Lianxing Wen -> Research -> Geodynamics -> Residual Tomography

Slabs, hotspots, cratons and mantle convection revealed from residual seismic tomography in the upper mantle

abstract

Upper mantle seismic tomography correlates with tectonic features on the surface of the Earth. Cratonic "roots", thickening oceanic plates and subducting slabs are the first order contributors to the tomography, at least above about 400 km depth. In order to further investigate the fate of slabs, the structure of hotspots and the style of mantle convection, we calculate residual maps by excluding from the tomography the first order effects of conductive cooling of oceanic plates, deep craton "roots", and partial melting or cooling caused by subducted lithosphere. The good correlations between residual tomography in the transition zone (400-650 km) with 0-30 Ma subduction, at degree l=2 can be explained by slab accumulation in this region. The correlations between residual tomography in the transition region and the subduction during earlier periods are poor. This may indicate that slabs reside near the 670 km discontinuity for only a certain period of time. Hotspots correlate with the residual tomography in the shallow mantle (100-400 km). Correlations decrease rapidly with depth. Good correlations occur at degree l=2, between hotspots and residual tomography in the shallow mantle and between hotspots and tomography in the lowermost mantle. Various correlations suggest that the correlations between hotspots and the lowermost mantle are intriguing. At degree 6, cratons correlate with residual topography, hotspots, and upper mantle seismic tomography. Cratonic "roots" may affect, or modulate, upper mantle convection. Subduction may affect the locations of both upwellings and downwelling, and may control mantle convection. The 670 km discontinuity may not be the place where long-lived mantle convective stratification takes place. Mantle convection may be decorrelated closer to 900 km, near a recently rediscovered mantle discontinuity.

Introduction

The relationship between subduction and seismic tomography has been studied widely. Richards and Engebretson (1992) interpreted the good correlations between the large-scale seismic heterogeneity , averaged over the whole lower mantle, and subduction during the Cenozoic and Mesozoic as the result of the cooling effects of the subduction. Scrivner and Anderson (1992) correlated subduction positions since the breakup of Pangea, with seismic tomography depth by depth throughout the whole mantle. They found correlations in the transition zone region. Ray and Anderson (1994) found good correlations between integrated slab locations since Pangea breakup and fast velocities in the depth range 220-1022 km. Wen and Anderson (1995) quantified the slab flux by estimating the subducted volume in the hotspot reference frame and correlated it with seismic tomography throughout the mantle. They found significant correlations in the depth interval 900-1100 km and attributed these to the accumulation of subducted lithosphere in this region. Correlations were also found in the upper mantle and transition zone for recent subduction. The relationship between hotspots and seismic tomography has also been investigated. Excellent correlations at degree l=2 were found in the lower mantle and, at degree l=6 in the upper mantle (Richards et al., 1988; Cazenave, et al., 1989; Kedar et al., 1992). Cazenave et al. (1989) interpreted their results in terms of degree 2 convection in the lower mantle and degree 6 dominated convection in the upper mantle. Richards et al. (1988) hypothesized that hotspots originated in the deep lower mantle, based on the good correlations at degree 2, whereas Montagner and Romanowicz (1993) speculated that hotspots came from the transition zone, based on the dramatic decrease in the correlation below the transition zone, at degree 6. Ray and Anderson (1994) pointed out that hotspot locations were no better correlated with lower mantle tomography than were ridge locations. Subduction history provides a powerful constrain on the fate of the slabs in the deep mantle, particularly as seismic tomography provides more realistic images of the interior of the Earth. One of the difficulties in relating tomography with subduction and hotspots in the upper mantle, is the complexity of the tomography in this region. There are large contributions from the cooling of the oceanic plates and craton "roots" and this makes it difficult to get meanful results from the hotspot and slab correlations with the seismic tomography. The role of deep craton "roots" in the upper mantle has been extensively discussed (Lerner-Lam and Jordan, 1987; Hara and Geller, 1994; Polet and Anderson, 1995). The contribution from downgoing slabs or possible stagnant slabs, may also prevent one from correctly relating the seismic tomography to hotspots. Some information about mantle convection may be revealed by studying the correlation between surface tectonic features and geophysical observables (e.g. topography, hotspots etc.). In this paper we construct residual tomography for the upper mantle by excluding effects from conductive cooling of oceanic plates, craton "roots", and the partial melting or cooling induced by downgoing slabs. The residual tomography is compared with the pattern of subduction history in the past 130 Ma (Wen and Anderson, 1995) and with the distribution of hotspots. The procedures for excluding these effects are given in the first section. The correlation between residual tomography and past subduction is presented in the second section. In the third section we use residual tomography to help constrain the origin of hotspots. In the fourth section, the degree 6 significance in geodynamics is discussed. Finally, the style of mantle convection will be discussed.

