M., D. Eberhart-Phillips, G. Stuart, and Y. Nishimura (2006), Imaging subduction from the trench to 300 km depth beneath the cen-tral North Island, New Zealand, with Vp and Vp/Vs, Geophys. J. Int., 165, 565–583.

Roth, E. G., and D. A. Wiens (2000), An empirical relationship between seismic attenuation and velocity anomalies in the upper mantle, Geo-phys. Res. Lett., 27(5), 601–604.

Schellart, W. P., J. Freeman, D. R. Stegman, L. Moresi, and D. May (2007), Evolution and diversity of subduction zones controlled by slabwidth, Nature, 446, 308–311.

Shaw, A. M., E. H. Hauri, T. P. Fischer, D. R. Hilton, and K. A. Kelley (2008), Hydrogen isotopes in Mariana arc melt inclusions: Implications forsubduction dehydration and the deep-Earth water cycle, Earth Planet. Sci. Lett., 275, 138–145.

Shiobara, H., and T. Kanazawa (2008), Development of a light weight and autonomic sensor system for ocean bottom seismometer, J. Seis-mol. Soc. Jpn., 61, 137–144.

Shiobara, H., H. Sugioka, K. Mochizuki, S. Oki, T. Kanazawa, Y. Fukao, and K. Suyehiro (2010), Double seismic zone at the north Marianarevealed by long-term ocean bottom array observation, Geophys. J. Int., 183, 1455–1469.

Shito, A., S.-I. Karato, K. N. Matsukage, and Y. Nishibara (2006), Towards mapping the three-dimensional distribution of water in the uppermantle from velocity and attenuation tomography, in Earth’s Deep Water Cycle, edited by S. D. Jacobsen and S. Van Der Lee, pp. 225–236, AGU, Washington, D. C.

Stern, R. J., M. J. Fouch, and S. L. Klemperer (2003), An overview of the Izu-Bonin-Mariana subduction factory, in Inside the Subduction Fac-tory, Geophys. Monogr. Ser., vol. 138, edited by J. Eiler, pp. 175–222, AGU, Washington, D. C.

Stolper, E., and S. Newman (1994), The role of water in the petrogenesis of Mariana trough magmas, Earth Planet. Sci. Lett., 121, 293–325.Sundberg, M., and R. F. Cooper (2010), A composite viscoelastic model for incorporating grain boundary sliding and transient diffusion

creep; correlating creep and attenuation responses for materials with a fine grain size, Philos. Mag., 90(20), 2817–2840.Syracuse, E. M., G. A. Abers, K. Fischer, L. MacKenzie, C. Rychert, M. Protti, V. Gonz�alez, and W. Strauch (2008), Seismic tomography and

earthquake locations in the Nicaraguan and Costa Rican upper mantle, Geochem. Geophys. Geosyst., 9, Q07S08, doi:10.1029/2008GC001963.

Takahashi, N., S. Kodiara, S. L. Klemperer, Y. Tatsumi, Y. Kaneda, and K. Suyehiro (2007), Crustal structure and evolution of the Marianaintra-oceanic island arc, Geology, 35, 203–206.

Takahashi, N., S. Kodaira, Y. Tatsumi, Y. Kaneda, and K. Suyehiro (2008), Structure and growth of the Izu-Bonin-Mariana arc crust: 1. Seismicconstraint on crust and mantle structure of the Mariana arc–back-arc system, J. Geophys. Res., 113, B01104, doi:10.1029/2007JB005120.

Takei, Y. (2002), Effect of pore geometry on VP/VS: From equilibrium geometry to crack, J. Geophys. Res., 107(B2), 2043, doi:10.1029/2001JB000522.

Tatsumi, Y., H. Shukuno, K. Tani, N. Takahashi, S. Kodaira, and T. Kogiso (2008), Structure and growth of the Izu-Bonin-Mariana arc crust: 2.Role of crust-mantle transformation and the transparent Moho in arc crust evolution, J. Geophys. Res., 113, B02203, doi:10.1029/2007JB005121.

Tibi, R., D. A. Wiens, and X. Yuan (2008), Seismic evidence for widespread serpentinized forearc mantle along the Mariana convergencemargin, Geophys. Res. Lett., 35, L13303, doi:10.1029/2008GL034163.

