Does Earth have a sixth ocean?


Editor's Introduction

Dehydration melting at the top of the lower mantle

annotated by

Water present on the surface of the Earth is easy to find and measure.  What about the water we can't see?  Scientists have always thought that massive amounts of water might exist deep beneath the Earth's surface, but they never knew how to measure this. To complicate things, water trapped in the Earth's crust is not found in the familiar forms of a liquid or vapor, but rather is contained inside the molecular structure of minerals called ringwoodite and wadsleyite.  In this paper, scientists describe how they were able to not only locate the water trapped in the Earth's crust, but to measure it as well.

Paper Details

Original title
Dehydration melting at the top of the lower mantle
Original publication date
Vol. 344 no. 6189 pp. 1265-1268
Issue name


The high water storage capacity of minerals in Earth’s mantle transition zone (410- to 660-kilometer depth) implies the possibility of a deep H2O reservoir, which could cause dehydration melting of vertically flowing mantle. We examined the effects of downwelling from the transition zone into the lower mantle with high-pressure laboratory experiments, numerical modeling, and seismic P-to-S conversions recorded by a dense seismic array in North America. In experiments, the transition of hydrous ringwoodite to perovskite and (Mg,Fe)O produces intergranular melt. Detections of abrupt decreases in seismic velocity where downwelling mantle is inferred are consistent with partial melt below 660 kilometers. These results suggest hydration of a large region of the transition zone and that dehydration melting may act to trap H2O in the transition zone.


The water content of the upper mantle as sampled by mid-ocean ridge basalts is 0.005 to 0.02 weight % (wt %) (1), but a potentially much larger, deep reservoir of water may exist in the mantle transition zone between 410- and 660-km depth owing to the 1 to 3 wt % H2O storage capacity of the major mineral phases wadsleyite and ringwoodite (23). Convective mass transfer across the boundaries of the transition zone could cause dehydration melting, and consequently filtering of incompatible elements, if water contents in the transition zone exceed that of the shallower or deeper mantle (4). An open question is whether transition zone water contents are sufficient to cause dehydration melting where there is downward flow into the lower mantle.

Dehydration melting due to downward flow across the 660-km discontinuity (660) would require both hydration of ringwoodite in the transition zone and low water storage capacity at the top of the lower mantle. The recent discovery of a ~1.5 wt % H2O hydrous ringwoodite inclusion in a diamond (5) demonstrates that, at least locally, the mantle transition zone may be close to water saturation. Regional detections of high seismic attenuation (6) and electrical conductivity (7) in the transition zone suggest hydration at larger scales. However, high-pressure experiments on the incorporation of H2O into silicate perovskite vary widely from 0.0001 wt % (8) to 0.4 wt % H2O (9), with other estimates in between (10). Recent experiments on coexisting phase assemblages indicate a high H2O partition coefficient between ringwoodite and silicate perovskite of 15:1 (11), suggesting a large contrast in water storage capacity at the boundary between the base of the transition zone and the top of the lower mantle.

We integrated laboratory experiments, seismic imaging, and numerical models of mantle flow to investigate mass transfer and melting at the interface between the transition zone and lower mantle beneath North America (12). We conducted in situ laser heating experiments to directly transform hydrous ringwoodite to form silicate perovskite and (Mg,Fe)O in a diamond-anvil cell (DAC) and analyzed the recovered sample with synchrotron–Fourier transform infrared (FTIR) spectroscopy and transmission electron microscopy (TEM) (Fig. 1). FTIR spectra taken away from the laser-heated spots are typical for hydrous Fe–bearing ringwoodite (13). Within the laser-heated spots where perovskite and (Mg,Fe)O formed, there is still strong absorption in the OH-stretching region, although notably different from absorption that occurred before heating (Fig. 1B). Within laser-heated spots, the maximum in OH-stretching absorbance occurs at ~3400 cm−1, and there is a sharp peak at ~3680 cm−1associated with brucite (8). Both features are common to the spectra of previous studies (910) that reported 0.1 to 0.4 wt % H2O in perovskite. Recovery of the sample from the DAC allowed detailed study by TEM (Fig. 1C), which shows that nanoscale, intergranular silicate melt was formed around single crystals of perovskite. The broad, asymmetric absorption band observed in Fig. 1B thus likely represents OH in the melt phase, because the partition coefficient of H2O between ringwoodite and silicate perovskite is about 15:1 (11).


