Geophysical Journal International Geophys. J. Int. (2011) 187, 1151–1174 doi: 10.1111/j.1365-246X.2011.05236.x

Switching between alternative responses of the lithosphere to continental collision

Marzieh Baes, Rob Govers and Rinus Wortel Faculty of Geosciences, Utrecht University, Utrecht, The Netherlands. E-mail: [email protected]

Accepted 2011 September 19. Received 2011 September 19; in original form 2011 January 2 Downloaded from https://academic.oup.com/gji/article/187/3/1151/612017 by guest on 29 September 2021 SUMMARY We study possible responses to arc–continent or continent–continent collision using numerical models. Our short-term integration models show that the initial stage of deformation following continental collision is governed by the competition between three potential weakness zones: (1) mantle wedge, (2) plate interface and (3) lower continental crust. Depending on which of these is the weakest zone in the system, three different responses can be recognized: (1) subduction polarity reversal, (2) continuation of subduction and (3) delamination and back stepping. Subduction polarity reversal occurs if the mantle wedge is the weakest zone in GJI Geodynamics and tectonics the system. This happens only if the viscosity of the mantle wedge is at least one order of magnitude lower than the average viscosity of the lithosphere. In continent–continent collision, one additional condition needs to be satisfied for subduction polarity reversal to occur: for the subducting lithosphere the ratio of the viscosity of the lower continental crust to the viscosity of the upper lithospheric mantle must be equal to or higher than 0.006. The time required for polarity reversal depends on several parameters: the convergence rate, the sinking velocity of the detached slab and the relative strength of the mantle wedge, arc and backarc. The response to collision is continued subduction if the plate interface is the weakest zone, and is delamination and back stepping if the lower continental crust is the weakest area in the system. Our finding that a low-viscosity wedge is a prerequisite for a reversal of subduction polarity agrees with inferences about regions for which subduction polarity reversal has been proposed. Key words: Subduction zone processes; Continental margins: convergent; High strain de- formation zones; Rheology: crust and lithosphere.

results in an oppositely dipping subduction zone, typically within 1 INTRODUCTION the arc or backarc. Subduction polarity reversal may presently be Arrival of continental lithosphere or buoyant oceanic plateaus (as occurring at the Wetar thrust belt in eastern Indonesia (Hamilton part of the subducting plate) at the trench of a convergent plate 1973; Curray et al. 1977). Along the Algerian margin, the presence boundary results in collisional tectonic settings. Understanding the of reverse faults has been proposed to be indicative of the earliest evolution of such settings is one of the great challenges in geody- stage of subduction polarity reversal (Deverchere et al. 2005). Sites namics. In exploring the possible scenarios for continuing conver- where a subduction polarity reversal may have occurred in the past gence following collision, not only the subducting plate but also include the San Cristobal trench in Solomon Islands (Cooper & the overriding plate should be considered. Referring to the part of Taylor 1985; Kroenke et al. 1986) and New Hebrides subduction the overriding plate adjacent to the trench as an arc, the nature of zone (Falvey 1975; Rodda & Kroenke 1984). the overriding plate behind an arc may be either (1) oceanic litho- Another possible response to continental collision is delamina- sphere, possibly an oceanic type backarc basin, or (2) continental tion. In this mechanism, the whole or part of the buoyant continental lithosphere. Previous studies have suggested different scenarios for crust separates from the rest of the lithosphere and is accreted to the lithospheric response following collision, including subduction the overriding plate (Bird 1978; Kerr & Mahoney 2007; De Franco polarity reversal and delamination. Subduction polarity reversal as a et al. 2008). Delamination has been proposed in several localities consequence of attempted continental subduction was first proposed including the Himalayas (Bird 1978; Mattauer 1986), the Aegean by McKenzie (1969). He suggested that the buoyancy of subducted region in Greece (Van Hinsbergen et al. 2005), the North American continental crust results in cessation of subduction. Subduction po- cordillera (Bird 1979; Ben-Avraham et al. 1981) and the collision larity reversal follows if continued convergence of the two plates zone of the North and South China blocks (Li 1994).

C 2011 The Authors 1151 Geophysical Journal International C 2011 RAS 1152 M. Baes, R. Govers and R. Wortel

We refer to arc–continent (or equivalently continent–arc) col- Another feature of subduction zones is the high temperature in lision for the collisional setting where the backarc is oceanic the shallow backarc mantle, further from the mantle wedge. The ob- lithosphere and to continent–continent collision when the backarc served surface heat flow and seismic velocities of the mantle reveal is continental. The two settings are jointly referred to as arc/ that high upper mantle temperature is not restricted to the arc but continent–continent collision or simply continental collision. We extends for several hundred kilometres across the backarc (Currie use numerical models to investigate the switches between different & Hyndman 2006). These observations indicate a thin lithosphere responses of the lithosphere to continental collision. Our particular (1200 ◦Cat∼60 km) over a backarc width of 250 km to more than focus is on subduction polarity reversal, which has received less 900 km. attention from a modelling point of view, compared to the other scenarios for collisional settings. Our study is composed of several parts. We first study the geological and geophysical evidence of 2.2 Sites of subduction polarity reversal subduction polarity reversal and of delamination. We then review previous studies of the mechanical response to collision and incip- 2.2.1 Wetar thrust ient subduction. Subsequently, we present our numerical models and their results. Finally, we end with a discussion of the results and Timor Island (eastern Indonesia) is the site of collision between the Downloaded from https://academic.oup.com/gji/article/187/3/1151/612017 by guest on 29 September 2021 their comparison with observations. Australian continental crust and the Banda arc. To the west, the Jurassic-aged Indian Ocean lithosphere is subducting northwards beneath the Sunda arc. GPS measurements reveal that most of the convergence between the Australian and Eurasian plates is now 2 OBSERVATIONS accommodated in the backarc along the Wetar thrust (e.g. Genrich et al. 1996; Kreemer & Holt 2000; Nugroho et al. 2009). 2.1 General observations: mantle wedge and backarc Prior to collision, the Indian Ocean had been subducting below characteristics of subduction zones the Banda arc since 15 Ma, when the Indo-Australian plate was Subduction carries a substantial amount of water into the Earth’s moving towards the Eurasian Plate at a rate of 7–8 cm yr−1 (Harris interior. This water is released by dehydration of the subducted 1991, and references therein). The continental crust arrived at the crust, serpentinized mantle and sediments. Hydration decreases the trench at about 3 Ma, leading to the formation of an arc–continent viscosity of the upper mantle (e.g. Chopra & Paterson 1984; Mei collision complex in the Timor region. Following collision, vol- & Kohlstedt 2000), and consequently changes wedge temperatures canic activity ceased in eastern Timor on the islands of Romang, (Van Keken 2003). This leads to a reduction of the viscosity of Wetar, Atauro and Alor. A south-dipping thrust, known as the We- the mantle wedge. Arcay et al. (2005) investigated the effect of tar thrust, was formed in the backarc at about 0.15 Ma (McCaffrey slab dehydration on the dynamics of the mantle wedge. In their 1996). Seismicity data shows high seismic activity and large earth- numerical experiments, the net convergence required to form a weak quakes in the backarc region (Silver et al. 1983; Kreemer & Holt mantle wedge varies from ∼550 to ∼900 km, which depends on 2000). Based on these geodetic, geologic and seismic observations, several parameters such as convergence rate and thermal structure. the deformation of the region can be explained by subduction po- Billen & Hirth (2005) indicated that a subduction rate of more than larity reversal, which was first proposed by Hamilton (1979). Using 2.5 cm yr−1 is required to prevent cooling of the low viscosity seismic data, McCaffrey et al. (1985) showed that continental crust mantle wedge. They also denoted that this subducting rate might has been subducted to a depth of about 150 km. They also proposed vary depending on parameters such as transition strain-rate, grain that the slab is detaching at a depth of ∼50–100 km in the east Savu size and water content. Sea. Observational constraints on the mantle wedge structure are heat flow, seismological data, composition of arc lavas, topogra- phy, gravity and geoid. The most direct observational constraint on 2.2.2 Reverse faults along the Algerian margin the thermal structure of the mantle wedge is surface heat flow. Surface heat flow generally decreases from the trench towards Algeria is located at the plate boundary between Africa and Eurasia. the arc, due to the cooling effect of the subducting slab. Further The Alpine orogen in this area formed as a consequence of closure into the fore-arc and into the backarc region, surface heat flow of the Ligurian ocean. Subduction of the Ligurian ocean beneath increases as a result of the mantle wedge flow (e.g. Wada et al. Iberia commenced about 30 Ma (Rosenbaum et al. 2002; Schettino 2008, and references therein). Seismological studies reveal a zone & Turco 2006), which was followed by southeastward retreat of the of low velocity and high attenuation (e.g. Hasegawa et al. 1991; trench at a rate of 3–4 cm yr−1 (Faccenna et al. 2001, and reference Barazangi & Isacks 1971) in the mantle wedge that is associated therein). Arc–continent collision along the north African (Alge- with the presence of volatiles such as water. Chemical analyses rian) margin occurred in the early-middle Miocene (∼15–18 Ma) of arc magmas provide constraints on the presence of volatiles in (Carminati et al. 1998). The slab broke off afterwards, as evidenced the mantle wedge. The arc-related rocks are characterized by the by a high velocity anomaly in the upper mantle in tomographic enrichment of highly mobile large ion lithophile elements relative images (Wortel & Spakman 2000). to high field strength elements, which demonstrates the presence Nowadays, Africa is converging with Eurasia at a rate of of slab-derived hydrous fluids in the mantle wedge (Ulmer 2001). 5mmyr−1 in the Western Mediterranean region (Fernandes et al. Geoid, gravity and topography are other observations for constrain- 2003; Serpelloni et al. 2007, and references therein). GPS obser- ing the mantle wedge structure. Billen & Gurnis (2001) showed vations show that about 3 mm yr−1 of this convergence rate is that the observed topography, gravity and geoid of the overriding accommodated offshore, off northern Algeria, where active reverse plate match well with model predictions only if a low viscosity faults dip towards the south (Stich et al. 2006). Deverchere et al. mantle wedge decouples the subducting plate from the overriding (2005) proposed that the presence of reverse faults dipping oppo- plate. sitely to the former subduction zone could be indicative of incipient

C 2011 The Authors, GJI, 187, 1151–1174 Geophysical Journal International C 2011 RAS Lithospheric responses to collision 1153 subduction. This would represent the earliest stage of subduction about a change in subduction polarity and the formation of a new polarity reversal in this region. subduction zone in the region.

2.3 Sites of delamination 2.2.3 San Cristobal trench in Solomon Islands The Solomon Island arc region as part of the Australia–Pacific plate 2.3.1 Himalayas boundary is one of the Cenozoic examples of subduction polarity Collision between the Indian and Eurasian plates as a result of clo- reversal. Seismological/seismicity data reveals the presence of dual, sure of the Neo-Tethys ocean started at about 40–50 Ma (Mattauer oppositely dipping Benioff zones beneath the Solomon Island arc 1986; Bilham et al. 1997). At least 6000 km of Neo-Tethyan (Cooper & Taylor 1985; Shinohara et al. 2003; Miura et al. 2004), ocean was subducted beneath Eurasia prior to collision (Hafken- which is consistent with a subduction polarity reversal scenario. scheid et al. 2006). Following collision, the rate of subduction de- Subduction of the Pacific Plate beneath the Australian Plate creased considerably from 10 cm yr−1 in the Upper Cretaceous to started in the Eocene (Wells 1989; Hathway 1993) when the 1–2 cm yr−1 today (Mattauer 1986; Bilham et al. 1997; DeCelles Downloaded from https://academic.oup.com/gji/article/187/3/1151/612017 by guest on 29 September 2021 convergence rate between the Pacific and Australian plates was & DeCelles 2001). Bird (1978) proposed that continued subduction ∼ −1 7.9 cm yr (Wells 1989). Following the arrival of the Ontong Java in the Himalayas was facilitated by peeling away of the subduct- plateau at the former North Solomon trench, reversal in subduction ing lithospheric mantle from the corresponding continental crust polarity occurred, resulting in formation of the San Cristobal trench. (delamination). Alternatively, Mattauer (1986) suggested that de- Although numerous studies (e.g. Kroenke 1989; Kroenke et al. formation in the Himalayas is governed by shear delamination of 1986; Yan & Kroenke 1993; Phinney et al. 2004; Mann & Taira the continental crust along the crust-mantle decollement, which 2004; Cowley et al. 2004; Schuth et al. 2009) agree that the Solomon leads to sliding of the lithospheric mantle below the crust along the arc underwent a reversal in subduction polarity, there are differences Moho. Stacking of thrust sheets in the central part of the Himalayas in the dating of this tectonic event. Regarding the timing, evolution- is indicative of shear delamination and continued subduction of the ary models are divided into two groups. The first group consists of lithospheric mantle (e.g. Johnson 2002). Parrish et al. (2006) found models in which the convergence between the Ontong Java plateau ultrahigh-pressure (UHP) metamorphic rocks in Northern Pakistan and the Solomon arc is proposed to have occurred in two stages with an average age of 45 Ma and pressure–temperature conditions of soft and hard docking (Kroenke et al. 1986; Kroenke 1989; Yan indicating that continental crust had been subducted to a depth of & Kroenke 1993). According to this group of models, the Ontong at least 100 km. Java plateau arrived at the trench at ∼22 Ma, which is referred to as soft docking. It was followed by detachment of the Pacific slab at ∼20–15 Ma. Subduction polarity reversal and initiation of 2.3.2 Aegean region in Greece northeast dipping subduction along the San Cristobal trench oc- curred at ∼12–6 Ma. Hard docking of the Ontong Java plateau, The stacked nappes in Greece resulted from collision of several which led to shortening and uplift in the Malaita area, occurred at microcontinental fragments during subduction of the African Plate about 5 Ma. The second group of models are those which propose beneath the Eurasian Plate since the Cretaceous. The average con- ∼ −1 that the Ontong Java plateau collided with the Solomon arc only at vergence rate since the Cretaceous was 2.1–2.5 cm yr (Van 5 Ma, resulting in slab break-off and formation of the San Cristobal Hinsbergen et al. 2005). Palaeogeographic reconstruction models trench (Cowley et al. 2004; Mann & Taira 2004; Miura et al. 2004; of the Aegean region (e.g. Stampfli & Borel 2004) show the presence Phinney et al. 2004). The authors of the second group of models of several oceanic basins (with an average width of 300–500 km) argue that the Solomon arc was an extensional intraoceanic arc prior separated by microcontinental plates. Van Hinsbergen et al. (2005) to 5 Ma and that there is no seismic evidence for the soft docking proposed that despite several episodes of collision, subduction con- of the Ontong Java plateau in the Early Miocene. tinued in the region. This could be possible only if delamination occurred following each phase of continental collision, resulting in the formation of a wedge of stacked nappes in the Aegean region. The slab imaged by tomography indicates continued subduction 2.2.4 New Hebrides of about 2100–2400 km of lithospheric mantle and lower crust Prior to 10 Ma, the Pacific Plate was subducting beneath the Aus- since the Cretaceous. Evidence for UHP metamorphic units in the tralian Plate at a rate of ∼8cmyr−1 (Wells 1989). This subduction Rhodope suggests subduction of continental crust to a depth of at zone had been initiated in the Eocene (Hathway 1993). Subduction least 70–80 km (Liati et al. 2002). polarity reversal occurred in the region at ∼10 Ma, resulting in subduction of the Australian Plate beneath the Pacific Plate along 2.3.3 North American Cordillera the New Hebrides trench. Rodda & Kroenke (1984) suggested that geological evidence of folding, uplift and erosion in the Fiji Is- The concept of terrane accretion originates from geological evi- lands at ∼10 Ma is associated with initiation of a new subduction dence in North America as the North American Cordillera was rec- zone in the New Hebrides at that time. Chen & Brudzinski (2001) ognized as the product of accretion of several terranes such as vol- indicated that seismological/seismicity data reveal a subhorizontal canic arcs, oceanic plateaus and seamounts to the continental plate remnant of slab beneath the Fiji basin, which was interpreted by the since the Mesozoic (e.g. Coney et al. 1980; Hoffman 1988). Most of authors as the detached slab of former southwest dipping subduc- the terranes, including the terrane of central British Columbia, ter- tion at the Vitiaz trench. Falvey (1975) suggested that subduction ranes in the Klamath Mountains and Sierra Nevada, were accreted polarity reversal was caused by the collision of the Melanesian Bor- to the North American continent in the Middle Jurassic–Early Cre- der plateau with the Vitiaz trench. Pysklywec et al.(2003) proposed taceous (Umhoefer 2003). There is evidence of imbricated thrust that an avalanche of slab material into the lower mantle brought sheets in several localities in North America including southern

