Main

Plate tectonic theory successfully describes how the lithosphere—Earth’s rigid outermost shell—consists of a mosaic of segments that move and interact across the surface of our planet. It also accounts for the majority of Earth’s volcanism, which is concentrated at plate boundaries. However, an important class of volcanism occurs within plates or across plate boundaries, often forming linear volcanic chains that grow older in the direction of plate motion. Because of their geometry, age distributions, topographic expression and geochemical characteristics, most of these so-called hotspots are believed to mark the surface expression of upwelling mantle plumes1,7.

Around 50 hotspots have been identified at Earth’s surface8,9. Of these, only ∼20% occur on continents and, hence, most of our understanding of mantle plumes comes from hotspot tracks in oceanic settings. However, oceanic lithosphere is regularly recycled into the mantle through subduction, so if we are to understand plume-related volcanism before ∼200 million years ago (Ma), which constitutes most of Earth’s geological record10, we must learn: (1) how plumes interact with continental lithosphere; and (2) how this interaction affects the chemical composition and erupted volume of lavas at the surface. Here, we first combine observational constraints from surface geology, geochronology and plate-motion histories to identify Earth’s longest continental hotspot track in eastern Australia. We subsequently integrate constraints from seismology and geochemistry to ascertain how regional lithospheric thickness variations influence the volume and composition of plume-derived magmas along this track.

Cenozoic era volcanism in eastern Australia represents one of the world’s most extensive intraplate volcanic regions11 (Fig. 1). Three types of volcano are identified in the widely used classification of Wellman and McDougall11, which we also adopt here: (1) central volcanoes, which are predominantly basaltic in composition but have felsic lava flows or intrusions, with lavas typically produced from central vents, often building large volcanic complexes; (2) lava fields, which are basaltic, extensive and thin and are often characterized by an abundance of small scoria, lava cones and maars; and (3) the leucitite suite, which is dominated by low-volume, potassium-rich, leucitite-bearing lavas. These volcanic classes are principally distinguished petrologically, with central volcanoes distinguished from lava-field volcanoes by the presence of felsic rock and both distinguished from the leucitite suite by the absence of leucitite12. However, these classes also show considerable differences in age trends: 40Ar–39Ar and K–Ar geochronological studies demonstrate that both the central volcanoes and the leucitite suite define age-progressive tracks that become younger to the south. The tracks identified, so far, include: (1) Comboyne—an ∼770-km-long track that extends from Fraser Island in Queensland to Comboyne in New South Wales, displaying a record of volcanic activity from ∼32 Ma to 16 Ma (ref. 13); (2) Canobolas—an ∼760-km-long track, extending from Bunya in Queensland to Canobolas in New South Wales, recording volcanism from ∼24 Ma to 12 Ma (ref. 14); (3) an ∼400-km-long track, extending across central Queensland from Cape Hillsborough to Buckland and recording volcanism from ∼34 Ma to 27 Ma (ref. 6); and (4) an ∼650-km-long leucitite-bearing track, extending from Bokhara River in New South Wales to Cosgrove in Victoria and displaying a record of volcanic activity from ∼17 Ma to 9 Ma (ref. 5). These tracks are widely believed to mark the passage of mantle plumes beneath the northwards-migrating Australian plate5,6,13,14. Lava-field volcanics, on the other hand, show no such age progression and are thought to be generated through an alternative process, with an edge-driven model being suggested for the formation of the Newer Volcanics Province (NVP)15,16.

Figure 1: The distribution and classification of eastern Australian Cenozoic volcanic centres and their relationship to regional lithospheric thickness variations.
figure 1

a, Volcanic centres, coloured following the classification of Wellman and McDougall11, where black, grey and red denote central volcanoes, basaltic lava fields and leucitite-bearing volcanics, respectively. Age-progressive hotspot tracks are denoted by black dashed lines. We identify Earth’s longest continental hotspot track, the Cosgrove track, which extends across the Australian continent from Cape Hillsborough (CH) to Cosgrove (C). Stars denote a selection of volcanic centres along this track that are coloured by age (in Ma), while squares mark predicted volcanic locations, based on reconstructed absolute motions of the Australian plate18. If the predicted locations fall within the circles surrounding each dated volcanic centre (our measure of the uncertainty in predicted locations), then we consider that volcanic centre to be the surface expression of the same mantle plume: eight of nine volcanic centres satisfy this criteria (filled squares), while one does not (open square). The dotted line, to the south of Cosgrove, marks the predicted extension of the Cosgrove hotspot track, towards its present-day location, which is illustrated in Extended Data Fig. 2. B, Buckland; BH, Begargo Hill; BR, Bokhara River; BU, Bunya; CA, Canobolas; CO, Comboyne; FI, Fraser Island; NVP, Newer Volcanics Province. b, The same volcanic centres, plotted above an estimate of lithospheric thickness, highlighting a clear correlation between lithospheric thickness and volcanic outcrop and composition along the Cosgrove hotspot track.

