Nature of the plate contact and subduction

zones diversity

Roberta De Franco a,∗, Rob Govers a, Rinus Wortel a

aDepartment of Earth Sciences, Utrecht University, Budapestlaan 4, 3508 TA

Utrecht, The Netherlands

Abstract

In recent studies we showed that the nature of the plate contact in subduction zones

is an important physical feature in both oceanic lithospheric subduction and con-

tinental collision. We investigated two fundamental states of the plate contact: one

based on a fault and the other based on a subduction channel. Using geodynamic

modeling, we determined the specific signatures of both states of the subduction con-

tact. In the present study, we combine results of our previous numerical experiments

with a re-analysis of published observations. Overall, our synthesis connects seismic

moment release with back-arc deformation and tectonic processes at the margin. It

leads us to identify four classes of subduction zones. The first two classes results

directly from our numerical experiments. In class 1, subduction zones are charac-

terized by a plate contact that is largely fault-like with an accretionary margin. In

class 2, the plate contacts are largely channel-type and have an erosive margin. Class

3, where the plate contact is entirely channel-like, consists of accretionary margins

with a high sediment supply. Subduction zones of class 4, mostly characterized by

an erosive convergent margin (northern Chili, Peru, Honshu and Kuril), are more

complicated. They can be explained by incorporating regional observations.

Key words: subduction zones; accretionary margins; erosive margins; large

Preprint submitted to Elsevier 7 April 2008

earthquakes; back-arc deformation

1 Introduction1

Earthquake intensity, back-arc state of stress, subduction velocity and sed-2

iment supply vary greatly from one subduction zone to the other. Under-3

standing the dynamics of the subduction process in view of these observed4

features represents a challenging problem of plate tectonics. In their classic5

study Uyeda & Kanamori (1979) group subduction zones into Marianas and6

Chili types. In the first case, the back-arc region shows extension, while in the7

second case compression prevails. Chilean type subduction zones are charac-8

terized by very strong thrust earthquakes, while Marianas type zones are not.9

Uyeda & Kanamori (1979) suggested that different subduction types result10

from different levels of coupling between the subducting and the overriding11

plate. The idea of Uyeda & Kanamori (1979) was confirmed by Conrad et al.12

(2004) who investigated the relation between back-arc stress and the magni-13

tude of the largest seismic events at each subduction zone and showed that14

there is a good correlation between these two observables.15

Convergent margins (i.e., the most shallow part of subduction contacts)16

can be roughly classified into accretionary margins and erosive margins (e. g.,17

Von Huene & Scholl, 1991, 1993; Clift & Vannucchi, 2004). Accretion transfers18

material from the subducting plate to the forearc wedge and plate boundary19

zone. It enlarges the accretionary prism and is accompanied by sediment sub-20

∗ Corresponding authorEmail address: defranco@geo.uu.nl, tel.: 0031 30 2535076, fax: 0031

30 2535030 (Roberta De Franco).

2

duction. The accretion rate depends on the balance between sediment influx21

and outflux through subduction and is favored at low subduction rate (Le Pi-22

chon et al., 1993). Subduction erosion is defined as the process that causes a23

net loss of material from the hanging wall. In this case the wedge volume is24

stationary or decreases with time (Lallemand et al., 1994). Lallemand et al.25

(1994) classified subduction erosion in two types, frontal and basal tectonic26

erosion. Frontal erosion takes place near the trench, and is thought to be due27

to the grabens or to ridges and seamounts. Basal tectonic erosion occurs at the28

base of the upper plate in a low friction environment, and might be due to a29

fast convergence rate that results in hydrofracturing by overpressuring. Such a30

first order classification into convergent and erosive margins neglects some of31

the observed variability of individual subduction plate contacts; a margin can32

evolve from one type into the other, or the plate contact may vary laterally.33

Accretion at the toe and erosion may occur simultaneously like, for example,34

in Japan and Peru (Von Huene & Lallemand, 1990).35

Trying to explain the dichotomy between the seismically active Chilean36

and the aseismic Marianas type subduction zones, Cloos & Shreve (1996) sug-37

gested that Chilean-type margins typically have thick trench fills and that38

the plate boundary zone thins arc-wards, while Marianas type margins have39

a thin or non-existent accretionary prism and the plate contact zone thickens40

with depth. They suggest that sea-mounts are subducted entirely to become41

seismogenic asperities in the Chilian type margin, while in the Marianas type42

just a truncated part of the sea-mount enters the subduction. This trunca-43

tion, in combination with the thick subduction channel, precludes subducted44

seamounts from becoming seismogenic asperities. Erosional margins are thus45

seismically quiescent, and accretionary margins exhibit high seismic moment46

release rates. Observational backup of this idea is good, although there exist47

3

some exceptions.48

Over the years several data sets have been systematically compiled and an-49

alyzed in order to understand the dynamics of subduction zones and to reveal50

the relation between different observables (e.g., Jarrard, 1986; Clift & Vannuc-51

chi, 2004; Lallemand et al., 2005; Sdrolias & Muller, 2006). Seismological stud-52

