Three-dimensional structure of the African superplume from waveform modelling

Transcription

1 Geophys. J. Int. (5) 161, doi: /j.1365-X x Three-dimensional structure of the African superplume from waveform modelling Sidao Ni, 1,2 Don V. Helmberger 2 and Jeroen Tromp 2 1 School of Earth and Space Sciences, University of Science and Technology of China 96 Jinzhai Road, Hefei, Anhui, 26, China 2 Seismological Laboratory , California Institute of Technology, Pasadena, CA Accepted 4 October 15. Received 4 September 27; in original form 4 February 25 INTRODUCTION In the upper mantle there is clear evidence for sharp lateral variations in velocity associated with subduction zones (Zhao et al. 1994; Widiyantoro et al. 1999). As for the lower mantle, various tomographic models have revealed large-scale 3-D structures (Su et al. 1994; Masters et al. 1996; Ritsema et al. 1999), but the resolution of these models, typically a few thousand kilometres, is not high enough to resolve sharp lateral variations. Studies based on rapid variations in traveltimes and anomalous seismic waveforms indicate the existence of small-scale lateral variations near the core mantle boundary (CMB) (Garnero & Helmberger 1998; Russell et al. 1999; Wen& Helmberger 1998; Wysession et al. 1; Luo et al. 1; Ni & Helmberger 1; Tkalcic & Romanowicz 2). The sharpness of large-scale lower mantle structures has been addressed by studies of the African superplume (Ritsema et al. 1998; Ni et al. 2; Ni & Helmberger 3a,b), which we will refer to as the African low-velocity structure (ALVS). The ALVS has a ridge-like structure with dimensions of approximately 7 km long, 1 km high and 1 km across (Fig. 1). The shear velocity in the ALVS is reduced relative to PREM (Dziewonski & Anderson 1981) by about 3 per cent, whereas the compressional velocity is quite normal (i.e. PREM-like). The edges of the structure are very sharp in some places, with a width of less than 5 km. This model of the ALVS explains the rapid variations of S, SKS, Sdiff and ScS traveltimes, as SUMMARY Previous 2-D modelling of seismic waveforms and traveltimes has revealed a large-scale ridge-like velocity anomaly beneath Africa, which is usually referred to as the African superplume. The structure starts from the southern Indian Ocean and extends northwestwards into the Atlantic Ocean, with its base on the core mantle boundary. The structure has relatively sharp lateral boundaries with the shear velocity inside 3 per cent lower than the ambient mantle, while the compressional velocity is almost normal. The 3-D structure is best illustrated by seismic waveforms recorded by the South Africa Array generated by earthquakes in the western Pacific Ocean. The diffracted S phase travels along the axis of the structure for over 6, with the northern- and southernmost paths sampling its edges. The S waveforms are simple but delayed by up to s for paths sampling the middle of the structure, whereas they display two arrivals for paths along the boundaries. These complex waveforms can be explained by 3-D multipathing due to rapid lateral variations in shear-wave velocity along the edges of the structure. These sharp features are confirmed by modelling broadband records associated with the proposed ridge structure with two independent methods: the spectral element method and a ray-based 3-D code (DWKM). Key words: DWKM, low-velocity zone, lower mantle, superplume. well as the complexity of SKS waveforms caused by multipathing effects due to the sharp boundaries. Waveform modelling has been restricted to paths roughly perpendicular to the strike of the ALVS, thus justifying 2-D simulations in cross-sections. For along-strike paths, e.g. for earthquakes in the western Pacific Ocean recorded by the South Africa Array, 3-D effects are expected to be significant. Fortunately, advances in computational and theoretical seismology have facilitated fully 3-D waveform modelling (Komatitsch et al. 2). Additionally, raybased techniques involving a combination of analytical approximations (DWKM) lead to much faster but less accurate 3-D algorithms (Helmberger & Ni 5). Such approximate algorithms provide intuition about how seismic waveforms behave for different structures, and the added speed makes it possible to investigate a wide variety of structures before performing a more costly spectral-element calculation. In this paper we use both the spectral element method (SEM) and the hybrid DWKM algorithm to compute 3-D synthetic seismograms for the ALVS, which we compare with observed S waveforms. DATA AND ANALYSIS The density of seismic stations in Africa used to be quite low, with about five WWSSN stations. Nevertheless, even with analogue records from these few WWSSN stations, Ni et al. (1999) were GJI Seismology C 5 RAS 283