Fig. 1.  Laboratory experiments on hydrous ringwoodite.  (A) Single-crystal of hydrous ringwoodite (blue crystal) containing 1 wt % H2O inside a DAC at 30 GPa. The sample was laser heated to 1600°C in several spots (orange circles) to perform direct transformation to perovskite and (Mg,Fe)O. (B) Synchrotron-FTIR spectra of the recovered sample in three locations: an unheated part of the crystal (spectrum 1) and two locations within laser-heated spots (spectra 2 and 3). FTIR spectra were collected with a 10 μm by 10 μm aperture, illustrated and numbered by white boxes in (A). (C) TEM within a laser-heated spot (position 2) shows crystals of perovskite and intergranular amorphous quench (melt).

Panel A:  High pressure

A photograph of the laboratory sample inside of a DAC.

The brownish regions have been heated to 1600°Celsius with a laser.

This is how scientists recreate conditions similar to the inside of Earth.

Panel B:  Absorption coefficients

The absorption coefficient is a measure of how far light can travel into a medium until it is dissipated.

A high absorption coefficient corresponds to a material that does not allow light to propagate far into it.

Panel C:  TEM

A transmission electron micrograph (TEM) of a laser heated region of hydrous ringwoodite.

Explaining GPa

GPa: Combination of the metric prefix “Giga” and the SI unit of pressure “Pascal”.

Giga = 109. A pascal is equivalent to 1 Newton of force applied over an area of 1 square meter.

1 GPa is equal to 1 billion newtons applied over 1 square meter or 1 newton applied over a billionth of a square meter.

The pressure increases by about 10 GPa for every 300 km depth (i.e. the pressure at 100 km depth is about 3 GPa).

Also, 1 GPa is about 10 kbar (10,000 bars) and the pressure at Earth’s surface is about 1 bar.

So, the experiment conducted here at 30 GPa is roughly equivalent to ~900 km depth, and, 30 GPa = 300 kbar, which is 300,000 times atmospheric pressure. 

Panel C small insert

This small inset figure is an electron diffraction pattern.

Only crystals produce a diffraction pattern like this, with regularly spaced spots (reflecting the ordered arrangement of atoms in a crystal).

This diffraction pattern can be analyzed to show that the spot from which it was taken (black arrow) is a crystal of silicate perovskite.

The area of the TEM image shown by the white arrows was molten during the experiment (note the irregular shape).

Now, cooled to room temperature, the melt became a glass (like obsidian is a glass made from rapidly cooling magma).

Unlike crystals, glasses do not posses a regular and periodic array of atoms and therefore do not produce diffraction patterns.

In this way, the researchers were able to show that a melt surrounded crystals of silicate perovskite (i.e., intergranular melt).

The interpretation of the results is that all the water partitioned into the melt, leaving the perovskite “dry.”

Partial melt in the mantle strongly affects seismic velocities and can create sharp velocity gradients where it is adjacent to subsolidus mantle. If dehydration melting occurs where ringwoodite is entrained downward across the 660, then in situ detection would be feasible in areas with dense seismic sampling. A major component of the EarthScope project (14) is the deployment of broadband seismometers with ~70-km spacing across the United States (Fig. 2A). These data enable imaging of geographic variations in seismic structure near the 660. We isolated conversions of earthquake seismic waves from P-to-S (Ps) as a result of sharp vertical gradients in seismic velocity with receiver function analysis (15). Ps receiver functions were then mapped to depth using P andS tomography models, creating a high-fidelity, common conversion point (CCP) image of vertical velocity gradients near the 660 beneath much of the continental United States (1216).


