tters 256 (2007) 473–483www.elsevier.com/locate/epsl

Earth and Planetary Science Le

Plate kinematics, slab shape and back-arc stress: A comparisonbetween laboratory models and current subduction zones

A. Heuret a,⁎, F. Funiciello b, C. Faccenna b, S. Lallemand a

a Géosciences Montpellier, UMR 5243 CNRS-UM2, Université Montpellier 2, CC. 60, place E. Bataillon, 34095 Montpellier cedex 5, Franceb Dip. Scienze Geologiche, Università degli Studi « Roma TRE », Largo S.Leonardo Murialdo 1, 00146 Roma, Italy

Received 22 May 2006; received in revised form 26 January 2007; accepted 1 February 2007

Available onlin

Editor: C.P. Jaupart

e 11 February 2007

Abstract

A combination of statistical studies on present-day subduction zones and three-dimensional (3D) laboratory models isperformed with the aim to clarify the way that plate kinematics control the geometry of the slab and the overriding platedeformation in subduction zones. In 3D laboratory models, the analogue of a two layer linearly viscous lithosphere–upper mantlesystem is achieved by means of silicon putty glucose syrup tank experiment. The subducting and overriding plate velocities aresystematically changed by exploring the variability field of natural plate kinematics. Both statistical and modelling approachesrecognize the importance of overriding plate motion on subduction process behavior: (1) trenches migrate at a rate close to theoverriding plate motion, but always move slower than the overriding plates. The mechanism at work is a direct consequence of“slab anchoring” opposed by both lithosphere and mantle viscous resistance and is responsible for overriding plate deformation andslab geometry variability. (2) An overriding plate shortens when the overriding plate moves toward the trench and conditions thatare favourable for overriding plate extension are created when the overriding plate moves away from the trench. (3) Shallow andsteep dips are found if the overriding plate moves toward and away from the trench, respectively.© 2007 Elsevier B.V. All rights reserved.

Keywords: subduction; plate kinematics; slab geometry; back-arc stress; laboratory models

1. Introduction

The common view of trench motion considers thatgravitational slab pull drives trenches to “roll back” [1–3].Only a few authors observed that some trenches advancedtowards the overriding plate [4–6]. However, a recent

⁎ Corresponding author. Tel.: +33 4 67143301; fax: +33 467143642.E-mail addresses: heuret@dstu.univ-montp2.fr (A. Heuret),

ffunicie@uniroma3.it (F. Funiciello), faccenna@uniroma3.it(C. Faccenna), lallemand@dstu.univ-montp2.fr (S. Lallemand).

0012-821X/$ - see front matter © 2007 Elsevier B.V. All rights reserved.doi:10.1016/j.epsl.2007.02.004

global study which considered all present-day oceanicsubduction zones has shown that the trench advancerepresents, in the HS3-Nuvel1A hotspot reference frame[7], about 50% of the cases [8,9]. This peculiar behaviorwas first reproduced by simplified laboratory modelswhere the variability in trench kinematics was obtained bychanging only the subducting plate parameters [10].

Recent statistical studies have shown that trenchmotion and, in turn, both the geometry of the subductedlithosphere and the overriding plate deformation style(compression or extension) are closely related to the ab-solute motion of the overriding plate (V

YOP) [8,9]. Trench

474 A. Heuret et al. / Earth and Planetary Science Letters 256 (2007) 473–483

retreat, shallow dip and overriding plate compressionmainly correlate with trenchwardmotion of the overridingplate (e.g., [5,8,11]). Trench advance, steep dip and back-arc extension are mostly observed if the overriding platemoves oceanward (e.g., [5,8,11]). However, the relation-ship between trench and overriding plate motions is nottrivial since trench motion is also related to the dynamicsof the slab and to the absolute motion of the subductingplate (VYSP).

In this paper, we examine trench migration, overrid-ing plate deformation and slab shape as a self-consistentconsequence of subducting plate and overriding platekinematics. We thus compare a statistical analysis of theworldwide present-day subduction system to the resultsof 3D laboratory models in which subduction isgenerated by forcing one plate downward beneathanother. The models explore the whole range of naturalkinematic configurations. In particular, they consider forthe first time the roles of imposed overriding platemotion and the consequences of trench advancing.

2. Plate kinematics, slab geometry and overridingplate deformation in current subduction zones

2.1. Data

The data used in the present paper (plate kinematics,slab geometries and overriding plate strain) are extractedfrom an updated database of subduction zone parametersthat was previously extensively described in [8,9,11]. The

Table 1Cumulated trench length, mean trench–normal plate kinematics (calculated frdeformation style of the 16 subduction segments with deflected slabs

Subduction zone Trenchlength

Mean subductingplate motion (VSP)

Mean overridingplate motion (VO

(km) (mm/yr) (mm/yr)