2 284 S. Ni, D. V. Helmberger and J. Tromp able to detect a large-scale low-velocity structure beneath Africa by analysing the change in SKS S crossover distances along various profiles. Ni et al. (1999) also found basal layer with a 3 per cent reduction in S velocity beneath the Atlantic Ocean. More recently, IRIS and GEOFON installed several broadband seismic stations which were used to determine the slope and height of the ALVS based upon modelling waveforms of SKS, ScS and S, taking ad- vantage of the abundance of earthquakes in South America (Ni & Helmberger 3a). However, more detailed studies are only feasible with dense broadband seismic arrays. S, SKS and ScS traveltimes measured at the Tanzania Array (about a dozen stations) were used to map a 2-D cross-section of the ALVS, which appears to be 1 1 km high, with a 3 4 per cent reduction in S velocity. The edge of this structure was estimated to be less than 3 km in width 25km km 13km 19km???? km 289km?? Figure 1. Tomographic model by Grand (2) (left) and the African Low Velocity Structure (ALVS) (right) displaying different depth slices as indicated by the numbers, with red colours (up to 3 per cent) representing slow anomalies and blue colours denoting fast anomalies. The ALVS starts from the mid-atlantic Ocean and ends into the western Indian Ocean, with a crescent shape extending about 1 km above the core mantle (CMB). The question marks denote the uncertain northwestern boundary. C 5 RAS, GJI, 161,

3 Structure of the African superplume Figure 2. Map view of great circle paths connecting the four events in the western Pacific (Table 1) to stations in the South African Array. Events and 995 sample the northern boundary of the ALVS, while event 99 samples the mid-section and event 9794 samples the southern boundary. Table 1. Earthquakes used in this study. Event ID Time Latitude Longitude Depth (deg.) (deg.) (km) :23: :5: :47: :8: based upon the observed rapid changes in traveltimes (Ritsema et al. 1998). Broadband waveform modelling of SKS yields better estimates of the sharpness, favouring a 3 per cent variation over less than 5 km (Ni et al. 2). The deployment of the South Africa Array (more than 5 broadband stations) provides much stronger constraints on the ALVS. Over a thousand measured SKS, ScS and SKKS traveltimes were used to model the ridge-like structure (Ni & Helmberger 3b). The boundaries of this structure can be easily detected from rapid variations in traveltimes. The northwest-to-southeast oriented structure provides an ideal geometry for observing 3-D effects, since the great circle paths connecting earthquakes in the western Pacific Ocean to the South Africa Array are almost parallel to the strike of the ALVS (Fig. 2). We chose four earthquakes with a magnitude of about 7. and a depth greater than 1 km (Table 1) to ensure high-quality Sdiff waveforms and to avoid complications due to the earthquake source and surface reflections. We used Harvard Centroid Moment Tensor (CMT) solutions for the locations and source mechanisms of the four earthquakes. The BHN and BHE channels of the broadband seismic records were rotated to the great circle path and deconvolved with the instrument response to obtain ground displacement. We only modelled the tangential component of S since signals on the radial component are weak. In Fig. 3 the S waveforms are displayed in two different ways: in the top panel the waveforms are organized by increasing epicentral distance using a reduced traveltime, i.e. each trace is aligned on the predicted S arrival time for PREM (Dziewonski & Anderson 1981); in the lower panel the seismograms are organized by increasing azimuth, again with each seismogram aligned on the theoretical S arrival time predicted by PREM. For the seismograms organized by epicentral distance in the top panel the traces seem quite complex for some epicentral distance ranges, e.g. for event between distances of 115 and 117, for event 99 between distances of 123 and 126, and for event 9794 between distances of 121 and 125. The apparent misalignment suggests that, at the same distance, stations show variable delay times for slightly different azimuths. Seismograms organized azimuthally in the bottom panel indeed show much higher coherence, confirming that the delay in traveltime depends more on azimuth than epicentral distance. Note that in exploration seismology it is common to look for vertical structures (e.g. salt domes) by conducting fan shots. Similarly, the South Africa Array data displayed above are more strongly controlled by vertical than horizontal structure, a rather unexpected result for the deep mantle. There are other consistent features among the three events (971222, 99 and 9794) which have epicentres located in a north-to-south trend in the western Pacific Ocean. For event , the northern stations (at larger azimuths) in the South Africa Array show early yet complicated waveforms, whereas the southern stations exhibit late but simple waveforms (delayed by up to 12 s). It seems that the secondary arrivals for the northern stations are of the same origin as the single arrivals for the southern stations (indicated by the thick line), thus suggesting multipathing. The waveform complexity for the northern stations may be caused by the interference of two arrivals: one is early, propagating in the normal mantle, and the other is late, propagating in the low-velocity ALVS. Since each arrival carries part of the energy, the amplitude of the complicated seismograms is expected to be smaller, as observed for the northern stations. For event 97, the traces are delayed by as much as 16 s, with the southern stations showing some waveform broadening. For event 9794, the pattern is reversed as compared with event , with the southern stations showing small delays and the northern stations showing large delays of up to 16 s, and complicated waveforms (at azimuths from to ). Again, the complexity in waveforms seems to be caused by the interference of two arrivals, indicated by the two thick lines, indicative of multipathing. For this event it is quite clear that the first arrivals denoted by the first thick line diminish in amplitude for larger azimuths (from to ), which is typical of diffraction effects (Helmberger & Ni 5; Ni et al. 2). The complexity in waveforms is probably not caused by complicated sources such as double events, which could produce azimuthal variations of traveltimes and waveforms. One reason is that