PowerPoint slide

Full size image

The central volcanoes of central Queensland and the leucitite suite of New South Wales and Victoria have been considered unrelated5,6,11,14,17, principally because the volcanic provinces identified in each differ dramatically in composition and eruptive volume, and they are separated by a volcanic gap of >650 km. However, their relative locations and ages suggest that they may be the surface expression of the same mantle plume, thus constituting a single hotspot track. In Fig. 1a we test this hypothesis by predicting volcanic locations along this track, using the reconstructed absolute motion of the Australian plate18. Specifically, we take the mapped locations of 15 volcanic centres (see Extended Data Table 1 and Extended Data Fig. 1), dated by 40Ar–39Ar techniques5,6, and predict their location at the time associated with the next dated volcanic centre (see Methods). Stars in Fig. 1a denote the locations of these dated volcanoes, while squares denote their predicted locations, based upon the reconstruction (note that for clarity, only 9 of the 15 dated volcanoes used in our analyses are shown: see Extended Data Fig. 2 for a comparable figure with all 15 volcanoes). Circles represent our estimate for the uncertainty in predicted locations, which arises through a combination of: (1) uncertainties in the underlying mantle plume diameter and the associated melt region’s lateral extent3,19; (2) the potential for plume drift8,20,21; (3) the unpredictability introduced through preferential melt extraction pathways22; and (4) uncertainties in the 40Ar–39Ar ages of dated volcanic centres (see Methods and Extended Data Fig. 3). If a predicted location falls within these circles, then we consider that volcanic centre to be the surface expression of the same mantle plume (filled squares in Fig. 1a): 8 of the 9 volcanic centres shown in Fig. 1a, and 14 of the 15 volcanic centres in our full analyses, satisfy this criterion. The only volcanic centre that fails to satisfy this criterion is Begargo Hill of the leucitite suite, which is consistently located further south than is predicted by our reconstruction. We speculate that this points towards either a rapid phase of southwards plume motion (exceeding 4 cm yr−1) from ∼17 Ma to 15 Ma, a change in the melt extraction pathway, or a combination of both. In support of these ideas, we note that: (1) variable plume migration rates have been observed elsewhere20 and are also predicted in global mantle convection simulations8,21; and (2) Begargo Hill lies to the south of a region of increased lithospheric thickness (discussed later), which may focus sub-lithospheric plume material and any associated melt southwards, as a result of Australia’s rapid motion towards the north. Most notably, however, the northernmost dated leucitite-bearing volcano satisfies the aforementioned location criterion, confirming that the central volcanoes of central Queensland and the leucitite suite of New South Wales and Victoria are the surface expression of the same mantle plume. Together, they constitute Earth’s longest continental hotspot track, which we call the Cosgrove track. This knowledge, however, leads to further questions. Specifically, since these volcanic centres are the surface expression of the same mantle plume, why does the Cosgrove track display large volcanic gaps? What drives the considerable variations in volume and chemical composition of plume-derived magmas between the central volcanoes and the leucitite suite? Are both of these characteristics related? To answer these fundamental questions, we combine observational constraints from seismology and geochemistry.

We first generate a map of lithospheric thickness, by combining constraints from recent three-dimensional body-wave tomography results15,23 with the regional Australian Seismological Reference Model (AuSREM)24 (Fig. 1b) (for further details, including uncertainties, see Methods and Extended Data Figs 4, 5, 6). The main features evident in Fig. 1b include: (1) the contrast between thicker lithosphere in the centre of Australia and thinner lithosphere to the east, which is consistent with a transition from Precambrian central to Phanerozoic eastern Australia and then oceanic lithosphere outboard of the continental margin; (2) a zone of thin lithosphere, bound to the east by a zone of intermediate-thickness lithosphere of similar width, which extends southwestwards from ∼30° S through central New South Wales into northern Victoria15; and (3) considerable changes in lithospheric thickness over relatively short horizontal distances. It is generally accepted that these lithospheric ‘steps’ will produce complex flows25 and, as noted previously, this led to the suggestion of an edge-related model for the formation of the NVP15,16,25. Indeed, such edge-related mechanisms are probably also applicable to other lava-field volcanism in the region: as illustrated in Fig. 1b, all lava-field volcanic provinces lie adjacent to substantial steps in lithospheric thickness, above comparatively thin lithosphere, thus providing a favourable setting15,16.