Andaman (And) 665 20 −10Sunda (Sun) 3310 63 −14Mariana (Mar) 1530 90 −63Izu-Bonin (Izu) 840 97 −47Japan (Jap) 860 110 −11South Kuriles (SKur) 950 97 −21Colombia (Col) 860 15 38Peru–Chile (Per) 2690 27 45Juan Fernandez (Juan) 630 29 43Antilles (Ant) 995 40 −29Sandwich (Sand) 925 27 −22Kermadec (Ker) 1035 103 −47Tonga (Ton) 1080 102 −37New Britain (NBrit) 680 92 −62

See the bar in Fig. 1 for labels E3 to C3. The seismic tomography studies usetable.

different parameters have been sampled every 220 km oftrench along all the present-day oceanic subduction zones.We restricted our analysis to slabs that have reached theupper mantle/lower mantle discontinuity. Indeed, inlaboratorymodels, such slabs show steady-state behaviors(see [12]) and we assume the same for natural subductionzones. We based our study on 173 transects andindividualized 14 different subduction sections as afunction of their plate kinematics, slab geometry, andback-arc deformation style (the data are averaged alongeach section; Table 1).

Plates and trench absolute motions are calculated in theHS3-NUVEL1A hotspot reference frame [7]. We use thetrench–normal component of velocities. Both subductingand overriding plate motions (VSP and VOP, respectively)are defined as positive when directed toward the trench(Fig. 1a). Plate convergence (VPC) is the sumof subductingand overriding plate velocities: VPC=VSP+VOP. Trenchmotion (VT) is calculated by adding the overriding platedeformation velocity (VD) to the overriding plate motion,under the assumption of negligible tectonic accretion anderosion rates (see also [8] for more details): VT=VOP+VD.Trench seaward motion and overriding plate extension aredefined as positive.

The “overriding plate strain” (OPS) is determinedfrom focal mechanisms of earthquakes occurring at adepth less than 40 km within the overriding plate and farfrom the subduction interface. Each subduction zone isclassified in a continuum of seven strain classes, fromhighly extensional (class E3, back-arc spreading) to

om [20]; see also [7,8,13] for a more detailed dataset), and the back-arc

P)Mean plate convergencerate (VPC)

Back-arcdeformation style

Referencefor tomography

(mm/yr)

10 E3 [22]48 0 [23]27 E3 [24]50 E2 [24]99 C3 [24]75 C2 [24]52 C3 [23]71 C3 [23]73 C3 [23]11 E1 [23]5 E3 [23]55 E2 [24]65 E3 [25]31 E3 [25]

d to constrain the deep geometry of the slabs are also referenced in this

Fig. 1. a) Structural pattern of subduction zone. V is for velocity, SP is for subducting plate, OP is for overriding plate, PC is for plate convergence, Tis for trench and D is for overriding plate deformation. Black arrows refer to positive velocities. b) Relations between the plate kinematics, the slabshape and the overriding plate deformation style of current subduction zones. The 174 transects of the database are represented by grey dots. The 14subduction zones whose slabs deflect at depth are also represented in this VSP/VOP space. VSP and VOP variability of each subduction zone issymbolized by error bars. Dot size is a function of the segment length (kilometres of trench). Dot colour refers to the overriding plate deformationstyle. Each slab shape is deduced by combining both hypocenter distribution [21] and seismic tomography profiles (references are given in Table 1).The dashed part of the profile is only constrained by tomographic data. The length scale is the same for all the profiles. The grey area refers to negativeVPC (no convergence between the subducting and overriding plate). Abbreviations for subduction zone names are given in Table 1.

475A. Heuret et al. / Earth and Planetary Science Letters 256 (2007) 473–483

highly compressive (class C3, back-arc shortening) as afunction of the dominant focal mechanism (Fig. 1b) (seealso [8] for more details). We use OPS rather than VD

because the latter value is not available with enoughaccuracy for all subduction zones, especially for thosecharacterized by the lowest strain rate values. However,the strain classes more or less correspond to a givennormal component of strain rate (E3 to E2: rifting orspreading rate N1 cm/yr; E1 to C1: strain rate b1 cm/yr;C2 to C3: shortening N1 cm/yr).

The slab geometry is described by using the trench–normal hypocenter distribution [13] and existing seismictomography profiles ([14–17]; see Table 1 for details).In particular, as the deepest part of many slabs is

aseismic, seismic tomography imaging has been used tovisualize the deepest geometry of the subductedlithosphere. The dip of the slab is measured between125 and 660 km in depth (see [11] for details).

2.2. Statistical observations

Fig. 1b summarizes the relationships obtainedbetween VSP, VOP, slab geometry and overriding platedeformation style. Two major striking features appear.First, overriding plate velocities are equally partitionedbetween retreat and advancing motion and correlatewith subducting velocity (R=−0.73), as fast movingsubducting plates are preferentially associated with

Fig. 2. Correlations observed in natural subduction zones for trench motion (row 1), Overriding Plate Strain (row2), and slab dip (row 3) vs.overriding plate motion (column a) and subducting plate motion (column b). Absolute motions are associated with errors bars of ±5 mm/yr. Slab dipsare measured with an accuracy of about ±5°.