Residual upper mantle tomographic models

The first order contributors to seismic tomography in the upper mantle are assumed to be oceanic plates, craton "roots", and on-going subduction. Other possible contributors, such as hotspots, stagnant slabs, and small scale mantle convection are assumed to be second order. If they are important, they may show up in the residual maps. The residual tomography is defined as the seismic tomography excluding the effects from those first order contributors.

Oceanic Plates

Our first job is to remove from the tomographic models the effects of the cooling oceanic plates. Velocity heterogeneities can be related to the temperature of the mantle by the temperature derivatives for the minerals in the mantle. The temperature distribution beneath oceans can be calculated from the age of oceanic lithosphere and thermal cooling models (de Jonge et al., 1994; Nataf and Ricard, 1995). We use the digital age map of M\"{u}ller et al. (1993). Several thermal models have been established in order to explain the bathymetry and heat flow in the ocean. Of particular importance are the half space cooling model (Turcotte and Schubert, 1982) and plate model ( Parsons and McKenzie, 1978). The plate model gives an adequate fit to these observations (e.g. Stein and Stein, 1992). However, the flattening of the oceanic age-depth relation for sea floor older than 70 Ma may be attributed to other effects, such as hotspots (Heestand and Crough, 1981; Schoreder, 1984) and other geodynamic process (Davies, 1988a, b; Cazenave and Lago, 1991; Morgan and Smith, 1992). Recent seismic studies also indicate that the oceanic lithosphere continues to cool after 70 Ma (e.g. Zhang and Tanimoto, 1991; Woodward and Masters, 1991). 3D global seismic tomography can also be used to constrain the oceanic thermal model. Fig. 1 shows the velocity perturbation variation with the age of the oceanic lithosphere at depths from 100 to 650 km, based on the seismic tomographic model SH12WM13 ( Su et al., 1994). The velocity perturbations are calculated by averaging the velocity perturbation along the positions corresponding to each isochron on the surface of the Earth. The velocity perturbation vs. age curves at different depths have the same characteristics; they are flat over a certain period of age, then increase rapidly with age. The "turning points" of these curves are strongly dependent on the thermal diffusivity. The traditional value of diffusivity used in the geodynamical literature is about $\kappa=1.0 mm^{2}s^{-1}$. With this value, the conduction is primarily confined to about the top 150 km of the mantle. Based on the tomographic models, cooling may extend deeper. The characteristics of the velocity-age curves at 200-450 km depth indicate that the cooling, possibly due to conduction, is still happening at these depths. The "turning points" at 250-300 km are almost the same as that of 200 km. This indicates there might be a diffusivity "jump" in this region near the 210 km discontinuity, which is possibly a chemical boundary. The pressure also plays an important role. We assume, for our present purposes, that the tomographic models are "exact" and do not suffer from any smearing. We are trying to remove the near-surface effects than interpret them. The thermal diffusivity is very anisotropic for olivine. We use the mean diffusivity for olivine, $\kappa_{1}=1.65 mm^{2}s^{-1}$ (Kobayashi, 1974), from 0-210 km. The diffusivity below 210 km is assumed to be doubled, possibly due to pressure and chemical differences. A cooling model with one layer (0-210 km) over a half space is used to explain the velocity-age curves at various depths. The Appendix gives analytical solutions for the temperature distribution. There is still one unknown, the conductivity ratio between that of the layer and that of the half space. The temperature derivatives which relate the temperature perturbation to velocity perturbation are uncertain. The experimental results are also uncertain (Estey and Douglas, 1986; Karato, 1993). Pressure and chemical differences would make this parameter even more uncertain. The conductivity ratio controls the temperature distribution as well. Since our purpose is to find the best fit model and remove it rather than to find the conductivity ratio or temperature derivatives, We therefore make no effort in guessing these parameters in the mantle, but fix the conductivity ratio and and find the best temperature derivatives at various depths. The best fit velocity models, in the oceanic regions, are found by adjusting the temperature derivatives, at various depths, in order to minimizing the difference between the thermal velocity model and SH12WM13 in the oceanic regions. Fig. 1 also shows the predicted velocity-age relation and comparisons with the SH12WM13. Any deviations from this predicted model, based on the age of the oceanic lithosphere on the surface will be regarded as anomalies. The predicted models are expanded into spherical harmonics and are truncated at degree 12, in order to compare with SH12WM13. The first residual tomography (RES1) is obtained by excluding the oceanic plate component from the seismic tomography (SH12WM13). It should be mentioned that tomographic modeling often use simplified assumptions about near surface condition and there can also be a problem with vertical smearing. We ignore these complications.

