A unified continental thickness from seismology and diamonds suggests a melt-defined plate

See allHide authors and affiliations

Science  11 Aug 2017:
Vol. 357, Issue 6351, pp. 580-583
DOI: 10.1126/science.aan0741

A coherent depth for continental plates

The thickness of the continental portion of Earth's cold and rigid surface plates is a source of debate. Tharimena et al. analyzed a specific type of seismic signal called SS precursors to provide a robust estimate of plate thickness under the continents (see the Perspective by Savage). The values range from 130 to 190 km, which lines up well with the depth where diamonds are stable—an independent line of evidence for the depth of continents.

Science, this issue p. 580; see also p. 549


Thick, rigid continents move over the weaker underlying mantle, although geophysical and geochemical constraints on the exact thickness and defining mechanism of the continental plates are widely discrepant. Xenoliths suggest a chemical continental lithosphere ~175 kilometers thick, whereas seismic tomography supports a much thicker root (>250 kilometers) and a gradual lithosphere-asthenosphere transition, consistent with a thermal definition. We modeled SS precursor waveforms from continental interiors and found a 7 to 9% velocity drop at depths of 130 to 190 kilometers. The discontinuity depth is well correlated with the origin depths of diamond-bearing xenoliths and corresponds to the transition from coarse to deformed xenoliths. At this depth, the xenolith-derived geotherm also intersects the carbonate-silicate solidus, suggesting that partial melt defines the plate boundaries beneath the continental interior.

The rigid lithospheric plate transitions to the weaker underlying asthenosphere at the lithosphere-asthenosphere boundary (LAB). The thickness and definition of the plate have not yet been established, and attempts to do so have proven particularly challenging beneath the continents. Here the tectonic plates are thought to be very thick, with deep, buoyant roots that enable their stability and alter mantle convection. However, the exact thickness of the continental lithosphere depends on the type of observation. Exhumed rock samples suggest a chemically depleted continent to ~175 km depth (1), whereas seismic tomography imaging indicates that continents are up to 400 km thick (2, 3). Reconciling this discrepancy is key for a better understanding of plate tectonics and mantle convection.

Xenoliths, pieces of rock brought up from depth during eruption, offer direct sampling of the deep continental lithosphere. Examination of xenoliths suggests a cold, dehydrated, and compositionally depleted lithosphere down to ~175 km depth, beneath which the mantle is more fertile in meltable components such as clinopyroxene and garnet (1, 4). The more fertile samples from greater depths are interpreted as the convecting mantle (5), as metasomatized lithosphere (6), or as an Archean slab underplated onto the base of the continent (7). Diamond-bearing xenoliths offer greater insight into the depth extent of continents, because the diamonds have a cold stability field that is thought to exist only within the rigid continents. Some diamonds are also billions of years old. Therefore, it has been suggested that the stable continents extend to depths at least as great as the origin depths of the samples (5). Although xenoliths provide important clues for where the continent ends and the convecting mantle begins, the challenges in interpreting mineralogical and geochemical features result in debate about how the observed compositional features are tied to the base of the continental plate (8).

Seismic tomography can also image the deep continental lithosphere, typically resolving structure on the scale of tens to hundreds of kilometers. Surface wave speeds under the continents argue for a thick continental lithosphere, as the wave speeds are fast to depths of 200 to 300 km (9, 10). Some body wave tomography studies suggest that the fast region extends to greater depths of up to 400 km (2, 3). The discrepancy between xenolith studies and seismic tomography may be due to a thinner chemical lithosphere underlain by a thermal root (1, 11), as seismic wave speeds have a strong sensitivity to temperature. Alternatively, xenolith depths may have a shallow bias if they sample only the edges of the continental interior, where the continental plate could be thinner (8). The seismic velocity gradient from the fast continental lithosphere to the underlying slower asthenosphere is often very gradual in surface wave tomography models (10), consistent with a thermally defined lithosphere. However, surface wave resolution is broad at these depths, impeding attempts to distinguish a gradual gradient from a sharp contrast and also to determine the thickness of the continental lithosphere. In addition, some studies question the existence of a velocity decrease with depth beneath continental interiors (12).

