Mantle Structure and Tectonic Evolution of the Region North and East of Australia

Geol. Soc. Australia Spec. Publ. 22, and Geol. Soc. America Spec. Pap. 372 (2003), 361–381

Mantle structure and tectonic evolution of the region north and east of Australia

R. HALL1* AND W. SPAKMAN2

1 SE Asia Research Group, Department of Geology, Royal Holloway University of London, Egham, Surrey TW20 0EX, UK. 2 Vening Meinesz Research School of Geodynamics, Faculty of Earth Sciences, Universiteit Utrecht, Budapestlaan 4, 3584CD Utrecht, Netherlands.

Tomographic images of the mantle beneath the region extending from the Molucca Sea eastward to Tonga, and from the Australian craton north into the Pacific, reveal a number of distinctive high seismic-velocity anomalies. The anomalies can be interpreted as subducted slabs and the positions of the slabs can be compared to predictions made by tectonic models for the region. Several strong anomalies are due to present-day subduction and the slab lengths and positions are consistent with Neogene subduction at the Tonga and the New Hebrides Trenches, where the anomalies suggest rapid rollback of subduction hinges since about 10 Ma, and beneath the New Britain and Halmahera Arcs. There are several generally flat-lying deeper anomalies which are not related to present subduction. Beneath the Bird’s Head and Arafura Sea is an anomaly which we interpret to be the result of north-dipping subduction beneath the Philippines–Halmahera Arc between 45 and 25 Ma. A very large anomaly, which extends from the Papuan peninsula to the New Hebrides and from the Solomon Islands to the east Australian margin, is interpreted as the result of south-dipping subduction beneath the Melanesian Arc between 45 and 25 Ma. Our interpretation implies that a flat-lying slab can survive for many tens of millions of years at the bottom of the upper mantle. There is a huge anomaly in the lower mantle which extends from beneath the Gulf of Carpentaria to Papua. We suggest this is a slab subducted before 45 Ma, which may be correlated with a Cretaceous slab beneath the Australian–Antarctic Discordance or an early Cenozoic slab subducted north of Australia. The anomaly is located above the position where there must have been a change in polarity in subduction at the boundary between the north- and south-dipping subduction zones north of Australia between 45 and 25 Ma. All of these have been overridden by Australia since 25 Ma. One subduction system predicted by the tectonic models, the Marumuni Arc of Papua New Guinea, is not seen on the tomographic images. KEY WORDS: island arcs, Melanesia, New Guinea, plate tectonics, subduction, tomography.

INTRODUCTION plate-convergence zones. In general terms, subducted lithosphere produces a strong temperature anomaly in the The northern Australia plate boundary (Figure 1) is a com- upper mantle (McKenzie 1970) which causes slabs to be plex and actively deforming region which includes some of seismologically detectable as regions with relatively fast the fastest relative plate motions on Earth (Bevis et al. seismic wave speeds. With seismic tomography the pre- 1995; Tregoning et al. 1998). Present-day plate motions served record of former plate convergence can be made vis- offer some insight into the recent history of the region but ible, albeit within spatial resolution limits. Thus, even a cursory study of the region shows that present tomography models provide an additional and independent motions provide only a snapshot of the tectonic history. source of information on tectonic evolution which can be Plate motion models indicate that Australia has moved exploited to test tectonic reconstructions (de Jonge et al. rapidly north since the Eocene and that this movement has 1994; Bunge & Grand 2000; Hafkenscheid et al. 2001). been accompanied by subduction and collision events. To Conversely, plate-tectonic models of a region can help in reconstruct the region in the past requires an interpretation developing new interpretations of imaged mantle structure of the evidence remaining and the nature of subduction (Richards & Engebretsen 1992; van der Hilst 1995; van der and collision means that some of the evidence is destroyed. Voo 1999a, b; Wortel & Spakman 2000) The comparison Thus, there are many different tectonic models and distin- between tectonic and tomographic models often implies guishing between them can be difficult. new constraints on tectonic evolution, which, however, rely Tomographic methods can provide new information on a correct interpretation of tomographic results. which can help to test and improve tectonic models. The interpretation of anomalously fast regions in terms Implicitly tectonic reconstructions predict where lithos- of a subducted slab is straightforward if the slab is seismi- phere has been subducted through geological time and consequently how much lithosphere has been consumed in * Corresponding author: [email protected] 362 R. Hall and W. Spakman

Figure 1 Digital elevation model showing principal surface features of the northern Australian plate margin based on Smith & Sandwell (1997) showing their bathymetry combined with topography from GTOPO30. cally active. However, interpretation of aseismic regions where plate-tectonic reconstructions come into play and with relatively fast seismic wave speeds as (remnants of) comparing predictions of such reconstructions with imaged subducted slabs is generally problematical because thermal mantle structure is at present the most powerful tool avail- and compositional heterogeneity may be caused by other able for the interpretation of tomographic models and for mantle processes/structure not associated with subduction. testing the reconstructions. Tomography provides a snapshot of mantle structure and In this paper we discuss tomography and tectonic inter- dynamics from which the space–time evolution of subduc- pretations and compare predictions of subduction with tion cannot be derived without invoking other information. imaged mantle structure of a region extending from the For instance, dynamic modelling (Zhong & Gurnis 1995) Molucca Sea eastward to Tonga, and from the Australian predicts that, depending on a number of factors, slab mor- craton north into the Pacific, in order to understand the phology and angle of subduction may change during the evolution of the northern Australian plate margin during subduction process. Because of varying subduction rates, the Cenozoic. Primarily, we invoke the tectonic model for variable surface area of an oceanic basin subducted, and the interpretation of the tomographic results which exhibit possible lateral variation of mantle viscosity on a global several enigmatic, spatially large, wave-speed anomalies at scale, the correspondence between the depth of subduction great depth in the mantle. We have deliberately avoided and time a slab takes to reach that depth may not be sim- extending the region of interest further west to the North ple. Furthermore, the plate-like character of subduction in West Shelf and east Indonesia because it is too compli- the upper mantle may be completely lost in the lower man- cated and will require a separate discussion. tle. Interpretation is additionally complicated by a variety of possible imaging errors some of which are extremely difficult to identify. All these factors may complicate interpre- PRESENT-DAY TECTONIC SETTING tations of high wave-speed anomalies in the deeper mantle, in particular in regions where the correspondence between Currently there is a long and complex boundary separating present-day surface tectonics and deeper mantle structure the Australian Plate from oceanic plates of the Pacific is lost as a result of tens to hundreds of millions of years of (Figure 2) which is marked by a zone of shallow seismicity plate-tectonic evolution and mantle convection. This is stretching from the Bird’s Head of New Guinea to the Australian Plate tomography and tectonics 363

Figure 2 Tectonic map of the same area as Figure 1 with principal geographical locations referred to in the text. Barbed lines are active subduction zones and double lines are active spreading centres. Other lines are major faults. Small black filled triangles are volcanoes from the Smithsonian database (http://www.nmnh.si.edu/gvp/), and bathymetry is from the GEBCO digital atlas (IOC et al. 1997). Bathymetric contours are at 200, 2000, 4000 and 6000 m.

Tonga–Kermadec Arc (Figure 3). To the north of this boundaries to the east in the Solomons, where the Pacific and ary are the Philippine Sea, the Caroline and Pacific Plates. Australian Plates are moving past one another, and to the To the south is the Australian Plate but within its margins west into a diffuse zone within the northern New Guinea and in the boundary zone with the Pacific are a number of margin which continues west to the Bird’s Head. GPS mea- microplates. Johnson and Molnar (1972) identified many of surements suggest that the Bird’s Head is now moving with these and the principal tectonic features of this boundary the Pacific Plate and therefore the Australia–Pacific bound- and later workers have identified additional small plates, ary must pass through New Guinea south of the Bird’s small spreading centres and backarc basins within this Head to link into the complexities of the Banda Sea and complex zone. southeast Asia. At the eastern edge of the Australian Plate (Figure 2) the At the present day the plate tectonics of the region is Pacific Plate is subducting west beneath the clearly complex and to some extent confused since in some Tonga–Kermadec Arc, behind which are the spreading cen- areas there is no clear connection between plate bound- tres of the Lau Basin. Further west is the New Hebrides Arc aries. For example, at the east end of the Manus spreading where subduction is in almost the opposite direction, centre (Figure 4) the link through New Ireland and into the towards the northeast. In between the New Hebrides and Solomons is a complex zone where GPS measurements the Tonga Arcs are the spreading centres of the North Fiji (Tregoning et al. 1998) indicate very high rates of Basin. In the sector between the Solomons and eastern Pacific–Australia motion but where there are no easily New Guinea subduction is essentially north-directed, at a identifiable major faults on which the motion has occurred. high rate beneath the New Britain Arc and at a lower rate In the Bismarck Sea region and to the south there are sev- beneath the Solomons. To the south of this subduction eral bathymetric troughs which are interpreted as active zone, within the Australian Plate, is the actively spreading and inactive trenches. The New Britain Trench has clear Woodlark Basin (Figure 4). To the north of the New Britain seismicity down to several hundred kilometres but Arc is the Bismarck Sea where the spreading centre of the although the Manus Trench, the Trobriand Trough and the Manus Basin links in a complex way to strike-slip bound- Moresby–Pocklington Trough may have some deformation 364 R. Hall and W. Spakman