Fig. 2.  Vertical cross-sections through the CCP image. (A) Map of the study region with broadband seismometers denoted by black triangles and the locations of vertical cross-sections shown in (B) to (E) denoted by white dashed lines. (B) Vertical cross-section through the CCP image. The location of the eastern end of the cross-section is shown as B? in (A). The 660, which separates the transition zone and lower mantle, is clearly imaged by 2.5 to 4% amplitude positive Ps/P arrivals in all cross-sections. Negative arrivals beneath the 660 have amplitudes ≤2.2%. (C to E) Vertical cross-sections through the CCP image beneath the three other lines labeled in (A).

Panel A: Seismic map

A map of the western United States. Each triangle is a seismometer location. It seems that the western U.S. is fairly well covered!

Pay attention to the dotted white lines in Fig. 2A! these indicate cross sections of Earth taken from a depth of 500 to 900 km.

Panel B

The red and blue regions represent changes in seismic wave speed.

The blue stripes represent velocity increases "with depth," and red represent velocity decreases "with depth." 

In a blue region, the wave speed increases.

In a red region, the wave speed decreases.

The blue line across the entire image (where velocities increase) is the 660 km seismic discontinuity, caused by the disappearance of ringwoodite to form the denser assemblage of minerals, silicate perovskite plus ferropericlase, and therefore is observed all over the Earth.

The red blobs below the 660 km discontinuity correspond to regions where velocity decrease with depth – something that does not happen in solid rock.

Because melts have much slower velocity than solid rocks, the researchers interpret these red features to indicate the presence of melt, caused by the dehydration of ringwoodite as mantle material flows downward across the 660 km seismic discontinuity. 

Panel C, D, and E

The X-axis of these graphs corresponds to the Longitude measurement.

Longitude only gives you a sense of location in one direction on the globe.

Arizona, Utah, Idaho, and Montana all lay around 110 degrees west (–110°  in these plots).

Positive-amplitude Ps conversions clearly define the abrupt velocity increase with depth at the 660 beneath the entire array (Fig. 2), and other, weaker discontinuities are sporadically detected in the mantle beneath the EarthScope array (1617). Common secondary features in the CCP image near the 660 are negative-amplitudePs conversions in the uppermost 120 km of the lower mantle. A map of the locations of sub-660 negative Psconversions with magnitude >1.25% of the direct P-wave shows that these features are prevalent beneath a large area including the Great Plains and near the northern margin of the array (Figs. 2 and 3). In contrast, negative Psconversions near the top of the lower mantle are absent beneath the southwestern United States. The amplitude of the negative Ps conversions from beneath the 660 is up to 2.2% of the direct P-wave amplitude. For a negativePs conversion near the top of the lower mantle, 730 km, an amplitude of 2% is consistent with a decrease in shear velocity of 2.6% over a depth interval of ≤20 km based on a synthetic calculation of receiver functions (12). The areas where negative Ps conversions are detected are not correlated with volumes of anomalously low-velocity mantle in tomography images (1618); hence, a thermal origin for the reduced velocities is unlikely.


Fig. 3.  Vertical flow between the transition zone and lower mantle.  (A) The background color shows the vertical flow velocity at the boundary between the transition zone and lower mantle predicted by a mantle circulation model using density structure inferred from the SH11-TX tomography model. White squares denote locations where the seismic CCP image detects velocity decreases with depth in the depth range of 670 to 800 km. (B) The background color shows the vertical flow velocity predicted using the S40RTS tomography model. In both models, velocity decreases near the top of the lower mantle are absent beneath the southwestern United States where upward flow is predicted and prevalent in areas where downward flow into the lower mantle is predicted.

Visualizing mantle flow

Visualizing mantle flow: What would happen if you dropped a lead weight into a swimming pool? Because the lead is denser than the water, you would expect it to sink. Using an initial density distribution, geophysicists can predict the time evolution of a fluid system.

Panel A:  How does the mantle flow?

Check the units on that color bar!

Do you think you could observe the mantle flow speeds with the naked eye?

Panel B:  Flow direction
Note the colorbar; the red end of the spectrum indicates mantle rock flowing upwards towards the crust.
Panel B: Velocity increasing with depth
Note the white squares, Under these regions, seismic velocity increases with depth between 670 km and 800 km.

