Seismological constraints on the South African superplume; could be the oldest distinct structure on earth

Transcription

1 Earth and Planetary Science Letters 206 (2003) 119^131 Seismological constraints on the South African superplume; could be the oldest distinct structure on earth Sidao Ni, Don V. Helmberger Seismological Laboratory , California Institute of Technology, Pasadena, CA 91125, USA Received 5 August 2002; received in revised form 1 November 2002; accepted 7 November 2002 Abstract A recent study of the lower mantle structure beneath Africa revealed strong lateral changes in S-velocity extending upward from the core^mantle boundary to about 1500 km. SKS travel times observed on the South African Array display jumps of about 6 s when ray paths cross these nearly vertical boundaries. Back projecting these delays onto the core^mantle boundary allows a clear image of the horizontal extent of this structure starting at mid-africa (15 S, 5 E) where it strikes roughly northwest to beyond the tip of South Africa (45 S, 55 E) where it bends toward the Indian Ocean. Waveform sections of S, ScS, SKS, and SKKS are modeled along two corridors, one along strike and one at right angles to establish its uniformity. The structure is about 1200 km wide and has about a 3% drop in S-velocity although some small-scale features are apparent in the roof structure and midsection. If this structure is stabilized by a localized viscosity condition or a dense core as suggested by some joint inversions, gravity, and free oscillations, it may be isolated from mantle stirring and therefore very old with unique chemistry. ß 2002 Elsevier Science B.V. All rights reserved. Keywords: superplumes; low velocity structure; chemical plume; sharp boundary; lower mantle 1. Introduction * Corresponding author. Tel.: ; Fax: address: stone@gps.caltech.edu (S. Ni). The presence of extensive low velocity structures beneath South Africa and the mid-paci c is the most robust feature of the many Global tomographic models [1]. These images are relatively weak, typically less than 1.5% variation in S-velocity, and are blurred both vertically and horizontally (Fig. 1). Higher resolution can be obtained by working with di erential times between (SKS3S) and (ScS3S). These di erentials depend less on the precision of source location and origin time and more on the lateral variation of structure compared to tomographic methods. Datasets can be assembled from a single station with multiple sources and/or a single event recorded by an array of stations. To explain the (SKS3S) and (ScS3S) datasets for a corridor of paths connecting South America to Africa, we modi ed a tomographic model such as displayed in Fig. 1A [2]. A consequence of this model is a jump in SKS travel times as paths cross these relatively sharp boundaries. If the boundary is sharp enough and the edge is parallel to the SKS ray path, the SKS waveform displays two X / 02 / $ ^ see front matter ß 2002 Elsevier Science B.V. All rights reserved. PII: S X(02)

2 120 S. Ni, D.V. Helmberger / Earth and Planetary Science Letters 206 (2003) 119^131 Fig. 1. The upper two panels display the tomographic results from Ritsema et al. [1] along a 2D section (A) and the DQ layer (B). The 2D section runs from South America to South Africa denoted in B as a dashed line. The heavy green line de nes the boundary of the anomalous shear-wave structure (33%) of the LVZ2 model. Red indicates slow velocities and blue relatively fast velocities ( þ 1.5%). Example ray paths are displayed in A with S, ScS (in red) and SKS (in blue). Note that SKS is nearly parallel to the eastern (green) boundary, which means that the travel time of SKS is predicted to change rapidly at this epicentral distance. (B) A map of events (stars) and the South African Array (triangles). The great circle paths are included indicating the various azimuths sampled. (C) SKS and SKKS exit points at the core^mantle boundary (CMB). The points are color-coded, with blue representing no delays (PREM-like) and red for delays more than 5 s. Note the rapid transitions which usually occur in less than 3 in distance (150km at the CMB). The various symbols indicate events arriving from some of the more important paths. The color background shows geoid anomalies (in meters). The geoid high around the tip of Africa correlates with SKS observations well.