Figure 3 Seismicity of the region selected from the data set of Engdahl et al. (1998) used in the tomographic modelling. The maps show the same areas as the tomographic images of Figure 6. Top: hypocentres between 0 and 100 km. Bottom: hypocentres between 500 and 700 km. Grey lines, major plate boundaries; black lines, great circle segments of the tomographic slices presented in Figure 9. Australian Plate tomography and tectonics 365

Figure 4 Geographical features of the Bismarck Sea and Woodlark Basin and surrounding regions. Barbed lines are active subduction zones and double lines are active spreading centres. Other lines are major faults. Small black filled triangles are volcanoes from the Smithsonian database (http://www.nmnh.si.edu/gvp/), and bathymetry is from the GEBCO digital atlas (IOC et al. 1997). Bathymetric contours are at 200, 2000, 4000 and 6000 m. associated with them, and may have been trenches in the been proposed and attempt to identify aspects of these past, they appear not to be active plate boundaries at the models which could distinguish them using tomographic present day. There is a similar confused situation in west- results. ern New Guinea and the Bird’s Head region (Figure 2) where to the north of the Bird’s Head there is an active Cenozoic tectonics spreading centre in the Ayu Trough between the Philippine Sea Plate and the Caroline Plate. At the south end of this Plate-tectonic models for the region fall into three groups. spreading centre there are no clear plate boundaries. The First, there are some regional plate models which show the New Guinea Trench is a relatively shallow and sediment- major plates of southeast Asia, the western Pacific and filled feature which is much shallower than a normal Australia although few cover the entire region of interest in oceanic trench and to the south of it is a poorly defined, this paper from New Guinea to the Tonga–Kermadec Arc broadly south-dipping, zone of diffuse seismicity which (Crook & Belbin 1978; Hamilton 1979; Kroenke 1984; extends only to depths of 100–200 km. The region south of Wells 1989; Jolivet et al. 1989; Rangin et al. 1990; Smith the Ayu Trough appears to be a wide zone of intraplate 1990; Daly et al. 1991; Yan & Kroenke 1993; Lee & Lawver deformation without a simple clear plate boundary. The 1995; Hall 1996, 1997, 1998, 2002). Second, there are complexity of present tectonics in part reflects very rapid many local plate models which deal with small sections of motion along the Pacific–Australia plate boundary, its vari- the Australian margin (Jaques & Robinson 1977; Taylor able geometry from east to west, accommodated for exam- 1979; Johnson & Jaques 1980; Cooper & Taylor 1987; ple by the creation of new plates or by short-lived Pigram & Davies 1987; Hill & Hegarty 1987; Hill et al. subduction of small plates, and intra-plate deformation. 1993; Struckmeyer et al. 1993; Abbott 1995; Musgrave & Firth 1999; Hill & Raza 1999; Charlton 2000; Taylor et al. 2000). Finally, there are many plate-tectonic cartoons CENOZOIC TECTONIC MODELS which typically show cross-sections or maps of subduction zones at different stages, but which do not give a complete In order to interpret the tomographic images of the region it history (Johnson 1976; Falvey 1978; Falvey & Pritchard is helpful to have tectonic models to which the images can 1982; Ridgeway 1987; Richards et al. 1990; Benes et al. be compared. Below we summarise the features of different 1994; Petterson et al. 1997; Monnier et al. 1999; Weiler & models of the Cenozoic tectonic development that have Coe 2000). Here we summarise the key features of different 366 R. Hall and W. Spakman groups of models and divide the Australian margin into two ceased and subduction transferred to other locations. The major segments. The first includes New Guinea and main difference between models is in the proposed site of extends to the east of the Papuan peninsula towards the initiation of the arc, and the timing of arc collision with the Solomons, and the second extends from the Solomons east- Ontong Java Plateau. Some models suggest that the wards to the Tonga–Kermadec Arc. We call these the New Melanesian Arc was rifted away from the Australian conti- Guinea segment and the Melanesian segment. nental margin by southward-subduction in the early Cenozoic, whereas others suggest the arc was initiated by New Guinea segment southward-subduction in an intra-oceanic setting north of the Australian margin. Initiation of subduction at the In the New Guinea segment models can essentially be Australian margin would have led to formation of a large divided into two. The majority of models advocate north- marginal basin behind the Melanesian Arc and implies very ward-subduction of oceanic crust north of Australia begin- significant rollback of the subduction hinge during the ning in the early Cenozoic leading to collision of the Palaeogene. The alternative intra-oceanic arc model Australian margin with a south-facing arc in the late requires that an ocean existed before subduction was initi- Cenozoic. The second category of models advocates that ated. In both models almost all the ocean crust postulated the northern Australian margin was an active margin and was subducted during the Neogene. The intra-oceanic arc there was southward-subduction beneath this margin models imply this crust was Mesozoic – Early Eocene before the active north Australia margin collided with an whereas the rollback models imply it was arc. The evolution of the New Guinea margin is discussed Eocene–Oligocene. in more detail by Hall (2002) and Hill & Hall (2003). Most models accept that arc–plateau collision took Many models show a single arc–continent collision. place in the Miocene but may have occurred over an Variations include the opening of marginal basins within extended period. The early stage is thought to have short- the Australian margin and their subsequent subduction, ened part of the leading edge of the Ontong Java Plateau and multiple subduction zones. In some models these sub- margin and is often referred to as a soft collision. When duction zones have a consistent polarity, for example, contraction of the Ontong Java Plateau could no longer many models suggest more than one subduction zone continue there was thrusting of the Solomons Arc over the north of Australia dipping north, but in other models there plateau during a hard collision phase. It is generally are subduction zones which have opposing polarities. More accepted that the Solomons were transferred to the Pacific complex models propose or imply multiple arc–continent Plate or partially coupled to the Pacific Plate in the Early to collisions (Pigram & Davies 1987; Daly et al. 1991; Lee & Middle Miocene and therefore the boundary ceased to be a Lawver 1995) with accretion of arc fragments to the north- subduction zone and instead became a broadly strike-slip ern New Guinea margin at several periods during the fault zone with the possibility of later local subduction to Cenozoic. Suggested dates of these events also vary. the southwest or to the northeast on either side of the A common theme in many of the models, particularly Solomons. The most recent stages in the development of those which involve multiple subduction zones and multi- the Melanesian segment in the last 10–12 million years ple collisions, is the diachronous collision of a volcanic arc involved the initiation of northeast-dipping subduction with the northern New Guinea margin. This is generally beneath the New Hebrides Arc, the relatively rapid rotation agreed to have started earliest in the west and progressed of the New Hebrides Arc to form the North Fiji Basin, and eastwards to the present collision between the New significant rollback of the subduction hinge as the Pacific Britain–Finisterre Arc and New Guinea. The collision is Plate was subducted west beneath the Tonga–Kermadec commonly interpreted to have followed subduction to both Arc. north and south leaving beneath Papua a doubly vergent subducted oceanic slab (Cooper & Taylor 1987; Pegler et al. New Guinea and Melanesian boundary 1995) similar to the inverted U-shape of the Molucca Sea slab (Hamilton 1979; Cardwell et al. 1980; McCaffrey Because nearly all the reconstructions of the region focus 1982). The Solomon Sea is usually interpreted as the rem- on either the New Guinea or the Melanesian regions there nant of this ocean (Richards et al. 1990; Daly et al. 1991; has been little attention paid to the change of polarity in Lee & Lawver 1995). subduction which is implied at about the boundary between the New Guinea and the Melanesian segments. In Melanesian segment a few models (Lee & Lawver 1995; Jolivet et al. 1989) the change is shown schematically but the changing length of There are fewer models for the Melanesian segment (Crook the section between the north- and south-dipping subduc- & Belbin 1978; Kroenke 1984; Wells 1989; Smith 1990; Yan tion zones is not shown, nor is it clear exactly how this & Kroenke 1993; Hall 1998, 2002) and many are in cartoon region is to be interpreted. Falvey and Pritchard (1982) sug- cross-section form or deal with only parts of this segment gested that before 45 Ma there was southeast-dipping sub- (Falvey & Pritchard 1982; Ridgeway 1987; Petterson et al. duction beneath the New Britain – New Ireland Arc which 1997; Musgrave & Firth 1999; Taylor et al. 2000). There is was orthogonal to the Australian margin. The southeast- broad agreement that from the early Cenozoic there was dipping subduction bends round into the conventional south- or southwest-directed subduction beneath the southwest-dipping subduction beneath the Solomons. They Solomons and this continued until the Solomon Arc col- postulated rapid rotation of the New Britain – New Ireland lided with the Ontong Java Plateau. Pacific–Australia plate section of the arc at about 30 Ma followed by development convergence by subduction in the Solomons region then of northeast-dipping subduction to give the present config- Australian Plate tomography and tectonics 367