ies became more accurate, able to show new features at the subduction zones53

like the presence of inter-plate channel-like units of about 1-8 km thickness54

(Eberhart-Phillips & Martin, 1999; Oncken et al., 2003; Abers, 2005; Tsuru55

et al., 2002). Through inversion of arrival times from local earthquakes, these56

authors found an anomalous zone of low velocity at the interface between the57

plates. However, the arrival-time inversion can not resolve the thickness and58

the character of the layer independently. Investigation of the crustal fore-arc59

structures with wide-angle seismic data showed narrow, low velocity zones at60

the base of the fore-arc wedge, suggesting the existence of subduction chan-61

nels in Costa Rica, Makran, Nankai, Chile, Peru (Christeson et al., 1999; Kopp62

et al., 2000; Takahashi et al., 2003; Patzig et al., 2002). These data do not have63

the resolution to identify accurate subduction-channel geometries. Reflection64

seismic profiles from some of these fore-arcs give more accurate results; the65

disadvantage of these experiments is that they trace the subduction channel66

just to shallow depth (e.g., Von Huene et al., 2004; Ranero & von Huene,67

2000).68

The increasing evidence for variability of the plate contact motivated us to69

study its imprints on the response of subduction processes (De Franco et al.,70

2007). Through geodynamic numerical modeling we demonstrated that the71

overall plate contact nature has a dominant control on the state of stress of72

the overriding plate. In these models the entire plate contact is considered.73

We further investigated the role of the plate contact in the dynamic evolution74

4

of the subducting and overriding plate following the arrival of a terrane at75

the trench (De Franco et al., 2008). In the present study we shed new light76

on the geodynamics of subduction, based on the physical insights from our77

numerical experiments (De Franco et al., 2007, 2008) in combination with78

a re-analysis of published observations (Abers, 2005; Lallemand et al., 2005;79

Clift & Vannucchi, 2004). Our synthesis elucidates relations between back-arc80

state of stress, maximum seismic moment magnitude, and the nature of the81

all subduction plate contact, that affects the coupling level between overriding82

and subducting plate. Building on this, we identify four classes of subduction83

zones using observed back arc strain, nature of the convergent margin (erosive84

or accretionary) and maximum seismic moment magnitude. In doing this, we85

combine large scale features of subduction zones, following the approach of86

Uyeda & Kanamori (1979), with more regional observations as compiled by87

Cloos & Shreve (1996).88

2 Data and correlations89

The average dip angle of slabs at both shallow depths (depth range 0-12590

km) and great depths (depths greater than 125 km) is correlated with other91

characteristic parameters like the back-arc deformation and the nature of the92

plate contact. Recently, it was established that the magnitude of the low ve-93

locity anomaly at the top of the slab correlates with the average slab dip94

angle (Abers, 2005). Bodywave speeds decrease with increasing subduction an-95

gle at seven subduction zones: Aleutian (Ale), Alaska (Ala), Hokkaido-South96

Kurile (Hok-S.Kur), North Honshu (N. Hon), Marianas (Mar), Nicaragua97

(Nic), North Kurile-Kamchatka (N. Kur-Kam). Waveguide thicknesses vary98

5

between 2-8 km with large uncertainties. The velocity anomaly is as large as99

14 %, indicating that the waves within this layer propagate much more slowly100

than in the surrounding mantle. Between different subduction zones, the ve-101

locity anomaly varies by a factor of 2-3 down to 150 km depth.102

The low velocities may be explained by either metastable gabbro, that rep-103

resents an essential part of the oceanic crust, or low temperature hydrated104

mafic rocks that represent subduction channel-type rocks. This last hypoth-105

esis is plausible because slabs seem to dewater constantly during subduction106

(Gerya et al., 2002; Rupke et al., 2004). Quantitative experimental studies on107

rheology show that plastic deformation of upper mantle minerals is signifi-108

cantly intensified by the presence of water (e.g., Karato et al., 1986; Mei &109

Kohlstedt, 2000), suggesting that at higher water fugacity conditions weaken-110

ing effects may be quite pronounced. In this way it is possible to relate the low111

velocity to the rheology of the material. Numerical modeling experiments sup-112

port also a correlation between viscosity reduction in the subduction channel113

and increasing slab dip angle (Fig. 1 and Manea & Gurnis, 2007). Note how-114

ever, that the thickness of the subduction channel trades off with the seismic115

velocity anomaly: a thicker subduction channel with a relatively low velocity116

anomaly is analogous to a thinner channel with a higher velocity anomaly.117

Obviously we do not exclude that other factors than the plate contact may118

influence the slab dip angle. Other studies showed that trench motion, slab119

strength and evolution of the slab in time have a control on the deep slab dip120

(e.g., Cizkova et al., 2002; Billen & Hirth, 2007; Goes et al., 2008).121

In addition to the correlation between nature of the plate contact and dip122

angle, a strong correlation between back-arc deformation and subduction dip123

angle has been inferred (Lallemand et al., 2005). Back-arc spreading is ob-124

served for deep dip angles larger than 51◦, whereas back-arc shortening occurs125