3 S. Ni, D.V. Helmberger / Earth and Planetary Science Letters 206 (2003) 119^ Fig. 2. Displays of South African Array data obtained from the shallow East Paci c Rise event before and after shifting. The column on the left indicates the data plotted relative to PREM where the upper half of the data has normal arrival times relative to late arrivals beyond 101. Heavy lines indicate the apparent o sets. Smaller-scale di erences are indicated in the bottom half as a combination of solid and dashed traces, with the latter indicating a small delay of a few seconds. arrivals, one early and one late. If the boundary is gradual (50km or more), the SKS waveform loses complexity [3]. However, the jump in travel time still occurs abruptly which can be easily recognized on array data. Here we report on such results obtained at the South African Array, sampling a 60 circular patch of lower mantle structure where a large-scale sharp feature is easily outlined. Anomalous geochemical data from young Indian Ocean basalts obtained directly above this structure contain the Dupal anomaly, thought to be samples of a very old primitive reservoir [4]. In addition to the anomalous geochemical data, there is the issue of the South African^Indian Ocean superswell [5] and its possible explanation [6,7]. Combining these various types of data into a self-consistent model appears possible, but here we will address the seismological constraints on this interesting structure.

4 122 S. Ni, D.V. Helmberger / Earth and Planetary Science Letters 206 (2003) 119^ Data and analysis Fig. 3. The upper panel displays the travel time delays determined by cross-correlation for event (Fig. 2), relative to PREM. Open triangles indicate the small-scale variation associated with the northernmost array stations as shown in the lower panel. Note how close these paths must be and still have di erent arrival times. The above study [2] concentrated on events beneath South America. Here we examine paths from events sampling other azimuths (Fig. 1B). Some of these events are rather small shallow events, i.e., from the Eastern Paci c Rise, and are not ideal for modeling but they do provide azimuthal coverage. Most of the events from the Western Paci c are deep with clear depth phases. SKS phases from these events display rapid changes in arrival times at the South African Array. This pattern is expressed in Fig. 1C and is the main contribution of this study. An example set of SKS records from one of the shallow events (970529) is displayed in Fig. 2. Even though the records are complicated, they can be aligned quite well by applying a waveform correlation routine (second column). The secondary phases are probably associated with crustal depth phases, ss, etc., and because they are relatively stable in di erential timing, they aid in suppressing noise in the alignment process. The two line segments in the rst column correspond to the times predicted by PREM (from 93 to 101 ) followed by a second line plotted with a 5 s o set (102 to 110 ). The station delay across the array is discussed in James et al. [8] where they nd relatively small variation. The largest delays occur for the southernmost array stations with values ranging from 0.5 to 0.75 s. These stations are at the top of Fig. 2. Thus, the o set near 101 cannot be attributed to upper mantle structure, which is relatively weak in structural variation [9] and appears to be associated with the ridge-like structure in the lower mantle as found for other datasets [2]. Their di erential time determinations relative to PREM are displayed in Fig. 3. We have subdivided the stations into two groups both to indicate the uniformity in large structure and also to emphasize the coherence of smaller-scale lateral variation. We will concentrate on the large-scale features rst and return to variability issues later in Section 3. From 95 to 99, the delays are about 1.5 s indicative of a slow DQ [10,11]. Near 100, the delays increase rapidly reaching about 5 s by 102. Such a feature is not easily explained without an abrupt change in structure as displayed earlier in Fig. 1A. A prediction from that 2D velocity section is displayed in Fig. 4 where the synthetics show a similar trend as data from the event although at a slightly di erent azimuth. The slope of the western boundary wall is not aligned with the SKS path and consequently the synthetics do not produce multipathing [3]. Some of the observations are relatively complex which could be caused by sloping sides, or perhaps a local ultralow velocity zone (ULVZ) at the core^mantle boundary (CMB) [12] as ob-

5 S. Ni, D.V. Helmberger / Earth and Planetary Science Letters 206 (2003) 119^ Fig. 4. Comparison of synthetics (dashed) predicted from the 2D model (LVZ2) derived earlier [32] as displayed in Fig. 1, with observations (solid) from event served in PKP near the edges of the Paci c Plume [13]. However, we have not identi ed anomalous SKP d S, indicative of ULVZ, in any of these pro- les of SKS, which would produce strong secondary arrivals near 110 [14]. Events (displayed in Fig. 2) and (displayed in Fig. 4) both show an abrupt jump in SKS. These delays are displayed in map view in Fig. 1C, where we have back projected paths to the CMB, piercing points denoted as triangles. Event has a similar pattern as event except more points plot in blue since the event is about 2 further west. Paths from events and are sampling the ridge structure farther north and are in agreement with our earlier results, although the SKS sharpness is less obvious.

