Dipping Low-Velocity Layer in the Mid-Lower Mantle: Evidence for Geochemical Heterogeneity

See allHide authors and affiliations

Science  19 Mar 1999:
Vol. 283, Issue 5409, pp. 1888-1892
DOI: 10.1126/science.283.5409.1888

This article has a correction. Please see:


Data from western United States short-period seismic networks reveal a conversion from an S to a P wave within a low seismic velocity layer (greater than or equal to the 4 percent velocity difference compared to the surrounding mantle) in the mid–lower mantle (1400 to 1600 kilometers deep) east of the Mariana and Izu-Bonin subduction zones. The low-velocity layer (about 8 kilometers thick) dips 30° to 40° southward and is at least 500 kilometers by 300 kilometers. Its steep dip, large velocity contrast, and sharpness imply a chemical rather than a thermal origin. Ancient oceanic crust subducted into the lower mantle is a plausible candidate for the low-velocity layer because of its broad thin extent.

Planetary differentiation and convection create heterogeneities in the mantle, which are ultimately related to the cooling of Earth over the age of the solar system. Seismically, these heterogeneities express themselves in velocity heterogeneity due to variations in temperature, bulk composition, and phase changes in the mantle. In the upper mantle, all three mechanisms act, accounting for the greater velocity variations there (±5%) in comparison to the lower mantle (±0.5%) (1). In the lower mantle, it is not clear which mechanisms act, because the lower mantle seems comparatively homogeneous except for the lowermost 200 to 300 km of the mantle (D") and upper mantle slab extensions (2). We used a seismic array–based technique to find what are likely to be the more subtle lower mantle heterogeneities, which should provide information on Earth's longer term evolution.

Evidence for seismic velocity anomalies smaller than 500 km comes from the discrepancy between shear-wave velocities, which were derived from body waves and normal mode studies (3), and from a recent study of global stacking of core phase PKP precursors (4), yielding statistical models of heterogeneity in the mantle but not their individual positions. We refined the size, shape, and velocity contrast of one previously recognized heterogeneity in the mid–lower mantle (5) by analyzing later arrivals afterP waves from intermediate to deep focus earthquakes at the Mariana trench to the south of the Izu-Bonin trench (Fig. 1).

Figure 1

Map of the study area, showing earthquakes, S-to-P wave conversion points, and present and past trench locations. Open diamonds indicate earthquake epicenters (Table 1). Thick solid lines represent the present trench lines, and broken lines indicate the reconstructed locations of the Indonesia trench during the Mesozoic era (13). The trench migrated southwestward during the Mesozoic era. Solid squares denote the S-to-P wave conversion point locations. The ray paths projected onto the horizontal plane of a direct P wave and of an S-to-Pconverted wave are shown with thin solid lines. (Inset) A cross section of the S-to-P wave conversion points along section A–B. Solid squares show the locations of the conversion points, and solid bars represent dip angles of the conversion interfaces, which were determined by applying Snell's law to the rays of incoming S waves and convertedP waves.

The data are short-period seismograms from western United States networks for intermediate (∼200 km) to deep (∼600 km) earthquakes that occurred from 1993 to 1996. The later arrival (indicated by arrows in Fig. 2 and called “later phase” hereafter) for these events shows a systematic focal depth-delay time trend (Table 1). This result indicates that the later phase was caused by the conversion of an Swave to a P wave at a mid–lower mantle velocity heterogeneity (5). We located the source of the later phase for the five events on the basis of maximizing the scattering likelihood, which was computed with the observed travel timeδt, slowness δp, and arrival azimuth δφ in relation to direct P waves (Table 1) (6). The conversion points, or scatterers, form a plane (500 km by 300 km) dipping ∼30° southward (Fig. 1, inset, and Table 1). The plane's dip angle is consistent with dips that were independently determined, assuming that Snell's law holds at the wave conversion point across the interface (Fig. 1, inset) (5). We conclude from this mapping that the lower mantle heterogeneity is a nearly planar interface extending at least several hundred kilometers and that the later phases are the S-to-P converted waves at this dipping interface (Fig. 3A).