4 286 S. Ni, D. V. Helmberger and J. Tromp Distance (degree) Figure 3. Observed Sdiff waveforms organized by epicentral distance (top) and azimuth (bottom) for the three events (971222, 99, 9794). Traces of seismograms cross each other when they are organized by distances whereas they appear to be more coherent when organized in azimuth. The thick lines indicate the postulated double arrivals caused by the sharp boundaries of the ALVS.

5 Structure of the African superplume 287 Aziumth (degree) Data Synthetic Figure 4. Comparison of low-pass filtered observed Sdiff waveforms and SEM synthetics. Both data and synthetics are low-pass filtered at 15 s. The synthetics reproduce the observed azimuthal dependence in the traveltimes: for event the northern stations show early arrivals whereas the southern stations have arrivals that are delayed up to 12 s; for event 99 most of the arrivals are delayed more than 12 s, and for event 9794 the northern arrivals are delayed while the southern ones are early. The second negative arrival occurring for event 99 is the ss phase, which apparently arrives too late in the SEM synthetics, indicating some depth discrepancy in the CMT solution or that S velocities beneath the source and the surface are too slow. 2

6 288 S. Ni, D. V. Helmberger and J. Tromp Data Synthetic Synthetic Figure 5. Comparison of broadband observations and hybrid 3-D synthetics (DWKM). DWKM synthetics reproduce the azimuthal variations in traveltime as well as the broadening effects due to the sharp boundaries for certain ranges of azimuths ( for event , for event 9794). the azimuth range involved is too small, only, to cause substantial variations in traveltime. Moreover, the source time functions for these events, as revealed from teleseismic records at IRIS broadband stations in other azimuths, are fairly simple. To convince ourselves that the complexity in waveforms is caused by structure, we investigated another event (995) which is fairly close to event (Fig. 2). The waveforms from event 995 is very similar to those of , with the southern stations showing delays of up to 12 s C 5 RAS, GJI, 161,