6 124 S. Ni, D.V. Helmberger / Earth and Planetary Science Letters 206 (2003) 119^131 Fig. 5. A comparison of 2D travel times from synthetics generated from the above models (solid for LVZ2 and dashed for tomography) against data (R). The upper panels display the ray paths contributing to these arrivals relative to the 2D structure. Paths from the east are more plentiful and sampling points are nearly continuous, some of these values are SKKS as addressed later. But rst, we address the issue of thickness or height above the CMB. To do this, we will examine two corridors in detail, namely paths along cross-section AAP and along strike BBP as displayed in Fig. 1C Corridor through Sandwich Island (AAP) Data along this corridor have been analyzed in detail by Ritsema et al. [15]. An event beneath the Drake Passage produced a SKS similar to those displayed above except with opposite sign since they were recorded farther north at the Tanzania array and sample the eastern boundary where the velocity jumps back to PREM. The 2D structure developed to explain their data is similar to that in Fig. 1A. Fig. 5 shows the SKS estimates from event along the same Great Circle path but sampling the western boundary. Included in this display are the predictions from tomography, which have the right trend along with those from a 2D model (upper panel). The 2D model explains the SKS delays nicely. The sharpness of this feature is also apparent in ScS travel times as displayed in Fig. 5 (right). Because of the sensitivity of geometry, the di erence in ray path angle between SKS and ScS becomes important. Moving the western boundary to make it more vertical allows ScS to sample the slow structure near the top earlier in distance and produces a gentle slope in ScS delays as displayed. ScS appears PREMlike from 40to 48 where it begins to sample the 2D structure then becomes delayed up to 9 s beyond 55. The S-travel times from this event remain PREM-like over these ranges since its path never reaches the lower mantle [16]. They interpreted the slow ScS relative to S as a strong DQ e ect involving melting, but such a feature is not apparent in SKS samples at this location, ranges 90 to 95 in Fig. 5. In contrast to a ULVZ interpretation, we think our ridge structure explains this cross-section quite well as suggested in Fig. 1C. The issue of height vs. percentage drops in velocity is addressed at length [2,3] where the combination of di erential phases (ScS3S) and (SKS3S) proves de nitive.

7 S. Ni, D.V. Helmberger / Earth and Planetary Science Letters 206 (2003) 119^ Fig. 6. Comparison of SKS and SKKS observations from a western Paci c event showing the di erence in delays apparently caused by the relative position of the CMB exit points. These points are denoted as triangles (SKS) and squares (SKKS) in B where events from the smaller azimuths begin as PREM, open symbols. The circles are SKS exit points for event which sample the same region Corridor along strike (BBP) The BBP corridor runs mostly along structure as displayed in Fig. 1. There are several events with similar patterns sampling this zone and we picked event as representative. Since the ridge structure is sampled along strike, it is more enlightening to plot the data as a function of azimuth as displayed in Fig. 6. The SKS data are aligned on predicted PREM as a straight line and similarly for the phase SKKS. The sweep in azimuth overlaps the western boundary for SKKS