476 A. Heuret et al. / Earth and Planetary Science Letters 256 (2007) 473–483

overriding plates moving away from the trench andvice-versa.

Second, as noticed in several other previous studies[5,11], the overriding plate motion is somehow related toslab geometry and to OPS. Shallow dipping slabs andback-arc compressions are found if the overriding platemoves toward the trench and, inversely, if the overridingplate moves away from trench. This is best illustratedalong the VSP∼100 mm/yr line (Fig. 1b): decreasing VOP

back-arc shortening (Japan, South Kuriles) evolves intoback-arc spreading (Tonga, Izu-Bonin, Kermadec, Mari-anas and New Britain) while flat and forward deflectedslabs (Japan) become steeper (Tonga, Izu-Bonin, Kerma-dec), and, finally, overturned (Marianas, New Britain).However, the boundary between the back-arc extensionand the compressive domains is not easily defined fromthis diagram. In fact, it should also be a function ofsubducting plate velocity as some subduction zones withslow overriding plate retreat and high VSP exhibit back-arc shortening and a shallow dipping slab (Japan andSouth Kuriles).

Fig. 2 shows in more details the relationshipsbetween VOP, VSP and the style of subduction. In fact,trench motion, OPS and slab dip all correlate better withVOP than with VSP.

3. Plate kinematics, slab geometry and overridingplate deformation in laboratory models

3.1. Experimental setup and procedure

Following a previous experimental setup (e.g.,[10,12,18–22]), the subduction system is reproduced bya linear viscous layered system arranged in a Plexiglasbox. Silicone putty (Rhodrosil Gomme, PBDMS+galenafillers) and glucose syrup were selected to model thelong-term behavior of the lithosphere and upper mantle,respectively (Fig. 3a). Silicone putty is a visco-elasticmaterial that, at experimental strain rates, only behaves ina viscous manner. This was proven by the fact that theexperimental time-scale (order of min) is always higherthan the Maxwell relaxation time (order of s) [23]. The

Fig. 3. a) Experimental setup. The bottom of the box is impermeable. It mimics the 660 km discontinuity. b) Relations between the plate kinematics,the slab shape and the overriding plate deformation style of the performed experiments in a VSP/VOP space. Dot colour refers to the overriding plateshortening rate. In each model, the shape of the slab is drawn when a total amount of 25 cm of silicone plate is subducted.

477A. Heuret et al. / Earth and Planetary Science Letters 256 (2007) 473–483

upper mantle is simulated by glucose syrup, a Newtonianlow-viscosity and high-density fluid. The bottom of thebox mimics an impermeable 660 km discontinuity. Thesetup already adopted in previous studies is based onunavoidable experimental assumptions whose limitationsare extensively described in [12,18].

To model the mutual interaction between plates at aconvergent margin, the system is implemented withoverriding plates. The role of the overriding platemotion has not been tested in previous experimentalsettings. Both subducting and overriding plate velocitiesare controlled by the motion of a piston, so that externalimposed velocities superimpose on internal gravitationalvelocities. This setting mimics the forcing imposed by alarge plate motion on the dynamic of a single portion ofthe subducting system.

Except for the sides that are fixed to the pistons, weassume that each plate is completely surrounded byweak fault zones (trench and transform faults) whosecorresponding viscosity is that of the upper mantle.

Overriding and subducting plates are decoupled by athin layer of glucose syrup which prevents plate stickingeffects. Hence, we assume that the subduction fault isweak since it has the same viscosity as the upper mantle.Moreover, the surface of the oceanic plate is alwayslubricated using Vaseline to minimize possible frictioneffects at the trench. The idea of a weak zone betweenthe overriding plate and the subducting plate issupported by possible low viscosity wedge in subduc-tion zones (e.g., [24]).

Experimental parameters and the scaling relation-ships are listed in Table 2. In particular, a scale factor forlength is 1.6×10−7 (1 cm in the experiment correspondsto 60 km in nature), the scaling density factor betweenthe oceanic subducting lithosphere and the upper mantleis 1.05, and the viscosity ratio between the slab (ηo) andthe upper mantle (ηm) is fixed at 1.2×104, respectingthe range of natural density and viscosity contrasts [25–28]. Since 75% of the overriding plates in nature con-sist of the continental lithosphere [11], we model the

Table 2Parameter scaling in nature and laboratory models

Parameter Nature Reference model

Gravitational acceleration (g), m×s−2 9.81 9.81Thickness, (m)h, oceanic lithosphere 72000 0.012H, upper mantle 600000 0.10

Scale factor for length Lmodel /Lnature=1.6×10−7

Density, (kg×m−3)ρo, oceanic lithosphere 3300 1495ρc, continental lithosphere 3140 1323ρm, upper mantle 3220 1415

Density contrast (ρo−ρm) 80 80Density contrast (ρc−ρm) −80 −92Density contrast (ρo−ρc) 160 172Density ratio (ρo /ρm) 1.02 1.05Density ratio (ρc /ρm) 0.97 0.93Density ratio (ρc /ρo) 0.95 0.88