Figure 5 Tectonic reconstructions at 40 and 10 Ma (Hall 2002). At 40 Ma Australia was moving north and south-dipping subduction beneath the Melanesian Arc had just begun, whereas there was north-dipping subduction beneath the Philippines–Halmahera Arc east of Papua. At 10 Ma eastward subduction of the Molucca Sea had begun beneath Halmahera. The New Guinea terranes moved in a wide left-lateral strike-slip zone. Subduction of the Solomon Sea was underway beneath eastern New Guinea margin to form the Maramuni Arc. To the east of New Guinea subduction began on the east side of the Solomon Sea to initiate the New Hebrides Arc. 368 R. Hall and W. Spakman

Figure 6 Twelve depth images for the southwest Pacific/Australia region. The sections are taken from the global tomographic model of Bijwaard & Spakman (2000). Section depths are indicated in the lower left corner of each panel. Colour contouring depicts the anomalous P-wave speed relative to the average speed at depth. The reference model is ak135 (Kennett et al. 1995) Negative (positive) anomalies are coded in red (blue) colours and represent relatively slow (fast) regions with respect to the average ak135 propagation speeds. The limits of the contouring scale vary between –X% and +X% where X=3 (a), 2.5 (b), 1.5 (c–i), and 0.75 (j–l). Australian Plate tomography and tectonics 369 370 R. Hall and W. Spakman uration. During part of this interval the former southeast- cant subduction at the north Australian margin in Irian dipping subduction zone acted as a transform. Their model Jaya. Further east, Pacific–Australia convergence was is schematic but is the only model in which there is a sub- accommodated by development of new subduction zones: duction zone orthogonal to the Australian margin. there was southwest-dipping subduction beneath east Papua forming the Maramuni Arc and northeast-dipping The tectonic model subduction beneath the New Hebrides. The two opposed subduction systems were linked by a transform crossing the Hall (1997, 1998, 2002) has described a plate-tectonic Solomon Sea. model which covers the whole of the region of interest in When Maramuni subduction ceased the New Hebrides this paper, between the Molucca Sea and the Trench propagated west to start new north-directed subduc- Tonga–Kermadec Arc. It synthesises a range of data, from tion beneath the Solomons and New Britain Arcs. Slab-pull spreading histories obtained from small ocean basins to forces at the New Britain Trench then led to the formation of more qualitative information such as that obtained from the Woodlark spreading centre at the former Solomon Sea land geology. It therefore has features in common with transform. Very young Woodlark Basin lithosphere is now many of models outlined above but differs from all of them being subducted at the South Solomon trench. Throughout in postulating a mainly strike-slip plate-boundary zone in this period west-dipping subduction of the Pacific Plate con- northern New Guinea during the Neogene. Computer ani- tinued at the Tonga–Kermadec Trench with rapid rollback of mations of this model are in Hall (2002). Because we can the subduction hinge in the last 10 million years. examine the model at 1 million year steps for the whole region of interest we have used it to compare its predictions with features observed in tomographic images. The essen- SEISMIC TOMOGRAPHY MODELS tial features of this model are as follows. From the Mesozoic the north Australian margin was a The mantle structure of the entire region is only imaged in passive continental margin produced by Jurassic rifting. global mantle models. Recent developments in data analy- There was diachronous arc–continent collision of intra- sis and tomographic techniques have led to a new genera- Pacific arcs with this passive margin emplacing ophiolitic tion of global models that are capable of resolving smaller rocks in between New Guinea and New Caledonia regions details of mantle structure (van der Hilst et al. 1997; Grand in the Eocene. At about 45 Ma, after the collision, there was et al. 1997; Bijwaard et al. 1998; Ritsema et al. 1999; a plate reorganisation; new north-dipping subduction Megnin & Romanowicz 2000; Bijwaard & Spakman 2000). began about 2000 km north of Australia beneath the In particular, the P-wave-speed model of Bijwaard et al. Philippines–Halmahera Arc. At about 45 Ma new south- (1998), based on the dataset of Engdahl et al. (1998), ward subduction began at the northeast and east shows details of slab morphology comparable to those pre- Australian margin forming the Melanesian Arc between viously seen only in studies of smaller regions of Australia New Britain and Tonga (Figure 5). and the southwest Pacific (Abers & Roecker 1991; van der Subduction beneath the Philippines–Halmahera Arc Hilst 1995). continued until about 25 Ma. During this period the Imaging small details on a global mantle scale has been Philippines–Halmahera Arc remained in approximately the possible by using a special cell-parameterisation technique same position. In the Melanesian segment rollback of the which allows the variation of cell dimensions as a function subduction hinge between 45 and 25 Ma led to the forma- of the local degree of data density (Spakman & Bijwaard tion of a wide backarc basin from the Solomon Sea to the 2001). The smallest cells used have dimensions of 0.6° lat- South Fiji Basin. The subduction direction changed to erally and 35 km in depth. In this paper we use an improved southwest-dipping as the Melanesian Arc rotated north. tomographic model (Bijwaard & Spakman 2000) which was From about 40 Ma the Caroline Sea opened as a backarc obtained by incorporating in addition 3-D ray tracing in basin by rollback of the Pacific subduction hinge at the east- the tomographic analysis in an attempt to account for ern margin of the Philippine Sea Plate. The South Caroline (de-)focussing effects due to 3-D mantle structure. Arc was east of this backarc basin and was the site of for- The tomographic results for the region are displayed as mation of arc terranes now found in northern New Guinea. selected layer sections down to a depth of 1040 km in At about 25 Ma arc–continent collision occurred when Figure 6 and sensitivity tests for spatial resolution in Figure the Philippines–Halmahera Arc collided with the Australian 7. Figure 8 shows the principal positive anomalies on the margin and this terminated the north-dipping subduction. tomographic images of Figure 6 and some vertical depth The zone of collision stretched from Sulawesi to eastern sections are shown in Figure 9. New Guinea. At about the same time the Ontong Java Plateau collided with the Melanesian Arc. This led to ter- Model confidence mination of the southwest-dipping subduction system. After these collisions the arcs between the Philippines Sensitivity tests (Spakman & Nolet 1988; Bijwaard et al. and Melanesia became broadly a single system which 1998, Bijwaard & Spakman 2000) are extremely important rotated clockwise at the leading edge of the Pacific Plate, for the interpretation of the tomographic results as they accommodated by intra-plate subduction at the eastern indicate the scale or features which can be resolved and edge of the Philippine Sea Plate. The South Caroline Arc hence interpreted. Because of the importance of these tests, terranes, which are now found in north New Guinea, which have been conducted for a variety of models, we pro- moved along the north Australian margin in a complex vide below an explanation of them which we hope will sat- strike-slip zone (Figure 5). There was therefore no signifi- isfy both the non-technical and expert reader. Australian Plate tomography and tectonics 371