8 126 S. Ni, D.V. Helmberger / Earth and Planetary Science Letters 206 (2003) 119^131 Fig. 7. Explanation of the complex behavior of SKKS travel times. (A) SKKS record section with SKKS aligned on arrivals. SKKS waveforms are coherent for all the azimuthal ranges suggesting that the scattering in Fig. 6C is not noise. (B) SKKS delay times plotted against epicentral distances and azimuths. The delay time is indicated with di erent sizes of circles with maximum delay up to 6 s. The general pattern is that the travel time anomaly depends strongly on azimuth. It clearly shows that the SKKS time anomaly does not depend on epicentral distance for azimuth and s 208, but only depends on epicentral distances for azimuth between 204 and 208. A schematic explanation for the behavior of SKKS travel times is displayed on the right. For azimuth around 200, almost all SKKS rays miss the structure, thus small delays, and yield no dependence on distances. For azimuth of 206, some SKKS rays miss the structure and some hit, thus yield strong dependence on distance. For azimuth around 213, all SKKS rays hit the structure (delayed up to 6 s), and again yield little dependence on distance.

9 S. Ni, D.V. Helmberger / Earth and Planetary Science Letters 206 (2003) 119^ where the top several records plot near PREM, whereas the SKS phases which sample further to the north fall beneath the structure as shown in Fig. 6. While the SKS phases are delayed at all azimuths, they are particularly late near the middle of the structure reaching values greater than 7 s. The travel time determinations by correlation show these trends clearly as displayed in Fig. 6. Note that again the tomographic model is too weak while a 3% model t these delays nicely with a height varying from about 800 km (4 s) to 1500 km (7 s). This suggests that the roof is bowed upward (dome-shaped) or the structure has a slower core while the SKS are consistently late relative to PREM (Fig. 6A). The SKKS section displays a complex pattern of lags especially at azimuths near 206. This feature is caused by a strong distance dependence of ray paths as displayed in Fig. 7. At azimuths beyond about 208, the arrivals are all consistently late and at azimuths less than about 204, they are only slightly delayed. The gray zone denotes the sensitivity to small changes in ray path, again supporting a relatively sharp structure to cause such rapid jumps in travel times. An example of other data (event ) sampling the structure is given in Fig. 8. Note that at these ranges, the SKS and S are sampling the structure about equally in path length and, thus, the crossover from S to SKS should be normal as displayed in Fig. 8. This is again a shallow event with a complex waveform consisting of SKS plus depth phase ssks, but the change in polarity, down for S before 82 and up for SKS beyond 83.3, is quite obvious. Both are delayed about 5 s relative to PREM. Thus, again a high structure proves necessary to explain the di erent phases as discussed earlier. 3. Discussion and conclusion We have investigated the sharpness of the socalled African superplume by analyzing the data recorded by the South African Array. Over 800 SKS travel time picks (Fig. 1C), determined by waveform correlations, were obtained by sweeping through 180 of azimuth. By folding in our Fig. 8. The upper panel displays the geometry of paths entering the ridge structure along corridor BBP. This seismic section (lower panel) is reduced by theoretical SKS times from PREM. The slow paths for S and SKS are about the same for event Crossover remains PREM-like (lower section) with both arrivals delayed about 5 s. earlier results, we obtain a large-scale picture of a ridge-like structure, extending about 7000 km from Central Africa into the Indian Ocean. Because of the constrains on geometry, we were only able to sample the western boundary for events arriving from the west. Events from the east provided better coverage since both SKS and SKKS were available. Event produced mostly PREM-like picks except for the northernmost paths. These waveforms were noisy but could be anomalous. Paths from events and