What remains poorly understood, however, is how mantle plumes interact with these lithospheric thickness variations and, specifically, how they influence the volume and composition of plume-derived magmas. Intriguing trends are evident along the Cosgrove track in Fig. 1b: (1) volcanic gaps occur in regions where lithospheric thickness exceeds ∼150 km; (2) the basaltic and felsic central volcanoes of central Queensland occur in regions where lithospheric thickness is less than ∼110 km; and (3) low-volume, leucitite-bearing volcanism to the south occurs, exclusively, in regions of intermediate lithospheric thickness, with volcanic gaps within the leucitite suite also coinciding with regions of thicker lithosphere. These unambiguous trends suggest that the thickness of overlying lithosphere is dictating the volume and composition of plume-derived magmas, by limiting the rise height of the underlying plume and, hence, the degree of partial melting. We infer that the underlying mantle plume: (1) cannot rise to shallow enough depths to induce decompression melting in regions where lithospheric thickness exceeds ∼150 km, thus providing an explanation for the volcanic gaps along the Cosgrove track and placing the first observational constraint on the maximum melting depth of mantle plumes beneath continents (excluding ultra-volatile melts that form kimberlites and carbonatites); (2) undergoes high-degree partial melting beneath comparatively thin lithosphere to produce basaltic and felsic central volcanoes along the northern segment of the Cosgrove track; and (3) undergoes very low-degree partial melting in regions of intermediate lithospheric thickness, thus facilitating the production of low-volume leucitite-bearing volcanics towards the southern end of the Cosgrove track.

To ascertain whether or not these inferences are compatible with geochemical observations, we collated previously published trace-element data from volcanic outcrops along the Cosgrove track17,26 (see Methods and Extended Data Tables 2, 3). Although this data set is limited, consisting of only eight data points (four from the central volcanoes of central Queensland and four from the leucitite suite), it provides a basis for testing our hypothesis. As mantle material undergoes partial melting, incompatible trace elements, such as barium (Ba), are preferentially transferred into the melt. Indeed, for the most incompatible elements, their concentration in the melt is, to a good approximation, inversely proportional to the melt fraction. As the degree of partial melting increases, their concentrations are subsequently diluted27. Accordingly, if the available trace-element data supported the inferences presented earlier, leucitite-bearing volcanics would display higher concentrations of incompatible trace elements when compared to the basaltic central volcano samples. Such trends are indeed apparent in the trace-element concentrations plotted in Fig. 2: leucitite samples display barium concentrations that exceed those of basaltic samples by a factor of ∼3. Our inference that the melt fraction along the Cosgrove track is controlled by lithospheric thickness, which limits the rise height of plumes, should also leave a discernible signature in the trace-element concentrations illustrated in Fig. 2. Melting at greater depth, in the presence of higher proportions of garnet28, will sequester the heavy rare-earth elements, such as lutetium (Lu), with respect to the middle rare-earth elements, such as gadolinium (Gd). Consequently, samples from the leucitite suite should display elevated Gd/Lu ratios when compared to the basaltic samples from central Queensland. As illustrated in Fig. 2, this is the case: leucitites typically possess Gd/Lu ratios of ∼6, compared to ∼2–3 for the basalts of central Queensland. Our hypothesis, therefore, is supported by the available trace-element data.

Figure 2: Trace-element abundances of volcanic samples along the Cosgrove hotspot track.
figure 2

a, Trace-element concentrations (in p.p.m.), normalized to primitive mantle values31, from published analyses of four leucitite-bearing and four mantle-derived basaltic samples along the southern and northern sections of the Cosgrove track, respectively (see Methods). b, A plot of the Gd/Lu ratio (a proxy for melting depth) against the barium (Ba) concentration (a tracer for the extent of melting) for these samples.