C 2011 The Authors, GJI, 187, 1151–1174 Geophysical Journal International C 2011 RAS 1154 M. Baes, R. Govers and R. Wortel

Alaska, British Columbia and southern California (Ben-Avraham of this configuration is the occurrence of reverse faults along the et al. 1981, and references therein) indicating delamination of the Algerian margin) and subduction polarity reversal occurring while crust from the lithospheric mantle following entrance of terranes the slab is still attached to the surface (Fig. 1B; the Wetar thrust is into the subduction zone. Despite agreement on the key role of a possible example of this mode). Fig. 1(A1) shows the situation subduction in terrane accretion in North America, it is still poorly prior to the arrival of continental crust at the trench. When buoyant known how many subduction zones were involved in the terrane continental crust arrives at the trench (Fig. 1A2), a considerable accretion process and how delamination and accretion occurred in amount of continental crust subducts despite its resistance to sub- the region. Trop & Ridgway (2007) proposed that terrane accre- duction (as it is evidenced by the presence of (ultra) high pressure tion in southern Alaska is the result of two main collisional events metamorphic rock in many collisional settings). Here, we assume along two subduction zones: collision of the Wrangellia terrane in that the first response to continental collision is slab detachment. the Mesozoic and collision of the Yakutat terrane in the Cenozoic. Fig. 1(A3) shows the last phase of this scenario, where continued Nokleberg et al. (2000) suggested that terranes in the North America convergence between two plates results in failure of the overriding Cordillera were accreted on the continental plate during collisional plate and formation of a new subduction zone dipping oppositely events along several paired subduction zones. to the ceased one. The scenario in Fig. 1(B) is similar to that of Fig. 1(A), except that here polarity reversal occurs without the slab Downloaded from https://academic.oup.com/gji/article/187/3/1151/612017 by guest on 29 September 2021 having been detached. 2.3.4 Collision zone of the North and South China blocks Another response to collision is delamination and back stepping The Qinling orogenic belt was formed in the collision zone of the (Fig. 1C). In this scenario, resistance to subduction of continental North and South China blocks. The North China block was separated crust causes delamination of subducted continental crust from the from the South China block by the Palaeo-Tethyan Qinling ocean rest of the subducting lithosphere and formation of a new plate during Carboniferous-Permian (Meng & Zhang 1999; Dong et al. boundary near the former one. This scenario is similar to that in 2011). Closure of Palaeo-Tethyan Qinling ocean by subduction led Chemenda et al. (1996) and Li (1994). We note that delamination to the collision of two blocks in the Middle-Late Triassic. Delami- in this study is different from the one proposed by Bird (1979) in nation and thrust accretion has been proposed for the deformation which the whole lithospheric mantle peels away from the crust. mechanism following continental collision (Li 1994; Kusky et al. The last scenario, shown in Fig. 1(D), is inspired by ample evi- 2007; Zheng et al. 2010). During the delamination process, the up- dence of subduction of continental crust to depths of greater than per crust of the South China Block was accreted to the North China 150 km: UHP minerals such as coesite and diamond in metamor- Block whereas the lower crust and lithospheric mantle of the South phic rocks in collision zones (Ye et al. 2000, and references therein). China Block continued to subduct. Presence of UHP metamorphic Continued subduction must eventually lead to one of the other re- rocks in the orogen in eastern China points to the subduction of sponses (Figs 1A, B or C). Since, in this study, we investigate the continental materials to great depths (Li 1994). deformation pattern during the early stages of collision we include this (unstable) configuration in the list of our hypothesized scenar- ios. 2.4 Common characteristics of observations We speculate that the proposed scenarios in Fig. 1 are controlled by the competition between three weakness zones (in the sense of First of all, in all regions considered convergence continues (or con- most highly deforming zones) in the system as candidate zones to tinued) after collision. Secondly, in almost all the aforementioned accommodate continuing convergence: the mantle wedge (Figs 1A (possible) localities of subduction polarity reversal the former sub- and B), the incoming continental lower crust (Fig. 1C) and the plate duction zone was active for a long time in the region, indicating that interface (Fig. 1D). In the following we use numerical modelling to the subduction zone was mature. Thirdly, a general characteristic test this speculation. We investigate the effect of different factors on −1 is that the subduction rate was more than 3 cm yr before colli- the response to continental collision. Our reference model simulates sion and polarity reversal. As mentioned in Section 2.1, slab length the configuration shown in Fig. 1(A). We study the other scenarios (Arcay et al. 2005) and subduction rate (Billen & Hirth 2005) can be by changing modelling parameters with respect to the reference taken as a proxy for the reduction in wedge viscosity. We thus infer model. that the mantle wedge was weak in regions where polarity reversal occurred. We further expand on this issue in Section 7. In all sites of delamination, there is evidence for crustal stacking near the plate boundary. The other common feature of these local- 3.2 Previous modelling studies ities is the presence of (ultra) high-pressure rocks which indicates Subduction polarity reversal has been the subject of few modelling subduction of continental crust to great depth. studies. Chemenda et al. (1997) suggested that if the average arc- trench distance exceeds a critical value (which depends on the flex- ural bending of the arc/backarc lithosphere), the overriding plate 3 POSSIBLE RESPONSES TO fails along a plane dipping towards the (former) trench that leads to CONTINENTAL COLLISION subduction polarity reversal. Tang & Chemenda (2000) proposed that the direction of failure in the arc depends on the competition 3.1 Scenarios between two opposing torques which are induced by forces along Inspired by observations, we envisage three different responses to the interplate surface. They claimed that subduction polarity rever- continental collision in a setting of continuing convergence: sub- sal occurs if the torque due to the flexural rigidity of the subducting duction polarity reversal, delamination and back-stepping and con- plate is dominant. Chemenda et al. (2001a,b) extended the work tinuation of subduction (Fig. 1). For subduction polarity reversal, of Tang & Chemenda (2000)) and argued that failure occurs in we propose two configurations: change of subduction polarity oc- the backarc region if a spreading centre is present in the backarc. curring after slab detachment (Fig. 1A; a possible natural example They suggested that the direction of failure in the backarc depends

C 2011 The Authors, GJI, 187, 1151–1174 Geophysical Journal International C 2011 RAS Lithospheric responses to collision 1155 Downloaded from https://academic.oup.com/gji/article/187/3/1151/612017 by guest on 29 September 2021

Figure 1. Schematic illustration of the hypothesized scenarios for arc/continent–continent collision. The horizontal white arrows indicate the (continuing) convergence between two plates and the black arrows show the direction of motion within the plates. The white arrow in the mantle shows the direction of flow in the mantle. The hatched areas show the mantle wedge which may be weakened by the subduction process. (A) As buoyant continental crust arrives at the trench (1), the interaction between the slab pull and resistance of continental crust to subduction leads to slab break-off (2). Continued convergence between two plates results in formation of a new subduction zone on the former overriding plate (subduction polarity reversal; 3). (B) This is similar to the one proposed in (A), except that here we assume that a new plate boundary forms while the slab is (still) continuous. (C) Following continental collision, continued convergence between two plates results in delamination of the crust from the lithospheric mantle (of the subducting plate) and back-stepping. (D) In this scenario, subduction of the continental lithosphere continues. only on the trench/backarc spreading centre distance. Regard et al. is controlled by the internal heat production rate of the crust and (2008) indicated that slab break-off following the entrance of buoy- sediments. ant continental crust in the trench causes a decrease in subduction Some modelling studies have been carried out to investigate de- rate and eventually a change in subduction polarity. Faccenda et al. lamination following continental collision. Van den Beukel (1992) (2008) investigated the effect of convergence rate, crustal rheology indicated that continental crust can be subducted to mantle depths and radiogenic heat production on the style of post-subduction col- (uptoadepthof∼70 km), depending on the thickness and lision orogeny. They concluded that the timing of polarity reversal the thermal and compositional structure of the continental crust.

C 2011 The Authors, GJI, 187, 1151–1174 Geophysical Journal International C 2011 RAS 1156 M. Baes, R. Govers and R. Wortel

Table 1. List of numerical experiments. Class of properties Code Model properties First configuration in Fig. 1 M1 Reference model Second configuration in Fig. 1 M2 Attached slab to the surface Strength of mantle wedge M3 Stronger mantle wedge (with viscosity of 1 × 1023 Pa s) Subducting continental crust characteristics M4a Stronger subducting crust M5 Stronger subducting crust + Stronger mantle wedge M6a Weaker subducting crust M7a Higher density for the subducting continental crust (with density of 2900 kg m−3) Strength of arc M8a Stronger crust for arc M9a Stronger crust for arc + Stronger mantel wedge (with viscosity of 1 × 1023 Pa s) M10a Stronger crust for arc + Stronger mantel wedge (with viscosity 23 of 1 × 10 Pa s) + Stronger subducting crust Downloaded from https://academic.oup.com/gji/article/187/3/1151/612017 by guest on 29 September 2021 Channel characteristics M11a Wider channel (with channel width of 16 km) M12a Steeper channel dip angle (with a dip angle of 55◦) M13 Steeper channel dip angle (with a dip angle of 55◦) + Shorter trench/backarc distance Strength of backarc M14a Younger oceanic lithosphere in the backarc (with an age of ∼25 Myr) M15 Continental backarc M16 Continental backarc + Weaker subducting crust Convergence rate M17 Slower convergence rate (with a convergence rate of 1 cm yr−1) Detached slab characteristics M18a Longer slab (which is stagnant in the transition zone at the depth of 660 km) M19 Higher sinking slab velocity (with a velocity of 3 cm yr−1) aIndicates the models which are presented in Appendix A.