Seismic travel-time tomography is an inversion tech- 4.2° are well resolved (595 km). At greater depths (980 nique that produces a 3-D mantle model of seismic-wave and 1040 km) resolution is on the scale of 3–4°. At all speed from surface observations of the travel times of seis- depths it is not possible to resolve anything within the mic waves. For a variety of reasons (e.g. data errors) the region beneath the Pacific in the northwest part of the model can only be considered as an estimate of true Earth map. Generally, the information contained in Figure 7 structure. One of the major difficulties in tomography is the (and at other levels than those shown in Figure 7) is determination of model errors. Techniques exist to estimate extremely helpful in selecting those parts of the actual model errors for relatively small-size inverse problems tomographic model and the spatial scales for which inter- (Menke 1989), but these methods are not yet applicable to pretation can be attempted. a huge inverse problem of the type discussed here. For large inverse problems one has to resort to so-called sensitivity analysis with synthetic wave-speed models (Spakman & FEATURES VISIBLE IN TOMOGRAPHIC IMAGES Nolet 1988). First, a synthetic model is created, for example, one which contains block-shaped anomalies of alter- For the purposes of description and discussion we number nating sign, like a 3-dimensional checkerboard pattern. the principal positive anomalies on the tomographic images Next, the same seismic rays used in the actual data inver- (Figure 8). Several are directly associated with existing sub- sion are used to compute synthetic travel times from the duction (A1–A4). A number of more enigmatic wave-speed synthetic model. The last step is to invert the synthetic data anomalies (A5–A8) are found in the deeper parts of the in the same way as the real data inversion. This leads to an upper mantle and lower mantle. Here we first we identify estimate (the tomographic view) of the synthetic model. It the anomalies and in the next section we discuss their can be shown that the relation between the synthetic model interpretation. and its tomographic image is given by the same spatial res- A1. Positive P-wave speed anomalies associated with olution operator as for the true Earth and its tomographic subduction are readily identified in the Tonga–Kermadec image derived from real data. Thus, comparison of the syn- region where the slab is visible to depths of 1500 km. The thetic model and its tomographic image can lead to infer- slab has a flat-lying portion below Tonga in the north which ences of spatial resolution errors (image blurring or is seismically active (Figures 6g–i, 9h). Below the western smearing) that are equally applicable to the actual tomo- part of the flat-lying slab subduction can be followed into graphic problem. the lower mantle where it has a dominant north–south In Figure 7 we display examples of sensitivity analysis strike. The upper and lower mantle anomalies are well for our region. The synthetic models are constructed from resolved (Figure 7) and confirm the earlier findings of van isolated blocks (some with an irregular shape) with differ- der Hilst (1995). ent spatial dimensions and which have designated syn- A2. To the northwest, subduction at the New Hebrides thetic wave-speed anomalies of alternating sign. In Figure 7 Trench yields a fragmented image (Figure 6a–f), but the the outlines of these blocks are plotted with a black contour slab is well delineated by intermediate and deeper seismic- line. The colour coding depicts the tomographic image of ity (Figure 9g, h). To the northeast, below the North Fiji the synthetic model. If the synthetic model was perfectly Basin, a flat-lying positive anomaly in the upper mantle recovered each of the black outline volumes (areas on the corresponds to a region of deep seismicity (Figures 3b, 6g–i, layer plots shown) would contain either entirely red or 9g). Although weak positive anomalies are imaged below entirely blue and all the intervening area would be shaded 700 km poor spatial resolution does not warrant an inter- in the central neutral colour on the colour scale (pale pretation of slab penetration into the lower mantle below green). If the colour is blurred outside the contoured areas, the North Fiji Basin. and the colour ‘leaks out’ of the blocks this indicates that A3. Further to the northwest, below the Solomon the synthetic model is less well recovered. In regions where Islands segment of the plate boundary, tomographic evi- the tomographic image correlates well with the synthetic dence for an upper mantle slab is absent above 500 km in model the conclusion is often made that the spatial resolu- the central and southern part (Figures 6a–f, 9f). Positive tion is high and thus the tomographic image determined anomalies appear only below 500 km with seismic activity from the real data is also of high quality. There is a risk of in the southeast (Figure 3, bottom). Clear evidence for an this reasoning being wrong (Lévêque et al. 1993) but this upper mantle slab is found in the northwest part of the risk can be made small by conducting sensitivity tests with Solomon segment where slab geometry follows the cusp in a range of different block sizes. the plate boundary toward the New Britain Arc (Figures Figure 7 shows that spatial resolution is highly variable 6a–i, 9e). The subducting Solomon Sea slab is seismically in the mantle volume of interest. In the shallow mantle active down to 600 km and north dipping. Laterally, the details on the scale of 1.2° are detectable in the region of Solomon Sea slab suddenly terminates beneath central high seismicity close to the Australia–Pacific plate bound- New Guinea and no upper mantle slab is imaged below ary (200 km) but in other regions this scale of resolution west New Guinea (Figures 6a–g, 9c, d). is absent. However, at a similar depth (230 km) resolution A4. Still further to the west, subduction-related anom- on a scale of 2.4° is possible over much more of the alies are found in the upper mantle beneath Sulawesi and region. Similarly, near the lower part of the upper mantle Halmahera (Figure 6a–e). Subduction of the Molucca Sea (530 km) blocks on the scale of 2.4° are well resolved over below both Halmahera to the east and Sulawesi and the most of the central region close to the Australia–Pacific Sangihe Arc to the west shows a slab with a clear inverted plate boundary but, for example, beneath the Coral Sea U-shape (Figure 9a) as previously recognised by resolution is not so good although blocks on the scale of Widiyantoro and van der Hilst (1997). Below the Caroline 372 R. Hall and W. Spakman

Figure 7 Sensitivity ‘spike’-test results for assessing image quality. The six panels show inversion results of the retrieval of synthetic ‘spike’ models of anomalous seismic wave speed (Bijwaard & Spakman 2000) for some selected depths and for different spike (or block) size. See text for full explanation. The square-like black contour lines indicate the outline of single anomaly blocks with a –5% or +5% amplitude. Synthetic block-anomalies are in all directions surrounded by 0%-synthetic anomalies. This is different from the so-called checker-board test (Lévêque et al. 1993) and allows a much better detection of image smearing. The colour contouring shows the retrieval of the synthetic spike pattern. The contour limits are –2.5% and +2.5% which is a compromise between showing large-amplitude spike response and small-amplitude resolution artefacts. In the well-recovered areas many of the block amplitudes exceed the contouring limits but 100% recovery of spike amplitudes is rarely obtained. Image smearing is an expression of lack of spatial resolution which is probably small in the well-recovered regions and is large in the poorly and not-recovered regions. The first two panels (200, 230 km) show test results for spike sizes of 1.2 and 2.4° at the top of the upper mantle. The two panels in the centre (595, 530 km) show the spike response for 2.4 and 4.2° spikes near the bottom of the upper mantle. The last two panels (1040, 980 km) give the results for 3.0 and 4.2° spikes in the lower mantle. Notice that poor recovery of small blocks at a given depth does not imply poor recovery of larger blocks. Anomalous mantle structure with a spatial dimension of about 2.4° near 600 km depth cannot be detected below the eastern margin of Australia but structures with larger dimensions (4.2°) probably can be detected. Australian Plate tomography and tectonics 373

Plate this anomaly broadens with depth and appears to From the sensitivity test (Figure 7) we infer that the merge with a large anomaly which is continuous into the anomalies A5–A8 are spatially sufficiently resolved to war- lower mantle (Figure 9a). The Sangihe slab dips to the rant interpretation. northwest (Figure 9a) and its anomaly also apparently links with a deeper anomaly, which is part of a huge positive anomaly that underlies a large part of southeast Asia INTERPRETATION OF TOMOGRAPHIC FEATURES (Widiyantoro & van der Hilst 1997; Bijwaard et al. 1998; IN TERMS OF TECTONICS Rangin et al. 1999). A5. A prominent positive anomaly is imaged below the We briefly comment on anomalies associated with the Caroline Ridge (Figure 6g–k). It broadens with depth and is Australian Plate, then deal with anomalies clearly related to primarily present in the upper mantle (Figure 9a, b) except present-day subduction, and finally discuss other mantle for its northern limit which corresponds to the deep anomalies which we interpret in terms of older subduction. Mariana subduction. A6. Another extensive positive anomaly is found between Lithosphere under Australia depths of 500 and 700 km with an elongate shape extending from north of the Bird’s Head, beneath the Arafura Sea, to The tomographic model shows a positive anomaly beneath the Gulf of Carpentaria (Figures 6g–j, 9c, d). Towards the much of northern Australia (Figures 6a, b, 9c, d). There is north, this anomaly appears to merge with that associated however a significant variation in lithosphere velocities with the young subduction of the Halmahera slab (A4). from fast to slow from west to east New Guinea and from A7. An important positive anomaly is imaged below 500 west to east across Queensland. This trend is better imaged km and predominantly in the upper mantle which extends in surface-wave tomography models and interpreted as from the Papuan peninsula to the New Hebrides and from slower wave speed below the Palaeozoic fold belts of east- the Solomon Islands to the east Australian margin (Figure ern Australia and increasing wave speeds toward the west 6g–i). Because of the large lateral extent of this flat anomaly below the Proterozoic and Archaean shields (Simons et al. (Figure 9f, h) it can be identified in many other global man- 1999; Debayle & Kennett 2000) tle models including those with a nominally lower spatial resolution (Ritsema et al. 1999; Megnin & Romanowicz 2000). Young and current subduction systems A8. There is a huge positive anomaly which stands out very clearly between 710 and about 1100 km and extends A1. At the Tonga Trench the Pacific Plate subducts west from beneath the Gulf of Carpentaria to Papua with a broadly at a high rate, decreasing southwards towards New north-northeast–south-southwest trend (Figures 6j–l, 9e). Zealand (Figure 9g, h). The age (>120 Ma) of the subducted

Figure 8 The anomalies discussed in the text. A1–A4 (green) are dipping anomalies in the upper mantle, A5–A7 (blue) are broadly flat anomalies near to the base of the upper mantle, and A8 (red) is a strong anomaly in the lower mantle. The anomaly labelled A1 and shown in red is that associated with the Tonga slab in the lower mantle and the outline is that identified on the 1040 km depth layer of Figure 6l. Grey lines, major plate boundaries; black lines, great circle segments of the tomographic slices presented in Figure 9. 374 R. Hall and W. Spakman