To assess the correlation between the locations of abrupt velocity decreases near the top of the lower mantle and convective flow patterns, we used numerical models of mantle circulation. Computations were solved for flow using radially varying viscosity and prescribed plate motions at the surface (1219). Regional mantle flow rates are controlled by the poorly constrained ratio between density anomalies and viscosity, but flow directions are mainly determined by density anomaly patterns. To explore this sensitivity, we used multiple tomography models (15, 17,19) to infer density fields. For length scales less than ~500 km, the resulting vertical flow patterns at 660-km depth vary widely between models, but circulation models yield a common long-wavelength pattern of vertical flow across the 660 (Fig. 3). Downward flow through the 660 is dominant beneath the Great Plains and along the U.S.-Canada border, and upward flow is dominant beneath the southwestern United States (Fig. 3). This pattern is driven by large volumes of anomalously high-velocity lower mantle, which are inferred to be subducted slabs sinking beneath central and eastern North America (2022).

Comparing the results of two flow models shows that nearly all the locations of negative Ps conversions beneath the 660 coincide with downward flow (Fig. 3). One flow model is based on an inversion of travel-time data from the EarthScope stations using the TX-2008 global tomography model (20) as a starting model (Fig. 3A, SH11-TX). The other flow model is based on density structure from the global tomography model S40RTS (18), which does not include EarthScope seismic data and hence only recovers long-wavelength structure (Fig. 3B). Consequently, the first flow model includes stronger short-wavelength variations (19), but the long-wavelength pattern is consistent. Comparing the distributions of vertical velocities across the 660 in the entire CCP image and the subset of the CCP image where sub-660 negative Ps conversions are detected demonstrates that the latter area is biased toward regions of downward flow (Fig. 4). For both flow models, less than 3% of sub-660 negativePs conversions are found in areas with upward flow across the 660. Comparisons with smooth three-dimensional velocity structure at 660-km depth, rather than vertical flow velocity, show some bias toward higher tomographically imaged shear velocities in areas with sub-660 negative Ps conversions, but less similarity between the two models than is observed for the comparison with vertical flow.


Fig. 4.  Distributions of vertical flow and seismic velocity variations at 660 km.  (A) Probability density functions (PDFs) of vertical flow velocity at 660-km depth from the convection model using SH11-TX tomography. The PDF in red represents the subset of the CCP image where negative velocity gradients are detected beneath the 660 (sub-660 NVG), and the PDF in black represents the cumulative area from the CCP image. The cumulative area of the CCP image includes the area within 200 km of the seismometers shown in Fig. 2A. (B) PDFs of vertical flow velocity from the convection model using S40RTS tomography. (C) PDFs of mantle shear velocity variations near the 660 from the tomography model SH11-TX. (D) PDFs of mantle shear velocity variations from the tomography model S40RTS.

Average depth dependency

These plots show an averaged depth dependency of mantle flow. Red plots are for mantle volume below 660 km. Black plots are for the cumulative observed mantle volume.

Graphs A and C

Graphs A and C refer to mantle convection models generated by density structure inferred from SH11-TX. 

Graphs B and D

Graphs B and D refer to mantle convection models generated by density structure inferred from S40RTS.

X-axis dimensions

Plots A and B show vertical velocity of mantle fluid (a cumulative look at Figure 3)

Plots C and D show probabilities of shear wave velocity variations (similar to Figure 2 B-E).

Probability distributions

Try to interpret the Probability Distribution Functions (PDFs).

For figure A, the average mantle flow velocity below 600 km is -1 cm/yr.

What do the other PDFs show you?

The correlation between abrupt seismic velocity decreases near the top of the lower mantle and areas of downwelling across the 660 is consistent with the occurrence of dehydration melting as observed in our laboratory experiments. An alternative bulk-compositional origin of low velocities near the top of the lower mantle is segregated basalt that may be neutrally buoyant (23) and would reduce seismic velocities (24).