He indicated that subduction of continental crust leads to the de- collision. We solve the momentum equation to obtain stresses and tachment of the upper crust or the whole crust from the deeper velocities. In this paper, we use a short-term (up to a few millions part of the subducting plate, due to the resistance to subduction and of years) integration for our models and show that the model re- the low strength of the continental crust. Chemenda et al. (1996) sults are representative of the initial stage of deformation following showed that continental crust delaminates from the lithospheric continental collision. As the models focus on a short timescale rel- mantle only if the interplate pressure between subducting and over- ative to that of thermal diffusion (the Peclet´ number is greater than riding plate, which controls the amount of coupling between plates, 20), heat conduction can be ignored in our models. We do account, is high. Morency & Doin (2004) indicated that delamination is ini- however, for advection of heat. To facilitate referring to the models tiated where the Moho temperature is the highest. They suggested later in this paper, we define a reference model: model M1. We then that in nature delamination is possible only if the Moho temperature present several experiments in which various model characteristics exceeds ∼800 ◦C. De Franco et al. (2008) noted that the required are changed, as listed in Table 1. Fig. 2 shows the model geometry, conditions for delamination are: a narrow plate contact zone, a steep boundary conditions and initial temperature field for the reference slope angle and a low crustal strength for the incoming continental model. The model has a width of 1200 km and a depth of 660 km. In margin. the reference model we evaluate the geodynamic evolution of a con- The numerical models of Chemenda et al. (2001a) and Tang vergent plate boundary zone where the slab has broken off recently & Chemenda (2000) are based on elastoplastic models in which along the subducted passive margin (Fig. 1A). As a consequence, the asthenosphere and subducting plate are incorporated implicitly, a 35 km-thick continental crust initially sits below the subduction that is, in these models the forces associated with the astheno- channel. At the start of our model calculation, the detached slab sphere and subducting plate were imposed as boundary conditions. (70 km thick and 600 km long) has sunk only 30 km below the sub- Chemenda et al. (1996), Morency & Doin (2004), De Franco et al. ducted margin which is probably enough to mechanically decouple (2008) and Faccenda et al. (2008) studied the deformation pat- the subducted margin and the slab by necking. We assume that plate tern resulting from continent–continent collision. In this study, we convergence subsequently continues. aim to investigate the possible responses to both arc–continent and The oceanic lithosphere in the backarc (250 km away from the continent–continent collision using visco–elastic–plastic models. trench) has a 7 km-thick crust and a 53 km-thick lithospheric man- We investigate how variations in model parameters such as strength tle, which corresponds to an oceanic lithosphere with an age of of mantle wedge, sinking velocity of the detached slab, convergence ∼40 Myr. Inclusion of a thin lithosphere in the backarc stems from rate, dip of the channel, trench/backarc distance and age of oceanic studies of the thermal structure of backarcs (Currie & Hyndman plate in the backarc lead to different responses to continental colli- 2006) which denote that high upper mantle temperatures usually sion. extend for several hundred kilometres across the backarc, regard- less of whether the backarc has been subject to extension or not. The arc itself is assumed to be continental; it has a continental crust of 4 MODEL SETUP 35 km-thick which is underlain by a weak mantle wedge. The man- We use the GTECTON finite element code (Govers & Wortel 1993) tle wedge lies above the subducting plate and beneath the arc, and to study the deformation patterns during early stages of continental extends 200 km farther beneath the backarc. The subducting plate

C 2011 The Authors, GJI, 187, 1151–1174 Geophysical Journal International C 2011 RAS Lithospheric responses to collision 1157 Downloaded from https://academic.oup.com/gji/article/187/3/1151/612017 by guest on 29 September 2021

Figure 2. Geometry, boundary conditions and initial temperature of model M1. A horizontal velocity of 1 cm yr−1 (convergence rate: 2 cm yr−1) is imposed on both lithospheric side boundaries and outflow boundary conditions are incorporated below the lithosphere on both sides to balance (lithospheric) inflow. A vertical velocity of 1.5 cm yr−1 is imposed at the top and the bottom of the detached slab. The top model boundary is a free surface and the model bottom is fixed. Colours represent initial model temperatures. Since the viscosity of the detached slab in the model is set to be a Newtonian viscosity of 1 × 1023 Pa s, we do not display its temperature in the figure. is composed of a 35 km-thick crust and a 75 km-thick lithospheric mantle below the lithosphere changes adiabatically with a gradient mantle. The base of lithosphere corresponds to the depth of the of 0.4◦ Kkm−1 (Stacey 1977). 1300 ◦C geotherm (Parsons & Sclater 1977), lying approximately The driving forces in the model are the gravitational forces as- at a depth of 110 km. sociated with density variations and the force corresponding to the The plate boundary is modelled as a channel with a dip angle of imposed convergence rate. We use the method introduced by Govers 30◦ and width of 8 km. Inclusion of a channel (instead of a fault) & Wortel (2005) for the implementation of gravity in the model. in the model is based on the results of De Franco et al. (2008) In this method, densities are separated into two components: (1) who showed that subduction of incoming continental crust (which 1-D reference density field, which does not contribute to the initial is assumed in our initiation model geometry) is possible only if the forcing and is applied as a hydrostatic pressure, and (2) density plate contact is a channel. Wethink that a channel along a subduction anomalies which drive model deformation. The sinking behaviour interface evolves due to a combination of strain localization and the of the detached slab will generally be complicated. For our first material flux from the trench and the overriding plate. In our models, assessment of the influence of the detached slab on the evolution we do not track this evolution and we choose a channel width and of the surface lithospheres, we simply impose a sinking velocity. a Newtonian viscosity. We consider this model subduction channel We impose a vertical sinking velocity of 1.5 cm yr−1, close to the to represent the steady state (end) result of these processes. range of ∼2–5 cm yr−1 found by Hafkenscheid et al. (2006) and We impose a velocity of 1 cm yr−1 to the lithosphere on both sides Van Hunen & Allen (2011). We investigate the influence of a higher (left and right, in opposite directions) of the model. To preserve sinking velocity on the results in model M19. mass balance, inflow at the lithospheric level is compensated by an The model has a visco–elastic–plastic rheology in which the equivalent outflow in the asthenosphere; we impose outflow to the domains of elastic, viscous and plastic behaviour are determined by side boundaries below the lithosphere to keep the net inflow and the state of stress, strain rate and temperature. The total strain rate outflow to zero. The top boundary is a Lagrangian free surface and is the sum of the elastic, viscous and visco–plastic strain rates. The the model bottom is fixed in both directions, simulating the presence viscosity, η,isgivenby of higher viscosity materials in the lower mantle (Mitrovica & Forte   1 Q + PV − − 1997). η = exp σ (n 1), (1) The initial temperature field in the subducting plate away from A RT E the trench is based on a steady-state geotherm with a surface heat − σ flow of 65 mWm 2, which is the average heat flow in continents. where n, A, Q, P, V, R, T and E the are power exponent, Near the trench the temperature in the subducting plate increases pre-exponential coefficient, activation energy, pressure, activation linearly with depth into the plate. The temperature in the oceanic volume, universal gas constant, temperature and effective stress lithosphere in the backarc is calculated based on the cooling half- σ = 1 σ σ ( E 2 ´ij ´ij), respectively. Our sign convention for stresses is space model for a lithospheric age of 40 Myr. The temperature in the that tension is positive and compression is negative.

C 2011 The Authors, GJI, 187, 1151–1174 Geophysical Journal International C 2011 RAS 1158 M. Baes, R. Govers and R. Wortel

Table 2. Physical parameters for the reference model. in size from 1.5 km in the central part of the model to 26 km near Description Model value the model lower, left and right boundaries. Time step size (100 yr) is determined by the minimum viscosity in our models. Since the × 11 Young’s modulus (E)110 Pa mesh becomes distorted during numerical calculation, we regrid ν Poisson’s ratio ( )0.3the mesh every 10 kyr. We use the remeshing procedure of De Pre-exponential constant (A)forthe 3.3 × 1021 Pa−n s−1 Franco (2008) and De Franco et al. (2008). The sensitivity of the crust (Aplite)a Pre-exponential constant (A)forthe 3.2 × 1024 Pa−n s−1 model to regridding is investigated by running models with different crust (Anorthosite)a regridding intervals. Results of these models show that 10 kyr is an Pre-exponential constant (A)forthe 1.3 × 10−14 Pa−n s−1 optimal remeshing interval. mantleb Stress exponent (n)forthecrust 3.1 (Aplite) 5 MODEL RESULTS Stress exponent (n)forthecrust 3.2 (Anorthosite) In this section, we first present the results of model M1. We then

Stress exponent (n) for the mantle 3.0 study the results of model M2 which resembles the configuration Downloaded from https://academic.oup.com/gji/article/187/3/1151/612017 by guest on 29 September 2021 Activation energy (Q)forthecrust 163 KJ mole−1 shown in Fig. 1(B). After that, we investigate the sensitivity of the (Aplite) results to model parameters such as: viscosity of mantle wedge − Activation energy (Q)forthecrust 238 KJ mole 1 (model M3), strength and buoyancy of the subducting continental (Anorthosite) crust (models M4-M7), strength of the arc (models M8-M10), chan- −1 Activation energy (Q) for the mantle 485 KJ mole nel properties (models M11-M13), strength of the backarc (models Activation volume (V)forthecrust 17 × 10−6 m3 mole−1 M14-M16), convergence rate (model M17) and length and sinking (Aplite) Activation volume (V)forthecrust 17 × 10−6 m3 mole−1 velocity of the detached slab (models M18 and M19). Table 1 lists (Anorthosite) the models and their brief descriptions. Here, we present the model Activation volume (V) for the mantle 25 × 10−6 m3 mole−1 description and results for models M1, M2, M3, M5, M13, M15, Plastic weakening parameter (a)0.1M16, M17 and M19. The remaining models are presented in Ap- Plastic weakening parameter (b)0.1pendix A. In Section 5.10, we summarize the results of models in Density of the crust 2800 kg m−3 Appendix A. Density of the mantle 3250 kg m−3 Cohesion (C)50Mpa μ Friction coefficient ( )0.425.1 Model M1; the reference model a Freed & Burgmann (2004). b Karato & Wu (1993). Figs 3(a) and (b) display the plastic strain rate and surface vertical displacement (upper panels) at t = 250 and 400 kyr, respectively. We use power-law plastic flow to represent brittle deformation. At t = 250 kyr, plastic failure occurs in the upper parts of the Yield stress is expressed by a Drucker–Prager criterion based on lithosphere along the arc/backarc boundary. Further convergence Byerlee’s law (Byerlee 1978) leads to failure of the lithosphere at greater depth. Failure of the √ − entire lithosphere occurs after 400 kyr of convergence (Fig. 3b). 3 (R 1) = σYield =− (1 − λ)P + C =−SP + C, (2) Effective strain and shear strain at t 400 kyr (Figs 3c and d) show 2  R dextral shear (here, dextral and sinistral shear refer to the sense of where R = ( 1 − μ2 − μ)−2 ≈ 4. μ, λ and C are the friction shear in a vertical cross-section) along the localized deformation coefficient, pore fluid factor and cohesion, respectively. Plastic yield zone in the arc/backarc boundary. The sense of shear along the strength thus increases linearly with pressure. channel is also dextral, indicating that subduction has ceased. This Plastic strain softening is used to localize deformation in the can also be inferred from the direction of velocity vectors in the model. Plastic strain softening is modelled by a decrease of param- (former) subducting plate in Figs 3(c) and (d). Horizontal velocities eter S in the yield strength formula as a function of strain in the arc are higher than those in the (former) subducting plate and ε2 the backarc lithosphere, which is due to the corner flow in the weak S = (a + (1 − a)exp(− E ))S , (3) mantle wedge. 2b2 0 To demonstrate that the deformation pattern at t = 400 kyr is ε where S0 is the initial value of S and E is the effective strain indicative of subduction of the backarc beneath the arc at higher ε = 1 ε ε ( E 2 ´ij ´ij). a and b are softening parameters which control net convergence, we run a model (M1ch) similar to M1 but here we the amount of plastic softening and the amount of strain to complete replace the newly formed shear zone by a narrow dipping channel. weakening, respectively. The a and b are set to 0.1 in our models. In A 4 km-thick sediment layer lies on top of the surface to fill (and Appendix B, we investigate the effect of this choice on the results. lubricate) the channel with sedimentary material as the oceanic Table 2 lists the material properties used in the models. The plate descends into the mantle. Figs 3(e) and (f) show the strain rheological parameters of the crust and mantle are adopted form and shear strain of model M1ch at t = 3.2 Myr. At this time, the Freed & Burgmann (2004) and Karato & Wu (1993), respectively. oceanic lithosphere has been subducted beneath the arc along the A uniform viscosity of 1 × 1023 Pa s is assigned to the detached slab newly developed plate boundary. The upward motion of the (former) (Royden & Husson 2009). The viscosity of the channel and mantle subducting lithosphere has increased. This is because the suction wedge is set to a constant value of 5 × 1020 Pa s, which is the average force associated with the detached slab—which opposes the positive viscosity of the asthenosphere in the model. We use a Lagrangian buoyancy of the continental crust—decreases with the sinking of formulation to study the deformation pattern following collision. the slab to greater depth. We tested that model results have converged at selected temporal Surface uplift (Figs 3a and b) includes a significant contribu- and spatial discretization. We use linear triangular elements ranging tion that is caused by model spin-up. Body forces from density

C 2011 The Authors, GJI, 187, 1151–1174 Geophysical Journal International C 2011 RAS Lithospheric responses to collision 1159 Downloaded from https://academic.oup.com/gji/article/187/3/1151/612017 by guest on 29 September 2021

Figure 3. Results of model M1. Plastic strain rate at (a) t = 250 kyr, (b) t = 400 kyr. The upper panels show the vertical surface displacement at the indicated times. (c) Effective strain and (d) shear strain at t = 400 kyr. (e) Effective strain and (f) shear strain at t = 3.2 Myr for model M1ch (a model similar to model M1 but with a dipping channel instead of shear zone along the arc/backarc boundary and a lubricated layer on the top surface). Black arrows represent the velocity field at the indicated times. The vector in the green box on the lower left side of the figure indicates the scale of velocity vectors. The arrows ofthe colour bar of total shear strain show the sense of shear as seen in the vertical model section. (g) Vertical surface displacement after removal of the effect of model spin-up for model M1 at t = 400 kyr (solid curve) and for model M1ch at t = 3.2 Myr (dotted curve). anomalies are applied instantaneously at the start of the model run. surface depression along the arc/backarc boundary deepens and The initial elastic response consists of displacements and stresses. the arc close to the newly formed shear zone experiences uplift Plastic deformation subsequently relaxes stresses wherever they while the area ∼70 km from the shear zone on the arc is subject exceed the local strength, and viscous stress flow occurs in low- to subsidence. Uplift and subsidence of the arc are in accordance viscosity regions. From preliminary experiments we find that the with findings of subduction initiation studies (e.g. Toth & Gurnis velocity field becomes stationary after about 250 kyr. This we take 1998; Hall et al. 2003; Gurnis et al. 2004), which indicate that as the model time when the spin-up signature has sufficiently di- following formation of a new trench, the overriding plate close minished. The surface topography at 250 kyr thus represents the to the trench starts to uplift while a depression develops on the reference for evaluating subsequent changes in surface topography overriding plate away from the trench. The results of model M1ch due to geodynamic processes in our models. Fig. 3(g) shows the show that the initial deformation pattern (model M1) is indicative vertical surface displacements at t = 400 kyr and 3.2 Myr rela- of the subsequent evolution. We exploit this in the following mod- tive to that at t = 250 kyr. This figure indicates that at higher net els by only showing results shortly after model runs are started convergence, uplift on the (former) subducting plate increases, the (∼0.5 Myr).