6

for deep dip angles smaller than 31◦.126

When taken together, these correlations suggest that there is also a correla-127

tion between back-arc strain state and the rheological properties of the plate128

contact (see Fig. 1a and see also Chemenda et al. (2000)). In order to quan-129

tify this relation we calculate the correlation coefficient (Fig. 2) between the130

nature of the plate contact and the back-arc strain (Abers, 2005). The plate131

contact nature is parameterized by the velocity anomaly dlnVp = Vp−V0

V0, where132

Vp is the P wave velocity and V0 is the reference P wave velocity). Nicaragua is133

not included because the back-arc state of strain is not provided by Lallemand134

et al. (2005). The correlation coefficient for the data points shown in Figure135

2 a is R = −0.83. However, the seismic velocity anomalies have uncertainties136

(see Abers (2005)). Therefore we test the robustness of the correlation by sim-137

ply adding random noise (see the error bars) to the data points and evaluate138

the correlation coefficient for 1000 possible realizations. From the resulting139

distribution of correlation coefficients we compute the cumulative probabil-140

ity distribution (Fig. 2 b). 83 % of the 1000 correlation coefficients have a141

value lower than R = −0.8, indicating that the velocity jump is negatively142

correlated with the back arc state of strain (see Fig. 2 a). We conclude that143

the weaker the rheology of the subducting channel the more extensional the144

back-arc region.145

Our numerical experiments show a similar correlation (De Franco et al.,146

2007). In these geodynamic models, we linked the nature of the plate contact147

and the back-arc state of stress (tensional stress corresponds to extensional148

strain). The type of plate contact controls the dynamic response of the upper149

plate, drives the displacement of the overriding plate and, as a consequence, the150

stress distribution. We adopted two descriptions for the active plate contact: a151

subduction fault and a subduction channel, with the following characteristics:152

7

1) A deformable subduction fault is described via updated slippery nodes, in153

which the fault slip is locally kept parallel to the fault (Buiter et al., 2001).154

2) A subduction channel separates the subducting and the overriding plate.155

The channel width is assumed to be approximately 6 km (Shreve & Cloos,156

1986; Beaumont et al., 1999). Channel viscosity is taken to be Newtonian,157

ranging from 7·1017 Pa·s (Shreve & Cloos, 1986; England & Holland, 1979,158

red and yellow curves in Fig. 3) to 10 21 Pa·s (Renner et al., 2001; Stockhert,159

2002; Gerya et al., 2002, blue curve in Fig 3).160

In all the models the right-hand side of the subducting plate is free to move161

in horizontal direction, while the overriding plate is fixed on the left-hand side162

(see Fig. 3 bottom part). In Figure 3 we show the horizontal stress at the163

surface of five different models. A clear trend from high compressive to tensile164

back-arc stress is visible moving from a fault model (black curve) to a weak and165

wide subduction channel model (yellow curve). The first four models (black,166

grey, blue and red curves) are taken from De Franco et al. (2007) whereas the167

yellow curve represents a new model in which the thickness of the channel is168

increased (to about 20 km), while the viscosity is the same as in the model169

shown by the red curve (7·1017 Pa·s). The effect of the increased thickness is170

to obtain the back-arc tension. We conclude that increasing the thickness of171

the channel or further decreasing the viscosity in the channel will eventually172

lead to a tensile back-arc state of stress. In all these models a low viscosity173

wedge (LVW) is adopted above the subducting slab in order to reduce the174

down-warping, and as a consequence the magnitude of compressive stress, of175

the overriding plate in the arc/back-arc region (e.g., Billen & Gurnis, 2001,176

2003; Kelemen et al., 2004; Currie et al., 2004). The LVW is likely due to the177

presence of water released from the subducting slab or to melting processes178

at the base of the arc crust.179

8

Subsequently we investigate the relation between the nature of the plate180

contact and seismicity. Recently, Conrad et al. (2004) found that there is a181

positive correlation between back-arc state of strain and the maximum seis-182

mic moment release, meaning that strong earthquakes are characteristic for183

subduction zones with a compressive back-arc. This correlation, in combina-184

tion with the relation between back-arc deformation and the nature of the185

plate contact, suggests that the nature of the plate contact and the maxi-186

mum seismic moment are linked (see Fig. 1 b). In order to assess the validity187

of this relation, we analyze 7 subduction zones for which the channel veloc-188

ity anomaly is given by Abers (2005). We use the maximum seismic moment189

magnitude for each of these subduction zones between 1904 and 2007 taken190

from the USGS catalog (http://earthquake.usgs.gov/eqcenter). The resulting191

data points are shown in Figure 4a together with the best fitting line. The192

correlation coefficient is R = 0.92. Repeating the same procedure as before we193

compute the cumulative probability distribution of the correlation coefficient194

from 1000 noisy realizations. We find that 80 % of the 1000 correlation coeffi-195

cients have a value higher than R = 0.82 (Fig. 4b), indicating that the velocity196

anomaly is robustly correlated with the greatest moment release (see Fig. 4a).197