Cratons

Polet and Anderson (1995) divided the cratons into two provinces, based on Sclater et al. (1981). Province 1 contains continents older than 1700 Ma (Archean and Early Proterozoic) and province 2 includes continents between 800 and 1700 Ma (Middle Proterozoic). We classify the cratons according to their geographic locations and ages. The cratons are divided into 13 groups; six for cratons between 800-1700 Ma, and seven for those older than 1700 Ma. Each group is related to its age and a major plate. For instance, cratons, between 800-1700 Ma, on the South American plate are placed in the same group. Cratons older than 1700 Ma in Eurasia fall into two groups. Based on the residual tomography (RES1), we calculate the average velocity perturbation beneath each group of cratons. Fig. 2 shows the velocity perturbations beneath each group of cratons vs. depth. The heavy lines are the velocity perturbations for cratons older than 1700 Ma. The light lines are for cratons between 800-1700 Ma. The contribution of craton "roots" to seismic velocity variation in spherical harmonic space is obtained by expanding the function, which, in cratonic regions for each group of craton, has the value of the average velocity perturbations from the residual tomography (RES1), and zero outside, into spherical harmonics. This set of spherical harmonic coefficients can be multiplied by an arbitrary constant ($ C $). This proportionality constant ($ C $) is the second parameter to be determined.

On-going Subduction

We assume that the subducting plates sink vertically into the upper mantle at the velocity of the plate at the trench. The ages of the slabs are reconstructed at every depth. We assume that the seismic velocity perturbation within the slab is constant ($\delta V_{s}$) at a certain depth. The width of a particular slab segment is equal to its thickness. Thickness is calculated from the age of the oceanic lithosphere at the time of subduction (Wen and Anderson, 1995). On-going subduction can cause low-velocities in the shallow mantle, because of volatile fluxed melting in the mantle wedge (Anderson et al., 1992), and high-velocities at greater depth due to low-temperatures in the slab. We permit the $\delta V_{s}$ to take on negative or positive values (negative values imply partial melting).

Residual tomographic Models

Synthetic Models are obtained by linear superposition of the contributions from slabs, oceanic plates, and cratons at various depths. The two parameters ($\delta V_{s}$ and $C$) at each depth are chosen by minimizing the quantity: $\sum \sqrt{V_{syn}^{2}-V_{tomo}^{2}}$ Where $V_{syn}$ and $V_{tomo}$ are the velocity perturbations of synthetic model and SH12WM13 at certain depth respectively. The summation is over every $1^{\circ} \times 1^{\circ}$ cell in a global grid. The residual models are obtained by subtracting synthetic tomography from SH12WM13. The residual models from 100-650 km depths are plotted in Fig. 3.