C 2011 The Authors, GJI, 187, 1151–1174 Geophysical Journal International C 2011 RAS 1160 M. Baes, R. Govers and R. Wortel Downloaded from https://academic.oup.com/gji/article/187/3/1151/612017 by guest on 29 September 2021

Figure 4. (a–d) Results of model M2a. (a) Effective strain and (c) shear strain at t = 1.05 Myr. (b) Vertical surface displacement for model M2a at t = 1.05 Myr (solid curve), in comparison with that of model M1 at t = 400 kyr (dashed curve). (d) Vertical surface displacement after removal of the effect of model spin-up at t = 1.05 Myr. (e–h) Results of model M2b. (e) Effective strain and (g) shear strain at t = 850 kyr. (f) Vertical surface displacement for model M2b at t = 850 kyr (solid curve), in comparison with that of model M1 at t = 400 kyr (dashed curve). (h) Vertical surface displacement after removal of the effect of model spin-up at t = 850 kyr. Black arrows represent the velocity field at the indicated times and the arrows of the colour bar of total shear strain show the sense of shear. The vector in the green box on the lower left side of the figure indicates the scale of velocity vectors.

5.2 Model M2; slab attached to the surface Model M2b has the same setting as model M2a but here we assume that following continental collision, subduction in the lower In Section 3.1, we propose two configurations for subduction po- parts of the slab continues with the same rate as the convergence rate larity reversal (Figs 1A and B). In model M1, we investigated sub- v ( subduction = 1). The results of model M2b are similar to those of duction polarity reversal following slab detachment. To investigate vconvergence the second configuration, we run a model similar to model M1, but model M2a, except that a shear zone develops along the arc/backarc here we replace the sinking detached slab by a dipping slab that is boundary after 850 kyr (Figs 4e and g). attached to the surface. We consider two cases: (1) In model M2a The vertical surface displacement (Figs 4f and h) shows a similar we investigate the response to continental collision with no imposed pattern as that in model M2a, except that in model M2b, the uplift velocities at the bottom of the slab; and (2) in model M2b we con- on the (former) subducting plate and the arc near the channel is less, sider the case in which the imposed velocity at the bottom of the which is due to the suction force associated with the tendency to slab is set equal to the convergence velocity. subduction in the deeper parts of the slab. Results of model M2a are shown in Figs 4(a)–(d). Failure along In both models M2a and M2b, continental collision results in the arc/backarc boundary occurs after 1.05 Myr of convergence subduction polarity reversal while the plate interface of the ma- which is 650 kyr later than in model M1. The reason for the higher ture subduction zone is still active to a lesser extent. These results net convergence required to rupture the lithosphere is that the suction suggest an oppositely dipping subduction zone develops following force associated with the slab is less efficient along the arc/backarc continental collision while the original slab is still attached to the boundary in this model, compared to that in model M1. The channel surface plate. has a sinistral shear motion, indicating that the plate interface is still active. The shear motion along the arc/backarc boundary is dextral, implying subduction of the backarc beneath the arc. The 5.3 Effect of the viscosity of the mantle wedge (including magnitude of the sinistral strain is low (especially in the shallow model M3; strong mantle wedge) parts), indicating that the plate interface is less active than the new In model M1, the mantle wedge is as weak as the underlying as- shear zone in the backarc. thenosphere (with a viscosity of 5 × 1020 Pa s). To investigate the Vertical surface displacement (Figs 4b and d) shows a similar sensitivity to this assumption, we set up a model (model M3) with pattern as that in model M1. Higher uplift on the subducting plate similar configuration as model M1 but with a higher mantle wedge and on the arc compared to model M1 is due to the lower suction viscosity of 1 × 1023 Pa s, which is the average viscosity of the force associated with the slab in model M2a. lithosphere in our experiments. Figs 5(a)–(d) show the results of

C 2011 The Authors, GJI, 187, 1151–1174 Geophysical Journal International C 2011 RAS Lithospheric responses to collision 1161 Downloaded from https://academic.oup.com/gji/article/187/3/1151/612017 by guest on 29 September 2021

Figure 5. (a–d) Results of model M3. (a) Effective strain and (c) shear strain at t = 1.65 Myr. (b) Vertical surface displacement for model M3 at t = 1.65 Myr (solid curve), in comparison with that of model M1 at t = 400 kyr (dashed curve). (d) Vertical surface displacement after removal of the effect of model spin-up at t = 1.65 Myr. (e–h) Results of model M5. (e) Effective strain and (g) shear strain at t = 1.35 Myr. (f) Vertical surface displacement for model M5 at t = 1.35 Myr (solid curve), in comparison with that of model M1 at t = 400 kyr (dashed curve). (h) Vertical surface displacement after removal of the effect of model spin-up at t = 1.35 Myr. (i–l) Results of model M13. (i) Effective strain and (k) shear strain at t = 400 kyr. (j) Vertical surface displacement for model M13 at t = 400 kyr (solid curve), in comparison with that of model M1 at t = 400 kyr (dashed curve). (l) Vertical surface displacement after removal of the effect of model spin-up at t = 400 kyr. Black arrows represent the velocity field at the indicated times and the arrows of the colour bar of total shear strain show the sense of shear. The vector in the green box on the lower left side of the figure indicates the scale of velocity vectors. model M3. At t = 1.65 Myr, a localized deformation zone develops from the magnitude of velocity vectors near these two shear zones. along the arc/backarc boundary. On the subducting lithosphere, two The velocity vectors within the arc are small which is due to the localized deformation zones, dipping parallel to the plate interface, reduced corner flow in the strong mantle wedge. develop which extend from the surface to the base of the continental Fig. 5(b) shows the vertical surface displacement at t = 1.65 Myr, crust (Fig. 5a). There is no shear motion within the channel (Fig. 5c) in comparison with that of model M1 at t = 400 kyr. At t = 1.65 Myr, indicating that the subduction contact is not operative. The motion depressions develop on the subducting plate as a result of the for- along the localized deformation zone in the arc/backarc boundary mation of localized deformation zones, the maximum uplift on the is dextral while the localized deformation zone on the subducting arc reaches ∼6.5 km and a depression forms along the arc/backarc plate, close to the trench has a sinistral motion. This shear zone boundary. The vertical surface displacement after removal of model connects to the sinistral shear zone which develops at the base of spin-up at t = 1.65 Myr (Fig. 5d) shows depressions on the sub- the subducted continental crust. Since the shear zone along the ducting plate, an uplift of ∼2 km on the arc, and a depression along arc/backarc does not extend throughout the whole lithosphere be- the arc/backarc boundary. neath the arc, it is not as active as the shear zone which develops on Results of this model suggest that the weakest zone in the system the subducting plate close to the trench. This can also be inferred is the lower continental crust. This leads to the delamination of the

C 2011 The Authors, GJI, 187, 1151–1174 Geophysical Journal International C 2011 RAS 1162 M. Baes, R. Govers and R. Wortel continental crust from the rest of the subducting plate and formation plate is due to the steeper plate interface which increases the imprint of a new plate boundary near the former trench. of crustal buoyancy on the surface deformation. Lower uplift on the Results of models with different mantle wedge viscosity (only arc results from the suction force associated with the detached slab results of model with mantle wedge viscosity of 1 × 1023 Pa s are which affects the arc more efficiently in this model. shown here) suggest that a single shear zone develops along the In this model, the net convergence needed to rupture the former arc/backarc boundary (which leads to subduction polarity reversal) overriding plate is equal to that in model M1; however, it is less as long as the viscosity of the mantle wedge is at least one order than that in model M12 (which is similar to this model except of magnitude less than the average viscosity of the lithosphere. The that its trench/backarc distance is wider, see Appendix A for more net convergence required to develop such a shear zone increases details). The reason for this difference is that the suction force of with increasing viscosity of the mantle wedge. If the viscosity con- the sinking detached slab affects the arc/backarc boundary more trast between the mantle wedge and surrounding lithosphere is less effectively in models M1 and M13 than in model M12. As a result, than one order of magnitude, the dominant deformation pattern is lithospheric failure occurs sooner in models M1 and M13. Results delamination and back stepping. of model M13 suggest that the trench/backarc distance does not influence the temporal evolution of subduction polarity reversal as long as the suction force associated with the detached slab affects Downloaded from https://academic.oup.com/gji/article/187/3/1151/612017 by guest on 29 September 2021 5.4 Model M5; strong mantle wedge and strong the arc/backarc boundary efficiently. subducting crust Model M5 is identical to model M3 except that it has a stronger subducting continental crust [with rheological parameters of 5.6 Model M15; continental backarc anorthosite (Freed & Burgmann 2004)]. In model M5, a dextral Model M15 is similar to model M1, except that the backarc litho- shear zone develops along the arc/backarc boundary after 1.35 Myr sphere is continental. In model M15, we assume an oblique com- of convergence (Fig. 5g). The motion along the channel is sinistral positional boundary between arc and backarc, similar to that in (Fig. 5g), indicating that the subduction process is active. The shear model M1. Results show that a single shear zone develops along the zone which develops along the arc/backarc boundary does not ex- arc/backarc boundary after 600 kyr of convergence (Figs 6a and c) tend throughout the whole lithosphere beneath the arc, therefore it which is 200 kyr later than in model M1. Higher net convergence cannot be considered as an active plate boundary. This can also be required to localize deformation is due to a lower (compositional) inferred from the velocity fields in Figs 5(e) and (g) which show strength contrast between the arc and backarc in model M15. Ver- higher velocities in the subducting plate near the channel than those tical surface displacement (Figs 6b and d) shows a similar pattern in the backarc near the newly formed shear zone. Similar to model to that of model M1, except that in model M15 the backarc surface M3 there is little motion within the arc as a result of a strong mantle has uplifted due to the presence of continental lithosphere in the wedge below the arc. backarc. In this model, the failure direction is controlled by the dip Vertical surface displacement at t = 1.35 Myr is shown as the of the arc/backarc boundary. This experiment suggests that follow- solid curve in Fig. 5(f). Compared to the vertical displacement of ing continent–continent collision, continued convergence between model M1 (dashed curve in Fig. 5f), uplift on the subducting plate two plates can lead to subduction polarity reversal if the composi- and on the arc near the arc/backarc boundary is less, while uplift on tional boundary between arc and backarc dips obliquely towards the the arc close to the trench is higher. The surface vertical displace- arc. ment after removal of model spin-up (Fig. 5h) shows subsidence in the subducting plate close to the trench which is associated with continuation of subduction. It also shows an uplift of ∼2kmonthe arc and a depression along the arc/backarc boundary. 5.7 Model M16; continental backarc and weak Results of this model show that the weakest zone in the system subducting crust is the plate interface. This indicates that entrance of a strong con- Model M16 differs from model M1 in having a continental backarc tinental crust into a subduction zone with a strong mantle wedge and a weather subducting crust (the ratio of the viscosity of the lower results in continuation of subduction and formation of a back-thrust continental crust to the viscosity of the upper lithospheric mantle fault along the arc/backarc boundary. (of the subducting lithosphere) is 0.005). Results of this model (Figs 6e–h) show that after 1.5 Myr of convergence a shear zone develops in the subducting plate near the former trench. The motion 5.5 Model M13; short trench/backarc distance along the original channel is dextral, indicating that there is no Here, we investigate the effect of trench/backarc distance on the subduction. The newly formed shear zone connects to the shear zone results. In model M13, the channel dip angle is 55◦ and the backarc which develops in the lower continental crust. The motion along is closer to the trench (trench/backarc distance is 200 km) than in this shear zone implies detachment of the subducting continental model M1. Adopting a shorter trench/backarc distance in model crust from the rest of subducting lithosphere (see the direction of M13 is in accordance with geological observations which indicate velocity vectors in Figs 6e and g). A new plate boundary thus forms that a steeper subduction zone has a smaller arc/trench distance and in the subducting plate near the former trench (see the location of smaller frontal wedge (e.g. Lallemand et al. 2005, and references development of surface depression in Figs 6f and h). therein). In this experiment the weakest zone in the system is the lower In model M13, lithospheric failure occurs along the arc/backarc continental crust of the subducting lithosphere, leading to delami- boundary after 400 kyr of convergence (Figs 5i and k). Surface nation and back stepping. The results of model M16 indicate that vertical displacement (Fig. 5j) shows that the subducting plate near the viscosity distribution (parameterized by the Moho temperature) the (former) trench experiences higher uplift, while the arc has less of the incoming continent can change the deformation pattern fol- uplift in comparison to model M1. Higher uplift on the subducting lowing continent–continent collision.

C 2011 The Authors, GJI, 187, 1151–1174 Geophysical Journal International C 2011 RAS Lithospheric responses to collision 1163 Downloaded from https://academic.oup.com/gji/article/187/3/1151/612017 by guest on 29 September 2021

Figure 6. (a–d) Results of model M15. (a) Effective strain and (c) shear strain at t = 600 kyr. (b) Vertical surface displacement for model M15 at t = 600 kyr (solid curve), in comparison with that of model M1 at t = 400 kyr (dashed curve). (d) Vertical surface displacement after removal of the effect of model spin-up at t = 600 kyr. (e–h) Results of model M16. (e) Effective strain and (g) shear strain at t = 1.5 Myr. (f) Vertical surface displacement for model M16 at t = 1.5 Myr (solid curve), in comparison with that of model M1 at t = 400 kyr (dashed curve). (h) Vertical surface displacement after removal of the effect of model spin-up at t = 1.5 Myr. (i–l) Results of model M17. (i) Effective strain and (k) shear strain at t = 500 kyr. (j) Vertical surface displacement for model M17 at t = 500 kyr (solid curve), in comparison with that of model M1 at t = 400 kyr (dashed curve). (l) Vertical surface displacement after removal of the effect of model spin-up at t = 500 kyr. (m–p) Results of model M19. (m) Effective strain and (o) shear strain at t = 300 kyr. (n) Vertical surface displacement for model M19 at t = 300 kyr (solid curve), in comparison with that of model M1 at t = 400 kyr (dashed curve). (p) Vertical surface displacement after removal of the effect of model spin-up at t = 300 kyr. Black arrows represent the velocity field at the indicated times and the arrows of the colour bar of total shear strain show the sense of shear. The vector in the green box on the lower left side of the figure indicates the scale of velocity vectors.

