Research Article

Seismostratigraphy and Thermal Structure of Earth's Core-Mantle Boundary Region

See allHide authors and affiliations

Science  30 Mar 2007:
Vol. 315, Issue 5820, pp. 1813-1817
DOI: 10.1126/science.1137867


We used three-dimensional inverse scattering of core-reflected shear waves for large-scale, high-resolution exploration of Earth's deep interior (D″) and detected multiple, piecewise continuous interfaces in the lowermost layer (D″) beneath Central and North America. With thermodynamic properties of phase transitions in mantle silicates, we interpret the images and estimate in situ temperatures. A widespread wave-speed increase at 150 to 300 kilometers above the coremantle boundary is consistent with a transition from perovskite to postperovskite. Internal D″ stratification may be due to multiple phase-boundary crossings, and a deep wave-speed reduction may mark the base of a postperovskite lens about 2300 kilometers wide and 250 kilometers thick. The core-mantle boundary temperature is estimated at 3950 ± 200 kelvin. Beneath Central America, a site of deep subduction, the D″ is relatively cold (ΔT = 700 ± 100 kelvin). Accounting for a factor-of-two uncertainty in thermal conductivity, core heat flux is 80 to 160 milliwatts per square meter (mW m–2) into the coldest D″ region and 35 to 70 mW m–2 away from it. Combined with estimates from the central Pacific, this suggests a global average of 50 to 100 mW m–2 and a total heat loss of 7.5 to 15 terawatts.

At a depth of ∼2890 km, the core-mantle boundary (CMB) separates turbulent flow of liquid metals in the outer core from slowly convecting, highly viscous mantle silicates. The 200- to 300-km-thick thermochemical boundary layer on the mantle side—the so-called D″ layer—is enigmatic (1, 2), but a recently discovered phase transition from perovskite (pv) to postperovskite (ppv) in (Mg,Fe)SiO3 (35) begins to explain seismologically observed complexity [e.g., (6)]. If the ppv transition occurs, one can, in principle, estimate in situ variations in temperature from the pressure-temperature dependence (that is, the Clapeyron slope) and the seismologically inferred location of the associated interface (7). Steep (conductive) thermal gradients in D″ can produce multiple crossings of the phase boundary, and identification of associated seismic signals offers new opportunities for constraining (local) core heat flux (8, 9).

Seismic (transmission) tomography delineates smooth changes in wave speed associated with mantle convection (Fig. 1A), but one must focus on the scattered wave field to image interfaces associated with transitions in mineralogy or composition. Scattering of PKP (the main P wave propagating through the core) in D″, first recognized in the early 1970s (10), has been used to constrain stochastic models of deep mantle structure [e.g., (11)], but the most detailed and accurate constraints on D″ structure to date have come from forward modeling of shear waves reflected at or near the CMB (12, 13). This approach has its drawbacks, however. First, it requires prior knowledge about the target structure and often assumes relatively simple geometries, the uniqueness of which is not easily established. Second, it relies on signal associated with near- and postcritical incidence, which limits radial resolution and the CMB regions that can be studied (14). The small distance window can also reduce the available source-receiver azimuths, which can degrade imaging in directions perpendicular to dominant sampling (15).

Fig. 1.

(A) In the mantle, tomography depicts smooth P-wave speed variations (45) associated with deep subduction (blue structure in center of section) beneath Central America; inverse scattering constrains deep mantle reflections in the lowermost 400 km. Superimposed on the tomography/scattering image are schematic ray paths of ScS waves reflecting at and above the CMB: a depicts specular CMB reflections, which contribute to the main ScS arrival in the seismogram shown, b depicts scattering above the CMB, which produces precursors, and c depicts nonspecular reflections (at CMB or above it), which arrive mainly in the coda of ScS (we consider precritical reflections only). (B) Geographic map of the study region with bounce points of the ∼80,000 ScS data points (specular CMB reflections) used in our inverse scattering study (17) (black dots), along with the lines of cross section for the 2D images shown in Figs. 2 and 3. At each yellow dot, the generalized Radon transform (SOM Text S1) produces radial profiles of contrasts in elastic properties (inset). The color does not scale with impedance contrast but with the reflection coefficient (14) (blue, R >0; red, R < 0). Only structure outside the 75% confidence level (thin lines) is discussed here.