Implications from the correlations between residual tomographic models with the subduction history

The subduction history for the past 130 Ma has been reconstructed in the hotspot reference frame, based on plate tectonic models (Wen and Anderson, 1995). We correlate the residual tomographic models, at various depths in the upper mantle, with subduction history. Since no significance correlations are found at other degrees, only correlations at degree 2 and 6 are plotted in Fig. 4. Positive correlations mean that the subduction material corresponds with high velocities. The good correlations between 0-30 Ma subduction and residual tomography, in the transition region, may imply that some slabs are trapped in this region. There are no significant correlations between the 30-130 Ma subduction history and residual tomography at degree 2. Significant negative correlations are found for 60-90 Ma subduction in the transition region (400-650 km) at degree 6. These results confirm the previous results of Wen and Anderson (1995). Slabs subducted prior to 30 Ma may have sunk into the lower mantle. It is surprising that good correlations occur for the recently subducted slabs, because the on-going subduction has been subtracted out of the seismic tomography. One possibility is that we didn't subtract out the effects of on-going subduction efficiently. In order to check this possibility, we subtract an additional slab effect, in order to make the correlation coefficient, at degree l=2, between resultant residual tomography and 0-30 Ma subduction, just below the 50\% confidence level. The resultant residual tomography shows a very low velocity ring along the subduction zones except the Kurile, Japan, Izu-Bonin, Mariana, New Hebrides and Philippine trenchs. This suggests that the effects of on-going subduction have been removed efficiently and also reveals that the trapping of slabs at the 670 km discontinuity can be localized. Different slabs in different subduction zones may have different behaviors in the transition zone. Some slabs may be stopped at the 670 km discontinuity for a period of time, while some slabs may penetrate into the lower mantle. The residual tomography shows high velocity beneath the Kurile, Japan, Izu-Bonin, Mariana, New Hebrides and Philippine trenchs. This implies that the subducted slabs may accumulate beneath these trenches at the 670 km discontinuity. We test this possibility by dividing the subduction in the past 30 Ma into two groups. Group 1 contains only the 0-30 Ma subduction in the Kurile, Japan, Izu-Bonin, Mariana, New Hebrides and Philippine trenches; group 2 includes subduction in the other convergence regions (e.g. Aleutian, Chile-Peru, Tonga-Fiju, Java trenches etc). We found excellent correlations at degree 2 and 3 for group 1 subduction and no correlation for group 2 subduction (Table 1). Correlations which are significant at greater than 90 \% confidence level are underlined. The good correlations between the 0-30 Ma subduction and residual tomography can be explained by the accumulation of slabs beneath the Kurile, Japan, Izu-Bonin, Mariana, New Hebrides and Philippine trenches in the transition zone. It should be pointed out that we cannot resolve individual subducted slabs in this study. Slabs beneath group 1 trenches have relatively shallow dip (except Mariana). The degree 2 heterogeneity is the most important component for many geophysical observables, such as seismic velocity (e.g. Masters et al., 1982; Nakanishi and Anderson, 1983; Woodhouse and Dziewonski, 1989) and geoid (Lerch et al., 1979). However, the origin of this degree is still controversial. Scrivner and Anderson (1992) suggested that the degree 2 heterogeneity might be caused by the history of subduction since the breakup of Pangea. In particular they suggested the presence of stagnant slabs in the mesosphere, based on the good correlations between time-integrated slab positions and seismic tomography. Our analysis supports their conclusion. We subtract effect of the stagnant slabs from the residual tomography in the transition zone (450-650 km), assuming that stagnant slabs are only beneath group 1 trenches. Fig. 5 shows the power spectra of the seismic tomography (SH12WM13), cratonic roots, on-going subduction and stagnant slabs, at degree l=2. In the shallow mantle (above 200 km ), the oceanic lithosphere, cratons and subducting slabs contribute most of the power at degree 2. In the transition zone, stagnant slabs are apparently responsible for the degree 2 lateral variation. Seismology suggests that the behavior of the slabs at the 670 km discontinuity is highly variable (e.g. Jordan and Lynn, 1974; Zhou and Anderson, 1989; van der Hist et al., 1991; Fukao, et al., 1992). Our results indicate that, indeed, the behavior of the slab at the 670 km discontinuity is region- dependent. But the 670 km discontinuity may not be the final destination of the slabs based on the poor correlations between the residual tomography and 30-130 Ma subduction. Stagnant slabs may sink into the lower mantle after a period of time of accumulation. Geodynamical stimulations indicate that trench migration and the motion of overriding plates are closely related to the behavior of slabs at the 670 km discontinuity (Gurnis and Zhong, 1995). Gurnis and Zhong (1995) also indicated that the 670-km phase change does not significantly influence the ultimate ability of subducted slabs to penetrate into the lower mantle, assuming a homogeneous mantle.