10 128 S. Ni, D.V. Helmberger / Earth and Planetary Science Letters 206 (2003) 119^131 Fig. 9. Travel time anomaly shape sensitivity to wall sharpness and dip angle. Two representative models (dashed and dotted) are displayed in A and B along with SKS and ScS ray paths to display the importance of geometry. Note that a sharp jump in timing occurs when the ray path is parallel to the interface denoting the boundary, see dotted ScS results on the right. The heavy lines in C are values obtained earlier in Fig. 5 for the preferred model. The SKS dataset has been expanded by including events (970903, star; , diamond; , solid triangle) with a few degrees of shift to correct for source locations relative to the ridge-shaped structure. Although there is small-scale variation present, the boundary appears to be crossed in about 2 or less than 100 km at the CMB. Experiments conducted on dynamic models suggest that this is too sharp to be thermal [3] although this is debated in Hansen and Yuen [31]. are along the edge of the structure and produced mostly delayed values again de ning a relatively sharp boundary although 3D e ects are expected to be strong. The overall 3D pattern presented here is somewhat compatible to that given by Wen [17] except that he used S diff and con ned the anomalous structure to DQ. We prefer the ridge interpretation because of the compatibility of other phases as determined by the corridor studies and the lack of anomalous SKP d S associated with very slow structures. Nevertheless, the agreement with the location of the ridge-like structure as projected on to the CMB is good. Some SKS recordings are relatively complex, and probably could be directly attributed to multipathing [3]. However, the delay time pattern appears quite sharp which also allows some constraint on boundary de nition. In Fig. 9, the travel time delays for three events are shifted a few degrees for general alignment in pattern and plotted together. The paths from these events sample the structure at right angles and approach a 2D geometry. Synthetics for several plausible models were generated and analyzed as data to produce the various curves. The data indicate a normal to delayed transition over 3, which suggests a boundary zone less than 150km wide where the slope probably varies considerably along the structure from north to south. Some of the scatter in the SKS delays at the larger

11 S. Ni, D.V. Helmberger / Earth and Planetary Science Letters 206 (2003) 119^ ranges appears to be real and associated with deep structure as mentioned earlier (Fig. 3). The open triangles have larger delays and are located near the middle of the ridge, not that far from the anomalously strong delays displayed in Fig. 6. Thus, there is evidence for internal structure. This variation appears to be about 10^20% and can be interpreted in several ways: (1) increased height in roof structure (bowed up in the middle), (2) enhanced percentage drop in velocity near the middle core (perhaps caused by excess temperature), and (3) some internal thin zones of ULVZ at its base. We did not observe any direct evidence for ULVZs so we prefer the combination of (1) and (2). The above data analyses provide convincing evidence for a large-scale lower mantle structure beneath South Africa with dimensions of about 1000 kmu1500 kmu7000 km, see Fig. 10 for a schematic 3D image. It appears to have increased density [18], although this result is still in dispute [19]. The P-velocities appear only slightly low as reported by Tkalc ic and Romanowicz [20]. One of the best P-wave constraint comes from the study by Simmons and Grand [16] who examined PcP3P along the same paths discussed in Fig. 5. They obtained a di erence in travel times in the ratio of 1 to 7 s. This means that P-velocities are only slightly decreased from normal mantle, at least for this sample. The large dimension and estimated volume of the plume suggest a low Rayleigh number [21] and its shape appears compatible with some laboratory experiments [22] and dynamic modeling [23]. Images produced by laboratory experiments on uids containing both density and viscosity variations with depth produced a large variety Fig. 10. Schematic 3D image of the African plume. With a northwest^southwest orientation, it is about 1000 km across, 1200^1500 km high, and at least 7000 km long, with a 3% low velocity reduction inside the ridge-like structure. of convective shapes and temporal behavior. Large values of thermal buoyancy, B s 0.5, produce very slow motions even on a scale of 1 Gyr, assuming a high viscosity lower layer. These results are compatible with Kellogg et al. [24] that produce a stable basal layer with B = 0.9 for the whole history of the Earth. Values of B = 0.2 with a high ratio of viscosity in the bottom layer (factor 100) produce images that look like our ridge structure. Such structures have been proposed to occur beneath Africa and the mid-paci c as explanations for superswells [22]. Small-scale plumes and tendrils extend towards the surface from the top of these large-scale more stable ridge-like structures as recently con rmed by Gonnermann et al. [25] in laboratory experiments. Numerical models containing variable thermal conductivity can also stabilize large-scale plume-like structures as addressed by Yuen et al. [26]. We have included the geoid with our data (Fig. 1C) totest these hypotheses. The isotopic data from this region consistently show high 87 Sr/ 86 Sr and 207 Pb/ 204 Pb ratios. A detailed analysis of the complete set of isotopes [27] led Hart [4] to argue that this isotope domain demonstrates very few signs of degassing and was produced early in Earth history by core^mantle processes. The overall ts of the Dupal contours with our structure are impressive and it is tempting to conclude that this material (our ridge structure and probably others) is indeed the hot primitive material needed to explain the extra heat ow and geochemistry as discussed in Kellogg et al. [24]. The above Davaille interpretation implies that this ridge-like structure remains as an isolated feature from early Earth history. Another interpretation is that it is formed by accretion of piles of blobs. These piles would produce ne-scale thermal upwelling to explain the various hot spots [23,28]. Both hypotheses assume a two-stage convection process, stage one consisting of low Rayleigh number sluggish motions in the lower mantle followed by stage two where fast moving high Rayleigh number upper mantle plumes bring materials to the surface [21,29] and in numerical models by Dubu et et al. [30]. A related lowerstage dynamic model tting a 2D cross-section (Fig. 1A) is developed in Ni et al. [3]. They also