Viscosity, (Pa×s)ηo, oceanic lithosphere 1024 3.5×105

ηc, continental lithosphere 1024 1.6×105

ηm, upper mantle 1020 30Viscosity ratio (ηo /ηm) 104 1.8×104

Characteristic time, (s) (tnature / tmodel= (Δρgh)model / (Δρgh)nature× (ηnature /ηmodel)) 3.1×1013 (1 Myr) ∼60 (1 min)Characteristic velocity (Unature /Umodel)= (Lnature× tmodel) / (Lmodel× tnature) 10 (cm/yr) 1.6 (cm/min)

478 A. Heuret et al. / Earth and Planetary Science Letters 256 (2007) 473–483

experimental overriding plates by means of “continen-tal-like” silicone putty that is more viscous and lessdense than the one used to simulate the oceanic sub-ducting plate.

In the initial configuration, the leading edge of theoceanic plate is forced downward to a depth of about3 cm within the experimental mantle to start the sub-duction process. Afterwards, the two rigid pistonsmoving perpendicularly to the trench are used to driveboth the subducting and the overriding plate at variablerates. Considering the imposed scale ratio for length,gravity, viscosity and density (Table 2), an experimentalvelocity of 15 mm/min corresponds to 100 mm/yr innature. The majority of the experiments are thus scaledto fall inside the representative natural velocity field.

Each experiment is monitored using a sequence ofdigital pictures taken in the lateral and top views.Afterwards, the motion of the plates (VSP, VOP), theresulting trench motion VT, and the dip of the slab aremeasured using image-processing tools. VSP, VOP andVT are given in a box reference frame. A regular gridinitially drawn on the overriding plate allows mea-surement of the finite displacement field of each singlenode of the grid along the x- and y-axes at each timeinterval. The cumulative deformation of each elementof the grid during the duration of the experiment isthen calculated while integrating the finite displace-ment field over time.

3.2. Experimental results

A total number of 19 experiments have beenperformed using various combinations of imposed VSP

and VOP (Table 3) in order to explore the variability fieldof natural plate velocities (Fig. 1). All of the experimentsshow the same typical sequence of three phases asextensively described earlier in [10,12]: (phase 1)sinking of the slab through the upper mantle, (phase 2)transient slab/660 km discontinuity interaction, and(phase 3) steady state subduction with the slab lying onthe 660 km discontinuity (Fig. 4). The slab geometriesand the overriding plate deformation velocities observedduring the steady-state phase are represented for eachexperiment in a (VSP; VOP) diagram (Fig. 3b).

Fig. 3b shows that VOP is a key-parameter whichdetermines two different styles of subduction. Style Ioccurs when VOPN0. Under this condition, the trenchretreats overriding plate undergoes compression; theslab attains a shallow dip while it bends at depth in theforward direction. Style II occurs when VOPb0. Underthis condition, the trench advances, the overriding plateis poorly deformed or slightly shortened. The slabassumes a steep dip and gets overturned onto the bottomof the box.

Experiment 4 (Fig. 4a, Table 3, VOP=+20 mm/min)is representative of Style I. During phase 1, theoverriding plate moves toward the trench and forces

Table 3Applied subducting and overriding plate velocities, resulting plate convergence and deformation of the overriding plate in the models and comparisonto nature

Model Subducting plate velocity(VSP)

Overriding plate velocity(VOP)

Plate convergence rate(VPC)

Overriding plate deformationvelocity

Model(mm/min)

Nature(cm/yr)

Model(mm/min)

Nature(cm/yr)

Model(mm/min)

Nature(cm/yr)

Model(mm/min)

Nature(cm/yr)

1 0.0 0 10.0 6.25 10.0 6.25 −0.6 −0.372 0.0 0 44.2 27.6 44.2 27.6 −10.4 −6.53 10.0 6.25 9.8 6.1 19.8 12.4 −2.3 −1.44 10.0 6.25 20.3 12.7 30.3 18.9 −6.0 −3.755 35.2 22 −20.0 −12.5 15.2 9.5 −0.0 −0.06 20.8 13 10.0 6.25 30.8 19.25 −2.4 −1.57 21.1 13.2 −10.0 −6.25 11.1 6.9 −0.2 −0.18 44.4 27.7 −20.0 −12.5 24.4 15.25 −0.8 −0.59 34.4 21.5 −10.0 −6.25 24.4 15.25 −2.5 −1.610 44.4 27.7 −10.0 −6.25 34.4 21.5 −3.1 −1.911 13.3 8.3 −5.0 −3.1 8.3 5.2 −0.8 −0.512 0.0 0 5.0 3.1 5.0 3.1 −0.5 −0.313 5.0 3.1 0.0 0 5.0 3.1 −0.4 −0.2514 −10.0 −6.25 23.8 14.9 13.8 8.6 −5.2 −3.2515 20.0 12.5 0.0 0 20.0 12.5 −2.8 −1.7516 10.0 6.25 0.0 0 10.0 6.25 −1.1 −0.717 0.0 0 30.0 18.75 30.0 18.75 −3.0 −1.918 0.0 0 20.0 12.5 20.0 12.5 −2.9 −1.819 48.4 30.25 −32.0 −20.0 16.4 10.25 5.5 3.4