Hotspots and mantle convection

Residual tomography, which excludes the near-surface features, may provide a constraint on convection in the mantle. Passive ridges are the first-order upwellings on the surface of the Earth. They possibly represent normal (uncooled) mantle. Hotspots are generally considered to be caused by deep, narrow, active upwellings. In this section, we use seismic tomography to constrain the characteristics of hotspots throughout the mantle. We use seismic tomography (SH12WM13) in the lower mantle and residual tomography in the upper mantle. The list of 47 hotspots, complied by Morgan (1981) and Crough and Jurdy (1980), is used. The overall distribution of hotspots correlates very well with low seismic velocities in the deep lower mantle (1700 km-CMB) at degree l=2. There is some correlation with l=3 as well in the lower mantle. However, the degree 2 correlations are poor in the depth region 700-1700 km (except 900-1000 km). This is consistent with the results of previous authors (e.g. Kedar et al., 1992). Table 2 gives correlation coefficients between hotspot distribution and residual tomography in the upper mantle. Positive values mean that hotspot positions favor low velocity regions. Correlations which are significant greater that 90 % confidence level are underlined. Hotspots correlate with slow seismic velocities at degree l=2 to 400 km depth and at degree l=4 to 150 km depth. The degree 2 correlations decrease very rapidly into the transition zone. Also, at l=4, hotspots correlate with fast velocities in part of the transition region ( 400-500 km) (Table 2). The good correlations between hotspots and low seismic velocities at degree l=2 in the top of the upper mantle have two possible interpretations. One interpretation is that the good correlations in the top 400 km and in the deep lower mantle may have the same cause. Residual tomography, which may be a reflection of the mantle convection pattern, has a pattern which is similar to the seismic tomography in the deep mantle. Hotspots may connect the thermal structures in the deep lower mantle and the shallow mantle. On the other hand, the good correlations at degree l=2, 4 between residual tomography and hotspots in the top 400 km, together with those at l=6 of tomography, hotspots and geoid (Tanimoto and Anderson, 1985; Cazenave and Thoraval, 1994) suggest that hotspots might originate in the top 400 km. The good correlations between hotspots and lower mantle heterogeneities at degree l=2 are intriguing, but are not necessarily evidence for a deep mantle origin for hotspots. It does suggest, however, that lower mantle convection may modulate, or control, upper mantle behavior, or vice versa. In order to check the correlations between hotspots and residual tomography, it is useful to discuss hotspots in the context of the three dimensional residual velocity structure of the upper mantle. Most of the hotspots in the Pacific and circum-Pacific area are in low velocity regions of the upper mantle. Some hotspots are also in low velocity regions in Africa, the Indian ocean and the North Atlantic. Most hotspots in the south Atlantic, around South America and near South Africa are in high velocity regions. The high velocities in some hotspots in Africa could be because that they are close to high-velocity cratonic "roots". This spatial visual check plus the good correlations at degree l=4 and some correlation at degree l=6 suggest that the correlations, at degree l=2, between hotspots and residual seismic tomography in the upper 400 km are meaningful. One way to check if the hotspot correlations with residual tomography are related to those in the deep mantle is to correlate the residual tomography with the seismic tomography in the deep lower mantle. Fig. 6 shows the correlations between seismic tomography at 2500 km and residual tomographic models, at degree l=2. No significant correlation is found. The correlation even becomes negative for residual tomography at 650 km. Tomographic models from 1700 km to CMB have similar correlations. However, residual tomographic models between 200-500 km depth correlate with the seismic tomography model in the 900-1000 km depth region. This is the only region in the lower mantle that correlates with residual tomography. Direct comparison among the degree 2 patterns of hotspots, residual tomography and lower mantle tomography is shown in Fig. 7. They all have low values in the central Pacific and Africa, but the residual tomography and hotspot highs (devoid of hotspots) extend to the south of South America and east of Asia. The lower mantle is affected by subduction. It is still not clear that the lowermost mantle tomography will correlate with the residual tomography and hotspots, when the subduction effects are excluded. The poor correlations between seismic tomography in the deep lower mantle and residual tomography, plus the cautions of Ray and Anderson (1994), suggest that the hotspot-lower mantle connection is intriguing. Other seismic models and other hotspot lists give different results (Kedar et al., 1992; Ray and Anderson, 1994). Scrivner and Anderson (1992) found strong negative correlations between hotspot positions and the 0-180 Ma slab locations. Hotspots do not originate in mantle that has been cooled or blocked by slab. Normal mantle, cooled by the subduction will, of course, correlate with the subduction history. But this mantle will also correlate with hotspots, even if only "normal" mantle is present. Downwellings in the deep lower mantle, whether they are caused by slab accumulations, or are indirectly related to present or past subduction, will make the seismic patterns in this region correlate with hotspots. Do slabs cool the mantle or do plumes heat the mantle, or both? Is normal mantle uncooled or not heated? The cause and the effect must still be disentangled.