Scattered wave imaging can also be used to tightly constrain velocity discontinuities such as those that might occur at the LAB, especially if the discontinuities are sharp. Sharp LABs are not consistent with the gradual transition predicted for a thermally defined plate, and they have been used to argue for the presence of melt or hydration (13). A sharp LAB has been detected in most tectonic environments, including oceans, Phanerozoic orogenic zones, and magmatic belts (13, 14). However, detecting a discontinuity at the depths predicted for the continental interior LAB (>150 km) has proven more challenging. Receiver function imaging shows a discontinuity at 150 to 300 km beneath South Africa and South America (table S1). However, no discontinuity at LAB depths has been reported beneath North America and Australia (table S1). The absence of a receiver function discontinuity could occur if the LAB transition occurs gradually. However, the reasons for sporadic detection of a LAB discontinuity beneath continents at 150 to 300 km depth by higher-frequency waveforms are not clear. The lack of a consistent sharp signature at expected LAB depths beneath continental interiors brings into question the definition of the tectonic plate beneath the continents, and whether or not it varies according to tectonic environment.

We used precursor arrivals to SS seismic waves to determine the lithospheric discontinuity structure beneath continental interiors. SS seismic waves are shear waves that bounce off the surface of Earth midway between the earthquake and the seismometer, whereas SS precursors are underside reflections off deeper discontinuities. The advantage of SS precursors is that they give tight constraints on discontinuity depths beneath the bounce point location, allowing systematic comparison among continents. We stacked SS waveforms into nine bounce point bins: North America (NA), South America (SA), Europe (EUR), Siberia (SIB), West Africa (WAF), South Africa (SAF), India (IND), Australia (AUS), and Antarctica (ANT). Eight of the bins correspond to the “Archean craton” classification in the 3SMAC regionalization (15, 16) (Fig. 1). Beneath ANT, we used the “craton” classification from (17) because 3SMAC regionalization is lacking in that region. We fitted the SS data stacks with synthetics generated by convolving a discontinuity operator with a global oceanic waveform stack, our starting synthetic (16). We implemented a range of discontinuity depths, amplitudes, and sharpnesses with the operator. Positive discontinuities correspond to velocity increases with depth; negative discontinuities correspond to velocity decreases with depth. We determined the best-fitting discontinuity(ies) beneath the continental interiors with a grid search algorithm (16).

Fig. 1 Continental SS bounce point map.

SS bounce points (colored circles) beneath the continental interiors were binned according to the 3SMAC regionalization (15). Solid colors indicate LAB depths. Pastel-colored diamonds indicate lithospheric thickness calculated from diamond-bearing xenolith PT data (Fig. 4), with a 15-km offset from our LAB results. Numbers of bounce points are in parentheses.

We imaged a positive discontinuity, the Moho at 35 to 48 (±4) km depth, beneath the nine continental interiors (Fig. 2, fig. S2, and table S2). The discontinuity depths are well-correlated with the weighted averages of CRUST1.0 (18), with a correlation coefficient of 0.87 (19) (Fig. 3A). This demonstrates the robustness of the method (20). Uncertainty in the SS depth, < ±4 km, likely reflects lateral variation in crustal thickness within the large bins, as evidenced by the variability in CRUST1.0 (horizontal error bars in Fig. 3A). The small number of waveforms from IND prevented us from resolving structure deeper than the Moho (fig. S2).

Fig. 2 Synthetic waveform fits to the SS data stacks.

SS stacks (black) were modeled using a global oceanic SS stack as a starting synthetic (blue) and discontinuity operators (16). The colored arrows show locations of discontinuities. The horizontal error bars indicate Gaussian widths (±2σ) over which the velocity drop occurs. The gray shaded region indicates 95% confidence bounds. We show the variability in our data and the robustness of SS modeling in fig. S1.

Fig. 3 Comparison of SS precursor results to previous studies.

(A) SS Moho depths compared to average crustal thickness from CRUST1.0 (18). (B) SS MLD depths compared to results of receiver function studies (table S1). (C to E) SS LAB depths compared to the depth of the maximum negative gradient in Voigt average shear velocity in SEMum2 (10) (C), receiver function studies (table S1) (D), and depth of the deepest diamond-bearing xenoliths (table S3) (E). Estimates for lithospheric thickness from diamond-bearing xenoliths: NA, 181 km; EUR, 204 km; SIB, 150 km; SA, 178 km; WAF, 170 km; SAF, 190 km; AUS, 180 km. The gray bars in (B) and (D) show the depth range of discontinuities reported using receiver functions.