12 130 S. Ni, D.V. Helmberger / Earth and Planetary Science Letters 206 (2003) 119^131 argue for a chemical origin because of its wall sharpness. Both types of structures are transient in nature and will be entrained with time, depending on buoyancy contrast (B factor). Thus some types of models imply a well-stirred mantle while the above laboratory-type model suggests a more stationary lower mantle, or perhaps a mixture of convective styles (partly stirred). This would imply that the Earth only recently started producing plates as the Earth s degassing continues to develop. However, it appears that sharpness can also be produced by a strong dependence of viscosity on temperature [31]. Clearly, a great deal of seismology, geodynamic and geochemical modeling is needed to understand the full implications of this new class of lower mantle structure and its in uence on upper mantle convection. Acknowledgements We thank reviewers Louise Kellogg, Dave Yuen, and Don Anderson for their helpful comments, and Evelina Cui for her help with preparation of the manuscript. This work was supported by NSF Grant EAR Contribution Number 8917 of the Division of Geological and Planetary Sciences, California Institute of Technology.[RV] References [1] J. Ritsema, H. Van Heijst, J. Woodhouse, Complex shear wave velocity structure related to mantle upwellings beneath Africa and Iceland, Science 286 (1999) 1925^1928. [2] S. Ni, D.V. Helmberger, Ridge-like lower mantle structure beneath South Africa, J. Geophys. Res. (2002) in press. [3] S. Ni, E. Tan, M. Gurnis, D.V. Helmberger, Sharp sides to the African super plume, Science 296 (2002) 1850^1852. [4] S. Hart, A large-scale isotope anomaly in the Southern Hemisphere mantle, Nature 309 (1984) 753^757. [5] A.A. Nyblade, S.W. Robinson, The African superswell, Geophys. Res. Lett. 21 (1994) 765^768. [6] B.H. Hager, R.W. Clayton, M.A. Richards, R.P. Comer, A.M. Dziewonski, Lower mantle heterogeneity; dynamic topography and the geoid, Nature 313 (1985) 541^545. [7] M. Gurnis, J.X. Mitrovica, J.S. Ritsema, H. Van Heijst, Constraining mantle density structure using geological evidence of surface uplift rates: The case of the African super plume, Geochem. Geophys. Geosyst. 1 (2000) 35. [8] D.E. James, M.J. Fouch, J.C. VanDecar, S. van der Lee, Kaapvaal Seismic Group, Tectospheric structure beneath southern Africa, Geophys. Res. Lett. 28 (2001) 2485^ [9] F.M. Freybourger, J.B. Gaherty, T.H. Jordan, Kaapvaal Seismic Group, Structure of the Kaapvaal craton from surface waves, Geophys. Res. Lett. 28 (2001) 2489^2492. [10] S.P. Grand, R.D. van der Hilst, S. Widiyantoro, Global seismic tomography: A snapshot of convection in the Earth, GSA Today 7 (1997) 1^7. [11] S. Ni, X. Ding, D.V. Helmberger, M. Gurnis, Low-velocity structure beneath Africa from forward modeling, Earth Planet. Sci. Lett. 170(1999) 497^507. [12] E.J. Garnero, D.V. Helmberger, Further structural constraints and uncertainties of a thin laterally varying ultralow-velocity layer at the base of the mantle, J. Geophys. Res. B 103 (1998) 12495^ [13] S. Luo, S. Ni, D. Helmberger, Ultra low velocity zone revealed from multipathed PKPab, Earth Planet. Sci. Lett. 189 (2001) 155^164. [14] L. Wen, D.V. Helmberger, A two-dimensional P-SV hybrid method and its application to modeling localized structures near the core-mantle boundary, J. Geophys. Res. 103 (1998) 17901^ [15] J. Ritsema, S. Ni, D.V. Helmberger, H.P. Crotwell, Evidence for strong shear velocity reductions and velocity gradients in the lower mantle beneath Africa, Geophys. Res. Lett. 25 (1998) 4245^4248. [16] N.A. Simmons, S.P. Grand, Partial melting in the deepest mantle, Geophys. Res. Lett. (2002) in press. [17] L. Wen, Seismic evidence for a rapidly varying compositional anomaly at the base of the Earth s mantle beneath the Indian Ocean, Earth Planet. Sci. Lett. 194 (2001) 83^ 95. [18] M. Ishii, J. Tromp, Normal-mode and free-air gravity constraints on lateral variations in velocity and density of Earth s mantle, Science 285 (1999) 1231^1236. [19] J.S. Resovsky, M.H. Ritzwoller, Regularization uncertainty in density models estimated from normal mode data, Geophys. Res. Lett. 26 (1999) 2319^2322. [20] H. Tkalc ic, B. Romanowicz, Short scale heterogeneity in the lowermost mantle: Insights from PcP-P and ScS-S data, Earth Planet. Sci. Lett. 201 (2002) 57^68. [21] D. Anderson, Top-down tectonics?, Science 293 (2001) 2016^2018. [22] A. Davaille, Simultaneous generation of hotspots and superswells by convection in a heterogeneous planetary mantle, Nature 402 (1999) 756^760. [23] P. Tackley, Mantle convection and plate tectonics: Toward an integrated physical and chemical theory, Science 288 (2000) 2002^2007. [24] L. Kellogg, B.H. Hager, R. van der Hilst, Compositional strati cation in the deep mantle, Science 283 (1999) 1881^ [25] H.M. Gonnermann, M. Manga, A.M. Jellinek, Dynamics and longevity of an initially strati ed mantle, Geophys. Res. Lett. 29 (2002) 1029.