Stacking of data from dense source and receiver arrays can enhance weak signals and has begun to produce tantalizing images of D″ stratification (9, 15, 16). However, most applications still rely on near-critical data and require strict data selection and visual inspection, which prohibits application to very large data sets

Large-Scale Deep Earth Exploration Seismology

Systematic three-dimensional (3D) imaging of large D″ regions becomes feasible if one can (i) rely on fewer a priori assumptions, so that structures of unknown geometries can be detected; (ii) use efficient methods to detect, extract, and interpret subtle signals in large volumes of broad-band waveforms available through modern data centers; (iii) use data from more source-receiver distances; and (iv) validate results by means other than labor-intensive forward modeling. To achieve this, we combine inverse scattering—specifically, a generalized Radon transform (GRT) of the wave field comprising core-reflected shear waves (ScS) (SOM Text S1)—with statistical methods (SOM Text S2) to produce images of D″ structure and to estimate their uncertainty. This approach benefits from the superior 3D sampling afforded by narrow- and wide-angle data (SOM Text S3) and requires few a priori assumptions about target structures and no visual inspection of data.

For the source-receiver distances considered, specular (mirror-like) reflections of interfaces above the CMB arrive before ScS proper, whereas nonspecular CMB reflections arrive after the main phase (Fig. 1A). The GRT uses data redundancy to identify coherent signal from (single) scattering and localizes the causative structures (17). Statistical inference, a nonlinear process that replaces stacking, suppresses the effects of uneven source-receiver distribution, incorrect wave-speed models, and multiple scattering, and yields rigorous Bayesian confidence levels (18). The latter is used as the initial tool for model validation, which allows routine analysis.

We used ∼80,000 broad-band (1 to 20 s) records of the ScS wave field for large-scale, high-resolution exploration of the D″ beneath Central and North America (Fig. 1B) (SOM Text S3). The GRT maps data from wide angles [epicentral distances between 43° and 80° (fig. S3)] into multiple images of the same structure (as illustrated in fig. S1), which are then combined into a radial profile of medium contrasts at a specified CMB location (Fig. 1B, inset). These reflectivity profiles are aligned on the CMB and corrected for 3D mantle heterogeneity inferred from global tomography (19). Juxtaposition of these radial profiles produces 2D (or 3D) images that reveal the lateral extent of medium contrasts and lowermost mantle stratification (fig. S2). Figure 2 clearly reveals the CMB beneath the Americas, but our subject is the rich pattern of scatterers and interfaces up to 400 km above it.

Fig. 2.

Three-dimensional exploration seismology of the lowermost mantle. Seismic images of the lowermost mantle (CMB to 400 km above it) are produced by lateral juxtaposition of radial general Radon transform profiles (fig. S2) calculated at image points along the sectionlines shown in Fig. 1. Structure outside 75% confidence bands (18) includes the CMB (at 0 km) and several scatter interfaces above it. Thinly dashed lines indicate scatter interfaces (L1, L2) highlighted in Fig. 3. This 3D rendition illustrates the large spatial scales over which inverse scattering with the ScS wave field can be used to explore the lowermost mantle. Points X1 and X2 indicate the section intersections. The background color depicts the prediction of the height above the CMB of a presumed phase transformation with a Clapeyron slope of 6 MPa/K (21). In most of this region, the correlation between L1 and the predicted values is very good

D″ Structure and Temperature

Figure 3 displays the GRT images (top) along with variations in S-wave speed, dlnVS, derived from tomography (bottom). To emphasize the relations among them, we superimpose the strongest scatterers (at more than 75% confidence) on the wave-speed maps. We focus on the structures labeled L1, L2, and L3.

Fig. 3.