Figure 9 Eight vertical sections through the tomographic model of Bijwaard & Spakman (2000) to a depth of 1500 km. All cross-sections are taken along a great circle segment (straight line in each map) of 30°. Section coordinates (starting point, azimuth) are given in the map. Colours in the map denote the Earth’s topography (ETOPO5). Each map has a small inset showing the location of the section. As in Figures 6 and 7 colours in the sections denote the anomalous P-wave velocity structure. It should be noted that anomaly amplitudes vary with depth from many percent in the top of the mantle to typical peak values around 0.5% at 1000 km or deeper (Bijwaard & Spakman 2000). All sections are contoured between –1% and +1% and relatively strong colours in the uppermost mantle or weak colours in the lower mantle should not be interpreted as an indication of significance or insignificance of imaged structure. White dots in the sections represent hypocentres (taken from Engdahl et al. 1998) of earthquakes that occurred within 50 km of the section plane. Positions of sections are shown in Figures 6 and 8. Australian Plate tomography and tectonics 375

lithosphere and high rate of subduction (>10 cm/y) result in northern end of the Tonga–Kermadec Arc where at least a deep negatively buoyant slab of lithosphere. Van der Hilst 2100 km of slab is imaged, in contrast to much slower roll- (1995) has shown that there are distinctive high-velocity back at the southern end of the system where <1000 km of anomalies beneath the Tonga–Kermadec Arc which make slab is visible. Assuming a constant rate of subduction the subducted slab visible down to more than 1200 km. these dimensions suggest that the anomalies are due to Beneath the Tonga Arc the slab dips uniformly at about 50° subduction since 25 Ma and that most rollback occurred and becomes almost horizontal in the transition zone and since about 10 Ma (Figure 10). uppermost lower mantle (Figure 9h). It then sinks into the A2. Beneath the North Fiji Basin there are high velocity lower mantle several hundred kilometres further west. Van anomalies visible in the upper mantle (Figure 9g). The slab der Hilst (1995) suggested that the shape of the subducted dips steeply in a northeasterly direction and becomes slab is due to rapid rollback of the subduction hinge at the almost horizontal in the transition zone. The tomographic 376 R. Hall and W. Spakman