Degree 6 in Geodynamics

The l=6 component is particularly important in geodynamics and tomography. Tanimoto and Anderson (1985) showed an excellent correlation between Love wave phase velocities and the geoid, suggesting that this part of the geoid originates in the upper mantle. Some hotspot lists exhibit a peak at l=6 (e.g. Richards, et al., 1988; Kedar et al., 1992). Residual topography at l=6 correlates very well with seismic velocities between 300 and 500 km depth (Cazenave and Thoraval, 1994). Highs of the degree 6 residual topography coincide with the south Pacific Superswell, the Afar region, the North Atlantic, the equatorial Atlantic and south of New Zealand (Cazenave and Thoraval, 1994). The l=6 hotspot map (Richards et al., 1988) is similar, as are l=6 upper mantle tomographic maps in the upper mantle. We have expanded our craton function into spherical harmonics. Fig. 8 shows the degree 6 map of the craton function, which has unit value in cratonic regions and zero outside. The l=6 expansion picks up most of the cratons, as might be expected (as fast regions) but also has low velocities in the North Atlantic, equatorial Atlantic, the Pacific superswell, south of New Zealand and the Afar; i.e., it looks very much like the l=6 hotspots and tomographic maps. This has several alternative explanations: \begin{enumerate} \item{Hotspots, swells and hotter-than-average mantle occur where there are no cratons and these regions are generally antipodal to cratons} \item{Hot upwelling mantle tends to drive cratons away and they settle in areas of colder mantle.} \item{l=6 convection is intrinsic to the upper mantle (Tackley et al., 1993) and cratons establish the phase of the pattern; upwellings occur in complementary locations.} \item{The pattern of convection in the mantle is controlled by cratons, and their associated subduction zones (and the history of subduction); hot upwellings are not particularly fundamental but occur in "normal" mantle.} \end{enumerate} Cratons probably have long-lived "roots" extending to about 200 km and associated high-velocity material, perhaps a thermal boundary layer, extends somewhat deeper. This is a large fraction of the depth of the upper mantle and the presence of thick cratons must influence mantle convection. In addition, a moving craton overrides cold oceanic lithosphere, placing a cold slab about 100 km thick under the craton. This cold dense downwelling also affects mantle convection, even after it settles "on the bottom." The upper mantle can be viewed as a system that is differentially cooled from below, cooled and dragged down at subduction zones and affected by lateral temperature gradients at the surface, with subcratonic isotherms at a deeper level than isotherms elsewhere, except in slabs. This general type of convection was treated by Pekeris (1935), Allan et al. (1967) and King and Anderson (1994). It is quite distinct from the Rayleigh-B\'{e}nard convection in a fluid system heated from below which forms the basis for most discussions of mantle convection. Of course, the mantle is, to some extent, heated from below and from within, and all these factors must be considered. The correlations discussed here and by Cazenave and Thoraval (1994) suggest that swells and high-temperature mantle are part of a global circulation pattern, rather than localized plumes that are independent of plate tectonics and background mantle convection. Previous authors have suggested that degree 6 is a dominant wavelength of plume formation. Our results, plus those of Tackley et al. (1993), suggest that l=6 convection could be intrinsic to the background flow, modulated by surface boundary conditions. It is of interest that two of the most intense thermal anomalies on Earth (as judged by tomographic maps) are in the Afar and South Pacific superswell regions. These are also two of the "anti-craton" maxima on the l=6 craton map. These maxima are even more intense if the actual tomographic craton function are expanded (i.e. giving different weights to the cratons, depending on the magnitude of their seismic anomalies). These l=6 maxima are in phase with, and reinforced by, the l=2 pattern. Other anti-craton maxima are in the western Pacific, the south equatorial Atlantic, the region between New Zealand, Australia and Antarctica, the North Atlantic, the central Indian Ocean, Asia, and the NE Pacific. These are areas that, in general, are not cratons (except Siberia) and often not subduction zones (except Chile-Peru, New Zealand and Sumatra). This again suggests that the intermediate wavelength convective field in the mantle, typified by l=6, whose pattern may be controlled by surface (cratons) and edge (subduction) boundary conditions, is controlling the locations of features which have been attributed to local and independent deep mantle plumes from an unstable lower thermal boundary layer.