479A. Heuret et al. / Earth and Planetary Science Letters 256 (2007) 473–483

the overall slab to retreat at a rate close to VOP. The slabdip progressively decreases and reaches the lowerboundary at a low angle of about 30°. A general slow

Fig. 4. Styles of subduction as a function of VOP: a) VOPN0 /Style I (Model 4represented for each model. The dashed trajectory of the slab between each ofslab on a total set of about 15 pictures. Vertical lines above the plates repres

down of the process is recorded when the slab reachesthe bottom of the box and begins to fold along thebottom after 4′20″, phase 2. Afterwards, the system

); b) VOPb0 /Style II (Model 7). 3 selected steps of slab evolution arethe represented steps is drawn by following some reference points of theent the trench position for each step.

Fig. 5. Same as Fig. 2, but for laboratory models. Each motion is measured with an accuracy of about ±0.5 mm/yr which corresponds to the circlediameter. The accuracy of the slab dip measurement strongly depends on the subduction style. Style I geometries are almost constant with depth andthe dip of the slab is easily measurable (the error bars are ±5° in average). As they vary a lot with depth, Style II geometries imply lower accuracy inthe dip measurement (±10° in average).

480 A. Heuret et al. / Earth and Planetary Science Letters 256 (2007) 473–483

reaches steady state conditions (phase 3). During thisphase, only 75% of the imposed VOP is transmitted tothe trench (VT=14 mm/min). The difference betweentrench and overriding plate motions is accommodatedby overriding plate shortening of about 6 mm/y. At thesame time, the horizontal distance between the trenchand the tip of the slab progressively increases: the trenchretreats while the draped part of the slab slides along thelower boundary at about 10 mm/min. The resultingeffect is recorded by the slab which adopts shallowdipping geometries of about 25°.

Experiment 7 (Fig. 4b, Table 3, VOP=−10 mm/min) isrepresentative of Style II. During the entire experiment,trench motion is driven in the forward direction by theapplied VSP. However, VT is regulated by imposed over-riding plate retreat and ultimately remains close to VOP.During phase 1, the overall slab translates with the trenchand it becomes progressively steeper (slab dip is almost90° when phase 2 begins). After reaching the lowerboundary, the slab deflects backward and drapes on thelower boundary while the trench continues to advance.

The tip of the slab becomes almost stationary (Fig. 4b)allowing it to maintain a constant cylindrical shapethroughout the experiment (phase 3). The overriding platedeformation is negligible (less than 0.5 mm/min).

The sensitivity of subduction style to VOP is wellillustrated while fixing VSP and increasing VOP in Fig. 3b.Slabs become shallower, overriding plate shorteningbecomes higher and subduction style changes fromStyle II to Style I. On the contrary, for a given VOP,subduction style is not sensitive to variations of VSP orVPC.

Likewise, Fig. 5 clearly shows that trench velocity,overriding plate deformation and slab dip correlatesbetter with VOP than with VSP, as observed for naturalexamples (Fig. 2).

4. Discussion

Although oversimplified, the comparison betweenlaboratory models (Figs. 3 and 5) and natural cases(Figs. 1 and 2) can be used as a guide to interpret the main

481A. Heuret et al. / Earth and Planetary Science Letters 256 (2007) 473–483

relationships between the different subduction elements.In particular, we observed that VT, VSUB and VOP are alllinked to each other both in the laboratory and in natureand their mutual variations produce changes in slabgeometry and back-arc deformation. However, the keykinematic parameter appears to be VOP which depends onexternal factors. This is the case for the Philippine Seaplate, for example, which retreats at a high rate away fromthe Izu-Bonin–Mariana Trench because it is pulled byRyukyu subduction [29]. We discuss our results whichshow similarities and discrepancies between natural andexperimental kinematic relationships of the subductionsystem and two key parameters: back-arc deformation andslab geometry.

4.1. Back-arc deformation

In our models, the relationships between VOP and VTis rather obvious (Fig. 5; R=0.99): trench retreats if theoverriding plate moves toward the subducting plate(Style I; VOPN0) and advances if the overriding platemoves the other way (Style II; VOPb0). The mechanismat work, however, is different for the two cases: the trenchis forced to move by the overriding plate only whenVOPN0. If VOPb0, the advancing motion of the trench isdriven by the subducting plate. A similar mechanism isexpected to be efficient in natural subduction zones asthey show comparable VT /VOP relationships (Fig. 2;R=0.66). However, in both experimental and naturalcases, the trench generally moves slower than theoverriding plate. The mechanism at work implies thatresistance is opposed to lateral slab migration by bothlithosphere and mantle viscous resistance (“slab anchor-ing” [30]). The delay between trench motion and theoverriding plate motion generates back-arc deformation.In particular, the comparison between experimental andnatural data reveals a similar negative correlationbetween VOP and the way the overriding plate deforms(Figs. 2 and 5). Such correlation has been previouslynoted and discussed in several other studies (e.g.,[8,30,31]). In particular, Uyeda and Kanamori [30]proposed a distinction between “Chilean-type” and“Marianas-type” subduction zones which are compara-ble, at some extent, to the two different subduction styleswe observed in the models.