We imaged a negative discontinuity at 80 to 121 km depth beneath six of the nine continental interior bins (Fig. 2, fig. S3, and table S2). The discontinuity is too shallow to represent the LAB as defined by surface waves and/or xenoliths and is more consistent with an internal lithospheric discontinuity, such as the mid-lithospheric discontinuity (MLD). The discontinuity is sharp, with an average velocity contrast of 5 ± 1.5% occurring over <14 km. The discontinuity agrees with the depth of negative discontinuities imaged by receiver functions beneath NA, WAF, and SIB (table S1). The discontinuity is slightly shallower beneath EUR (86 ± 7 km versus 102 to 116 km) and slightly deeper beneath ANT (110 ± 3 km versus 102 km) (Fig. 3B and tables S1 and S2). Beneath NA, we recovered a discontinuity when we included only bounce points in the Hudson Bay, perhaps owing to lateral variability (fig. S2). Beneath AUS, our MLD discontinuity is deeper (121 ± 3 km) than that from receiver functions (69 ± 8 km to 85 ± 14 km) (Fig. 3B and table S1). We resolved no MLD discontinuity beneath SAF, where one receiver function study reports a discontinuity at 85 km depth (table S1). Discrepancies in depth or existence could be caused by differences in the sensitivity of the waveforms. Our SS waveforms represent an average of a broad lateral area, whereas receiver functions give information over a smaller area beneath seismic station locations.

We imaged a negative discontinuity at 130 to 190 km depth, likely the LAB, beneath eight continental interior bins (Figs. 1 and 2). The discontinuity represents a strong velocity decrease with depth of 8 ± 2%. The signal is very sharp beneath WAF, SAF, and AUS, occurring over 15 km or less. The signal is less sharp beneath NA, SA, SIB, and ANT, occurring over <30 km. Beneath EUR the discontinuity is more gradual, occurring over 52 km (table S2). Broader signals could be caused by gradients in depth or lateral variability (21).

Our discontinuity depths at 130 to 190 km (table S2) agree with depths reported by receiver functions in some locations, including SA (141 to 155 km here versus 130 to 160 km) and ANT (159 to 167 km here versus 130 to 177 km) (Fig. 3D and table S1). Comparisons are difficult in other locations where receiver functions conflict in depth and/or existence. For instance, beneath SAF, our result (176 to 188 km) agrees with reported discontinuities at 140 to 200 km [e.g., (22)], whereas other reported values are much deeper at 257 to 300 km (23, 24) and another found no discontinuity at >150 km (25) (table S1). Our result is in a similar depth range as some reports from NA, AUS, and EUR (24, 26, 27), although it is at odds with other reports of very weak or nonexistent discontinuities at these depths beneath the same continental interiors (2729) (table S1). Depth discrepancies are likely explained by the different sensitivities of SS in comparison to receiver functions. Also, Ps crustal reverberations mask true discontinuities in the 150- to 200-km depth range. In addition, the depth range over which this discontinuity occurs in our result (~30 km) could make it undetectable to both Ps and Sp. The strength of the discontinuity would be diminished by ~60% in high-frequency Ps and Sp imaging, making it difficult to detect above the noise (30). Overall, our SS precursors suggest a discontinuity that is consistently present beneath continental interiors, in contrast to the sparse and sometimes discrepant detection by receiver function imaging.

A gradual reduction in seismic velocity at depths of 100 to 220 km occurs in some surface wave models beneath continental interiors (10), a region where our SS imaging would likely detect discontinuities. Our result correlates with the depth of the sharpest gradient in Voigt shear velocity (10) when SIB (the outlier) is excluded, with a correlation coefficient of 0.76; this coefficient is higher, 0.80, when we consider the possible deeper SIB discontinuity (Fig. 3C and fig. S7) (16, 19). This agreement suggests that our result reflects the base of the seismically fast continental lithosphere detected by surface waves, but also that the transition is sharper (<30 km) than that of the surface wave models (~100 km) (10). Surface waves have broad resolution at these depths and cannot discriminate between a gradual and a sharp discontinuity. This is in contrast to some surface wave and body wave studies suggesting that a low-velocity zone may not exist beneath some continents (2, 12), although this absence is potentially an artifact of the broad depth resolution of the waveforms.

Our results correspond well to lithospheric thickness that we calculated from diamond-bearing xenolith pressure-temperature (PT) estimates (Figs. 1 and 3E). Our SS Moho results provided a crustal thickness, and we calculated lithospheric thickness from diamond-bearing xenoliths assuming a crustal density of 2600 kg/m3 and a mantle density of 3330 kg/m3. We used the highest reported pressure in continental interiors where there are multiple analyses of diamond-bearing xenoliths. We assumed that continental lithospheric diamonds are not stable at greater depths than the highest reported pressure. Our estimates of lithospheric thicknesses from diamond-bearing xenoliths range from 130 to 190 km (Fig. 3E and table S3); these values correlate with the depth extent of the lithosphere from diamond thermobarometry (correlation coefficient of 0.93) (19) (Fig. 3E). Our SS results are shallower than those from diamonds by ~15 km (Fig. 3E). This is much smaller than the previously large discrepancy in which the geophysical lithosphere was up to hundreds of kilometers deeper than the geochemical lithosphere. If this small difference between our seismic discontinuity depths and diamond origin depths is robust, it could reflect a subtle difference in sensitivity between the seismic wave speeds and diamond stability to the underlying mechanism (Fig. 3E). The good overall correlation between the seismic discontinuity and the depth extent of the diamonds suggests that our result beneath the continents corresponds to the chemical depth extent of the tectonic plate.