PowerPoint slide

Source data

Full size image

An aspect of the Cosgrove track that has not been addressed is why basaltic and felsic central volcanoes do not re-emerge to the south of Cosgrove, in a region of comparatively thin lithosphere. As mantle plumes have finite lifetimes8,21, it is possible that the underlying plume faded at ∼8 Ma, although this would be an unlikely coincidence. We speculate that an alternative mechanism is at play: it has previously been demonstrated15 that regional three-dimensional lithospheric thickness variations, coupled with rapid northwards motion of the Australian plate, gives rise to a focused edge-driven convection (EDC) cell to the west of the Cosgrove track, in the vicinity of the NVP, thus providing a mechanism for the localization of lava-field volcanism to this region. However, an explanation for the onset of NVP volcanism, at ∼5 Ma (ref. 11), has remained elusive: the lithospheric thickness variations driving EDC in this region are probably long-lived29, which is difficult to reconcile with comparatively recent volcanism. Our reconstructions, however, place the mantle plume that generated the Cosgrove hotspot track <50 km to the east of the NVP, from ∼6.5 Ma to 5 Ma. We speculate that the capture and entrainment of this plume, into a pre-existing EDC cell, was the trigger for magmatism within the NVP and explains the absence of a hotspot track to the south of Cosgrove. In support of these ideas, we note that although EDC is expected to occur on all lithospheric steps, including those to the east of the plume’s predicted passage, the dominant cell in this region lies directly beneath the NVP15. Accordingly, preferential westwards flow (and, hence, entrainment of plume material) into the NVP region, is to be expected, and was indeed evident in our previous models15. To our knowledge, the interaction between mantle plumes and EDC has not been documented elsewhere. However, this process: (1) has important implications for the surface expression of mantle plumes in the vicinity of step changes in lithospheric thickness; and (2) provides a solution to the global puzzle of why step changes in lithospheric thickness, which occur along craton edges and at passive margins, produce volcanism only at isolated locations, thus complementing our previous study15. Finally, we note that the predicted present-day location of the mantle plume that generated the Cosgrove hotspot track lies to the northwest of Tasmania, which coincides with a region of recent seismicity and is at the western limit of the so-called East Australia Plume System imaged previously using finite frequency tomography30 (see Extended Data Fig. 2).

Methods

40Ar–39Ar dated volcanic centres

Of the volcanic centres that make up the Cosgrove hotspot track, 15 have recently been dated via 40Ar–39Ar geochronology. These include nine volcanic centres from the central volcanoes of Central Queensland6 and six volcanic centres from the leucitite suite of New South Wales and Victoria5. These data are summarized in Extended Data Table 1, with localities also illustrated in Extended Data Fig. 1.

Reconstructed hotspot track

In Fig. 1a, we test the hypothesis that the central volcanoes of central Queensland and the leucitite suite of New South Wales and Victoria represent the surface expression of the same mantle plume, thus constituting a single hotspot track, by predicting volcanic locations along this track, using the reconstructed absolute motion of the Australian plate18. Specifically, we take the mapped location of 15 volcanic centres5,6 (Extended Data Table 1) and predict their location at the time associated with the next (in a temporal sense) dated volcanic centre. Stars in Fig. 1a denote the locations of these dated volcanoes, while squares denote their predicted locations, based upon the reconstruction. Note that for illustrative purposes, only 9 of the 15 dated volcanoes used in our analyses were shown in Fig. 1a. A comparable figure, for all 15 dated volcanoes, is shown in Extended Data Fig. 2.

In doing this, we have implicitly assumed that the underlying mantle plume is a fixed point source at the base of the lithosphere and that melt transport towards Earth’s surface is not influenced by pre-existing lithospheric structure. However, palaeomagnetic data indicate that mantle plumes drift slowly, at a fraction of surface plate velocities20,32,33, an inference that is supported by global mantle convection simulations8,21. In addition, plume tails have an upper-mantle diameter that probably exceeds 200 km and produce a sub-lithospheric melt region that may be up to half of this size3,19, while lithospheric structure has an important role in controlling the lateral transport of plume material and melt-extraction pathways22. We therefore estimate the uncertainty in our predicted locations (Fig. 1a, circles, and Extended Data Fig. 2), by allowing for plume motion at a modest rate of up to 1 cm yr−1, and adding 100 km to account for the melt region’s lateral extent and the unpredictability introduced through preferential melt extraction pathways. Furthermore, we propagate the uncertainty on the measured ages of volcanic samples onto the predicted volcanic locations5,6. Specifically, we use the measured age uncertainties5,6 to shorten/lengthen the duration of the temporal stages over which the instantaneous angular velocities/poles18 are applied. This results in two different predictions of geographic positions. We take the geodesic distance between these two as a representative value of the uncertainty contribution, on the predicted locations, which is associated with age. If a predicted location falls within the resulting uncertainty circles, we consider that volcano to be the surface expression of the same mantle plume (Fig. 1a, filled squares, and Extended Data Fig. 2): 8 of the 9 volcanic centres shown in Fig. 1a and 14 of the 15 shown in Extended Data Fig. 2 satisfy this criterion. The sensitivity of our results to the uncertainty parameters specified (that is, the rate of plume motion and the melt zone’s lateral extent) is illustrated in Extended Data Fig. 3.