For Style I (trench retreat), the modelling slowdown ofthe trench and the resultant overriding plate deformationrepresents between 5% to 25% of the imposed VOP,respectively (Fig. 3). In nature this situation is wellillustrated by the Peru–Chile subduction zone where theoverriding plate (South America) moves toward theNazca subducting plate at a 45–50 mm/yr rate (Fig. 1).

The trench retreats more slowly and the back-arc shortensat about 10–20% of the VOP (Fig. 1).

For Style II (trench advance), back-arc extension hasbeen experimentally tested only at the highest VOP value(Fig. 3, Table 2; model 19; VOP=−32 mm/min). In thiscase, the extension rate represents about 18% of theimposed VOP. In natural subduction zones, this case isprobably best expressed by the Mariana subduction zonewhere the overriding plate (Philippine Sea Plate) movesaway from the Pacific subducting plate at a rate of about60mm/yr. The trench advancesmore slowly and the back-arc spreads at about 50–70% of the VOP (Fig. 1). Forintermediate/low VOP, experimental overriding platedeformation is also sensitive to VSP and upper plateshortening is always observed (Fig. 3). If VOP is keptconstant, the compressive stresses transmitted through theplate interface increase with increasing VSP (Fig. 3). Thispeculiar behavior is observed in nature only in Japan,where back-arc compression is associated with a slowlyretreating Amur overriding plate and a fast Pacificsubducting plate (Fig. 1).

A striking difference between the natural andlaboratory data is that the extensional process in naturealso occurs for low overriding plate retreating rates.Hence, the delay between trench motion and upper platemotion in nature is higher than that in the laboratoryenvironment. This evidence indicates that lateralmigration of the trench–slab system is easier in thelaboratory system than in the natural mantle because ofthe low resistance offered by the experimental lowviscous mantle. In comparison with previous laboratorymodels [18,19], we deduce that a lower experimentalslab/upper mantle viscosity ratio (102–103) would bemore appropriate to simulate the natural behavior of thesubduction system.

4.2. Slab geometry

Both natural and experimental subduction systemsreveal similar relationships betweenVOP and the dip of theslab obtained in the steady state phase (i.e. after the slab-660 km discontinuity interaction). If the overriding platemoves toward the trench (VOPN0), the slab is shallowdipping, while steep geometry is observed if the overridingplatemoves the otherway (VOPb0). This result has alreadybeen explored in several laboratory models (e.g.[12,22,32,33]) and 2D numerical simulations (e.g., [34–37]) which show that slab dip is inversely proportional totrench retreat velocity. In agreement with these previousstudies, the relationships between trench motion, upperplate motion and slab dip can be best explained by the factthat a slab is almost perfectly anchored at depth so that any

482 A. Heuret et al. / Earth and Planetary Science Letters 256 (2007) 473–483

trench motion has to be necessarily accommodated by achange in the slab geometry.

With respect to previous models, however, we havefound that overriding plate motion not only controls thedip of the slab but also controls its shape; in particular itaffects the direction of deflection on the bottom of thetank. Advancing and retreating overriding plates arealways associated with backward and forward reclinedslabs, respectively (Fig. 3). This is apparently not the casein current subduction zones, where backward reclinedsystems appear to be sparse (Fig. 1). This discrepancy canbe related either to experimental limitations or to naturalassumptions.

The geometry of the experimental slab depends onseveral parameters as the rheology simulating thesubduction system, the selected slab/upper mantleviscosity ratio, slab thickness or slab thermal parameter(e.g., [5,18,38,39]). Comparison with other models showthat the backward reclined configuration in an advancingsubduction system can be achieved only under a limitedrange of slab/mantle coupling and buoyancy conditions[10,18,19]. For example, a backward reclined configura-tion in an advancing system cannot be achieved if the slabis stiff [10] or if the slab/mantle viscosity ratio is lowerthan 103 [21]. Low slab/mantle viscosity contrasts innatural subduction systems might thus explain why suchoverturned slabs are so rare relatively to what is observedin our experiments (ηp /ηm=10

4).As far as the natural data set is concerned, at least two

explanations can be put forward. The first is related to thetime scale of the observations used in Fig. 1. While slabshape as inferred by Wadati–Benioff zones and tomo-graphic images traces out only a snapshot of the present-day slab geometry, the kinematic data are integrated overthe last 5.7 Myr [7]. In addition, the HS3 hotspot absolutereference frame we used to describe the natural platemotions may include a net rotation of the lithosphere thatis too large [7] as compared to other absolute referencemodels. Decreasing the net rotation value would indeedshift the origin of the axis of Fig. 1 towards the left–bottom corner and the horizontal zero line would becloser to the change in subduction style, as observed inFig. 3 for laboratory models.