Reflectivity from inverse scattering, at more than 75% confidence (18) (top) and S-speed (dlnVS) from tomography (19) (bottom). Scatter images are obtained by interpolation between GRT profiles (Fig. 1B, inset) calculated every 1° (∼60 km at CMB) along sections A-A′, B-B′, and C-C′ in Fig. 1. For the frequencies and incidence angles used, the radial resolution is ∼10 km. The color scale for tomography is given between B-B′ and C-C′. (A) L1, L2, and L3 label the scatter interfaces (thinly dashed). (B and C) The associated scatterers (visually enhanced) are superimposed on the tomography profiles, with dark gray depicting positive reflections and pink/red negative contrasts. Interface L1 aligns increases of wave speed with increasing depth; L2 delineates a decrease; L3 is more ambiguous but generally coincides with a wave-speed increase. Whereas L1 and L2 are piecewise continuous, L3 has an intermittent, en echelon appearance. The solid (dashed) blue lines in the bottom panels depict the phase transition location predicted by (21) (22). Points A1 and A2 and points B1 to B4 on L1 and L2 are used for temperature calculations (Fig. 4). The gray scale below B-B′ depicts the lateral variation in temperature gradient along the CMB (for γppv = 10 MPa/K) estimated from (29). In the central portion of the section, dT/dz cannot be determined directly because the occurrence of the double crossing cannot be resolved.

Postperovskite transition. Structure L1 delineates an increase of wave speed with depth. In most places L1 is unambiguous because the amplitude of the pulses that define it can be a substantial fraction of the CMB peak (Fig. 1B, inset), depending on the angle at which L1 is sampled (fig. S1). Our GRT results pertaining to L1 corroborate more localized previous studies of the main D″ discontinuity beneath Central America [see table 1 in (15)], but they also constrain structure far into previously uncharted D″ territory. The depth variation of L1 correlates strongly with changes in S-wave speed (20). If the latter is (partly) of thermal origin, the GRT images provide direct evidence for a widespread presence of a D″ interface with a positive pressure-temperature relationship.

Sidorin et al. (21) postulated that a phase transition with Clapeyron slope γ = 6 MPaK–1 marks the top of D″. A recent analysis (22) puts the main discontinuity ∼95 km deeper (dashed blue line, Fig. 3), but the original prediction (solid blue lines) is quite close to L1. The larger topography on the latter suggests that γ is more than 6 MPaK–1, which is consistent with estimates from mineral physics for the ppv transition (7), or that the actual wave-speed variations are larger than inferred tomographically. If, indeed, interface L1 marks the ppv transition, we can estimate in situ mantle temperatures from its topography (Fig. 4). For γppv = 10 MPaK–1, the temperatures at B1 and B2 are ∼2000 and ∼2900 K, respectively, and the difference between A1 and A2 is ∼600 K.

Fig. 4.

(A) Temperature (x axis) versus distance above CMB (y axis). As a point of reference for producing absolute temperature and depth (pressure), we assume Pppv = 124 GPa and Tppv = 2500K, and we consider Clapeyron slopes γppv = 6, 8, 10, and 12 MPa/K. If L1 in Fig. 3 represents the ppv transition, the temperature at points B1 and B2 can be estimated as γppv = 10 MPa/K, Tppv,B2 = 2000K, and Tppv,B1 = 2900 K. Dotted lines depict estimated geotherms through B1 and B2; the shallow part is adiabatic, but toward the CMB the conductive geotherm is described as an error function (the change of adiabat to conductive is not known, however). (B) Cold, intermediate, and hot geotherms. As the mantle temperature decreases, a ppv transition occurs at increasing distances above CMB, and a second crossing would occur at decreasing height above CMB. The geotherm through the ppv transition (B3) and the back-transformation (B4) is calculated (29) using a thickness of the hypothetical boundary layer, H, of 150 km. (C) Mantle temperature, Tcmb, at the CMB (dashed lines) and the temperature change, ΔTTBL, across the thermal boundary layer (solid lines) as a function of TBL thickness, H, and for γppv = 8, 10, and 12 MPa/K. Stable estimates are obtained for H = 100 to 200 km (gray shading).