Note that these correlations display a basic trend that expose a relationship198

between two parameters. We do not expect perfect correlation because of the199

complexity of the subduction process and the small amount of data.200

In summary, compression in the back-arc region and strong earthquakes201

prevail for a high viscosity subduction channel, or for a very thin channel202

(represented by a fault) in which the body wave velocity anomaly is small.203

Lower compression or eventually tension and weak earthquakes prevail for a204

weak or wide channel with relative weak material fill, where the body wave205

9

velocity anomaly is more pronounced.206

3 A new subduction zone classification207

In a global view of subduction zones, another key discriminating physical fea-208

ture is the accretionary or erosive nature of the margin. Accretionary and209

erosive margins have characteristic shallow features which we aim to link with210

the deeper part of the plate contact. Cloos & Shreve (1996) suggested that211

in accretionary margins, characterized at shallow depth by thick trench fills,212

tall sea mounts are subducted entirely. In erosive margins, characterized at213

shallow depth by thin trench fills, only truncated sea-mounts are subducted.214

Our numerical models (De Franco et al., 2008) also shed light on the behav-215

ior of sea mounts upon subduction. A topographic feature (in the numerical216

model it is represented as a continental fragment) approaching the trench is217

entirely subducted in the presence of a wide and weak subduction channel218

(see Fig 5b and 5 d). In a fault type model, subduction of a steep topographic219

feature does not occur without previous delamination of the upper part of220

the incoming fragment (see Fig 5c and 5e and De Franco et al. (2008)). Our221

models confirm the idea of Cloos and Shreve (1996) if we equate the shallow222

subduction channel in the model to an accretionary margin, and if we inter-223

pret the shallow part of the model subduction fault as an erosive margin.224

According to Cloos & Shreve (1996), as we move deeper down along the225

plate contact, the shear zone becomes thinner in accretionary margins, while226

in erosive margins the shear zone thickens. Using the correlation between227

the back-arc state of stress and the nature of the plate contact found in the228

previous section, we infer that the back-arc state of stress is compressive in229

10

accretionary margins and tensional in erosive margins.230

To investigate the validity of such an interpretation we link the type of con-231

vergent margins and the back-arc state of strain of several subduction zones232

using the data of Clift & Vannucchi (2004) and Lallemand et al. (2005) (see233

Fig. 6a). We identify four different classes, in which the back-arc region is234

either in extension (T) or in compression (C), in combination with the margin235

being either accretionary (A) or erosive (E) (see Fig. 6). As expected, most236

of the accretionary margins are characterized by back-arc compression (class237

1-CA), while the erosive margins mostly have an extensive back-arc state of238

strain (class 2-TE). These two classes directly follow from our numerical re-239

sults and from the previously established correlation between back-arc state240

of strain and nature of the plate contact. We show cartoons of these two types241

of margins in Figure 7 a and Figure 7 b, respectively, in which the entire plate242

contact is shown.243

Not all the accretionary margins show the same behavior and there are244

some evident exceptions (see Fig. 6 a). For instance, Aegean arc, Makran and245

Barbados are characterized by back-arc extension (class 3-TA). Albeit less246

straightforwardly than for the first two classes, the results of the previous247

section and of De Franco et al. (2007, 2008) also account for the behavior248

of subduction zones of class 3-TA. Our results imply that a weak and wide249

subduction channel characterizes the entire plate contact. For the margins of250

class 3-TA, the fact that the sediment delivery rate and the material subduc-251

tion rate are higher than for the other margins (the sediment delivery rate is252

between 131 and 179 km3/yr and the subduction material rate is between 109253

and 150 km3/yr (Clift & Vannucchi, 2004)), explains the presence not only of254

an accretionary wedge at the inlet, but also of a thick sedimentary channel255

that decouples the two plates along the entire plate contact (Figure 7 c).256

11

Class 4-CE represents a more complicated case. In order to offer a possible257

explanation we need to supplement our results with additional information258

from various regional observational studies. The erosive margins of this class,259

contrary to the ones of class 2-TE, are characterized by very strong compres-260

sion (Fig. 6 a). As a consequence we would expect the deeper part of the plate261

contact to be described by a fault or, alternatively, by a channel with a small262

velocity anomaly (equivalent to a high viscosity channel). In the past, this263

kind of erosive margins has been envisaged as strongly coupled subduction264

zones where the erosion is due to high frictional abrasion. For instance, in265

the North Chile subduction zone, characterized by the downgoing plate cov-266

ered with less than 100 m of pelagic sediment, a sediment starved trench, and267

earthquakes of Mw = 8.0 was indicated as a typical example of such highly268

frictional margins. However, recent studies on North Chile have shown that269

some water is incorporated in the sediments at the plate contact (e.g., Sallares270