The discontinuity at 130 to 190 km depth is not defined by temperature or anisotropy alone. The sharpness of the discontinuity (<30 km depth) is inconsistent with thermal gradients between the lithosphere and asthenosphere in geodynamic models, which generally occur over >70 km depth (31). Anisotropy does not likely define our observation on its own. A change in azimuthal anisotropy would be accompanied by a change in polarity with back-azimuth (32), which was not observed. An increase in radial anisotropy with depth is also not a likely explanation, because recent surface wave models beneath the continents suggest radial anisotropy (VSH > VSV) decreases in magnitude with depth at 130 to 190 km beneath the continents (10).

Nor is the discontinuity at 130 to 190 km defined by composition alone. Xenolith compositions suggest that the continents are depleted at shallow depths and more fertile at greater depths. However, our velocity drop is also too large to be explained by typical continental bulk compositional depletion, which can only explain up to ~1% velocity contrast (33). The deep continental mantle may be enriched in volatiles, and it has been suggested that this could lower seismic velocity substantially by enhancing the effect of elastically accommodated grain boundary sliding. However, the predicted depths for this effect (~60 to 150 km) are shallower than our discontinuity at up to 190 km depth (34).

Another possible mechanism to explain strong, sharp seismic velocity gradients is a transition from a melt-free shallow layer to a deeper layer containing a small degree of partial melt. To investigate the melt hypothesis, we compared the carbonate-silicate solidus (35) to collated PT estimates for xenoliths with textural information (table S3). We examined three regions: the Slave province, South Africa, and Siberia. Textural information provides an indication of mantle deformation. Coarse-grained samples are interpreted as being within the continental lithosphere, whereas the deformed xenoliths are generally believed to have been affected by eruptions that brought them to the surface and related processes (36). These deeper samples indicate high adiabatic eruption temperatures and show evidence of multiple events of metasomatism, which produced enrichment in incompatible trace elements, volatiles, and SiO2 (6, 37). Geothermobarometry of coarse-grained samples indicates a geotherm consistent with steady-state conductive cooling down to depths of 150 to 200 km. Our discontinuity depths lie at the intersection of the conductive geotherm and the carbonate-silicate solidus (Fig. 4), where the mantle is likely metasomatized (6, 37). The presence of metasomatism and enrichment in volatiles would cause the mantle to be partially melted at greater depths.

Fig. 4 SS LAB depths lie at the intersection of the conductive geotherm and the carbonate-silicate solidus.

Depth-temperature relations for xenoliths from NA, SAF, and SIB (table S3). Continental conductive geotherms (dashed gray lines) were calculated after (4). Geotherms terminate in a mantle adiabat from 1200° to 1300°C (shaded gray region). The graphite (G)–diamond (D) equilibrium (solid black line) is from (41). The carbonate-silicate solidus (solid red line) is from (35). SS LAB results are indicated by dashed green lines, with the deeper SIB estimate shown as a dashed magenta line (16).

If small degrees of carbonated silicate melts are present today beneath the continents, this could explain the seismic discontinuity at 130 to 190 km and also define the base of the continental lithosphere. Melt has a strong effect on shear velocity, with 1% partial melt reducing velocity by 7.9% (38), consistent with our observations. The presence of partial melt would also enhance mantle deformation, effectively lowering mantle viscosity (39, 40). The reduced viscosity would likely enhance convection, raising temperature and affecting the stability of diamonds. Overall, our result suggests a unified petrological and seismic continental thickness and a tectonic plate defined by partial melt.

Supplementary Materials

Materials and Methods

Figs. S1 to S7

Tables S1 to S3

References (4272)

References and Notes

  1. See supplementary materials.
  2. We use a Pearson correlation coefficient to describe the linear dependence between our results and those from previous studies.
  3. Acknowledgments: We thank H. Grutter for providing data sources from his extended abstract from the 8th International Kimberlite Conference, T. Gernon and M. De Wit for their comments on an earlier version of the manuscript, and the anonymous reviewers for their insightful comments. The seismic data used in this study are available at the Incorporated Research Institutions for Seismology (IRIS) data management center ( and were downloaded using the Standing Order for Data (SOD) package ( Supported by Natural Environment Research Council grant NE/M003507/1 and European Research Council grant GA 638665.

Stay Connected to Science

Navigate This Article