Lithospheric thickness estimate

The lithospheric thickness estimate given in this study is derived from recent three-dimensional seismic tomography results15,23, which exploit body-wave arrival time information from the WOMBAT transportable array in eastern Australia. WOMBAT is the largest experiment of its type in the southern hemisphere, which, to date, comprises a cumulative total of ∼700 stations, with spacings varying from 50 km on the mainland to 15 km in Tasmania. Extended Data Figure 4 shows the location of the 12 sub-arrays used in this study. Note that although two additional arrays have been deployed to the northeast, the associated data are not yet available for use.

FMTOMO34,35 is used to invert relative arrival time residuals of various global P-wave phases for three-dimensional velocity variations. Owing to the dominance of short-period seismometers (1 Hz natural frequency) in WOMBAT, S-wave arrivals are difficult to identify, which is why we favour P-wave arrivals. Two limitations of using relative P-wave arrival times in seismic tomography are that absolute velocities are not constrained, and the long-wavelength structure (greater than the aperture of the sub-arrays) is filtered out during the inversion. We overcome these limitations by using the AuSREM mantle model24 as a starting model in the inversion, which has a horizontal resolution of ∼200–250 km. Assuming that AuSREM is accurate at this structural wavelength, the problem of suturing together relative arrival time data from different arrays is overcome36 and both the absolute and relative velocities can be considered (although the latter will be more reliable).

The use of three-dimensional mantle velocity models to derive estimates of lithospheric thickness is common. In Australia, a number of different estimates have been made based upon S-wave velocity models obtained from surface wave tomography37,38,39. Proxies for the depth to the base of the lithosphere include a particular choice of velocity contour, a decrease in velocity within a certain depth range, and the behaviour of the velocity gradient. In our case, we have a three-dimensional P-wave velocity model rather than a three-dimensional S-wave velocity model, which is likely to be less sensitive to the presence of a mechanically weak sub-lithosphere mantle layer. Moreover, the limited depth resolution caused by the sub-vertical nature of the incident paths means that using the depth to the base of fast velocity anomalies, or some chosen contour, is unlikely to yield robust results, particularly in terms of absolute values. Extended Data Figure 5 shows a slice through the P-wave velocity model at 120 km depth, which reveals a clear division between higher velocities to the west and lower velocities to the east, which is consistent with a transition from Precambrian central and western Australia, to Phanerozoic eastern Australia and finally oceanic lithosphere outboard of the continental margin. Regions of elevated velocities probably correspond to thicker lithosphere, as the pattern of velocity variations shares a first-order resemblance to the lithospheric thickness estimates of previous studies39,40.

The AuSREM model41 also includes a lithospheric thickness estimate, based on velocity gradients from surface and body-wave tomography and refracted waves in the mantle. This estimate shows a broad consistency with other studies39,40, with the main difference being in the absolute depths to the base of the lithosphere rather than the pattern of depth variations. Minimum depths vary between ∼50 km and 70 km in the vicinity of the eastern seaboard, whereas maximum depths vary between 180 km and 220 km in the central western section of our model region. On the basis of changes in the vertical gradient of SV-wave velocity, one study39 provides upper and lower bounds on the transition from lithosphere to sub-lithospheric mantle. This transition exceeds 50 km at several locations. Both the models described in refs 40 and 41 correspond more closely to the lower bound estimates of ref. 39.