C 2011 The Authors, GJI, 187, 1151–1174 Geophysical Journal International C 2011 RAS 1164 M. Baes, R. Govers and R. Wortel

5.8 Model M17; slow convergence rate (g) A slab that is stagnant at the transition zone has no effect on subduction polarity reversal (model M18). Viscous (rate-dependent) resistance to lithospheric motion is con- trolled by convergence rate in our models, due to the adoption In Appendix B, we present results of an analysis of sensitivity to of a power-law viscosity. In model M17, we investigate the influ- weakening parameters (expression 3). They show that increasing a ence of the convergence rate on the deformation pattern following or b reduces the length of strain localization zone and, hence, plastic arc–continent collision. In this model, the convergence velocity is failure of the entire lithosphere requires higher net convergence. taken to be half of that in model M1. A shear zone develops along the arc/backarc boundary after 500 kyr of convergence—100 kyr later and 3 km less net convergence than in model M1 (Figs 6i and k). 6 MODEL ANALYSIS The requirement of less net convergence to rupture the lithosphere in model M17 indicates that when the convergence rate is slow the 6.1 Possible responses to arc/continent–continent collision suction force associated with slab pull becomes the dominant force Fig. 7 summarizes the results of our experiments indicating to fail the lithosphere. Vertical surface displacement in this model the switches between three types of responses to arc/continent–

(Figs 6j and l) is similar to that in model M1. Results of this model continent collision and their key parameters. In this figure, the hor- Downloaded from https://academic.oup.com/gji/article/187/3/1151/612017 by guest on 29 September 2021 indicate that a lower convergence rate delays the development of a izontal axis indicates the relative strength of the mantle wedge, new plate boundary on the overriding plate. which is shown by ratio of the mantle wedge’s viscosity (ηMW)to the backarc lithosphere’s viscosity (ηLith). The vertical axis shows the degree of coupling between the incoming continental crust and 5.9 Model M19; high sinking velocity its lithospheric mantle, which is defined by ratio of the viscosity η To investigate the effect of the sinking velocity of the detached of the lower continental crust ( LCC) to the viscosity of the upper η slab, we perform an experiment similar to model M1 but with a lithospheric mantle ( ULM). The values of 0.02 and 0.006 on the higher sinking velocity of 3 cm yr−1. The effective strain and shear vertical axis are the threshold values for the upper right and upper strain for this model (Figs 6m and o) show that failure along the left panels, respectively. The value of 0.1 on the horizontal axis is arc/backarc boundary occurs after 300 kyr of convergence (which the threshold value for the right and left panels in the figure. The is 100 kyr earlier than in model M1). Vertical surface displacement threshold values on the horizontal and vertical axes are determined is similar to that of model M1 (Fig. 6n), except that now, due to the by stepwise changes in the viscosity of the mantle wedge and vis- × 20 stronger suction force associated with the sinking detached slab, the cosity of the lower continental crust (with a 1 10 Pa s stepsize), respectively. If the coupling between subducting continental crust uplift on the arc is less than that in model M1. Results of this model η and lithospheric mantle is high ( LCC ≥ 0.006), continental colli- indicate that the drag force associated with the sinking detached ηULM slab speeds up subduction polarity reversal. sion results in subduction polarity reversal only if the viscosity of the mantle wedge is at least one order of magnitude less than the average viscosity of the lithosphere (upper left panel of Fig. 7). In 5.10 Summary of models in Appendices this case, the weakest zone in the system is the mantle wedge. If the incoming continental crust and lithospheric mantle are η Results of the models in Appendix A indicate that LCC ≥ . highly coupled ( η 0 02) and the mantle wedge is strong η ULM ( MW > 0.1; upper right panel of Fig. 7), the response to colli- (a) in arc–continent collision, the response to collision is subduc- ηLith tion polarity reversal regardless of the coupling between subducting sion is continuation of subduction and formation of a back-thrust continental crust and lithospheric mantle as long as a weak mantle along the arc/backarc boundary. Subduction continuation implies wedge is present (models M4 and M6); that the weakest zone in the system is the plate interface, allowing (b) Our experiments with different crustal density values in the subduction of buoyant continental crust to the greater depths. −3 If the subducting continental crust and lithospheric mantle are range of 2800–2900 kg m indicate that the density of the subduct- η LCC < . ing crust does not have a significant effect on subduction polarity poorly coupled ( η 0 02) and the mantle wedge is strong η ULM reversal (model M7); ( MW > 0.1; lower right panel of Fig. 7), the response to collision ηLith (c) One of the parameters affecting the time of lithospheric rup- is delamination and back stepping; the subducting continental crust ture along the arc/backarc is the strength contrast between the arc is separated from the lithospheric mantle and a new plate boundary and backarc (models M8–M10); forms near the former trench (the black arrow in Fig. 7 shows the (d) When the channel is wide, the integrated resistance to shear location of new plate boundary). Therefore, the response to collision along the channel is lower than that in a narrow channel. In Model is delamination and back stepping if the weakest zone is the lower 11 we find that if we increase the channel width to 16 km, the continental crust. η LCC < . deformation still concentrates in the arc and backarc regions like in If the crust–lithospheric mantle coupling is weak ( η 0 006) η ULM our reference model, that is, the mantle wedge remains being the and the mantle wedge is also weak ( MW ≤ 0.1; lower left panel ηLith weakest part of the system. of Fig. 7), the response to collision depends on the nature of the (e) The channel dip angle affects the temporal evolution of sub- backarc. In the case of arc–continent collision (oceanic backarc), duction polarity reversal if the change in channel dip results in collision results in subduction polarity reversal, while the response decreasing the effect of suction force associated with the detached is delamination and back stepping when the backarc is continen- slab on the arc/backarc boundary (model M12); tal (continent–continent collision). The reason for the dependence (f) Higher temperature along the arc/backarc boundary reduces of the collisional response on the nature of the backarc is that the strength of the lithosphere, resulting in failure of the lithosphere when both the mantle wedge and the incoming lower continental at lower net convergence. For instance, young oceanic lithosphere crust are weak, the strength (composition) contrast between arc in the backarc speeds up the subduction polarity reversal process and backarc plays a key role in determining the weakest zone in (model M14); the system. In arc–continent collision, there is sufficient strength

C 2011 The Authors, GJI, 187, 1151–1174 Geophysical Journal International C 2011 RAS Lithospheric responses to collision 1165 Downloaded from https://academic.oup.com/gji/article/187/3/1151/612017 by guest on 29 September 2021

Figure 7. Three possible responses to arc/continent–continent collision. The horizontal axis indicates the relative strength of the mantle wedge which is expressed by ratio of the viscosity of the mantle wedge (ηMW) to the backarc lithosphere’s viscosity (ηLith). The vertical axis shows the coupling between continental crust and lithospheric mantle (of the subducting plate) which is defined as ratio of the viscosity of the lower continental crust (ηLCC) to the viscosity of the upper lithospheric mantle (ηULM). The horizontal white arrows indicate the direction of plate convergence and the small black arrows show the direction of motion within the plates. The large black arrows point to the newly formed (except for the upper right panel, in which the subduction interface continues to be active) plate boundaries, indicated in red. The values of 0.02 and 0.006 on the vertical axis are the threshold values for the upper right and upper left panels, respectively. The value of 0.1 on the horizontal axis is the threshold value for the right and left panels in the figure. Following arc–continent collision, η subduction polarity reversal occurs if the MW is equal to or lower than 0.1 (upper left and lower left panels). In continent–continent collision, one extra ηLith ηLCC condition must be satisfied: η ≥ 0.006 (upper left panel). Delamination and back stepping occur if either (1) the mantle wedge is strong and lower crust is η ULMη η η weak, that is, MW > 0.1and LCC < 0.02 (lower right panel), or (2) the mantle wedge is weak ( MW ≤ 0.1) and LCC < 0.006 (lower left panel). The latter ηLith ηULM ηLith ηULM is only valid in continent–continent collisional settings (continental backarc). Continued subduction occurs if the mantle wedge and coupling between crust η η and lithospheric mantle are both strong, that is, MW > 0.1and LCC ≥ 0.02 (upper right panel). ηLith ηULM

(composition) contrast between arc and backarc. Therefore, in the backarc starts to descend below the arc. Upon further development competition between mantle wedge and lower continental crust, the of the incipient subduction interface the buoyancy of the young mantle wedge is the weakest, resulting in subduction polarity rever- lithosphere may lead to strong interplate coupling. Possibly, the old sal. In continent–continent collision, the weakest zone is the lower plate contact will turn out to be the weaker zone and as such it continental crust of the subducting lithosphere as in this case there may resume accommodating most of the convergence. In model is not enough strength contrast between arc and backarc. M2 the subducting slab is still attached to the surface part of the We note that the configurations illustrated in Fig. 7 show the plate. In that configuration it may hamper later stages of the newly initial stage of deformation following continental collision. Further developing undertrusting in the backarc. convergence between the two plates may result in switches between In the case of delamination and back stepping, continued con- quadrants. For instance, following the entrance of strong continental vergence leads to the subduction of the continental crust along the crust into a subduction zone with a strong mantle wedge (upper new plate boundary. The pattern of deformation at continuing con- right panel of Fig. 7), continued subduction of continental crust vergence in this case is the formation of a series of thrust faults may lead to the formation of low-viscosity mantle wedge, due to parallel to each other near the plate boundary and stacking of con- the partial melting and dehydration of the crust at depth. Therefore, tinental fragments onto the overriding plate. With respect to all further convergence may eventually result in a switch of the scenario four situations considered above—and others that can be added to shown in the upper right panel of Fig. 7 to that of subduction these—we note that our modelling study addresses the early stages polarity reversal. In model M14 young oceanic lithosphere from the of plate boundary transitional developments, each with their specific

C 2011 The Authors, GJI, 187, 1151–1174 Geophysical Journal International C 2011 RAS 1166 M. Baes, R. Govers and R. Wortel competition between the possible weakness zones. Subsequent tion polarity reversal, or after, is not clear. For instance below the stages may have their own specific competing aspects, either simi- Algerian margin, the depth of the high velocity anomalies (in seis- lar to or in addition to the ones considered (e.g. in the case of the mic tomography results) warrants the conclusion that slab break-off structural complexity in model M2). These later stages are beyond occurred prior to the formation of reverse faults (Carminati et al. the scope of this study. 1998). However, for Timor, it has been proposed that subduction polarity reversal is occurring while the slab is still attached to the surface. Studies on slab detachment (e.g. Wong A Ton & Wortel 6.2 Factors that affect the time of lithospheric failure 1997; Van de Zedde & Wortel 2001) show that there is a time lag between arrival of continental crust to the trench and slab break-off. Sensitivity analysis of model M1 indicates that the time of litho- spheric failure in subduction polarity reversal depends on different Wong A Ton & Wortel (1997) concluded that slab break-off takes parameters. In this section, we summarize the results of our ex- place on a timescale of 1.7–39 Myr, depending on several param- eters such as age of the previously subducted oceanic lithosphere, periments concerning the parameters which influence the temporal evolution of subduction polarity reversal. frictional heat and convergence rate. Van Hunen & Allen (2011) showed that delay time between arrival of continental crust and the occurrence of slab break-off depends mostly on the strength of the Downloaded from https://academic.oup.com/gji/article/187/3/1151/612017 by guest on 29 September 2021 6.2.1 Viscosity of the mantle wedge previously subducted oceanic plate. They indicated that this delay time ranges from 10 Myr for young and weak slabs to more than 20 Our experiments have shown that the presence of high viscosity ma- Myr for old and strong slabs. In this study, we have investigated both terial in the mantle wedge delays the failure of the overriding plate the modes of (1) slab detachment prior to the subduction polarity along the arc/backarc boundary. This is valid only if the viscosity reversal and (2) subduction polarity reversal while the slab is still contrast between the mantle wedge and its surrounding lithosphere attached to the surface (Figs 1A and B). In both cases, continental is higher than one order of magnitude. collision results in subduction polarity reversal. However, results of models with the slab attached to the surface show that as long as the slab is continuous the plate interface remains active to a lesser 6.2.2 Strength of continental crust in the arc extent. In Section 6.1, we already noted that our modelling of the The strength of the continental crust in the arc affects the rheological early stages of the collisional process does not address the (later) (strength) contrast between arc and backarc. A strong continental complexities that may arise from the geometrical conflict between crust in the arc reduces this rheological contrast, and therefore in- two slabs subducting in opposing directions. creases the net convergence needed to rupture the former overriding plate. 7.2 Comparison with previous studies

6.2.3 Strength of the lithosphere in the backarc There are two differences between our results and those of Chemenda et al. (2001a) The overriding plate fails at higher net convergence if we increase the age of the oceanic lithosphere in the backarc. In the case of a (1) They concluded that the direction of failure depends on the continental backarc setting, the net convergence required to rupture trench/backarc distance. However, in our experiments the vergence the overriding plate is higher than that for the case of an oceanic of the new shear zone is independent of the trench/backarc distance. backarc. This is due to the compositional (strength) difference be- This difference arises from inclusion of a symmetric weakness zone tween the arc and backarc which is higher when oceanic lithosphere in the backarc in models of Chemenda et al. (2001a), which causes is present in the backarc. the plastic failure to be dependent on the wavelength of flexural bending. (2) They found that failure occurs at the (extinct) backarc spread- 6.2.4 Convergence rate and sinking velocity of the detached slab ing centre, while in our models failure occurs along the arc/backarc A slow convergence rate delays the failure of the overriding plate. boundary, which is a weak zone as a result of weak mantle The net convergence needed to rupture the lithosphere is not con- wedge and compositional (strength) difference between the arc and stant in models with different convergence rates. The reason is that backarc. when the convergence rate is low, the suction force associated with Our model results show that a weak mantle wedge is a vital the detached slab becomes the dominant force to fail the lithosphere. component in subduction polarity reversal. This is consistent with The sinking velocity of the detached slab depends on several factors the results of Goren et al. (2008) which indicate that one of the such as length (volume) of the detached slab and average viscos- required conditions for subduction nucleation at passive margins ity of the surrounding mantle. The suction force associated with is the presence of a weak low-viscosity continental lithosphere. the sinking detached slab within the mantle helps to speed up the There is a difference between our results and those of Goren et al. failure of the overriding plate. (2008); in our models the density difference between oceanic and continental lithospheres in the arc/backarc boundary does not play a significant role in the formation of a new subduction, whereas in 7 DISCUSSION Goren et al. (2008) the lateral pressure gradient in a passive margin is a key factor in the subduction initiation process. The reason for 7.1 Subduction polarity reversal with or without slab this difference is that in our models the forces induced from the attached to the surface convergence rate and sinking detached slab are more dominant than Our reference model explores subduction polarity reversal follow- the force associated with the lateral density difference along the ing slab break-off. Whether in fact the slab detaches before subduc- arc/backarc.