Figure 10 Tectonic reconstructions from 45 to 5 Ma (Hall 2002), omitting 25 Ma for layout reasons. The reconstructions use the Indian–Atlantic hotspot frame of Müller et al. (1993). On each map the present positions of the deep mantle anomalies (A6–A8) are shown in yellow. There is a good correspondence of the anomaly A6 with the predicted position of a slab produced by north-dipping subduction beneath the east Philippines–Halmahera Arc between 45 and 25 Ma. There is also a good correspondence between anomaly A7 and the predicted position of a slab produced by south-dipping subduction beneath the Melanesian Arc between 45 and 25 Ma. The younger anomalies are shown in pale green (A1–A5). Note in particular the deep part of the Tonga slab (anomaly A1 on the 1040 km layer) compared to the position of the anomaly in the upper mantle. These suggest an almost stationary trench from about 25 to 10 Ma and fast rollback since 10 Ma at the Tonga Trench. Australian Plate tomography and tectonics 377 results do not exclude slab penetration into the lower man- shaped Molucca Sea Plate is moving north in mantle with tle because of poor resolution. Beneath the northern part of Australia. On some sections the Molucca Sea Plate anom- the North Fiji Basin are deep earthquakes at 550–650 km alies appear to merge with deeper anomalies and Rangin et which have been attributed to events in the flat-lying slab al. (1999) postulated that both the Sangihe and Halmahera (Okal & Kirby 1998). The seismicity and tomography sug- slabs can be traced deeper into the mantle. They connected gest a similar interpretation to that of the Tonga subduction the Sangihe slab to a broad high-velocity zone beneath the with rapid rollback causing development of a kink and flat Celebes Sea at depths between 700 and 1000 km, and segment in the subducted slab. The tectonic model (Hall joined the Halmahera slab to another broad high-velocity 2002) predicts about 1000 km of subduction since 15 Ma zone beneath the Bird’s Head at depths between 400 and at the New Hebrides Trench (Figure 10). Assuming a con- 600 km. However, if this is the case there has been about stant subduction rate the orientation and length are consis- 3500 km of slab subducted (Rangin et al. 1999) which tent with a slab subducted at the New Hebrides Trench would be much greater than the amount of subduction pre- since about 12 Ma, and clockwise rollback of the subduc- dicted by the tectonic model (Hall 2002) or other models tion hinge during this period. (Jolivet et al. 1989; Rangin et al. 1990) and would place A3. Beneath the Solomons there is no seismic or tomo- Halmahera much further east than suggested on any recon- graphic evidence of a significant length of subducted slab structions of any age. In contrast, we therefore interpret (Figure 9f). Seismicity to depths of 200–300 km suggests these broad and flat anomalies as the remnants of a slab subduction locally to north and south on the south and subducted beneath the Philippines–Halmahera Arc north sides of the Solomons respectively. There are a few between 45 and 25 Ma (Figure 10). scattered events at depths greater than 500 km. At the western end of the Solomons the subducted slab of the New Older subduction systems Britain Arc becomes visible and is associated with seismicity to depths of 450 km (Figure 9h). Just east of the longi- A5. Centred on the Caroline Ridge and extending south tude of New Ireland there is a marked arcuate kink in the towards the Bismarck Sea is a broad high-velocity anomaly high-velocity anomaly at depths below 200 km which at depths between 560 and 710 km (Figure 9b, d). This is becomes smoother at greater depths (Figure 6a–f). This the southern end of a Pacific high-velocity anomaly trend- may be due to deformation of the slab in the mantle. In the ing approximately north–south which continues to the tectonic model (Hall 2002) the New Britain subduction is north beneath the eastern Philippine Sea Plate. We inter- interpreted to have propagated northwest from the New pret this anomaly to be the result of Pacific subduction Hebrides Trench after subduction ceased at the Maramuni beneath the eastern margin of Philippine Sea Plate since 25 Trench. An alternative possibility suggested by the tomog- Ma. This interpretation implies rollback of the Pacific sub- raphy is that subduction initiated beneath the New Britain duction hinge and overriding of the subducted slab by the Arc and propagated east with the eastern Solomon Sea Philippine Sea Plate as shown in the tectonic model (Hall remaining partly coupled to the Pacific. This would account 2002) (Figure 10). for the absence of seismicity and a high-velocity anomaly A6. At depths between 500 and 710 km is an extensive, beneath the Solomons. However, the absence of seismicity elongate high-velocity anomaly extending from north of the and anomaly beneath the Solomons may be due to the very Bird’s Head, beneath the Arafura Sea, to the Gulf of young age of the Woodlark Basin crust being subducted; Carpentaria (Figure 9a–d). Towards the north this anomaly south of New Georgia the Woodlark spreading centre is appears to merge with the young Halmahera slab anomaly being subducted and therefore the subducted slab may be east of the Molucca Sea. However, the Halmahera slab has hotter than the mantle into which it is sinking. North of the a clear consistent dip to the east and the anomaly below Woodlark Rise, where a high-velocity anomaly is very clear, 500 km appears to be separated from it and flat lying. The the subducted slab descending beneath Bougainville is sig- deeper anomaly also extends much further south. It also nificantly older, and is probably Oligocene (Figure 10). becomes deeper to the north, to the north of Halmahera. A4. Beneath the Molucca Sea region high-velocity We interpret this anomaly to be the result of northward anomalies are clearly visible to the west and east (Figure subduction between 45 and 25 Ma beneath the 9a) and these were interpreted by Rangin et al. (1999) as Philippines–Halmahera Arc north of Australia. According the Molucca Sea slab subducted during the Neogene. to the tectonic model (Hall 2002) this would have been the Seismicity shows the west-dipping Sangihe slab is identifi- subduction zone at which approximately 2000 km of able to at least 650 km and the east-dipping Halmahera oceanic lithosphere was subducted between 45 and 25 Ma slab to about 200 km (Cardwell et al. 1980; McCaffrey (Figure 10). The length of slab estimated from the tomo- 1982). Tomography suggests the Halmahera slab goes graphic cross-sections is of a similar order. deeper to at least 400 km (Rangin et al. 1999). Taking into A7. At depths from about 500 to 700 km there is an account the dip on the slabs the total length of subducted extensive high-velocity anomaly extending from the lithosphere is therefore about 1500 km, very close to the Papuan peninsula to the New Hebrides and from the length predicted by the tectonic model (Hall 2002) since Solomons to the east Australian margin (Figure 9f, h). this subduction began on the Sangihe side of the Molucca Below the Coral Sea and the eastern margin of Australia Sea at 25 Ma. As both the Sangihe and Halmahera slabs this anomaly is not well defined at scales of a few hundred are still connected to the surface, and the southern bound- kilometres. The segment under Queensland may not be ary of the Molucca Sea, which is the Philippine Sea – connected to the extensive anomaly that is imaged from Australia plate boundary, is also moving northward, the beneath the Papuan peninsula to the New Hebrides. We continuity of these slabs implies that the entire inverted U- interpret the latter anomaly to be the result of south- and 378 R. Hall and W. Spakman southwest-directed Pacific subduction at the Melanesian prefer the first or second hypotheses since to produce this Arc between 45 and 25 Ma. Ma. As noted above, the for- anomaly with slab penetration a more or less stationary mation of a large marginal basin behind the Melanesian subduction zone would probably be required. Arc implies very significant rollback of the subduction hinge during the Palaeogene. By analogy with the present- Subduction systems which are not seen day Tonga system, rapid rollback would produce an extensive area of flat-lying slab which would be expected to sink Besides identifying mantle anomalies with old subduction to the base of the mantle after subduction ceases. A large systems the tomography also offers the opportunity to dis- high-velocity anomaly is consistent with both the Yan and tinguish between models which postulate different subduc- Kroenke (1993) model and the models which predict signif- tion histories. As reviewed above, most authors postulate icant rollback although the amount is different in the two north-dipping subduction north of Australia during the models. Figure 10 shows there is excellent agreement Palaeogene. We see no evidence for the south-dipping sub- between the prediction of the Hall (2002) tectonic model duction suggested in some interpretations. If this had and the tomography. If this interpretation is correct, a flat- occurred there should be distinctive high-velocity anom- lying slab can survive for many tens of millions of years at alies beneath the craton of western Australia (Figure 9c, d). the bottom of the upper mantle. Slowly increasing buoy- Their absence and the presence of high-velocity anomalies ancy due to temperature increase may prevent slumping further north (anomaly A6) cause us to prefer north-dipping into the lower mantle. subduction with subducted slabs now overridden by A8. One of the clearest anomalies in the region is an Australia. extensive volume of low velocities with a north-north- We see no evidence of continuation of subduction at the east–south-southwest trend visible between 920 and 1040 New Britain Arc very far to the west (Figure 9c, d). This km which extends from beneath the Gulf of Carpentaria to implies the ‘doubly vergent system’ did not extend signifi- Papua (Figure 9d, e). Young subduction in the Tonga Arc cantly west of Papua New Guinea and there has been no has caused a slab to enter the lower mantle but elsewhere Neogene subduction beneath western New Guinea. This is in the region there is little evidence for an active subduction consistent with the absence of subduction-related volcanic zone which has penetrated so deep. The very high rate of activity in Irian Jaya (Hall 2002) and the unusual chemical subduction and the great age of the subducted Pacific character of the volcanic rocks that are known (Housh & lithosphere beneath Tonga may be the cause of such deep McMahon 2000). penetration (van der Hilst 1995). Even the slabs postulated There is one subduction zone which is widely postu- above to have been subducted between 45 and 25 Ma lated, and included in the tectonic model (Hall 2002) (anomalies 5 to 7) do not extend significantly into the lower (Figure 10), but for which we see little evidence in the mantle. This anomaly is therefore unusual. We have a tomography, and that is the Maramuni Arc (Dow 1977; number of hypotheses for this anomaly. Gurnis et al. (2000) Rogerson & Williamson 1985; Cullen 1996; McDowell et al. have suggested that an approximately north–south seismic 1996; Hill & Raza 1999). Southwest-dipping subduction is anomaly beneath the Australian–Antarctic Discordance commonly suggested to have occurred during the Miocene south of Australia may be the result of a slab in the mantle beneath the Papuan peninsula. In the tectonic model (Hall subducted east of Australia during the Cretaceous and sub- 2002) the trench would have advanced as the Australian sequently overridden by Australia as it has moved east. Our Plate moved north from about 20 Ma to 7 Ma (Figure 10) anomaly 8 is at a similar depth and along strike from the and would have overridden the subducted slab. It would Gurnis et al. (2000) Australian–Antarctic Discordance therefore be expected that the subducted slab would anomaly and may have the same explanation if this sub- become flat-lying in the upper mantle and the slab should duction zone had continued north with a similar orienta- be visible at relatively shallow depths beneath the Coral tion. The anomaly is located above the position where there Sea and Queensland. About 1000 km of subduction is pre- must have been a change in polarity in subduction at the dicted by the tectonic model (Hall 2002) but no clear boundary between the New Guinea and the Melanesian anomaly is visible. It is possible that part of anomaly A7 segments, as noted above. Based on palaeomagnetic argu- under Queensland is due to this subduction. ments Falvey and Pritchard (1982) suggested that in this There are several possible explanations. First, the lithos- region before 45 Ma there was southeast-dipping subduc- phere being subducted was relatively young. According to tion which was orthogonal to the Australian margin. Thus a the tectonic model (Hall 2002) the age of the lithosphere second possibility is that the anomaly represents a slab arriving at the trench would have been about 20 million subducted before 45 Ma. This would be consistent with its years greater than the age of its subduction. Visibility on depth, and the orientation of subduction. A final possibility tomographic images would have been reduced because the is that the anomaly represents a slab subducted between seismic-wave speed contrast with the ambient upper man- 45 and 25 Ma which has sunk to a grater depth than other tle would be smaller. There is no lithosphere of this age cur- slabs subducted during the same period. The Tonga region rently being subducted in the region with which to make a shows that this is quite possible if there is high rate of sub- comparison but south of the Solomons, lithosphere of the duction and/or the age of the slab is great. The tectonic Woodlark Basin, which is less than 10 Ma in age, is being model (Hall 2002) suggests this region was a zone of very subducted and no high-velocity anomaly is seen. A second rapid subduction in this period in which there was fast roll- possibility is that the resolution of tomographic model in back of at least two subduction hinges (Figure 10), one the region beneath the Coral Sea and Queensland may be related to subduction beneath the Melanesian Arc and a inadequate. This is possible but one would expect to see second related to subduction beneath the Caroline Arc. We 1000 km of slab because resolution is of the order of 4°. A Australian Plate tomography and tectonics 379 third possibility is that the tectonic interpretation is wrong: systems, and it may well be impossible to image young for example volcanism was of shorter duration than postu- lithosphere being subducted. All of these points need to be lated based on the few K–Ar dates (Page 1976; Rogerson & considered in a region in which the tectonic setting is capa- Williamson 1985); the volcanic activity generally inter- ble of changing rapidly as indicated by the youth of several preted as subduction-related was due to another cause large marginal basins, and suggested by the complexity of (some is reported to be shoshonitic); or that more subduc- the region. It is also essential to have a dynamic view. Slabs tion occurred on the north side of the Solomon Sea and less can move in mantle, as illustrated best by the Molucca Sea on the south side at the Maramuni Arc [for example Abbott Plate which must have moved north with the Australian (1995) questions the existence of the Maramuni Arc]. Plate, and some deform in the mantle. Nonetheless, tomography can clearly make an important contribution by testing and improving tectonic models, and in some cases CONCLUSIONS questioning assumptions in a region where there is still a relatively limited knowledge of the surface geology. The rapidly improving tomographic images offer significant opportunities to distinguish between, test and improve tectonic interpretations of the region. In order to test models it ACKNOWLEDGEMENTS is important to be able to examine a model at closely spaced intervals of time, and it is an advantage to be able Part of this work (W. Spakman) was conducted under the to see how trenches move. Most tectonic models are not yet programmes of the Vening Meinesz Research School for easy to compare to tomographic results because they are Geodynamics (Utrecht University) and of the Netherlands incomplete, or because there are large gaps in time between Research Centre of Integrated Solid Earth Sciences. reconstructions. Financial support to R. Hall has been provided by NERC, We have compared the tomographic images to the pre- the Royal Society, the London University Central Research dictions of one model, that of Hall (2002) and in general Fund, and the Royal Holloway SE Asia Research Group found good agreement. The positions of most interpreted supported by a number of industrial companies. subduction zones are consistent with high-velocity mantle anomalies as are the predicted slab lengths. For the region of northern Australia and immediately north in the Pacific REFERENCES we conclude that there is evidence for north-directed sub- ABBOTT L. D. 1995. Neogene tectonic reconstruction of the duction north of Australia and an absence of evidence for Adelbert–Finisterre–New Britain collision, northern Papua New south-directed subduction beneath New Guinea between Guinea. Journal of Southeast Asian Earth Sciences 11, 33–51. 45 and 25 Ma. As Australia has moved north it has over- ABERS G. A. & ROECKER S. W. 1991. Deep structure of an arc-continent ridden former north-dipping subduction zones and the collision; earthquake relocation and inversion for upper mantle P and S wave velocities beneath Papua New Guinea. Journal of effects of such a passage may be evident in the stratigraphy. Geophysical Research 96, 6379–6401. For example, Müller et al. (2000) identified tectonic subsi- BENES V., SCOTT S. D. & BINNS R. A. 1994. Tectonics of rift propaga- dence in Oceanic Drilling Program sites off northeast tion into a continental margin: western Woodlark Basin, Papua Australia to be the result of dynamic topography induced as New Guinea. Journal of Geophysical Research 99, 4439–4455. the Australian margin passed over a subducted slab burial BEVIS M., TAYLOR F. W., S CHUTZ B. E., RECY J., ISACKS B. L., HELU S., SINGH R., KENDRICK E., STOWELL J., TAYLOR B. & CALMANT S. 1995. ground. We agree with the suggestion but prefer to interpret Geodetic observations of very rapid convergence and back-arc mantle anomalies in this region as the result of Palaeogene extension at the Tonga Arc. Nature 374, 249–251. south-directed subduction rather than north-directed sub- BIJWAARD H. & SPAKMAN W. 2000. Nonlinear global P-wave tomogra- duction. Our interpretation of a huge area of flat slab phy by iterated linearized inversion. Geophysical Journal International 141, 71–82. underlying the Coral Sea toward the New Hebrides implies BIJWAARD H., SPAKMAN W. & E NGDAHL E. R. 1998. Closing the gap that a flat-lying slab can survive for many tens of millions between regional and global travel time tomography. Journal of of years at the bottom of the upper mantle. Geophysical Research 103, 30055–30078. We see no evidence for significant subduction beneath BUNGE H-P. & GRAND S. P. 2000. Mesozoic plate-motion history below the northeast Pacific Ocean from seismic images of the subducted most of New Guinea since 25 Ma. This is consistent with Farallon slab. Nature 405, 337–340. the strike-slip model of Hall (2002) and it is consistent with CARDWELL R. K., ISACKS B. L. & KARIG D. E. 1980. The spatial distrib- other geological evidence, for example, the absence of ution of earthquakes, focal mechanism solutions and subducted Neogene subduction-related volcanic activity in Irian Jaya. lithosphere in the Philippines and northeast Indonesian islands. However, it is necessary to be cautious. One widely In: Hayes D. E. ed. The Tectonic and Geologic Evolution of Southeast Asian Seas and Islands, pp. 1–35. American accepted subduction system, the Maramuni Arc of Papua Geophysical Union Geophysical Monograph 23. New Guinea, is not visible on the tomographic images. This CHARLTON T. R. 2000. Tertiary evolution of the Eastern Indonesia could be because the lithosphere being subducted was rel- Collision Complex. Journal of Asian Earth Sciences 18, 603–631. atively young, the resolution of tomographic model is inad- COOPER P. & TAYLOR B. 1987. Seismotectonics of New Guinea; a model for arc reversal following arc continent collision. Tectonics 6, equate, or the tectonic interpretation is wrong. In many 53–67. critical areas, the tomographic model still has insufficient CROOK K. A. W. & BELBIN L. 1978. The Southwest Pacific area during resolution. Slabs may not be imaged even if they are there the last 90 million years. Journal of the Geological Society of due to a lack of sampling by ray paths. Tomography may Australia 25, 23–40. CULLEN A. G. 1996. Ramu Basin, Papua New Guinea: a record of Late not be able to resolve sufficient detail to distinguish Miocene terrane collision. American Association of Petroleum between different models that are similar but differ in Geologists Bulletin 80, 663–685. detail, the images may not reveal short-lived subduction DALY M. C., COOPER M. A., WILSON I., SMITH D. G. & HOOPER B. G. D. 380 R. Hall and W. Spakman