Owing to the potential for vertical smearing in the three-dimensional WOMBAT model, we instead take the vertically averaged velocity over a prescribed depth range and calibrate this against minimum and maximum regional lithospheric thickness estimates from previous studies. Thus, the maximum average velocity will equate to the thickest lithosphere, while the minimum average velocity will equate to the thinnest lithosphere, with a linear relationship assumed between depth and average velocity. This approach essentially assumes that variations in the thickness of a higher-velocity lithosphere above a lower velocity sub-lithospheric mantle is responsible for the observed variations in arrival time. Although this is a relatively crude assumption, if we apply this scheme to the P-wave component of the AuSREM model over a depth range of 50–200 km, then the resultant estimate of lithospheric thickness is similar to the AuSREM lithosphere thickness estimate. Extended Data Figure 6a shows the result of using the same 50–200 km depth range with the three-dimensional WOMBAT tomography model (equivalent to Fig. 1), noting that north of ∼28° S, we revert to the AuSREM model, as there is no additional data coverage in this region. The pattern of depth variations is remarkably similar to the pattern of velocity variations at 120 km depth in Extended Data Fig. 5, demonstrating that velocities do not vary significantly with depth in the uppermost mantle. Compared to previous lithospheric thickness estimates for the Australian continent, the main difference is that the new model includes a zone of (on average) intermediate thickness lithosphere that extends southwards from about 30° S through central New South Wales before terminating in northern Victoria. This feature is crucial to the results of this study, as it hosts the leucitite-bearing volcanics.

To investigate the robustness of our lithospheric thickness estimate, we perform the same calculation but vary the input parameters. In particular, we vary the minimum thickness of the model between 50 km and 70 km, and the maximum thickness between 180 km and 220 km, which reflects the range of extremes in our geographic region, between the models of refs 39, 40 and 41. Moreover, in an attempt to account for limited vertical resolution in the mantle and the potential for unresolved crustal structure to smear into mantle structure, we vary the minimum depth over which we take the velocity average between 50 km and 100 km and the maximum depth from 180 km to 230 km. In the former case, the three-dimensional tomographic model explicitly includes the AuSREM crustal and Moho depth model41 to minimize this trade-off effect, so this range is probably quite generous. However, with a minimum depth set at 100 km, we are likely to ignore shallower anomalies that are unrelated to lithospheric thickness. The range of maximum depths was chosen to account for smearing effects at depth; for example, if high velocity lithosphere terminating at 200 km depth is recovered as slightly lower velocity lithosphere terminating at 230 km depth owing to smearing, then it may manifest as thicker lithosphere if the vertical averaging is taken to a depth of 230 km rather than 200 km. We generate a total of 540 models by iterating over these four depth ranges, using an increment of 10 km. Extended Data Figure 6b shows the standard deviation of this model ensemble. In general, the uncertainty is dominated by our chosen range of minimum and maximum lithosphere thicknesses, but there is also a contribution from our choice of depth range over which to vertically average the velocity. The end result indicates that, in most regions of the lithosphere thickness model, the standard deviation of the uncertainty is ∼10 km or less, implying that the first-order features of Extended Data Fig. 6a (and Fig. 1b) are probably correct. Moreover, the pattern of thickness variations is more robust than the absolute variations, which are dependent on the minimum and maximum thickness bounds that we obtain from previously published models.

Geochemical samples

Leucitites of New South Wales and Victoria: trace-element data from published analyses of leucitite samples along the Cosgrove track are summarized in Extended Data Tables 2 and 3. These include three samples from Cosgrove in Victoria26 (acquired by solution ICPMS) and one sample from Condobolin in New South Wales17 (acquired by XRF). Limited trace-element data for three samples at Begargo Hill, Flagstaff and El Capitan in New South Wales, from ref. 42 (acquired by isotope dilution), are also included. Given that trace-element concentrations are only provided for four elements (Rb, Sr, Nd and Sm) in ref. 42, they are not plotted in Fig. 2. Nonetheless, the limited trace-element concentrations provided are consistent with more complete data sets17,26.

Basaltic samples from the central volcanoes of central Queensland: we restrict our choice of trace-element data to samples with a composition that is likely to reflect their mantle source (that is, SiO2 < 50%; MgO > 9%). The selected samples also have no negative Ni/Ta or Eu anomalies, which indicates no detectable crustal contamination or fractional crystallization within the crust.

Geochemical data, with the exception of ref. 17, can be downloaded from the GEOROC database at EarthChem (http://www.earthchem.org/).