Figure 2

Record sections of the vertical components of short-period seismograms. Seismograms are bandpass filtered from 0.2 to 2 Hz. The horizontal axes are the delay times (in seconds) after the onsets of direct P waves. The vertical axes are the epicentral distances in degrees. Event 3 in Table 1, recorded at the University of Washington Network (left). Event 6 in Table 1 (24 August 1995, 0628 UT), recorded at the Northern California Earthquake Center Network (middle). Event 6 in Table 1 (24 August 1995, 0155 UT), recorded at the University of Washington Network (right). The isolated and impulsive phases denoted by arrows (not predicted by any standard Earth models) are the later phases we studied.

Figure 3

(A) Schematic models ofS to P–wave conversion at the mid–lower mantle heterogeneity. The conversion occurs at an interface (first-order discontinuity), across which seismic velocities increase (top). The conversion occurs at a thin low-velocity layer (bottom). The incident angle of the incoming S wave is 20°. (B) Comparisons between the observed waveforms of the later phase (solid lines) and synthetics (22). These examples show the cases recorded at the Northern California array (NC) or at the University of Washington array (UW) (5) (event 3 and two events included in event 6, Table 1). Broken lines indicate a 4% velocity increase across a single interface; dotted lines indicate a 4% low-velocity layer that is 8 km thick. Numbers at the right of the linearly stacked waveforms are the amplitudes of the later phase in relation to direct P. Arrows indicate the downswing onsets of the later phases. Radiation intensities were computed on the basis of the Harvard Centroid Moment Tensor solutions, allowing a 5° uncertainty of the nodal planes and taking the maximum possible values (9), which can be nearly four times larger than the minimum values. Waveforms of the deeper and shallower events are both consistent with the low-velocity thin-layer model. (C) Cross correlation between the observed and synthetic waveforms with normalized amplitudes of synthetic waveforms averaged over deep events. Broken line indicates cross correlations as a function of layer thickness for a low-velocity thin-layer model. Cross correlation reaches a maximum at a thickness of 8 km, and the maximum for the velocity increase at a single interface is shown with a open square at a layer thickness of 0 km. The dotted line indicates the amplitudes of the S-to-P converted waves, which were normalized to the maximum that was obtained for a thickness of 6 to 7 km. The closed circle at a thickness of 0 km represents the relative amplitude for the single-interface model and t* represents the value of δt* defined in (22).

Table 1

Event list and conversion point locations. Lat., latitude; Long., longitude; deg., degree; mb, body-wave magnitude.

View this table:

The stacked later phase waveforms differ slightly from the directP waveform stacks. To model the interface's properties, we compared synthetic waveforms with the stacked waveforms of the later phase (7) (Fig. 3B). Two classes of interface models in which elastic property discontinuously changes across planar boundaries can explain the waveforms (Fig. 3A). The first has a ≥8% velocity increase from above the interface to below (5). The other has a thin ∼8-km-thick layer of ≥4% lower velocity than the surrounding mantle. This thin low-velocity layer model gives systematically larger correlation values by 0.05 to 0.1 than the first model for almost all array-event pairs (Fig. 3C). The synthetic waveforms for the second model yield the subtle but characteristic downswing onset of the later phase, which is not produced by the first model (Fig. 3B, arrows). On this account, we favor the second, thin low-velocity layer model for the source of S-to-Pconverted waves in the mid–lower mantle, which also requires a velocity heterogeneity that is half that of the first model (8).

The relative amplitudes of the later phase to direct P waves depend on its frequency content. At higher frequencies, amplitudes fall off above 0.25 Hz, whereas at lower frequencies, they fall off below 0.15 Hz (5). The lower frequency falloff is qualitatively consistent with the model of a layer that is not thick in comparison to the wavelength of incoming waves (5 to 10 km). The estimated velocity anomaly has a large uncertainty (at least a factor of 2) because of the factors that influence the observed amplitude ratios. These factors include the attenuation structure of the mantle between the foci and the conversion points and the relative radiation intensity of theS wave to the direct P wave, resulting from uncertainties in the focal mechanisms of the earthquakes (9). The shear velocity anomaly of 4% is a lower bound when we use lower attenuation and higher estimates of S-wave radiation intensity (see the caption of Fig. 3B). If we adopt the preliminary reference Earth model's (PREM's) attenuation model (10) or small estimates of the radiation intensity or both, the amount of the velocity anomaly can exceed 8%.