& Ranero, 2005). Despite sediment starvation, a frontal prism, constructed of271

debris, elevates pore pressure to reduce interplate friction. Therefore, processes272

other than high frictional abrasion are required to explain subduction erosion273

along northern Chile (e.g. hydrofracturing) and as a consequence a thick plate274

contact is expected. Therefore, the presence of strong back-arc compression275

instead of extension might be due to a change of the physical properties along276

the subduction channel: the material becoming stronger with depth due to de-277

watering, results in a strong coupling between the plates. Based on our model278

results (De Franco et al., 2007, 2008) and the correlation between back-arc279

state of strain and nature of the plate contact, we propose that the North280

Chile-type plate contact can be described by a subduction fault at shallow281

depth, followed deeper down by a subduction channel in which the viscosity282

increases with depth (Fig. 7 d).283

12

We propose another possibility that explains the anomalous behavior of284

class 4-CE. In Peru and North Japan, accretion and erosion are simultane-285

ously occurring (Von Huene & Lallemand, 1990). Accretion at the toe is active286

and erosion consumes the overriding plate generating subsidence further back.287

High friction erosional mechanisms generally proposed for these margins are288

not sufficient to explain the measured amount of eroded material (Von Huene289

& Lallemand, 1990). As a consequence, frontal erosion -due to the Nazca Ridge290

for Peru and the Daiichi Kashima seamount for Japan- and basal hydrofrac-291

turing are invoked. Subduction of a seamount or of a ridge is indicated as one292

of the reasons that can make an accretionary margin erosive. Since accretion293

is still active, part of the incoming sediments are used to create the accre-294

tionary wedge, decreasing the already low percentage of subducted sediments.295

We may speculate that in these cases the plate contact is characterized by296

a relatively wide inlet (accretionary margin) in which topographic features297

are subducted, followed at shallow depth by a thin plate contact where the298

coupling between the plates results in back-arc compression; deeper down, the299

plate contact enlarges again as a consequence of the sediment produced by300

basal erosion. Based on our numerical results (De Franco et al., 2007, 2008),301

we suggest that the Peru type of margin can best be described as changing302

with increasing depth from a subduction channel to a fault (that couples the303

plates), back to a subduction channel (Fig. 7 e).304

From the previous section, we know that strong earthquakes correlate with305

a thin or wide-and-strong plate contact and back-arc compression, whereas306

small earthquakes correlate with a wide-and-weak plate contact and back-arc307

extension. Therefore we expect to have strong earthquakes in classes 1-CA308

and 4-CE, and small earthquakes in classes 2-TE and 3-TA. In order to test309

this theory, we compile a list of the greatest seismic moment magnitudes for310

13

the considered subduction zones (Fig. 6 b). The results substantiate our pre-311

dictions: class 1-CA has 9.0 ≤ M ≤ 9.5 with the only exception of Japan ,312

class 2-TE has 7.2 ≤ M ≤ 8.1, class 3-TA has 7.1 ≤ M ≤ 8.0 and class 4-CE313

has 8.4 ≤ M ≤ 8.5.314

According to Cloos & Shreve (1996), class 1-CA is characterized by strong315

earthquakes, since the subducted seamounts become seismogenic asperities316

when they enter in contact with the upper plate. In class 2-TE the subducted317

truncated sea-mounts generate small earthquakes, since they do not touch the318

roof of the wide channel (see Fig. 7 a and b).319

In the North Chilean subduction zone (class 4-CE), Sallares & Ranero (2005)320

propose the release of elastic energy stored in the high viscosity channel as321

an explanation for the strong earthquakes. In the Peru/Japan type margin322

(class 4-CE), the strong earthquakes are possibly caused by the nucleation323

of seamounts where the plate contact becomes thinner, in a similar fashion324

than in the accretionary margins. This can even happen at very shallow depth325

resulting in shallow earthquakes that cause tsunamis. Without excluding the326

influence of large scale processes, in this study we highlight the importance327

of local features (rheological properties of the plate contact, geometry of the328

plate contact) in controlling the dynamic evolution of subduction zones.329

4 Conclusions330

We highlight the nature of the plate contact as an important physical feature331

of subduction zones by establishing correlations between the back-arc state of332

stress/strain, slab dip angle, and the maximum seismic moment. We identify333

four subduction zone classes. The first two classes (1-CA and 2-TE) directly334

14

follow from our numerical results and the established correlations in combi-335

nation with the observed nature of the convergent margin (accretionary or336

erosive). Although less straightforwardly than for the first two classes, our337

results also account for the behavior of subduction zones of class 3-TA. Class338

4-CE represents a more complicated case. The main characteristics of each339

class are listed below (denoting back-arc deformation as extensional (T) or340

compressional (C), and the nature of the margin as accretionary (A) or ero-341

sive (E)):342

Class 1-CA: The entire plate contact is well described by a wide channel at343

shallow depth and a fault at greater depth. This class contains accretionary344

margins mostly in back-arc compression with very strong earthquakes.345

Class 2-TE: The entire plate contact is a combination of a fault (at shallow346

depth) and a subduction channel (at deeper depth). This class contains erosive347

margins generally characterized by back-arc extension and small earthquakes.348

Class 3-TA: The entire plate contact is represented by a weak subduction349

channel. This class contains accretionary margins that exhibit back-arc ex-350

tension and small earthquakes. These margins are characterized by a high351

sediment subduction rate.352

Class 4-CE: We propose two explanations for the subduction zones in this353

class. One possibility is that the eroded material at the plate contact becomes354