However, long-term accumulation of basalt near the top of the lower mantle is not expected to be preferentially present where there is downwelling across the 660 and absent where there is not. The areas of downward flow across 660 do not all coincide with local presence of subducted slabs, so a direct link to composition of the sinking Farallon slab cannot explain the negative velocity gradients below 660. Assuming that the velocity reductions result from partial melt, and that the shear-velocity decrease per percent of melt is between 2.6 and 3.8%, as predicted for partial melt near 400-km depth (25), then 0.68 to 1% melt could explain a 2.6% shear velocity reduction indicated by negative Ps conversions with amplitude of 2% in the CCP image.

Prediction of partial melt percentages at 660-km depth for various H2O contents requires knowledge of water partition coefficients between minerals and melts at relevant pressure-temperature (P-T) conditions in the peridotite-saturated compositional system. At present, experiments in the hydrous peridotite system at conditions near the 660 have not been performed. However, using experimental results for partial melting near the 410-km discontinuity (410) in a bulk peridotite system with 1 wt % H2O indicates that ~5% partial melt at 410 km is expected (2627) where the partition coefficient of H2O between wadsleyite and olivine is at least 5:1 (11). We can expect at least 5% partial melt in a bulk 1 wt % H2O peridotite system where the partition coefficient between ringwoodite and silicate perovskite is 15:1 (11). Thus, production of up to 1% melt by dehydration melting of hydrous ringwoodite viscously entrained into the lower mantle is feasible.

The density of hydrous melt near the top of the lower mantle is uncertain, but it is likely buoyant with respect to the top of the lower mantle (28). Hence, we expect that the velocity decreases imaged beneath the 660 are transient features resulting from ongoing downward flow through the 660 that is driven by sinking slabs in the lower mantle. Eventually, the slightly buoyant hydrous melt would percolate upward, returning H2O to the transition zone (4). Dehydration melting has also been suggested to occur where hydrous wadsleyite upwells across the 410 and into the olivine stability field (327). Experiments indicate that hydrous melt is gravitationally stable atop the 410 (28), so once melt is generated, it may remain or spread laterally rather than maintaining a clear correlation with ongoing vertical flow patterns. Seismic detections of a low-velocity layer atop the 410 are common but laterally sporadic beneath North America and globally (2930). The combination of dehydration melting driven by downwelling across the 660 and upwelling across the 410 could create a long-term H2O trap in the transition zone (4).

Supplementary Materials

Materials and Methods

Figs. S1 to S4

References (3138)

Additional Data Tables S1 to S3

References and Notes

  1. E. Saal, E. H. Hauri, C. H. Langmuir, M. R. Perfit, Vapour undersaturation in primitive mid-ocean-ridge basalt and the volatile content of Earth’s upper mantle. Nature 419, 451–455 (2002).

  2. J. R. Smyth, β-Mg2SiO4 a potential host for water in the mantle? Am. Mineral. 72, 1051 (1987). Abstract

  3. D. L. Kohlstedt, H. Keppler, D. C. Rubie, Solubility of water in the α, β and γ phases of (Mg, Fe)2SiO4. Contrib. Mineral. Petrol. 123, 345–357 (1996).

  4. D. Bercovici, S. Karato, Whole-mantle convection and the transition-zone water filter. Nature 425, 39–44 (2003).

  5. D. G. Pearson, F. E. Brenker, F. Nestola, J. McNeill, L. Nasdala, M. T. Hutchison, S. Matveev, K. Mather, G. Silversmit, S. Schmitz, B. Vekemans, L. Vincze, Hydrous mantle transition zone indicated by ringwoodite included within diamond. Nature 507, 221–224 (2014).

  6. H. Zhu, E. Bozdag, T. S. Duffy, J. Tromp, Seismic attenuation beneath Europe and the North Atlantic: Implications for water in the mantle. Earth Planet. Sci. Lett. 381, 1–11 (2013).

  7. Kelbert, A. Schultz, G. Egbert, Global electromagnetic induction constraints on transition-zone water content variations. Nature 460, 1003–1006 (2009).