In summary, our results clearly show that overridingplate deformation and slab geometry (dip and shape)are the consequences of “slab anchoring” [30]. Theresistance opposed to lateral slab migration by both lith-osphere and mantle viscous resistance produce an un-avoidable delay between trench/overriding plate motionand trench/slab motion accommodated by overridingplate deformations and by changes in the geometry of theslab, respectively.

5. Conclusions

Trench motion, overriding plate deformation and slabshape have been studied as the consequences of thekinematics of the subducting and overriding plates.Comparable trends have been found for both naturalsubduction and laboratory models in which the litho-sphere/upper mantle subduction system is modelled bymeans of two linearly viscous scaled layers of siliconeputty and glucose syrup. These relationships recognizeVOP as the key-parameter influencing subduction processbehavior: (1) trenches migrate at a rate close to theoverriding plate motion. Trenches always move slowerthan the overriding plates and the delay is accommodatedby back-arc deformation and slab geometry; (2) theoverriding plate shortens or extends if the overriding platemoves trenchward or oceanward, respectively; and (3)shallow dip and deep dip are found if the overriding platemoves trenchward and oceanward, respectively. Suchtrends make physical sense considering the resistanceopposed to lateral slab migration by both lithosphere andmantle viscous resistance.

Acknowledgments

This research was supported by the CNRS-INSUDyETI program “Dynamics of Subduction”. We thankour colleagues who participated in this program,especially Joseph Martinod and Nicolas Espurt, for thenumerous discussions and debates. Laboratory modelshave been performed in the “Laboratory of Experimen-tal Tectonics”Dip. Scienze Geologiche Univ. Roma Tre,Roma, Italy. We are grateful to Syral srl for providingthe glucose syrup used in our experimental models.Thanks to two anonymous reviewers for their useful andconstructive comments.

References

[1] P. Molnar, T. Atwater, Interarc spreading and cordillerantectonics as alternates related to the age of subducted oceaniclithosphere, Earth Planet. Sci. Lett. 41 (1978) 827–857.

[2] J.F. Dewey, Episodicity, sequence and style at convergent plateboundaries, in: D.W. Strangway (Ed.), The Continental Crust andIts Mineral Deposits, Geol. Ass. Canada Spec. Paper, vol. 120,1980, pp. 553–573.

[3] Z. Garfunkel, D.L. Anderson, G. Schubert, Mantle circulationand lateral migration of subducting slabs, J. Geophys. Res. 91(1986) 7205–7223.

[4] R.L. Carlson, P.J. Melia, Subduction hinge migration, Tectono-physics 102 (1984) 1–16.

[5] R.D. Jarrard, Relations among subduction parameters, Rev.Geophys. 242 (1986) 217–284.

[6] K. Otsuki, Empirical relationships among the convergence rate ofplates, rollback rate of trench axis and island–arc tectonics: «

483A. Heuret et al. / Earth and Planetary Science Letters 256 (2007) 473–483

laws of convergence rate of plates », Tectonophysics 159 (1989)73–94.

[7] A.E. Gripp, R.G. Gordon, Young tracks of hot spot and currentplate velocities, Geophys. J. Int. 150 (2002) 321–361.

[8] A.Heuret, S. Lallemand, Platemotions, slab dynamics and back-arcdeformation, Phys. Earth Planet. Inter. 1491–1492 (2005) 31–51.

[9] A. Heuret, Dynamique des zones de subduction: Etude statistiqueglobale et approche analogique, PhD, Université de MontpellierII, 2005.

[10] N. Bellahsen, C. Faccenna, F. Funiciello, Dynamics ofsubduction and plate motion in laboratory experiments: insightsinto the “plate tectonics” behavior of the Earth, J. Geophys. Res.110B01401, (2005), doi:10.1029/2004JB002999.

[11] S. Lallemand, A. Heuret, D. Boutelier, On the RelationshipsBetween Slab Dip, Back-Arc Stress, Upper Plate AbsoluteMotion and Crustal Nature in Subduction Zones, Gcubed, 2005,doi:10.1029/2005GC000917.

[12] F. Funiciello, C. Faccenna, D. Giardini, K. Regenauer-Lieb,Dynamics of retreating slabs (part 2): insights from 3D laboratoryexperiments, J. Geophys. Res. 108B4 (2003).

[13] R. Engdahl, R. van der Hilst, R. Buland, Global teleseismicearthquake relocation with improved travel times and proceduresfor depth determination, Bull. Seismol. Soc. Am. 88 (1998)722–743.

[14] A. Replumaz, H. Karason, R. van der Hilst, J. Besse, P.Tapponnier, 4-D evolution of SE Asia's mantle from geologicalreconstructions and seismic tomography, Earth Planet. Sci. Lett.221 (2004) 103–115.