stronger with depth, due to dewatering. In this case the plate contact is repre-355

sented by a fault at shallow depth followed by a subduction channel character-356

ized by an increasing viscosity with depth. Alternatively, in case accretion is357

active together with erosion (e.g., Peru and North Japan), we suggest that the358

plate contact can best be described as changing with increasing depth from a359

subduction channel to a fault (that couples the plates), back to a subduction360

15

channel. This class contains erosive margins that show back-arc compression361

and strong earthquakes.362

Acknowledgments363

Partial support for RDF and computational resources for this work were pro-364

vided by the Netherlands Research Center for Integrated Solid Earth Science365

(ISES). We thank an anonymous reviewer and Serge Lallemand for construc-366

tive comments on an earlier version of this manuscript.367

REFERENCES368

Abers, G. A., 2005. Seismic low-velocity layer at the top of subducting slabs: ob-369

servations, predictions, and systematics, Phys. Earth Planet. Inter., 149(1-2),370

7–29.371

Beaumont, C., Ellis, S., & Pfiffner, A., 1999. Dynamics of sediment subduction-372

accretion at convergent margins: Short-term modes, long-term deformation, and373

tectonic implications, J. Geophys. Res., 104(B8), 17573–17601.374

Billen, I. B. & Gurnis, M., 2001. A low viscosity wedge in subduction zones, Earth375

Planet. Sci. Lett., 193(1-2), 227–236.376

Billen, M. I. & Gurnis, M., 2003. Comparison of dynamic flow models for the Central377

Aleutian and Tonga-Kermadec subduction zones, Geochem. Geophys. Geosyst,378

4, 1035.379

Billen, M. I. & Hirth, G., 2007. Rheologic controls on slab dynamics, Geochem.380

Geophys. Geosyst., 8, Q08012, doi:10.1029/2007GC001597.381

Buiter, S., Govers, R., & Wortel, M. J. R., 2001. A modelling study of vertical surface382

displacements at convergent plate margins, Geophys. J. Int., 147, 415–427.383

16

Chemenda, A., Lallemand, S., & Bokun, A., 2000. Strain partitioning and interplate384

friction in oblique subduction zones: Constraints provided by experimental mod-385

eling, J. Geophys. Res., 105(B3), 5567–5581.386

Christeson, G. L., McIntosh, K. D., Shipley, T. H., Flueh, E. R., & Goedde, H., 1999.387

Structure of the Costa Rica convergent margin, offshore Nicoya Peninsula, J.388

Geophys. Res., 104(B11), 25443–25468.389

Cizkova, H., van Hunen, J., van den Berg, A. P., & Vlaar, N. J., 2002. The influence390

of rheological weakening and yieldstress on the interaction of slabs with the 670391

km discontinuity, Earth Planet. Sci. Lett., 199, 447–457.392

Clift, P. & Vannucchi, P., 2004. Controls on tectonic accretion versus erosion in393

subduction zones: Implications for the origin and recycling of the continental394

crust, Rev. Geophys., 42(2), RG2001.395

Cloos, M. & Shreve, R. L., 1996. Shear zone thickness and the seismicity of Chilean-396

and Mariana-type subduction zones, Geology, 24(2), 107–110.397

Conrad, C., Bilek, S., & Lithgow-Bertelloni, C., 2004. Great earthquakes and slab398

pull: interaction between seismic coupling and plate-slab coupling, Earth Planet.399

Sci. Lett., 218(1-2), 109–122.400

Currie, C. A., Wang, K., Hyndman, R. D., & He, J. H., 2004. The thermal effects401

of steady-state slab-driven mantle flow above a subducting plate: the cascadia402

subduction zone and backarc, Earth Planet. Sci. Lett., 223(1-2), 35–48.403

De Franco, R., Govers, R., & Wortel, R., 2007. Numerical comparison of different404

convergent plate contacts: subduction channel and subduction fault, Geophys.405

J. Int., 271(1), 435–450.406

De Franco, R., Govers, R., & Wortel, M. J. R., 2008. Dynamics of continental407

collision: Influence of the plate contact, Geophys. J. Int., submitted.408

Eberhart-Phillips, D. & Martin, R., 1999. Plate interface properties in the north-409

east Hikurangi subduction zone, New Zeland, from converted seismic waves,410

Geophys. Res. Lett., 26(16), 2565–2568.411

17

England, P. C. & Holland, T. J. B., 1979. Archimeded and Tauren eclogites: the412

role of buoyancy in the preservation of exotic eclogite blocks, Earth Planet. Sci.413

Lett., 44, 287–294.414

Gerya, T. V., Stockhert, B., & Perchuck, A. L., 2002. Exhumation of high-pressure415

metamorphic rocks in a subduction channel: a numerical simulation, Tectonics,416

21(6), 227–236.417

Goes, S., Capitanio, F., & Morra, G., 2008. Evidence of lower-mantle slab penetra-418

tion phases in plate motions, Nature, 451, 981–984.419

Jarrard, R. D., 1986. Relation among subduction parameters, Rev. Geophys., 24(2),420