C 2011 The Authors, GJI, 187, 1151–1174 Geophysical Journal International C 2011 RAS Lithospheric responses to collision 1167

Results of our models involving delamination (models M3, M9 results. Domzig et al. (2006) reported uplift of Alboran–Algerian and M16) are consistent with previous studies such as De Franco margins, which can be interpreted as the uplift due to the formation et al. (2008), Morency & Doin (2004) and Van den Beukenl (1992), of a new subduction zone along the Algerian margin. indicating that delamination occurs when the lower continental crust is weakly coupled to the lithospheric mantle. They are also in good agreement with findings of Chemenda et al. (1996) regarding the 7.3.2 Delamination and continued subduction occurrence of delamination at high interplate pressure regimes. The high interplate pressure between subducting and overriding plate The Himalayas are the best natural example for the delamination implies that the shear resistance along the plate interface is high. process. Based on the slab length (which is about 6000 km) and As a result, in the competition between the continental lower crust convergence rate (∼10 cm yr−1), it is inferred that the Neo-Tethyan and the plate interface the former one is the weakest. subduction in the Himalayas had a weak mantle wedge. Our nu- merical experiments suggest that in the presence of a weak mantle wedge, delamination occurs only if the Moho temperature is higher than 800 ◦C. This is consistence with Singh & Negi (1982) who

7.3 Comparison with observations Downloaded from https://academic.oup.com/gji/article/187/3/1151/612017 by guest on 29 September 2021 showed that the northern part of Indian shield has a high Moho temperature of 850–900 ◦C. 7.3.1 Subduction polarity reversal In the Aegean region, the slab length (2100–2400 km) and con- In our experiments, subduction polarity reversal occurs if the weak vergence rate (∼2.1–2.5 cm yr−1) are in the favour of presence of a mantle wedge is the weakest zone in the system. The presence of weak mantle wedge. Since delamination and crust stacking in this a strength (compositional) contrast boundary between the arc and region occurred in the past (during the Cretaceous), the geological backarc plays a role as a secondary factor in subduction polarity re- evidence regarding the strength of the subducting continental crust versal. In all investigated regions in Section 2, subduction polarity in this region is poorly preserved. Similarly, in the North America reversal has resulted in formation of an oceanic subduction zone. and China collision zones, the geological evidence for the compo- This indicates the presence of an oceanic lithosphere in the backarc sition of the crust and the strength of the mantle wedge has been and, hence, the existence of a compositional (strength) boundary overprinted by the subsequent tectonic processes. Therefore, it is between the arc and backarc prior to the subduction polarity rever- not possible to compare the model results with observations in these sal. To verify the presence of the weak mantle wedge, we consider regions. two criteria: slab length [based on study of Arcay et al. (2005)] There is no natural example for the continued subduction sce- and subduction rate [based on study of Billen & Hirth (2005)]. In nario at the present-day, which may result from this configuration the following, we summarize the geophysical and geological evi- not being a stable tectonic setting. Several studies have shown ge- dence concerning these two criteria for subduction polarity reversal ological evidence for subduction of continental crust to the depth localities described in Section 2. more than 100 km in several collision zones (e.g. Chopin 1984; Ye One of the powerful tools to verify the slab length within the et al. 2000). This implies that during the earliest stages of collision, mantle is seismic tomography, which images the velocity structure prior to the subduction polarity reversal or delamination, the plate of the Earth’s crust, mantle and core. Tomographic images have interface was the weakest zone in the system. revealed a nearly 2000 km long, flat-lying anomaly below the New Hebrides and Solomon Islands, that has been interpreted to be the remnant of past subduction zones in these regions (Hall & Spakman 8 CONCLUSIONS 2002). Spakman & Hall (2010) showed that there is a north-dipping slab beneath Timor Island which extends from the surface till depth Our numerical models show three possible responses to arc/ of ∼660 km. Beneath the north African margin, the slab length is continent–continent collision in a setting of continuing conver- estimated to be between 700 and 800 km (Spakman & Wortel 2004). gence: (1) subduction polarity reversal, (2) delamination and back In all the aforementioned sites, the length of the slab is more than stepping and (3) continued subduction. The switches between these 650 km, which seems to be long enough to allow weakening of the responses are controlled by the competition between three (poten- mantle wedge due to the slab dehydration. The other criterion is the tial) weakness zones: (1) the mantle wedge, (2) the lower continental subduction rate which is fast enough [i.e. more than 2.5 cm yr−1, crust and (3) the plate interface. We find: based on Billen & Hirth (2005)] in all the investigated regions to − (1) If the mantle wedge is the weakest zone in the system, the produce a weak mantle wedge: 7.9 cm yr 1 convergence rate in New − response to collision will be subduction polarity reversal. This hap- Hebrides and Solomon Islands (Wells 1989), 7–8 cm yr 1 conver- pens only if the wedge viscosity is at least one order of magnitude gence rate in Timor Island (Harris 1991, and references therein). smaller than that of the backarc lithosphere. In continent–continent In Algeria, the convergence rate between Africa and Eurasia was − collision, one additional condition must be satisfied: for the sub- slow during Late Oligocene–Early Miocene [∼2cmyr 1, (Jolivet ducting lithosphere the ratio of the viscosity of the lower continental & Faccenna 2000, and references therein). However, since the slab − crust (η ) to the viscosity of the upper lithospheric mantle (η ) was subject to a rapid roll-back of 3–4 cm yr 1 (Faccenna et al. LCC ULM must be equal to or greater than 0.006. 2001, and references therein), the rate of subduction in the region − (2) The response to collision is delamination and back stepping had to be in the range of 3–4 cm yr 1. if the incoming lower continental crust is the weakest zone in the The model results have shown uplift for the overriding plate dur- η η MW > . LCC < system. We find that this occurs if either η 0 1and η ing early stages of subduction formation which is in accordance to η η Lith ULM 0.02, or if MW ≤ 0.1and LCC < 0.006. The latter is only valid for the previous studies on subduction initiation (Toth & Gurnis 1998; ηLith ηULM Hall et al. 2003; Gurnis et al. 2004). De Smet et al. (1990) show continent–continent collision. (3) Subduction of continental lithosphere continues if the plate an uplift of 750 m in west Timor, one of the regions for which po- η η interface is the weakest zone ( MW > 0.1and LCC ≥ 0.02). larity reversal has been proposed. This value agrees with our model ηLith ηULM

C 2011 The Authors, GJI, 187, 1151–1174 Geophysical Journal International C 2011 RAS 1168 M. Baes, R. Govers and R. Wortel

Subduction polarity reversal occurs as a result of failure of the Chopin, C., 1984. Coesite and pure pyrope in high-grade blueschists of the overriding plate along the arc/backarc boundary. Several factors af- western Alps: a first record and some consequences, Contrib. Mineral. fect the time of lithospheric failure including strength of the mantle Petrol., 86, 107–118. wedge, strength of the arc and backarc, convergence rate and the Chopra, P.N. & Paterson, M.S., 1984. The role of water in the deformation sinking velocity of the detached slab. of dunite, J. geophys. Res., 89, 7861–7876. Coney, P.J., Jones, D.L. & Monger, J.W., 1980. Cordillera suspect terranes, The regions where there are indications for subduction polarity Nature, 288, 329–333. reversal are characterized by long-lasting subduction. This warrants Cooper, P.A. & Taylor, B., 1985. Polarity reversal in the Solomon Islands the inference that the mantle wedge weakened through hydration. arc, Nature, 314, 428–430. This is in agreement with our model results, which only predict Cowley, S., Mann, P., Coffin, M.F. & Shipley, T.H., 2004. Oligocene to Re- polarity reversal if such a weak wedge exists. cent tectonic history of the Central Solomon intra-arc basin as determined from marine seismic reflection data and compilation of onland geology, Tectonophysics, 389, 267–307. ACKNOWLEDGMENTS Curray, J.R., Shor, G.G., Raitt, R.W. & Henry, M., 1977. Seismic refraction and reflection studies of crustal structure of the eastern Sunda and western

We thank Andrea Argnani and an anonymous reviewer for construc- Banda arcs, J. geophys. Res., 82, 2479–2489. Downloaded from https://academic.oup.com/gji/article/187/3/1151/612017 by guest on 29 September 2021 tive reviews. This study is a contribution to ESF-EUROCORES Currie, C.A. & Hyndman, R.D., 2006. The thermal structure of subduction Programs EUROMARGINS (Project WESTMED) and TOPO- zone back arcs, J. geophys. Res., 111, doi:10.1029/2005JB004024. EUROPE (Project TopoMed). M.B. was supported by the Nether- DeCelles, P.G. & DeCelles, P.C., 2001. Rates of shortening, propagation, lands Organization for Scientific Research (NWO). Computational underthrusting, and flexural wave migration in continental orogenic sys- resources for this work were provided by the Netherlands Research tems, Geology, 29, 135–138. De Franco, R., 2008. Dynamics of continental collision: influence of Center for Integrated Solid Earth Science (ISES 3.2.5 High End the nature of the plate contact, Geologica Ultraiectina, http://igitur- Scientific Computation Resources). archive.library.uu.nl/dissertations/2008-0402-200601/defranco.pdf. De Franco, R., Govers, R. & Wortel, R., 2008. Dynamics of continental REFERENCES collision: influence of the plate contact, Geophys. J. Int., 174, 1101–1120. De Smet, M.E.M., Forttuin, A.R., Troelstra, S.R., Van Marie, L.J., Karmini, Arcay, D., Tric, E. & Doin, M.-P.,2005. Numerical simulations of subduction M., Tiokrosapoetro, S. & Hadiwasastra, S., 1990. Detection of collision- zones. Effect of slab dehydration on the mantle wedge dynamics, Phys. related vertical movements in the Outer Banda Arc (Timor, Indonesia), Earth planet. Inter., 149, 133–153. using micropaleontological data, J. South. Asian Earth, 4, 337–356. Barazangi, M. & Isacks, B., 1971. Lateral variations of seismicwave at- Deverchere, J. et al., 2005. Active thrust faulting offshore Boumerdes, Alge- tenuation in the upper mantle above the inclined earthquake zone of the ria, and its relations to the 2003 Mw 6.9 earthquake, Geophys. Res. Lett., Tonga Island Arc: deep anomaly in the upper mantle, J. geophys. Res., 32, doi:10.1029/2004GL021646. 76, 8493–8515. Domzig, A. et al., 2006. Searching for the Africa-Eurasia Miocene boundary Ben-Avraham,Z., Nur, A., Jones, D. & Cox, A., 1981. Continental accretion: off shore western Algeria (MARADJAO03˜ cruise), C. R. Geosci., 338, from oceanic plateaus to allochthonous terranes, Science, 213, 47–54. 80–91. Bilham, R., Larson, K. & Freymueller, J., 1997. GPS measurements of Dong, Y., Zhang, G., Neubauer, F., Liu, X, Genser, J. & Hauzenberger, present-day convergence across the Nepal Himalaya, Nature, 386, 61–64. C., 2011. Tectonic evolution of the Qinling orogen, China: review and Billen, M.I. & Gurnis, M., 2001. A low viscosity wedge in subduction zones, synthesis, J. Asian Earth Sci., 41, 213–237. Earth planet. Sci. Lett., 193, 227–236. Faccenda, M., Gerya, T.V. & Chakraborty, S., 2008. Styles of post- Billen, M.I. & Hirth, G., 2005. Newtonian versus non-Newtonian upper subduction collisional orogeny: influence of convergence velocity, crustal mantle viscosity: implications for subduction initiation, Geophys. Res. rheology and radiogenic heat production, Lithos, 103, 257–287. Lett., 32, doi:10.1029/2005GL023457. Faccenna, C., Funiciello, F., Giardini, D. & Lucente, P., 2001. Episodic Bird, P., 1978. Initiation of intracontinental subduction in the Himalaya, J. back-arc extension during restricted mantle convection in the Central geophys. Res., 83, 4975–4987. Mediterranean, Earth planet. Sci. Lett., 187, 105–116. Bird, P.,1979. Continental delamination and the Colorado Pateau, J. geophys. Falvey, D.A., 1975. Arc reversals and a tectonic model for the South Fiji Res., 84, 7561–7571. Basin, Aust. Soc. Explor. Geophys. Bull., 6, 47–49. Byerlee, J., 1978. Friction of rocks, Pageoph, 116, 615–626. Fernandes, R.M.S., Ambrosius, B.A.C., Noomen, R., Bastos, L., Wortel, Carminati, E., Wortel, R., Spakman, W.& Sabadini, R., 1998. The role of slab M.J.R., Spakman, W. & Govers, R., 2003. The relative motion between detachment processes in the opening of the western-central Mediterranean Africa and Eurasia as derived from ITRF2000 and GPS data, Geophys. basins: some geological and geophysical evidence, Earth planet. Sci. Lett., Res. Lett., 30, doi:10.1029/2003GL017089. 160(3–4), 651–665. Freed, A.M. & Burgmann, R., 2004. Evidence of power-law flow in the Chemenda, A.I., Mattauer, M. & Bokun, A.N., 1996. Continental subduction Mojave desert mantle, Nature, 430, 548–551. and a mechanism for exhumation of high-pressure metamorphic rocks: Genrich, J.F., Bock, Y., McCaffrey, R., Calais, E., Stevens, C.W. & Subarya, new modelling and field data from Oman, Earth planet. Sci. Lett., 143, C., 1996. Accretion of the southern Banda arc to the Australian plate 173–182. margin determined by Global positioning System measurments, Tecton- Chemenda, A.I., Yang, R.K., Hsieh, C.-H. & Groholsky, A.L., 1997. Evo- ics, 15, 288–295. lutionary model for the Taiwan collision based on physical modelling, Goren, L., Aharonov, E., Mulugeta, G., Koyi, H.A. & Mart, Y., 2008. Duc- Tectonophysics, 274, 253–274. tile deformation of passive margins: a new mechanism for subduction Chemenda, A.I., Hurpin, D., Tang, J.-C., Stephan, J.-F. & Buffet, G., 2001. initiation, J. geophys. Res., 113, doi:10.1029/2005JB004179. Impact of arc- continent collision on the conditions of burial and exhuma- Govers, R. & Wortel, M.J.R., 1993. Initiation of asymmetric extension in tion of UHP/LT rocks: experimental and numerical modelling, Tectono- continental lithosphere, Tectonophysics, 223, 75–96. physics, 342, 137–161. Govers, R. & Wortel, M.J.R., 2005. Lithosphere tearing at STEP faults: Chemenda, A.I., Yang, R.-K., Stephan, J.-F., Konstantinovskaya, E.A. & response to edges of subduction zones, Earth planet. Sci. Lett., 236, Ivanov, G.M., 2001. New results from physical modelling of arc-continent 505–523. collision in Taiwan: evolutionary model, Tectonophysics, 333, 159–178. Gurnis, M., Hall, C. & Lavier, L.L., 2004. Evolving force bal- Chen, W.-P. & Brudzinski, M.R., 2001. Evidence for a large-scale remnant ance during incipient subduction. Geochem.Geophys. Geosyst., 5, of subducted lithosphere beneath Fiji, Science, 292, 2475–2479. doi:10.1029/2003GC000681.