On long wavelengths (1000 to 2000 km), the changes in temperature along L1 reflect the shoaling of the phase boundary toward D″ regions where relatively cool slab debris can be expected. Changes on a scale of ∼450 km relate to the margins of these regions and to localized slow regions. Steeper topography is observed locally, for instance at ∼800 km along section B-B′, which is close to a steplike jump previously attributed to folding of subducted slab (23). Seismic tomography (Fig. 1A) and the above temperature differences both support the presence of slab debris, and slab buckling is indeed likely to occur in this region (24). Drawing conclusions from isolated structures can be misleading, however, and partial reconstruction of multiple scatter interfaces can produce ambiguity. Interface L1 aligns the strongest scatterers and is consistent with (21), but we cannot rule out that between 0 and 800 km (in section B-B′) it continues southeastward with little topography and that the stronger structure at ∼150 km represents interface L3.

Postperovskite lens? The GRT images reveal much structural complexity between the CMB and L1, the presumed top of D″. Particularly intriguing is a wave-speed reduction (producing the negative pulse, L2, in Fig. 1B, lower left) that delineates a large-scale (∼2300 km) concave-up interface in the lowermost 100 km or so of the mantle. Because of side lobes of the CMB reflection, L2 cannot be established between 1000 and 2000 km along B-B′, but interfaces that dip toward the seismologically fast central region are resolved on either side of this patch. In this region, (15) reported a wave-speed decrease further above CMB and more parallel to the top D″ interface.

Published estimates of CMB temperature Tcmb vary from 3750 K to 4800 K (2527), and most exceed Tppv. The latter implies that ppv is instable near the CMB and that there is a second crossing of the phase boundary. The back transition (ppv→pv) occurs deeper in colder regions (Fig. 4), thus creating ppv lenses beneath mantle downwellings (8, 16). This transition is hard to detect (28), but recent studies provide evidence for it in the lowermost mantle beneath the Pacific (9) and the Cocos plate (22). In our GRT profiles, the negative pulses are statistically significant (Fig. 1 and fig. S2), and both the sign and the deepening toward regions of presumably lower D″ temperatures is consistent with ppv→pv. The presence of L2 and the anticorrelated topographies of L1 and L2 provide new evidence that ppv-rich lenses do indeed exist. The lens beneath Central America has a width of ∼2300 km and a thickness of ∼250 km; L1 and L2 merge (near B1) ∼125 km above the CMB (T = 2900 K; P = 128 GPa). Outside this lens, the temperature may be too high for ppv to be stable, except, perhaps, in regions of iron enrichment (9).

Mantle temperature near CMB. Whereas the ppv transition constrains temperatures some 150 to 300 km above the CMB, double crossings provide insight into the thermal structure closer to the core (8, 9). We know neither Tcmb nor the thickness H and temperature change ΔTTBL across a thermal boundary layer (TBL), but with reasonable assumptions we can estimate Tcmb and ΔTTBL by fitting error functions to temperatures at L1 and L2 as a function of H (29). H increases from colder to warmer D″ regions. If the CMB is isothermal, and not arbitrarily hot, we obtain stable estimates of Tcmb and ΔTTBL for H such that the ppv transition (e.g., at B3) occurs above or near the top of the TBL and the back transition (e.g., at B4) occurs well within it. For H= 100 to 200 km and γppv = 10 MPa/K, we infer Tcmb = 3950 ± 200 K and ΔTTBL = 1600 to 1400 K (Fig. 4C). These values depend on the Clapeyron slope and, in particular, on the phase-boundary temperature at the CMB; we estimate that TcmbTppv,cmb + 290 K. Our Tcmb is lower than, but within error consistent with, the estimate of 4100 K in (9).

Despite large uncertainties, Tcmb constrains the melting curve of iron in the liquid core because the temperature just below the CMB must exceed the melting temperature of core materials. Experimental and computational estimates of the melting point of pure Fe at CMB pressures span a range from 3200 ± 200 K (30) to 4800 ± 200 K (31). Minor elements can reduce this by as much as 700 to 1000 K (27), but experiments reveal that for oxygen and sulfur the effect is small (32). If Tcmb = 3950 ± 200 K, the melting point in (31) may thus be too high.