The geometry and properties of the heterogeneity constrain possibilities for its origin. A shear velocity heterogeneity (at least 4% slower than the surrounding mantle) that is this sharp probably cannot be solely due to temperature, because a high-temperature anomaly exceeding 500 K is required and the diffusive time scale for heat loss τ for a 8-km-thick slab is only 2 million years (My) (τ = L2/D and D ≈ 10−6 m2 s−1, where L and Dare the layer thickness and thermal diffusivity, respectively). The dipping feature of the object rejects a pressure-driven phase transition (11). Thus, it represents a chemically distinct region even if a thermal anomaly is also present. The heterogeneity's shape and thickness suggest oceanic crust that was subducted into the lower mantle, with its different bulk composition expressed as the velocity contrast with the surrounding mantle. We assess the plausibility of the model in the following. Slabs subducted during the Cenozoic era do not have any relation with this heterogeneity (12) nor do recent tomographic models reveal velocity anomalies in this region (2). On the other hand, a reconstruction of paleosubduction zones indicates that the Indonesia slab was located above the heterogeneity with an age of 160 to 170 million years ago (Fig. 1) (13). Velocities in plausible subducted basaltic mineral assemblages in the mid–lower mantle vary by only ∼1.5% in relation to pyrolite (14), but the shear modulus behavior at lower mantle conditions is too uncertain to exclude a contrast with subducted basalt. It is therefore uncertain at this stage if the observed velocity heterogeneity represents subducted oceanic crust, but other Earth structures with this thickness and lateral extent are difficult to envisage.

If the detected heterogeneity represents the oceanic crustal part of a Mesozoic slab, several consequences for mantle dynamics and geochemistry follow. The heterogeneity's residence time in the mantle is of the order of 160 My. Its planar shape indicates that buckling or folding of slabs is insignificant at scales smaller than 500 km. The dip of the observed layer is opposite to the northward dipping subduction of the Indonesia slab predicted from plate reconstructions (13), which may indicate that the slab lay horizontally stagnant above the 660-km discontinuity before descending into the lower mantle (15). The near absence of western Pacific hot spots (16) might also be a consequence of the same phenomenon if their rise from the lower mantle is found to be blocked by stagnant slabs such as this one. If the characteristic plume rise time was 15 to 30 My (17) and the slab remained there for at least 160 My, it would effectively screen hot spot rising and may become entrained in plume ascent, a feature entailed in some geochemical models (18).

The reservoir dimensions suggested by these observations constrain some models of lower mantle geochemical heterogeneity as well. A sheet of subducted oceanic crust can deliver the238U that is required for the mantle's high μ (HIMU) (μ = 238U/204Pb) component. Uranium decay produces 4He, which if sequestered in an ∼8-km-thick layer, would be nearly closed to He diffusion for the ≥1-billion-year time scale required to generate the observed207/204Pb–206/204Pb arrays in mid-ocean ridge basalt (MORB) and oceanic island basalt (OIB) (19, 20). Consequently, if these objects were to be sampled during OIB genesis, say by plumes, they would link a low 3He/4He component to HIMU. However, the generally 5 to 7R A (R A, atmospheric ratio) 3He/4He ratios of HIMU OIBs suggest segregation times of ∼100 My (19), which are shorter than the ∼160-My residence time of the objects estimated above. One way to break the link between HIMU and low 3He/4He is to postulate the diffusive loss of He out of a thin HIMU source body that is 0.2 to 3 km thick (19), thinner than the sheet we observe. Alternatively, a thicker body would require a lower internal3He/4He ratio and mixing with the MORB reservoir to attain the observed HIMU OIB ratio. If the sheet is a relatively new feature in the mantle, whose shape will be disrupted by future stirring, the diffusive model may be compatible with the sheet's thickness. However, the strong elastic contrast may plausibly be linked to an equally strong rheological contrast, which would inhibit further mixing. Thus, geochemical heterogeneity longevity may be longer than envisaged in mixing simulations in viscously homogeneous fluids (20, 21).


View Abstract

Stay Connected to Science

Navigate This Article