217–284.421

Karato, S., Paterson, M. S., & Fitzgerald, J. D., 1986. Rheology of synthetic olivine422

aggregates: influence of grain size and water, J. Geophys. Res., 91, 8151–8176.423

Kelemen, P. B., Rilling, J. L., Parmentier, E., L., M., & Hacker, B. R., 2004. Thermal424

structure due to solid-state flow in the mantle wedge beneath arcs, in Inside425

the Subduction Factory, Geophysical Monograph, edited by J. M. Eiler, vol. 138,426

American Geophysical Union, Washington, DC.427

Kneller, E. A., van Keken, P. E., Karato, S., & Park, J., 2005. B-type olivine fabric428

in the mantle wedge: Insights from high-resolution non-Newtonian subduction429

zone models, Earth Planet. Sci. Lett., 237(3-4), 781–797.430

Kopp, C., Fruehn, J., Flueh, E. R., Reichert, C., Kukowski, N., Bialas, J., &431

Klaeschen, D., 2000. Structure of the Makran subduction zone from wide-angle432

and reflection seismic data, Tectonophysics, 329(1-4), 171–191.433

Lallemand, S., Schnurle, P., & Malavieille, J., 1994. Coulomb theory applied to434

accretionary and nonaccretionary wedges: Possible causes for tectonic erosion435

and/or frontal accretion, J. Geophys. Res., 99(B6), 12033–12055.436

Lallemand, S., Heuret, A., & Boutelier, D., 2005. On the relationships between437

slab dip, back-arc stress, upper plate absolute motion, and crustal nature in438

subduction zones, Geochem. Geophys. Geosyst, 6(9), Art. No. Q09006.439

18

Le Pichon, X., Henry, P., & Lallemand, S., 1993. Accretion and erosion in subduction440

zones - the role of fluids, Annu. Rev. Earth Planet. Sci., 21, 307–331.441

Manea, V. & Gurnis, M., 2007. Subductione zone evolution and low viscosity wedges442

and channels, Earth Planet. Sci. Lett., doi: 10.1016/j.epsl.2007.08.030.443

McCall, G. J. H., 1997. The geotectonic history of the Makran and adjacent areas444

of southern Iran, Journal of Asian Earth Sciences, 15(6), 517–531.445

Mei, S. & Kohlstedt, D. L., 2000. Influence of water on plastic deformation of olivine446

aggregates 1. diffusion creep regime, J. Geophys. Res., 105(B9), 21457–21469.447

Oncken, O., Sobolev, S., & Sitller, M., 2003. Seismic imaging of a convergent conti-448

nental margin and plateau in the central Andes (Andean Continental Research449

Project 1996 (ANCORP’ 96)), J. Geoph. Res., 108(B7), Art. No. 2328.450

Patzig, R., Shapiro, S., Asch, G., Giese, P., & Wigger, P., 2002. Seismogenic plane451

of the northern Andean Subduction Zone from aftershocks of the Antofagasta452

(Chile) 1995 earthquake, Geophys. Res. Lett., 29(8), Art. No. 1264.453

Ranero, C. R. & von Huene, R., 2000. Subduction erosion along the Middle America454

convergent margin, Nature, 404(6779), 748–752.455

Renner, J., Stockhert, B., Zerbian, A., K., R., & Rummel, F., 2001. An experimen-456

tal study into the rheology of synthetic polycrystalline coesite aggregates, J.457

Geophys. Res., 106, 19411–19429.458

Rupke, L. H., Morgan, J. P., Hort, M., & Connolly, J. A. D., 2004. Serpentine and459

the subduction zone water cycle, Earth Planet. Sci. Lett., 223(1-2), 17–34.460

Sallares, V. & Ranero, C. R., 2005. Structure and tectonics of the erosional conver-461

gent margin off Antofagasta, north Chile (23 degrees 30 ’ S), J. Geophys. Res.,462

110(B6), 113–136.463

Sdrolias, M. & Muller, R. D., 2006. Controls on back-arc basin formation, Geochem.464

Geophys. Geosyst, 7, Q04016.465

Shreve, R. L. & Cloos, M., 1986. Dynamics of sediment subduction, melange for-466

mation, and prism accretion, J. Geophys. Res., 91(B10), 10299–10245.467

19

Stockhert, B., 2002. Stress and deformation in subduction zones- insight from record468

of exhumed high pressure metamorphic rocks, Geol. Soc. spec. Publ., 200, 255–469

274.470

Takahashi, N., Kodaira, S., Park, J. O., & Diebold, J., 2003. Heterogeneous structure471

of western Nankai seismogenic zone deduced by multichannel reflection data and472

wide-angle seismic data, Tectonophysics, 364(3-4), 167–190.473

Tsuru, T., Park, J. O., Miura, S., Kodaira, Y., & Hayashi, T., 2002. Along-arc struc-474

tural variation of the plate boundary at the Japan Trench margin: Implication475

of intraplet coupling, J. Geophys. Res., 107(B12), Art. No 2357.476

Uyeda, S. & Kanamori, H., 1979. Back-arc opening and the mode of subduction, J.477