C 2011 The Authors, GJI, 187, 1151–1174 Geophysical Journal International C 2011 RAS Lithospheric responses to collision 1169

Hafkenscheid, E., Wortel, M.J.R. & Spakman, W.,2006. Subduction history McCaffrey, R., 1996. Slip partitioning at convergent plate boundaries of SE of the Tethyan region derived from seismic tomography and tectonic Asia, Geol. Soc. Lond. Spec. Publ., 106, 3–18. reconstructions, J. geophys. Res., 111, doi:10.1029/2005JB003791. McCaffrey, R., Molnar, P., Roecker, S. & Joyodiwiryo, Y., 1985. Mi- Hall, C.E., Gurnis, M., Sdrolias, M., Lavier, L.L. & Muller, R.D., 2003. croearthquake seismicity and fault plane solutions related to arc-continent Catastrophic initiation of subduction following forced convergence across collision in the eastern Sunda arc, Indonesia, J. geophys. Res., 90, fracture zones, Earth planet. Sci. Lett., 212, 15–30. 4511–4528. Hall, R. & Spakman, W., 2002. Subduction slabs beneath the eastern McKenzie, D.P.,1969. Speculations on the consequences and causes of plate Indonesia-Tonga region: insights from tomography, Earth planet. Sci. motions, Geophys. J. R. astr. Soc., 18, 1–32. Lett., 201, 321–336. Mei, S. & Kohlstedt, D.L., 2000. Influence of water on plastic deformation Hamilton, W., 1973. Tectonics of the Indonesian region, Bull. Geol. Soc. of olivine aggregates: 2. Dislocation creep regime, J. geophys. Res., 105, Malays., 6, 3–10. 21 471–21 481. Hamilton, W., 1979. Tectonics of the Indonesian region, U.S. Geol. Surv. Meng, Q-R, & Zhang, G.-W., 1999. Timing of collision of the North Prof. P., 1078, 1–345. and South China blocks: controversy and reconciliation, Geology, 27, Harris, R.A., 1991. Temporal distribution of strain in the active Banda 123–126. orogen: a reconstruction of rival hypotheses, J. South. Asian Earth Sci., Mitrovica, J.X. & Forte, A.M., 1997. Radial profile of mantle viscosity:

6, 373–386. results from the joint inversion of convection and postglacial rebound Downloaded from https://academic.oup.com/gji/article/187/3/1151/612017 by guest on 29 September 2021 Hasegawa, A., Zhao, D., Hori, S., Yamamoto,A. & Horiuchi, S., 1991. Deep observables, J. geophys. Res., 102, 2751–2769. structure of the Northeastern Japan Arc and its relationship to seismic and Miura, S., Suyehiro, K., Shinohara, M., Takahashi, N., Araki, E. & Taira, volcanic activity, Nature, 352, 683–689. A., 2004. Seismological structure and implications of collision between Hathway, W.B., 1993. The Nadi Basin: neogene strike-slip faulting and the Ontong Java Plateau and Solomon Island Arc from ocean bottom sedimentation in a fragmented arc, western Viti Levu, Fiji, J. Geol. Soc. seismometer-airgun data, Tectonophysics, 389, 191–230. London, 150, 563–581. Morency, C. & Doin, M.-P.,2004. Numerical simulations of the mantle litho- Hoffman, P.F., 1988. United plates of America: the birth of a craton, Annu. sphere delamination. J. geophys. Res., 109, doi:10.1029/2003JB002414. Rev. Earth Planet. Sci., 16, 543–604. Nokleberg, W.J. et al., 2000. Phanerozoic tectonic evolution of the circum- Johnson, M.R.W., 2002. Shortening budgets and the role of continental North Pacific, U. S. Geol. Surv. Professional Paper, issue 1626. subduction during the India-Asia collision, Earth-Sci. Rev., 59, 101–123. Nugroho, H., Harris, R., Lestariya, A.W.& Maruf, B., 2009. Plate boundary Jolivet, L. & Faccenna, C., 2000. Mediterranean extension and the Africa- reorganization in the active Banda Arc-continent collision: insights from Eurasia collision, Tectonics, 19, 1095–1106. new GPS measurements, Tectonophysics, 479, 52–65. Karato, S. & Wu, P., 1993. Rheology of the upper mantle: a synthesis, Parrish, R.R., Gough, S.J., Searle, M.P. & Waters, J., 2006. Plate veloc- Science, 260, 771–778. ity exhumation of ultrahigh-pressure eclogite in the Pakistan Himalaya, Kerr, A.C. & Mahoney, J.J., 2007. Oceanic plateau: problematic plumes, Geology, 34, 989–992. potential paradigms, Chem. Geol., 241, 332–353. Parsons, B. & Sclater, J.G., 1977. An analysis of the variation of ocean floor Kreemer, C, & Holt, W.F., 2000. Active deformation in eastern Indonesia bathymetry and heat flow with age, J. geophys. Res., 82, 803–827. and the Philippines from GPS and seismicity data, J. geophys. Res., 105, Phinney, E.J., Mann, P., Coffin, M.F. & Shipley, T.H., 2004. Sequence 663–680. stratigraphy, structural style, and age of deformation of the Malaita ac- Kroenke, L.W.,1989. Interpretation of a multichannel seismic reflection pro- cretionary prism (Solomon arc-Ontong Java plateau convergent zone), file northeast of the Solomon Islands from the southern flank of the On- Tectonophysics, 389, 221–246. tong Java Plateau across the Malaita anticlinorium to the Solomon Islands Pysklywec, R.N., Mitrovica, J.X. & Ishii, M., 2003. Mantle avalanche as a arc, in Geology and Offshore Resources of Pacific Island Arcs-Solomon driving force for tectonic reorganization in the southwest Pacific, Earth Islands and Bougainville, Papua New Guinea region, Earth Science Se- planet. Sci. Lett., 209, 29–38. ries, Vol. 12, pp. 145–148, eds Vedder, J.G., Bruns, T.R., Circum-Pacific Regard, V., Faccenna, C., Bellier, O. & Martinod, J., 2008. Labratory ex- Council for Energy and Mineral Resources, Houston, TX. periments of slab break-off and slab dip reversal: insight into the Alpine Kroenke, L.W.,Resig, J.M. & Cooper, R.A., 1986. Tectonicsof the southeast- Oligocene reorganization, Terra Nova, 20, 267–273. ern Solomon Islands: formation of the Malaita anticlinorium. in Geology Rodda, P. & Kroenke, L.W., 1984. Fiji: a fragmented arc, in Cenozoic and Offshore Resources of Pacific Islands, Earth Science Series, Vol. 4, Tectonic Development of the Southwest Pacific, Vol. 6, pp. 87–109, ed. pp. 109–115, eds Vedder, J.G., Pound, K.S., Boundy S.Q., Circum-Pacific Kroenke, L.W., CCOP/SOPAC Technical Bull., Fiji. Council for Energy and Mineral Resources, Houston, TX. Rosenbaum, G., Lister, G.S. & Duboz, C., 2002. Reconstruction of the Kusky, T.M., Windley, B.F. & Zhai, M.-G., 2007. Tectonic evolution of the tectonic evolution of the western Mediterranean since the Oligocene, J. North China Block: from orogen to craton to orogen, Geol. Soc. London, Virtual Explor., 8, 107–130. Spec. Publ., 280, 1–34. Royden, L.H. & Husson, L., 2009. Subduction with variations in slab buoy- Lallemand, S., Heuret, A. & Boutelier, D., 2005. On the relationships ancy: models and application to the Banda and Apennine systems, in between slab dip, back-arc stress, upper plate absolute motion, and Subduction Zone Geodynamics, pp. 35–45, eds. Lallemand, S. & Funi- crustal nature in subduction zones, Geochem. Geophys. Geosyst., 6, ciello, F., Springer, Berlin. doi:10.1029/2005GC000917. Schettino, A. & Turco, E., 2006. Plate kinematics of the Western Mediter- Li, Z.-X., 1994. Collision between the North and South China blocks: a ranean region during the Oligocene and Early Miocene, Geophys. J. Int., crustal-detachment model for satuting in the region east of the Tanlu 166, 1398–1423. fault, Geology, 22, 739–742. Schuth, S., Muenker, C., Koenig, S., Qopoto, C., Basi, S., Garbe- Liati, A., Gebauer, D. & Wysoczanski, R., 2002. U-Pb SHRIMP-dating of Schoenberg, D. & Ballhaus, C., 2009. Petrogenesis of Lavas along the zircon domains from UHP garnet-rich mafic rocks and late pegmatoids Solomon Island Arc, SW Pacific: coupling of compositional variations in the Rhodope zone (N Greece); evidence for Early Cretaceous crys- and subduction zone geometry, J. Petrol., 59, 781–811. tallisation and Late Cretaceous metamorphism, Chem. Geol., 184, 281– Serpelloni, E., Vannucci, G., Pondrelli, S., Argnani, A., Casula, G., Anzidel, 299. M., Baldi, P., & Gasperini, P., 2007. Kinematics of the Western Africa- Mann, P. & Taira, A., 2004. Global tectonic significance of the Solomon Eurasia plate boundary from focal mechanisms and GPS data, Geophys. Island and Ontong Java Plateau convergent zone, Tectonophysics, 389, J. Int., 169, 1180–1200. 137–190. Shinohara, M., Suyehiro, K. & Murayama, T., 2003. Microearthquake seis- Mattauer, M., 1986. Intracontinental subduction, crust-mantle decollement micity in relation to double convergence around the Solomon Islands and crustal-stacking wedge in the Himalayas and other collision belts, arc by ocean-bottom seismometer observation, Geophys. J. Int., 153, Geol. Soc. Lond. Spec. Publ., 19, 37–50. 691–698.