  8. N. Bolfan-Casanova, H. Keppler, D. C. Rubie, Water partitioning at 660 km depth and evidence for very low water solubility in magnesium silicate perovskite. Geophys. Res. Lett. 30, 1905 (2003).

  9. M. Murakami, K. Hirose, H. Yurimoto, S. Nakashima, N. Takafuji, Water in Earth’s lower mantle. Science 295, 1885–1887 (2002).

  10. K. Litasov, E. Ohtani, F. Langenhorst, H. Yurimoto, T. Kubo, T. Kondo, Water solubility in Mg-perovskites and water storage capacity in the lower mantle. Earth Planet. Sci. Lett. 211, 189–203 (2003).

  11. T. Inoue, T. Wada, R. Sasaki, H. Yurimoto, Water partitioning in the Earth's mantle. Phys. Earth Planet. Inter. 183, 245–251 (2010).

  12. Materials and methods are available on Science Online.

  13. S. D. Jacobsen, J. R. Smyth, H. A. Spetzler, C. M. Holl, D. J. Frost, Sound velocities and elastic constants of iron-bearing hydrous ringwoodite. Phys. Earth Planet. Inter. 143-144, 47–56 (2004).

  14. R. A. Kerr, Solid-Earth science. Geophysical exploration linking deep Earth and backyard geology. Science 340, 1283–1285 (2013).

  15. J. P. Mercier, M. G. Bostock, A. M. Baig, Improved Green’s functions for passive source structural studies. Geophysics 71, SI95–SI102 (2006).

  16. Schmandt, K. G. Dueker, E. D. Humphreys, S. M. Hansen, Hot mantle upwelling across the 660 beneath Yellowstone. Earth Planet. Sci. Lett. 331-332, 224–236 (2012).

  17. Tauzin, R. D. van der Hilst, G. Wittlinger, Y. Ricard, Multiple transition zone seismic discontinuities and low velocity layers below western United States. J. Geophys. Res. Solid Earth 118, 2307–2322 (2013).

  18. J. Ritsema, A. Deuss, H. J. van Heijst, J. H. Woodhouse, S40RTS: A degree-40 shear-velocity model for the mantle from new Rayleigh wave dispersion, teleseismic travel-time and normal-mode splitting function measurements. Geophys. J. Int. 184, 1223–1236 (2011).

  19. T. W. Becker, C. Faccenna, E. D. Humphreys, A. R. Lowry, M. S. Miller, Static and dynamic support of western U.S. topography. Earth Planet. Sci. Lett. (2014).

  20. N. A. Simmons, A. M. Forte, S. P. Grand, N., A. Forte, S. P. Grand, Joint seismic, geodynamic and mineral physical constraints on three-dimensional mantle heterogeneity: Implications for the relative importance of thermal versus compositional heterogeneity. Geophys. J. Int. 177, 1284–1304 (2009).

  21. B. Steinberger, Slabs in the lower mantle–results of dynamic modeling compared with tomographic images and the geoid. Phys. Earth Planet. Inter. 118, 241–257 (2000).

  22. M. Forte, R. Moucha, N. A. Simmons, S. P. Grand, J. X. Mitrovica, Deep mantle contributions to the surface dynamics of the North American continent. Tectonophysics 481, 3–15 (2010).

  23. T. A. Lee, W. P. Chen, Possible density segregation of subducted oceanic lithosphere along a weak serpentinite layer and implications for compositional stratification of the Earth’s mantle. Earth Planet. Sci. Lett. 255, 357–366 (2007).

  24. W. B. Xu, C. Lithgow-Bertelloni, L. Stixrude, J. Ritsema, The effect of bulk composition and temperature on mantle seismic structure. Earth Planet. Sci. Lett. 275, 70–79 (2008).

  25. S. Hier-Majumder, A. Courtier, Seismic signature of small melt fraction atop the transition zone. Earth Planet. Sci. Lett. 308, 334–342 (2011).

  26. M. M. Hirschmann, T. Tenner, C. Aubaud, A. C. Withers, Dehydration melting of nominally anhydrous mantle: The primacy of partitioning. Phys. Earth Planet. Inter. 176, 54–68 (2009).