Geophys. Res, 84(B3), 1049–1061.478

Von Huene, R. & Lallemand, S., 1990. Tectonic erosion along the japan and peru479

convergent margins, Geol. Soc. Am. Bull., 102(6), 704–720.480

Von Huene, R. & Scholl, D. W., 1991. Observations at convergent margins concern-481

ing sediment subduction, subduction erosion, and the growth of continental-482

crust, Rev. Geophys., 29(3), 279–316.483

Von Huene, R. & Scholl, D. W., 1993. The return of sialic material to the mantle484

indicated by terrigeneous material subducted at convergent margins, Tectono-485

physics, 219(1-3), 163–175.486

Von Huene, R., Ranero, C. R., & Vannucchi, P., 2004. Generic model of subduction487

erosion, Geology, 32(10), 913–916.488

20

Figure Captions489

Fig. 1 a) Schematic representation of the relations between back-arc strain,490

slab dip angle and nature of the plate contact. b) Schematic representation491

of the relations between back-arc state of strain, earthquake magnitude and492

nature of the plate contact.493

Fig. 2 a) Best fitting line in the least square sense between the back-arc state494

of strain (Lallemand et al.,2005) and the relative velocity variation in the plate495

contact zone (dlnVp, Abers, 2005). b) Cumulative probability distribution of496

R computed from 1000 noisy data points, showing that 83 % of the 1000497

correlation coefficients have a value less than or equal to -0.8.498

Fig. 3 Horizontal stress at the free surface of five subduction models. Com-499

pression is negative. The solid black line represents frictionless fault model,500

the grey line a fault model with friction, the blue line a high viscosity channel501

model, the red line low viscosity channel model and the yellow line a low vis-502

cosity channel with a greater thickness (see text for a detailed explanation).503

Bottom: subduction geometry used in the numerical experiments and velocity504

field from a flow model corresponding to the frictionless fault model (corre-505

sponding to the black line in the top panel). The green region represents a506

low viscosity wedge. The plate contact reaches 100 km depth for both channel507

and fault models (red line). This is in agreement with Kneller et al. (2005).508

On the basis of heat flow data, seismic attenuation, and velocity tomography,509

they conclude that the plate boundary zone needs to extend to at least 70 km510

depth.511

Fig. 4 a) Best fitting line in the least square sense between the greatest seis-512

mic moment (M) and the relative velocity variation within the plate contact513

21

(dlnVp, Abers, 2005). The dlnVp value of Hok-S.Kur-N.Hon is an average be-514

tween the value given for Hokkaido-S. Kurile and N. Honshu. b) Cumulative515

probability distribution of R computed from 1000 noisy data points, showing516

that 80 % of the 1000 correlation coefficients have a value greater than or517

equal to 0.82.518

Fig. 5 a) General representation of a topographic feature approaching a sub-519

duction zone. The dashed rectangle is shown enlarged in b and c. b) Schematic520

representation of an approaching topographic feature in a subduction channel-521

type model: subduction of the entire continental crust takes place. c) Ap-522

proaching topographic feature in a subduction fault-type: truncation of part523

of the continental crust takes place. d) Effective strain rate distribution of a524

continental margin arriving at the trench in a subduction channel-type model525

(corresponding to cartoon b). e) Effective strain rate distribution of a conti-526

nental margin arriving at the trench in a subduction fault-type (corresponding527

to cartoon c). The black dashed line show the initial position of the slab. The528

grey region represents the LVW, in which no strain rate is shown.529

Fig. 6 a) Level of back-arc strain for different types of convergent margins.530

On the horizontal axis C means compression, 0 means neutral E means exten-531

sion. The scale from strong back-arc compression to strong back-arc spreading532

is according to Lallemand et al. (2005) C3, C2, C1, 0, E1, E2, E3, with the533

exception of the Aegean arc and Makran taken from Jarrard (1986) and Mc-534

Call (1997). The vertical axis classifies the margins in accretionary and erosive535

and does not include any scale. The letters in the class nomenclature mean:536

C=Compression, T=(ex)Tension, A=Accretion, E=Erosion. b) A list of sub-537

duction zones with their maximum seismic moment magnitude.538

22

Fig. 7 Summary of our subduction zone classification (subducting oceanic539

lithosphere on the lower left, and continental margin on the upper right).540

a) Accretionary margin with back-arc compression. b) Erosive margin with541

a back-arc in extension. c) Accretionary margin with back-arc extension. d)542

Erosive margin with back-arc compression and increasing viscosity in the chan-543

nel. e) Erosive margin with back-arc compression and active accretion process.544

Note that the scale of the deeper part of the model differs from the upper part.545

23

Fig. 1.

24

Fig. 2.

25

Fig. 3.

26

Fig. 4.

27

Fig. 5.

28

Fig. 6.

29

Fig. 7.

30