Um, J., and C. Thurber (1987), A fast algorithm for two-point seismic ray tracing, Bull. Seismol. Soc. Am., 77(3), 972–986.



http://dx.doi.org/10.1029/2011GL048776

http://dx.doi.org/10.1029/2011JB008384

http://dx.doi.org/10.1029/2004GC000895

http://dx.doi.org/10.1029/2008GC002313

http://dx.doi.org/10.1029/2009JB00697

http://dx.doi.org/10.1029/2007GL030176

http://dx.doi.org/10.1029/2008GC001963

http://dx.doi.org/10.1029/2008GC001963

http://dx.doi.org/10.1029/2007JB005120

http://dx.doi.org/10.1029/2001JB000522

http://dx.doi.org/10.1029/2001JB000522

http://dx.doi.org/10.1029/2007JB005121

http://dx.doi.org/10.1029/2007JB005121

http://dx.doi.org/10.1029/2008GL034163

Van Decar, J. C., and R. S. Crosson (1990), Determination of teleseismic relative phase arrival times using multi-channel cross-correlationand least squares, Bull. Seismol. Soc. Am., 80, 150–169.

van der Hilst, R. D., E. R. Engdahl, W. Spakman, and G. Nolet (1991), Tomographic imaging of subducted lithosphere below northwestPacific island arcs, Nature, 353, 37–43.

van Keken, P. E. (2003), The structure and dynamics of the mantle wedge, Earth Planet. Sci. Lett., 215, 323–338.Wada, I., C. A. Rychert, and K. Wang (2011), Sharp thermal transition in the forearc mantle wedge as a consequence of nonlinear mantle

wedge flow, Geophys. Res. Lett., 38, L13308, doi:10.1029/2011GL047705.Wade, J. A., et al. (2005), The May 2003 eruption of Anatahan volcano, Mariana Islands: Geochemical evolution of a silicic island-arc vol-

cano, J. Volcanol. Geotherm. Res., 146, 139–170.Wang, Z., and D. Zhao (2005), Seismic imaging of the entire arc of Tohoku and Hokkaido in Japan using P-wave, S-wave and sP depth-

phase data, Phys. Earth Planet. Inter., 152, 144–162.Webb, S. C., T. K. Deaton, and J. C. Lemire (2001), A broadband ocean-bottom seismometer system based on a 1-Hz natural period geo-

phone, Bull. Seismol. Soc. Am., 91(2), 304–312.Wiens, D. A., and G. P. Smith (2003), Seismological constraints on structure and flow patterns within the mantle wedge, in Inside the Sub-

duction Factory, Geophys. Monogr. Ser., vol. 138, edited by J. Eiler, pp. 59–82, AGU, Washington, D. C.Wiens, D. A., N. Seama, and J. A. Conder (2006a), Mantle structure and flow patterns beneath active back-arc basins inferred from passive

seismic and electromagnetic methods, in Back-Arc Spreading Systems: Interactions Among Physical, Chemical, Biological, and GeologicalProcesses, Geophys. Monogr. Ser., vol. 166, edited by D. M. Christie et al., pp. 43–62, AGU, Washington, D. C.

Wiens, D. A., K. A. Kelley, and T. Plank (2006b), Mantle temperature variations beneath the back-arc spreading centers inferred from seis-mology, petrology, and bathymetry, Earth Planet. Sci. Lett., 248, 30–42.

Wiens, D. A., J. A. Conder, and U. H. Faul (2008), The seismic structure and dynamics of the mantle wedge, Annu. Rev. Earth Planet. Sci., 36,421–455, doi:10.1146/annurev.earth.33.092203.122633.

Wright, S. J., and J. N. Holt (1986), A new non-linear least squares algorithm for the seismic inversion problem, Geophys. J. R. Astron. Soc.,87(3), 1041–1056, doi:10.1111/j.1365-246X.1986.tb01982.x.

Zhao, D. (2001), Seismological structure of subduction zones and its implications for arc magmatism and dynamics, Phys. Earth Planet.Inter., 127, 197–214, doi:10.1016/S0031–9201(01)00228-X.

Zhao, D., A. Hasegawa, and H. Kanamori (1994), Deep structure of Japan subduction zone as derived from local, regional, and teleseismicevents, J. Geophys. Res., 99(B11), 22,313–22,329.



http://dx.doi.org/10.1029/2011GL047705

http://dx.doi.org/10.1146/annurev.earth.33.092203.122633

http://dx.doi.org/10.1111/j.1365-246X.1986.tb01982.x

http://dx.doi.org/10.1016/S0031-9201(01)00228-X

p and s velocity tomography of the mariana subduction...

Documents