[15] H. Bijwaard, Seismic travel-time tomography from detailedglobal mantle structure, PhD Thesis, Utrecht University, 1999.

[16] Y. Fukao, S. Widiyantoro, M. Obayashi, Stagnant slabs in theupper and lower mantle transition region, Rev. Geophys. 393(2001) 291–323.

[17] R. Hall, W. Spakman, Subducted slab beneath the easternIndonesia–Tonga region: insights from tomography, EarthPlanet. Sci. Lett. 201 (2002) 321–336.

[18] F. Funiciello, C. Faccenna, D. Giardini, Role of lateral mantleflow in the evolution of subduction system: insights from 3-Dlaboratory experiments, Geophys. J. Int. 157 (2004) 1393–1406.

[19] W.P. Schellart, Influence of the subducting plate velocity on thegeometry of the slab and migration of the subduction hinge, EarthPlanet. Sci. Lett. 2313–2314 (2005) 197–219.

[20] F. Funiciello, M. Moroni, C. Piromallo, C. Faccenna, A.Cenedese, H.A. Bui, Mapping the flow during retreatingsubduction: laboratory models analyzed by Feature Tracking,J. Geophys. Res. 111, (2006) B03402, doi:10.1029/2005JB003792.

[21] W.P. Schellart, Kinematics of subduction and subduction-induced flow in the upper mantle, J. Geophys. Res. 109B7(2004) B07401.

[22] W.P. Schellart, Quantifying the net slab pull force as a drivingmechanism for plate tectonics, Geophys. Res. Lett. 31L07611(2004).

[23] R. Weijermars, H. Schmeling, Scaling of newtonian and nonnewtonian fluid dynamics without inertia for quantitativemodelling of rock flow due to gravity (including the concept ofrheological similarity), Phys. Earth Planet. Inter. 43 (1986)316–330.

[24] M.I. Billen, M. Gurnis, A low viscosity wedge in subductionzones, Earth Planet. Sci. Lett. 193 (2001) 227–236.

[25] B.H. Hager, Subducted slab and the geoid: constraints on mantlerheology and flow, J. Geophys. Res. 89 (1984) 6003–6015.

[26] M. Cloos, Lithospheric buoyancy and collisional orogenesis —subduction of oceanic plateaus, continental margins, island arcs,spreading ridges, and seamounts, Geol. Soc. Am. Bull. 1056 (1993)715–737.

[27] G.F. Davies, M.A. Richards, Mantle convection, J. Geol. 1002(1992) 151–206.

[28] J.X. Mitrovica, A.M. Forte, Radial profile of mantle viscosity:results from the joint inversion of convection and postglacialrebound observables, J. Geophys. Res., Solid Earth 102B2(1997) 2751–2769.

[29] K.M. Pacanovsky, D.M. Davies, R.M. Richardson, D.D.Coblentz, Intraplate stresses and plate-driving forces in thePhilippine Sea Plate, J. Geophys. Res. 104 (1999) 1095–1110.

[30] S. Uyeda, H. Kanamori, Back-arc opening and the mode ofsubduction, J. Geophys. Res. 84 (1979) 1049–1061.

[31] C.P. Conrad, S. Bilek, C. Lithgow-Bertelloni, Great earthquakesand slab pull: interaction between seismic coupling and plate–slab coupling, Earth Planet. Sci. Lett. 218 (2004) 109–122.

[32] R.W. Griffiths, R.I. Hackney, R.D. van der Hilst, A laboratoryinvestigation of effects of trench migration on the descent ofsubducted slabs, Earth Planet. Sci. Lett. 1331–1332 (1995) 1–17.

[33] L. Guillou-Frottier, J. Buttles, P. Olson, Laboratory experimentson structure of subducted lithosphere, Earth Planet. Sci. Lett. 133(1995) 19–34.

[34] U.R. Christensen, The influence of trench migration on slabpenetration into the lower mantle, Earth Planet. Sci. Lett. 140(1996) 27–39.

[35] D. Olbertz, M.J.R. Wortel, U. Hansen, Trench migration andsubduction zone geometry, Geophys. Res. Lett. 243 (1997)221–224.

[36] M. Gurnis, B. Hager, Controls on the structure of subducted slab,Nature 335 (1988) 317–321.

[37] S. Zhong, M. Gurnis, Towards a realistic simulation of platemargins in mantle convection, Geophys. Res. Lett. 22 (1995)981–984.

[38] S. Karato, D. Yuen, M.R. Riedel, Rheological structure anddeformation of subducted slabs in the mantle transition zone:implications for mantle circulation and deep earthquakes, Phys.Earth Planet. Inter. 12711274 (2001) 83–108.

[39] H. Cizkova, J. van Hunen, A.P. van den Berg, N.J. Vlaar, Theinfluence of rheological weakening and yield stress on theinteraction of slabs with the 670 km discontinuity, Earth Planet.Sci. Lett. 199 (2002) 447–457.