  27. T. J. Tenner, M. M. Hirschmann, A. C. Withers, P. Ardia, H2O storage capacity of olivine and low-Ca pyroxene from 10 to 13 GPa: Consequences for dehydration melting above the transition zone. Contrib. Mineral. Petrol. 163, 297–316 (2012).

  28. T. Sakamaki, A. Suzuki, E. Ohtani, Stability of hydrous melt at the base of the Earth’s upper mantle. Nature 439, 192–194 (2006).

  29. B. Schmandt, K. G. Dueker, S. M. Hansen, J. J. Jasbinsek, Z. Zhang, A sporadic low-velocity layer atop the western U.S. mantle transition zone and short-wavelength variations in transition zone discontinuities. Geochem. Geophys. Geosyst. 12, Q08014 (2011).

  30. B. Tauzin, E. Debayle, G. Wittlinger, Seismic evidence for a global low-velocity layer within the Earth's uppermantle. Nat. Geosci. 3, 718–721 (2010).

  31. L. Vinnik, Detection of waves converted from P to SV in the mantle. Phys. Earth Planet. Inter. 15, 39–45 (1977).

  32. S. Hansen, K. Dueker, P- and S-wave receiver function images of crustal imbrication beneath the Cheyenne belt in southeast Wyoming. Bull. Seismol. Soc. Am. 99, 1953–1961 (2009).

  33. B. Efron, R. Tibshirani, Bootstrap methods for standard errors, confidence intervals, and other measures of statistical accuracy. Stat. Sci. 1, 54–75 (1986).

  34. B. Schmandt, Mantle transition zone shear velocity gradients beneath USArray. Earth Planet. Sci. Lett. 355-356, 119–130 (2012).

  35. J. Park, Surface waves in layered anisotropic structures. Geophys. J. Int. 126, 173–183 (1996).

  36. B. H. Hager, R. J. O'Connell, A simple global model of plate dynamics and mantle convection. J. Geophys. Res. 86, 4843–4867 (1981).

  37. Ghosh, T. W. Becker, E. D. Humphreys, Dynamics of the North American continent. Geophys. J. Int. 194, 651–669 (2013).

  38. J. R. Smyth, C. M. Holl, D. J. Frost, S. D. Jacobsen, F. Langenhorst, C. A. McCammon, Structural systematics of hydrous ringwoodite and water in Earth’s interior. Am. Mineral. 88, 1402–1407 (2003).

Acknowledgments: Seismic data were acquired from the IRIS Data Management Center. This work was supported by NSF grants EAR-0748707 to S.D.J. and EAR-1215720 to T.W.B., and by the David and Lucile Packard Foundation and Carnegie/DOE Alliance Center (CDAC) to S.D.J. Portions of this work were performed at GSECARS (Sector 13), Advanced Photon Source (APS), Argonne National Laboratory. GSECARS is supported by the NSF (EAR-1128799) and U.S. Department of Energy (DOE) (DE-FG02-94ER14466). Use of the APS was supported by the DOE-BES (Basic Energy Sciences) (DE-AC02-06CH11357). Portions of this work were performed at beamline U2A of the National Synchrotron Light Source (NSLS), Brookhaven National Laboratory. U2A is supported by COMPRES (Consortium for Materials Properties Research in Earth Sciences) under NSF Cooperative Agreement EAR 11-57758 and DOE-NNSA (National Nuclear Security Administration) (DE-FC-52-O8NA28554, CDAC). Use of the NSLS was supported by the DOE-BES (DE-AC02-98CH10886). We thank S. Demouchy, D. J. Frost, E. H. Hauri, M. M. Hirschmann, F. Langenhorst, J. F. Lin, G. Shen, V. B. Prakapenka, and J. R. Smyth for discussions and help with experiments. B.S. and S.D.J. designed the research and wrote the paper. B.S. conducted the seismological research, and S.D.J. performed the experiments. T.W.B. produced the mantle circulation models, Z.L. contributed to the FTIR experiments, and K.G.D. contributed to seismic imaging. All authors participated in data interpretation and contributed to the manuscript.