Core heat flux. To estimate heat flux, we need to know thermal conductivity κ and the temperature gradient dT/dz across the TBL. A lower bound on the latter is given by γ ppv–1 (8); for γppv = 7 to 10 MPa/K, this implies dT/dz|min = 5.8 to 8.3 K/km. The gradient must be steeper than γ ppv–1 for a double crossing to occur. Our temperature estimates imply gradients of 7 to 16 K/km (Fig. 3B).

Even if we assume that these gradients are reasonably accurate, the regional and, in particular, global heat flux remains subject to considerable uncertainty. First, the region where dT/dz can be estimated does not span a diverse geodynamical environment. We consider the steep gradients as representative of cold D″ areas. Our study region does not include large-scale low wave-speed regions, where higher mantle temperatures and lower gradients are expected. We could speculate that our regional minimum (∼7 K/km) is close to the global average, but this would be a lower bound if the anomalously low wave speeds are (in part) due to chemistry [e.g., (33)]. Lay et al. (9) consider effects of iron and estimate the thermal gradient beneath the central Pacific at 8.5 ± 2.5 K/km.

Second, the value of thermal conductivity is debated. The often-used κ = 10 Wm–1K–1 (25) would imply a regional variation in heat flux qcmb of 115 ± 45 mWm–2, compared with 85± 25 mWm–2 below the Pacific (9). If one assumes that these regional values represent extremes, a reasonable average flux would be qcmb ≈ 100 mWm–2 (corresponding to a global heat loss across the CMB of ∼15 TW). However, κ may be as small as 5 Wm–1K–1 (34, 35). Depending on the amount of radiative conduction, the conductive component to the heat loss across the CMB is thus uncertain by (at least) a factor of two (Qcmb ≈ 7.5 to 15 TW). It is likely, however, that the flux into the base of the mantle exceeds the heat conducted down the core adiabat, which has implications for models of thermal coremantle coupling, core evolution, and the distribution of heat-producing elements (36, 37).

Other D″ interfaces? Over the entire geographical region studied here, the GRT images suggest the presence of scatterers (L3 in Fig. 3) some 100 km below the presumed top of the D″. Further study is required to establish whether it is the partial detection of a continuous interface (or multiple interfaces) or whether it represents a depth range with intermittent, en echelon scatter zones. The sign of change in elastic properties is not unequivocal because of ambiguity between main pulse and side lobes. Positive and negative changes may both occur, but variations in S-wave speed in this depth interval suggest the preponderance of radial increases in wave speed.

Slab folding (16, 24), along with preserved compositional heterogeneities, can perhaps provide the complexity detected here. Alternatively, local changes in temperature or chemical composition [e.g., in iron content (38) or partitioning between pv and magnesiowüstite (39)] can readily create multiple crossings of the phase boundary. Alternating, irregular ppv and pv lenses (or layers) may be more realistic than a single ppv lens (40) and can explain some of the multiple, but intermittent, scatter interfaces. This calls for further investigation by means of joint imaging and geodynamical and petrological modeling.

New Opportunities for Deep-Earth Imaging

The results presented here should be regarded as first steps, but they show the potential of using inverse scattering techniques originally developed to search for hydrocarbon reservoirs for the systematic, high resolution investigation of the lowermost mantle over large regions, including hitherto uncharted D″ territory. Application to Central America allows verification against previous results, but similar high-resolution D″ imaging should be possible elsewhere, for instance, beneath large parts of Eurasia. As we suppose was the case when seismologists pondered over the first reflection profiles of Earth's crust, image quality is not as good everywhere as one would wish, and even first-order observations can be puzzling. However, we expect that refinement of inverse scattering, accounting for increasingly many complexities and using exponentially growing data sets will remove existing ambiguities and provide insight into the structure, dynamics, and evolution of this remote frontier of cross-disciplinary research.

Supporting Online Material

SOM Text

Figs. S1 to S3


References and Notes

View Abstract

Navigate This Article