The stratification of mantle convection

The thick craton "roots" might have strong feedback on mantle flow, even if they don't have a major thermal effect on driving the flow. However, subduction is probably the most important control on mantle convection. Mantle convection can be viewd as a system, which is, at first order, mainly cooled by the subduction inside and modulated by the thick craton "roots" in the top of the mantle. That only some slabs are trapped at the 670-km discontinuity for a period of time indicates that the 670-km phase change may not preclude penetrative convection. However, this doesn't mean that we are ruling out layered mantle convection. Evidence for a deeper boundary has accumulated. The degree 2 patterns of the shallow and deep mantle seismic tomography are spatially shifted across a depth of about 800 km (Tanimoto, 1990). Kawakatsu and Niu (1994) presented seismic evidence for a 920 km discontinuity. Ritzwoller and Lavely (1995) showed that there is a significant decorrelation of mantle tomography at a depth of about 1000 km. Wen and Anderson (1995) found good correlations between the subducted slabs and seismic tomography in the 900-1100 depth region. The mantle does not become radially homogeneous and adiabatic until about 800 km depth (Dziewonski and Anderson, 1981). The velocity discontinuity at 670 km is primarily due to a phase change and the chemical change at this depth may be small. If so, the endothermic nature of the phase change may delay slab penetration but cannot prevent large masses of cold material from episodically cascading across the boundary (Honda et al., 1993; Tackley et al., 1993). Chemical discontinuities in the mantle are harder to detect if they are not associated with changes in mineralogy. In a convecting mantle chemical interfaces will tend to have high relief and will therefore be difficult to detect by standard seismological means. However, a small change in density can effectively suppress penetration of slabs. A variety of evidence suggests that there might be a barrier to convection at a depth of about 900-1000 km. The existence of a chemical boundary at this depth might induce convective stratification.

Conclusion

The residual tomography, which excludes the effects of oceanic lithosphere cooling, craton "roots" and partial melting or cooling related to subducted slabs, correlates with 0-30 Ma subduction, especially the subduction in the Kurile, Japan, Izu-Bonin, Mariana New Hebrides and Philippine trenchs. Some slabs may be trapped at the 670 km discontinuity for a period of time due to the endothermal phase change. Many areas of past subduction do not have evidence for cold slab at the base of the transition region. Slabs in these areas may penetrate the 670 km discontinuity into the top part of the lower mantle. Most hotspots appear to be in hot, or uncooled, regions of the upper mantle. The overall pattern of hotspots correlates with residual tomography in the top 400 km at degree l=2 and 4. Correlations decrease very rapidly in the transition zone region. Some hotspots occur in regions of the mantle that are slower than average down to 650 km depth. Hotspots also correlate with seismic tomography in the 900-1000 km region and the deep lower mantle at degree l=2. However the correlation between residual tomography in the upper mantle and seismic tomography in the deep lower mantle is poor. Subduction is probably the most important control on mantle convection. The thick "roots" of craton may have strong feedback to mantle convection in the upper mantle. Mantle convective stratification does not happen at the 670 km discontinuity. However, the possibility that mantle convection is stratified at deeper chemical boundary cannot be ruled out at this stage.

Acknowledgements: We thank Jascha Polet for providing digital data of cratons. This work was funded by NSF grant EAR 92-18390. Contribution No. 5554, Division of Geological and Planetary Sciences, California Institute of Technology.