1991. Cenozoic plate tectonics and basin evolution in Indonesia. JOHNSON R. W. 1976. Late Cainozoic volcanism and plate tectonics at Marine and Petroleum Geology 8, 2–21. the southern margin of the Bismarck Sea, Papua New Guinea. In: DEBAYLE E. & KENNETT B. L. 2000. The Australian continental upper Johnson R. W. ed. Volcanism in Australasia, pp. 101–116. Elsevier, mantle: structure and deformation inferred from surface waves. Amsterdam. Journal of Geophysical Research 105, 25423–25450. JOHNSON R. W. & JAQUES A. L. 1980. Continent–arc collision and rever- DE JONGE M. R., WORTEL M. J. R. & SPAKMAN W. 1994. Regional scale sal of arc polarity: new interpretations from a critical area. tectonic evolution and the seismic velocity structure of the lithos- Tectonophysics 63, 111–124. phere and upper mantle: the Mediterranean region. Journal of JOHNSON T. & MOLNAR P. 1972. Focal mechanisms and plate tectonics Geophysical Research 99B, 12091-12108. of the southwest Pacific. Journal of Geophysical Research 77, DOW D. B. 1977. A geological synthesis of Papua New Guinea. 5000–5032. Bureau of Mineral Resources Bulletin 201. JOLIVET L., HUCHON P. & R ANGIN C. 1989. Tectonic setting of Western ENGDAHL E. R., VAN DER HILST R. & BULAND R. P. 1998. Global tele- Pacific marginal basins. Tectonophysics 160, 23–47. seismic earthquake relocation with improved travel times and KENNETT B. L. N., ENGDAHL E. R. & BULAND R. 1995. Constraints on procedures for depth determination. Bulletin of the Seismological seismic velocities in the Earth from traveltimes. Geophysical Society of America 88, 722–743. Journal International 122, 108–124. FALVEY D. A. 1978. Analysis of paleomagnetic data from the New KROENKE L. W. 1984. Cenozoic tectonic development of the Southwest Hebrides. Bulletin of the Australian Society of Exploration Pacific. CCOP/SOPAC Technical Bulletin 6. Geophysics 9, 117–123. LEE T. Y. & LAWVER L. A. 1995. Cenozoic plate reconstruction of south- FALVEY D. A. & PRITCHARD T. 1982. Preliminary paleomagnetic results east Asia. Tectonophysics 251, 85–139. from northern Papua New Guinea: evidence for large microplate LÉVÊQUE J-J., RIVERA L. & WITTLINGER G. 1993. On the use of the rotations. In: Watson S. T. ed. Transactions of the 3rd Circum- checker-board test to assess the resolution of tomographic inver- Pacific Energy and Mineral Resources Conference, Honolulu, pp. sions. Geophysical Journal International 115, 313–318. 593–600. American Association of Petroleum Geologists, Tulsa. MCCAFFREY R. 1982. Lithospheric deformation within the Molucca sea GRAND S. P., VAN DER HILST R. & WIDIYANTORO S. 1997. Global seismic arc–arc collision: evidence from shallow and intermediate earth- tomography: A snapshot of convection in the Earth. GSA Today 7, quake activity. Journal of Geophysical Research 87, 3663–3678. 1–7. MCDOWELL F. W., MCMAHON T. P., WARREN P. Q. & CLOOS M. 1996. GURNIS M., MORESI L. & MÜLLER R. D. 2000. Models of mantle convec- Pliocene Cu–Au bearing igneous intrusions of the Gunung Bijih tion incorporating plate tectonics: the Australian region since the (Ertsberg) district, Irian Jaya, Indonesia. Journal of Geology 104, Cretaceous. In: Richards M. A., Gordon R. G. & van der Hilst R. D. 327–340. eds. The History and Dynamics of Global Plate Motions, pp. 211–238. MCKENZIE D. P. 1970. Temperature and potential temperature American Geophysical Union Geophysical Monograph 121. beneath island arcs. Tectonophysics 10, 357–366. HAFKENSCHEID E., BUITER S. J., WORTEL M. J., SPAKMAN W. & B IJWAARD MEGNIN C. & ROMANOWICZ B. 2000. The three-dimensional shear H. 2001. Modelling the seismic velocity structure beneath velocity structure of the mantle from inversion of body, surface Indonesia: a comparison with tomography. Tectonophysics 333, and higher-mode waveforms. Geophysical Journal International 35–46. 143, 709–728. HALL R. 1996. Reconstructing Cenozoic SE Asia. In: Hall R. & MENKE W. 1989. Geophysical Data Analysis: Discrete Inverse Theory. Blundell D. J. eds. Tectonic Evolution of SE Asia, pp. 153–184. Academic Press, San Diego. Geological Society of London Special Publication 106. MONNIER C., GIRARDEAU J., PUBELLIER M., POLVE M., PERMANA H. & HALL R. 1997. Cenozoic tectonics of SE Asia and Australasia. In: BELLON H. 1999. Petrology and geochemistry of the Cyclops ophi- Howes J. V. C. & Noble R. A. eds. Proceedings of the International olites (Irian Jaya, East Indonesia): consequences for the Cenozoic Conference on Petroleum Systems of SE Asia and Australia, pp. evolution of the north Australian margin. Mineralogy and 47–62. Indonesian Petroleum Association Jakarta. Petrology 65, 1–28. HALL R. 1998. The plate tectonics of Cenozoic SE Asia and the distri- MÜLLER R. D., LIM V. S. L. & ISERN A. R. 2000. Late Tertiary tectonic bution of land and sea. In: Hall R. & Holloway J. D. eds. subsidence on the northeast Australian passive margin: response Biogeography and Geological Evolution of SE Asia, pp. 99–131. to dynamic stratigraphy. Marine Geology 162, 337–352. Backhuys Publishers, Leiden. MÜLLER R. D., ROYER J-Y. & LAWVER L. A. 1993. Revised plate motions HALL R. 2002. Cenozoic geological and plate tectonic evolution of SE relative to hotspots from combined Atlantic and Indian Ocean Asia and the SW Pacific: computer-based reconstructions, model hotspot tracks. Geology 21, 275–278. and animations. Journal of Asian Earth Sciences 20, 353–434. MUSGRAVE R. J. & FIRTH J. V. 1999. Magnitude and timing of New HAMILTON W. 1979. Tectonics of the Indonesian region. USGS Hebrides Arc rotation: paleomagnetic evidence from Nendo, Professional Paper 1078. Solomon Islands. Journal of Geophysical Research 104, HILL K. C., GREY A., FOSTER D. & BARRETT R. 1993. An alternative 2841–2853. model for the Oligo-Miocene evolution of northern PNG and the OKAL E. A. & KIRBY S. H. 1998. Deep earthquakes beneath the Fiji Sepik–Ramu Basin. In: Carman G. J. & Carman Z. eds. Petroleum Basin, SW Pacific: Earth’s most intense deep seismicity in stag- Exploration in Papua New Guinea, Proceedings of the 2nd Papua nant slabs. Physics of the Earth and Planetary Interiors 109, New Guinea Petroleum Convention, Port Moresby, pp. 241–259. 25–63. Papua New Guinea Chamber of Mines, Port Moresby. PAGE R. W. 1976. Geochronology of igneous and metamorphic rocks HILL K. C. & HALL R. 2003. Mesozoic–Cenozoic evolution of in the New Guinea Highlands. Bureau of Mineral Resources Australia’s New Guinea margin in a west Pacific context. Bulletin 162. Geological Society of Australia Special Publication 22 and PEGLER G., DAS S. & WOODHOUSE J. H. 1995. A seismological study of Geological Society of America Special Paper 372, 265–289. the eastern New Guinea and the western Solomon Sea regions HILL K. C. & HEGARTY K. A. 1987. New tectonic framework for PNG and its tectonic implications. Geophysical Journal International and the Caroline plate: implications for the cessation of spreading 122, 961–981. in back-arc basins. PACRIM ‘87 Congress, pp. 179–182. PETTERSON M. G., NEAL C. R., MAHONEY J. J., KROENKE L. W., SAUNDERS Australasian Institute of Mining and Metallurgy, Melbourne. A. D., BABBS T. L., DUNCAN R. A., TOLIA D. & MCGRAIL B. 1997. HILL K. C. & RAZA A. 1999. Arc–continent collision in Papua Guinea: Structure and deformation of north and central Malaita, Solomon constraints from fission track thermochronology. Tectonics 18, Islands: tectonic implications for the Ontong Java Plateau 950–966. Solomon arc collision, and for the fate of oceanic plateaus. HOUSH T. & MCMAHON T. P. 2000. Ancient isotopic characteristics of Tectonophysics 283, 1–33. Neogene potassic magmatism in western New Guinea (Irian Jaya, PIGRAM C. J. & DAVIES H. L. 1987. Terranes and the accretion history Indonesia). Lithos 50, 217–239. of the New Guinea orogen. BMR Journal of Australian Geology & IOC, IHO & BODC 1997. GEBCO-97: The 1997 Edition of the Gebco Geophysics 10, 193–212. Digital Atlas. British Oceanographic Data Centre, Birkenhead. RANGIN C., JOLIVET L. & PUBELLIER M. 1990. A simple model for the tec- JAQUES A. L. & ROBINSON G. P. 1977. The continent/island-arc collision tonic evolution of southeast Asia and Indonesia region for the in northern Papua New Guinea. BMR Journal of Australian past 43 m.y. Bulletin de la Société géologique de France 8, Geology & Geophysics 2, 289–303. 889–905. Australian Plate tomography and tectonics 381