C 2011 The Authors, GJI, 187, 1151–1174 Geophysical Journal International C 2011 RAS 1170 M. Baes, R. Govers and R. Wortel

Silver, E.A., Reed, D. & McCaffrey, R., 1983. Back arc thrusting in the southwest Pacific, 100–0 Ma, Proc. Ocean Drill. Program, Sci. Results, eastern Sunda arc, Indonesia: a consequence of arc-continent collision, J. 130, 697–710. geophys. Res., 88, 7429–7448. Ye, K., Cong, B. & Ye, D., 2000. The possible subduction of continental Singh, R.N. & Negi, J.G., 1982. High Moho temperature in the Indian shield, material to depth greater than 200 km, Nature, 407, 734–736. Tectonophysics, 82, 299–306. Zheng, J.P. et al., 2010. Tectonic affinity of the west Qinling ter- Spakman, W. & Hall, R., 2010. Surface deformation and slab-mantle inter- rane (central China): North China or Yangtze? Tectonics, 29, action during Banda arc subduction rollback, Nat. Geosci., 3, 562–566. doi:10.1029/2008TC002428. Spakman, W. & Wortel, M.J.R., 2004. Tomographic view on Western Mediterranean geodynamics, in The TRANSMED Atlas: The Mediter- ranean Region from Crust to Mantle, pp. 31–52, eds Gavazza, W., Roure, APPENDIX A F.M., Spakman, W., Stampfli, G.M. & Ziegler, P.A., Springer Verlag, Berlin. A1 Model M4 Stacey, F.D., 1977. A thermal model of the Earth, Phys. Earth planet. Inter., 15, 341–348. In the reference model, the rheology of the subducting continental Stampfli, G.M. & Borel, G.D., 2004. The TRANSMED transects in space crust is taken to be aplite (Freed & Burgmann 2004). In model M4,

and time: constraints on the paleotectonic evolution of the Mediterranean we investigate the sensitivity of our reference model to the assumed Downloaded from https://academic.oup.com/gji/article/187/3/1151/612017 by guest on 29 September 2021 domain, in The TRANSMED Atlas: The Mediterranean Region from crustal rheology. Model M4 has a set up similar to that of model M1, Crust to Mantle, pp. 53–76, eds Gavazza, W., Roure, F.M., Spakman, except that a stronger continental crust with anorthosite rheology W., Stampfli, G.M. & Ziegler, P.A., Springer Verlag, Berlin. (Freed & Burgmann 2004) has been adopted for the subducting Stich, D., Serpelloni, E., Mancilla, F.L. & Morales, J., 2006. Kinematics of continental lithosphere. The results of this model are shown in the Iberia-Maghreb plate contact from seismic moment tensors and GPS Figs A1(Aa–Ad). They are similar to those for model M1. The only observations, Tectonophysics, 426, 295–317. difference is that in model M4, there is almost no shear motion at Tang, J.-C. & Chemenda, A.I., 2000. Numerical modelling of arc-continent the base of the subducting continental crust, as a consequence of collision: application to Taiwan, Tectonophysics, 325, 23–42. Toth, J. & Gurnis, M., 1998. Dynamics of subduction initiation at preexisting the strong coupling between the crust and the lithospheric mantle. fault zones, J. geophys. Res., 103, 18 053–18 067. This experiment indicates that higher coupling between continental Trop, J.M. & Ridgway, K.D., 2007. Mesozoic and Cenozoic tectonic groth crust and the lithospheric mantle has no effect on the subduction of southern Alaska: a sedimentary basin perspective, Geol. Soc. Am. Spec. polarity reversal following arc–continent collision. Paper, 431, 55–94. Ulmer, P., 2001. Partial melting in the mantle wedgethe role of H2O in the A2 Model M6 genesis of mantle-derived arc-related magmas, Phys. Earth planet. Inter., 127, 215–232. Here, we investigate the effect of Moho temperature on the defor- Umhoefer, P.J., 2003. A model for the North America Cordillera in the mation pattern following continental collision. Model M6 is the Early Cretaceous: tectonic escape related to arc collision of the Guerrero same as model M1 except that the ratio of the viscosity of the lower terrane and a change in North America plate motion, in Tectonic Evolution continental crust to the viscosity of the upper lithospheric mantle of Northwestern Mexico and the South-Western USA, GSA Special Paper is 0.005. Results of model M6 (Figs A1Ba–Bd) show that a shear Vol. 374, pp. 117–134, eds Johnson, S.E., Paterson, S.R., Fletcher, J.M., zone along the arc/backarc boundary forms after 600 kyr (200 kyr Girty, G.H., Kimbrough, D.I., and Martin-Barajas, A., Geological Society of America, Boulder, CO. later than in model M1). The reason for this higher net conver- Van de Zedde, D.M.A. & Wortel, M.J.R., 2001. Shallow slab detachment gence is that in model M6 the convergence rate is partially taken as a transient source of heat at midlithospheric depths, Tectonics, 20, by intraplate deformation of the subducting plate. Vertical surface 868–882. displacement of model M6 (Fig. A1Bb) shows that uplift of the Van den Beukel, J., 1992. Some thermomechanical aspects of the subduction subducting plate is less than that in model M1, which is due to the of continental lithosphere, Tectonics, 11, 316–329. intraplate deformation of the subducting plate. Van der Hilst, R.D., Engdahl, E.R. & Spakman, W., 1993. Tomographic Similar to model M1, in model M6 the weakest zone is the mantle inversion of P and pP data for aspherical mantle structure below the wedge, which leads to subduction polarity reversal. Our experiments northwest Pacific region, Geophys. J. Int., 115, 264–302. with different Moho temperatures show that the lower continental Van Hinsbergen, D.J.J., Hafkenscheid, E., Spakman, W., Meulenkamp, J.E. crust becomes the dominant weakness zone if the Moho temperature & Wortel, R., 2005. Nappe stacking resulting from subduction of oceanic ◦ and continental lithosphere below Greece, Geology, 33, 325–328. is higher than 1000 C, which is an unlikely case in nature. These Van Hunen, J. & Allen, M.B., 2011. Continental collision and slab break-off: results suggest that the Moho temperature of incoming continental a comparison of 3-D numerical models with observations, Earth planet. lithosphere does not affect the deformation pattern following an Sci. Lett., 302, 27–37. arc–continent collision. Van Keken, P.E., 2003. The structure and dynamics of the mantle wedge, Earth planet. Sci. Lett., 215, 323–338. A3 Model M7 Wada, I., Wang, K., He, J. & Hyndman, R.D., 2008. Weakening of the subduction interface and its effects on surface heat flow, slab de- To investigate the effect of density of the subducted continental hydration, and mantle wedge serpentinization, J. geophys. Res., 113, crust on the results, we run a model similar to model M1 but with doi:10.1029/2007JB005190. an average density of 2900 kg m−3 (instead of 2800 kg m−3 in the Wells, R.E., 1989. Origin of the oceanic basalt basement of the Solomon model M1) for the continental crust. Results of this model are shown Islands arc and its relationship to the Ontong Java Plateau-insights from in Figs A1(Ca)–(Cd). The main difference between results of this Cenozoic plate motion models, Tectonophysics, 165, 219–235. Wong A Ton, S.Y.M.& Wortel, M.J.R., 1997. Slab detachment in continental model and those of model M1 is that here, there is less uplift on the collision zones: an analysis of controlling parameters, Geophys. Res. Lett., subducting plate and on the arc close to the trench (Fig. A1Cb). This 24, 2095–2098. is a direct consequence of the entrance of less positively buoyant Wortel, M.J.R. & Spakman, W., 2000. Subduction and slab detachment in crust into the trench. This experiment indicates that the density the Mediterranean-Carpathian region, Science, 290, 1910–1917. of incoming continental blocks or fragments does not have a very Yan, C.Y. & Kroenke, L.W., 1993. A plate-tectonic reconstruction of the significant effect on the subduction polarity reversal.

C 2011 The Authors, GJI, 187, 1151–1174 Geophysical Journal International C 2011 RAS Lithospheric responses to collision 1171 Downloaded from https://academic.oup.com/gji/article/187/3/1151/612017 by guest on 29 September 2021

Figure A1. (Aa–Ad) Results of model M4. (Ba–Bd) Results of model M6. (Ca–Cd) Results of model M7. (Da–Dd) Results of model M8. (Ea-Ed) Results of model M9. (Fa-Fd) Results of model M10. (Ga-Gd) Results of model M11. (Ha-Hd) Results of model M12. (Ia-Id) Results of model M14. (Ja-Jd) Results of model M18. In all figures: (a) is the effective strain, and (b) is the shear strain at the indicated times. (c) is the vertical surface displacement for the indicated model at indicated time (solid curve), in comparison with that of model M1 at t = 400 kyr (dashed curve). (d) is the vertical surface displacement after removal of the effect of model spin-up at the indicated time. Black arrows represent the velocity field at the indicated times and the arrows of the colour bar of total shear strain show the sense of shear. The vector in the green box on the lower left side of the figure indicates the scale of velocity vectors.

C 2011 The Authors, GJI, 187, 1151–1174 Geophysical Journal International C 2011 RAS 1172 M. Baes, R. Govers and R. Wortel Downloaded from https://academic.oup.com/gji/article/187/3/1151/612017 by guest on 29 September 2021

Figure A1. (Continued.)

of strength contrast between the arc and backarc. Vertical surface A4 Model M8 displacement (Fig. A1Db) shows more uplift on the arc close to Model M8 differs from model M1 in having a stronger continental the trench and less uplift near the newly developed shear zone, crust in the arc [with anorthosite rheology (Freed & Burgmann compared to those of model M1. The higher uplift on the arc near 2004)]. In this model, a shear zone develops along the arc/backarc the trench is attributed to the suction force associated with sinking boundary after 550 kyr of convergence—150 kyr later than in model detached slab which decreases as the detached slab sinks to greater M1 (Figs A1Da–Dd). The requirement of higher net convergence depths. The lower uplift on the arc near the shear zone can be to rupture the lithosphere in this model is due to the reduction associated with the strength of the arc, as the stronger plate deflects

C 2011 The Authors, GJI, 187, 1151–1174 Geophysical Journal International C 2011 RAS Lithospheric responses to collision 1173 Downloaded from https://academic.oup.com/gji/article/187/3/1151/612017 by guest on 29 September 2021

Figure A1. (Continued.) less than the weaker one. This experiment suggests that the strength arc/backarc boundary after 400 kyr (Figs A1Ga and Gc). The hori- of the arc influences the temporal evolution of subduction polarity zontal velocities within the arc are higher than those in model M1, reversal. suggesting that a wider channel results in higher corner flow veloci- ties in the mantle wedge. Vertical surface displacement (Fig. A1Gb) shows higher uplift on the arc as a result of higher upward motion A5 Model M9 in the mantle wedge. Similar to model M1, the mantle wedge is the Model M9 is similar to model M3 except that in this model, the weakest zone, leading to subduction polarity reversal. These results arc has a stronger continental crust [with anorthosite rheology suggest that in the competition between the channel strength and (Freed & Burgmann 2004)]. Results of this model are shown in weak mantle wedge, the latter is the weakest. Figs A1(Ea)–(Ed). These results are similar to those of model M3, except that in this model, no shear zone develops along the arc/backarc boundary, as a result of reduction in the strength con- trast between the arc and backarc. This experiment indicates that A8 Model M12 when the response to collision is delamination and back stepping, Model M12 has a similar set up as model M1, except that here development of a back-thrust depends on the strength contrast be- the channel dip angle has increased to 55◦. In this model, the tween the arc and backarc. trench/backarc distance is kept the same as in model M1 (250 km). A shear zone develops along the arc/backarc boundary after 500 kyr A6 Model M10 of convergence (Fig. A1Hc). The requirement of higher net conver- gence for the development of a shear zone in this model (compared Model M10 differs from model M5 in having a stronger continental to model M1) is due to the suction force associated with the sink- crust in the arc [anorthosite rheology (Freed & Burgmann 2004)]. ing detached slab which does not act efficiently in the arc/backarc Results of this model (Figs A1Fa–Fd) are similar to those of model boundary in this model. M5, except that the shear zone along the arc/backarc boundary forms In this model, uplift on the subducting plate is higher than that at higher net convergence. This suggests that a stronger continental in model M1 (Fig. A1Hb), while the maximum elevation on the crust in the arc reduces the strength contrast between the arc and arc is lower. Higher uplift on the subducting plate is due to the backarc, which leads to a delay in formation of the back-thrust on steeper channel which increases the effect of buoyancy forces on the overriding plate following arc–continent collision. the surface deformation. The low uplift on the arc is due to the suction force associated with the sinking detached slab. This ex- periment denotes that the channel dip angle can affect the temporal A7 Model M11 evolution of subduction polarity reversal if the suction force associ- In model M11, we increase the width of channel from 8 km to 16 ated with detached slab does not effectively act on the arc/backarc km. Similar to model M1, a dextral shear zone develops along the boundary.

C 2011 The Authors, GJI, 187, 1151–1174 Geophysical Journal International C 2011 RAS 1174 M. Baes, R. Govers and R. Wortel

of model M18 are shown in Figs A1(Ja)–(Jd). These results are A9 Model M14 similar to those of model M1. The only difference is that in this In model M14, we investigate the effect of age (strength) of the model, the detached slab collapses towards the left (see the direction oceanic lithosphere by employing a younger oceanic lithosphere of the velocities within the detached slab in Figs A1(Ja) and (Jd). (with an age of 25 Myr) in the backarc. A shear zone develops As a result of westward collapse of the detached slab the surface along the arc/backarc boundary after 300 kyr, which is 100 kyr depression, which develops along the arc/backarc boundary, moves earlier than in model M1 (Figs A1Ia and Ic). The requirement slightly towards the left (Fig. A1Jb). This experiment suggests that of less net convergence is due to the higher temperature along the the slab stagnant at the 660 km discontinuity has no significant arc/backarc boundary, which facilitates rupturing of the lithosphere. effect on subduction polarity reversal. The main difference between topography of models M1 and M14 (Fig. A1Ib) is in the bathymetric depth of oceanic lithosphere in the backarc, which is less in model M14 due to the age-dependence APPENDIX B of ocean depth. The results of this experiment indicate that one of Deformation localization in our experiments depends on the weak- the parameters affecting the net convergence required to change ening parameters a and b (see expression 3 in Section 4). Here, we Downloaded from https://academic.oup.com/gji/article/187/3/1151/612017 by guest on 29 September 2021 the subduction polarity is the age of the oceanic lithosphere in the present the results of models aimed at assessing the sensitivity to backarc. the plastic weakening parameters a and b. Fig. B1(K) shows the logarithm of plastic strain rate at t = 400 Kyr for a model similar to model M1 but with a = 0.5 (instead of a = 0.1). Compared to A10 Model M18 model M1 (Fig. 3b) the zone of plastic failure along the arc/backarc A phase transition and a viscosity contrast at the depth of 660 km boundary is shorter. This is because an increase in a leads to a leads to the deflection and accumulation of subducting slab material reduction in plastic strain softening. Increasing of b has a similar within the transition zone (Van der Hilst et al. 1993, and references effect (Fig. B1L); with increasing b plastic strain weakening occurs therein). To study this effect, we run a model (M18) similar to at higher strains (see the expression 3 in Section 4). Therefore, plas- model M1 but with a long detached slab which extends from the tic failure of the entire lithosphere requires higher net convergence, depth of 140 to 660 km (bottom boundary of the model). Results compared to a model with lower b.

Figure B1. Logarithm of plastic strain rate of a model similar to the reference model but (K) with higher a and (L) with higher b (see the expression (3) in Section 4 for the definition of these parameters).

C 2011 The Authors, GJI, 187, 1151–1174 Geophysical Journal International C 2011 RAS