13 S. Ni, D.V. Helmberger / Earth and Planetary Science Letters 206 (2003) 119^ [26] A.S. Yuen, F. Dubu et, E.O. Sevre, E.S.G. Rainey, T.K.B. Yanagawa, Dynamical in uence of thermal conductivity on plume dynamical and the thermal structure of the lower mantle, Proceedings of Superplume International Workshop, Tokyo, Japan, 2002, p [27] B. Dupre, C.J. Alle'gre, Pb-Sr Isotope variation in Indian Ocean basalts and mixing phenomena, Nature 303 (1983) 142^146. [28] P. Tackley, in: M. Gurnis, M.E. Wysession, E. Knittle, B.A. Bu ett (Eds.), The Core-Mantle Boundary Region, AGU Monograph, Washington, DC, 1998, pp. 231^253. [29] D. Anderson, A statistical test of the two-reservoir model for helium isotopes, Earth Planet. Sci. Lett. 193 (2001) 77^82. [30] F.W. Dubu et, D.A. Yuen, E.S.G. Rainey, Controlling thermal chaos in the mantle by positive feedback from radiative thermal conductivity, Nonlinear Process. Geophys. 9 (2002) 311^323. [31] U. Hansen, D.A. Yuen, Modeling constraints from boundary layer estimates of sharpness of superplumes in the lower mantle, AGU Fall Meeting, [32] S. Ni, X. Ding, D.V. Helmberger, Constructing synthetics from deep earth tomographic models, Geophys. J. Int. 140 (2000) 71^82.