RANGIN C., SPAKMAN W., PUBELLIER M. & BIJWAARD H. 1999. 7, 171–174. Tomographic and geological constraints on subduction along the TAYLOR G. K., GASCOYNE J. & COLLEY H. 2000. Rapid rotation of Fiji: eastern Sundaland continental margin (South-East Asia). Bulletin paleomagnetic evidence and tectonic implications. Journal of de la Société géologique de France 170, 775–788. Geophysical Research 105, 5771–5781. RICHARDS J. P., CHAPPELL B. W. & MCCULLOCH M. T. 1990. Intraplate TREGONING P., LAMBECK K., STOLZ A., MORGAN P., MCCLUSKY S. C., VAN type magmatism in a continent island arc collision zone: Porgera DER BEEK P., MCQUEEN H., JACKSON R. J., LITTLE R. P., LAING A. & intrusive complex, Papua New Guinea. Geology 18, 958–961. MURPHY B. 1998. Estimation of current plate motions in Papua RICHARDS M. A. & ENGEBRETSEN D. C. 1992. Large-scale mantle con- New Guinea from Global Positioning System observations. vection and the history of subduction. Nature 355, 437–440. Journal of Geophysical Research 103, 12181–12203. RIDGEWAY J. 1987. Neogene displacements in the Solomons island VAN DER HILST R. 1995. Complex morphology of subducted lithos- arc. Tectonophysics 133, 81–93. phere in the mantle beneath the Tonga Trench. Nature 374, RITSEMA J., VAN HEIJST H-J. & WOODHOUSE J. H. 1999. Complex shear 154–157. wave velocity structure imaged beneath Africa and Iceland. VAN DER HILST R. D., WIDIYANTORO S. & ENGDAHL E. R. 1997. Evidence Science 286, 1925–1928. for deep mantle circulation from global tomography. Nature 386, ROGERSON R. & WILLIAMSON A. 1986. Age, petrology and mineraliza- 578–584. tion associated with two Neogene intrusive types in the eastern VAN DER VOO R., SPAKMAN W. & B IJWAARD H. 1999a. Mesozoic sub- highlands of Papua New Guinea. Buletin Persatuan Geologi ducted slabs under Siberia. Nature 397, 246–249. Malaysia 20, 487–502. VAN DER VOO R., SPAKMAN W. & B IJWAARD H. 1999b. Tethyan sub- SIMONS F. J., ZIELHUIS A. & VAN DER HILST R. D. 1999. The deep struc- ducted slabs under India. Earth and Planetary Science Letters 171, ture of the Australian continent from surface wave tomography. 7–20. Lithos 48, 17–43. WEILER P. D. & COE R. S. 2000. Rotations in the actively colliding SMITH R. I. 1990. Tertiary plate tectonic setting and evolution of Finisterre Arc Terrane: paleomagnetic constraints on Plio- Papua New Guinea. In: Carman G. J. & Carman Z. eds. Petroleum Pleistocene evolution of the South Bismarck microplate, north- Exploration in Papua New Guinea, Proceedings of the 1st PNG eastern Papua New Guinea. Tectonophysics 316, 297–325. Petroleum Convention, Port Moresby, pp. 229–244. Papua New WELLS R. E. 1989. The oceanic basalt basement of the Solomon Guinea Chamber of Mines, Port Moresby. Islands arc and its relationship to the Ontong Java Plateau: SMITH W. H. F. & SANDWELL D. T. 1997. Global seafloor topography insights from Cenozoic plate motion models. In: Vedder J. G. & from satellite altimetry and ship depth soundings. Science 277, Bruns T. R. eds. Geology and Offshore Resources of Pacific Islands 1956–1962. Arcs—Solomon Islands and Bougainville, Papua New Guinea SPAKMAN W. & B IJWAARD H. 2001. Optimization of cell parameteriza- Regions, pp. 7–22. Circum-Pacific Council for Energy and Mineral tions for tomographic inverse problems. Pure and Applied Resources Earth Science Series 12. Geophysics 158, 1401–1423. WIDYANTORO S. & VAN DER HILST R. D. 1997. Mantle structure beneath SPAKMAN W. & N OLET G. 1988. Imaging algorithms, accuracy and res- Indonesia inferred from high-resolution tomographic imaging. olution in delay time tomography. In: Vlaar N. J., Nolet N., Wortel Geophysical Journal International 130, 167–182. M. J. R. & Cloetingh S. A. P. L. eds. Mathematical Geophysics: A WORTEL M. J. R. & SPAKMAN W. 2000. Subduction and slab detach- Survey of Recent Developments in Seismology and Geodynamics, ment in the Mediterranean–Carpathian region. Science 290, pp. 158–188. Reidel, Dordrecht. 1910–1917. STRUCKMEYER H. I. M., YEUNG M. & PIGRAM C. J. 1993. Mesozoic to YAN C. Y. & KROENKE L. W. 1993. A plate tectonic reconstruction of the Cainozoic plate tectonic and palaeogeographic evolution of the SW Pacific 0-100 Ma. Proceedings of the Ocean Drilling Program, New Guinea region. In: Carman G. J. & Carman Z. eds. Petroleum Scientific Results 130, 697–709. Exploration in Papua New Guinea, Proceedings of the Second PNG ZHONG S. & GURNIS M. 1995. Mantle convection with mobile, faulted Petroleum Convention, Port Moresby, pp. 261–290. Papua New plate margins. Science 267, 838–843. Guinea Chamber of Mines, Port Moresby. TAYLOR B. 1979. Bismarck Sea: evolution of a back-arc basin. Geology Received 15 August 2001; accepted 12 June 2002