7 Structure of the African superplume 289 and the northern stations having complicated waveforms (at an azimuth of ). Another reason to discount complex source time functions is that the complexity occurs at the same position where the traveltimes are changing most rapidly. MODELLING The azimuthal behaviour of the data presented above has been noted by Wen (1). He subdivided the data into corridors and modelled the secondary arrivals as late arriving ScS pulses produced by introducing ultra low-velocity zones (ULVZ) at the CMB with various strengths and thicknesses. Our interpretation is different, in that we suggest that the late arrivals are caused by 3-D multipathing. In particular, when the great circle paths are close to the strike of the structure, 3-D effects are expected to be profound (Helmberger & Ni 5). To test this hypothesis, we use the SEM to construct 3-D synthetic seismograms numerically. The SEM has been demonstrated to be accurate in computing 3-D synthetic seismograms (Tromp & Komatitsch ) and is fast compared with other numerical approaches. In Fig. 4, we display both data (top) and synthetics (bottom) of SH waveforms for the three events, with all the seismograms being low-pass filtered with a cut-off period of 15 s. The synthetic seismograms reproduce the general characteristics of the traveltimes, but the waveform complexity is less well explained. For event , the synthetic waveforms show early arrivals for northern stations and late arrivals for southern stations. For event 99, sampling the middle of the structure, most of the traces are delayed by up to 14 s, with the southern traces somewhat broadened, just as observed in the data. For event 9794, the northern stations exhibit delays of up to 18 s while the southern stations do not show much delay. The waveform complexity in the synthetic seismograms is less pronounced because most of the observed complexity in the data occurs within a 1 s window and is thus subdued in the synthetic seismograms with periods of 15 s and longer. The frequency content of the SEM synthetics can be greatly increased by utilizing bigger supercomputers, e.g. on the Earth Simulator global simulations at periods of 3.5 s and longer are now feasible (Tsuboi et al. 3). To help further develop the ALVS model, we use an approximate ray-based hybrid approach which combines an analytical solution with numerical computation to obtain broadband seismograms. This so-called DWKM method (Helmberger & Ni 5) approximates 3-D wave propagation by generating a family of 2-D responses bracketing the station of interest using the WKM code (Ni et al., 3). We sample the inner Fresnel zone, i.e. the lit zone, with two responses calculated at distances x =±164 km away from the true station, and we sample the diffracted zone with two responses at distances y =±396 km. These four responses are weighted and combined by a diffraction operator controlled by the source duration. Synthetics generated based upon this approximation are displayed in Fig. 5 along with the data (top panel). The DWKM SEM Figure 6. Comparison between DWKM synthetics and SEM synthetics for event DWKM synthetics are computed with a dominant period around 15 s, comparable to that of the SEM synthetics. Both synthetics show the same azimuthal variation in traveltimes and broadened waveforms for the northern stations.

8 29 S. Ni, D. V. Helmberger and J. Tromp multipathing is more obvious at these shorter periods, as expected. For event , the northern stations show two arrivals with the first arrival near the PREM prediction, whereas the southern stations display late arrivals. The amplitude for the northern stations is only about half of that for the southern stations, as predicted by the synthetics. For event 99, most of the synthetics are delayed by up to 14 s, with the southern ones slightly early yet complicated, as observed in the data. This occurs at an azimuth of and probably means that either there is a ULVZ at the CMB which delays ScS,as proposed by Wen (1), or the southern ridge boundary has some fine structure. For event 9794, the synthetics are gradually delayed from south to north, with some complexity in the waveforms appearing around,but not strong enough to match the data. The boundaries are currently sharp in the ALVS model, so that the multipathing as approximated by the hybrid method cannot be enhanced. More waveform complexity could be generated by adding avery anomalous structure such as ULVZ along the boundaries, as proposed by Wen (1). To validate the hybrid 3-D algorithm, we constructed seismograms using a source with a dominant period of 16 s and compared those with synthetics generated by the SEM, as displayed in Fig. 6 for the ALVS model and in Fig. 7 for the tomographic model proposed by Ritsema et al. (1999). The synthetics agree quite well, except for a difference in shape probably caused by the 1-D part of the calculation (WKM). Note that the DWKM synthetics were generated from an earth-flattened model which underestimates the decay rate of diffracted core arrivals (Gilbert & Helmberger 1972). However, the differential shape change across the array is expected to be small and can be absorbed into an effective source time function. DISCUSSION AND CONCLUSION We have demonstrated that a ridge-like African low-velocity structure (ALVS) explains the rapid azimuthal variations of observed SH waveforms for earthquakes in the western Pacific Ocean recorded by the South Africa Array. The double SH pulses observed for events and 9794 are interpreted as multipathing effects, with the 3-D wavefield split along the relatively sharp boundaries. To demonstrate the effects of multipathing, we plotted the great circle paths along with a map view of the ALVS for different depths (Fig. 8). The coloured segments of the paths denote the actual paths traversed in the each layer. It is obvious that for the northernmost stations the great circle paths are very close to the boundary of the ALVS. Therefore, for the northernmost stations the wavefield is split, with one part travelling in the normal mantle and the other part travelling in the ALVS. For the southern stations the rays are well confined to the bulk of the low-velocity region, and only one delayed pulse is produced. Although this model shows promise, one can wonder about its uniqueness. If we consider only the data displayed in Fig. 3, we can easily trade off the low-velocity distribution with thickness. That is, we could change the velocities in the patches (Fig. 8) while DWKM SEM Figure 7. Comparison of SEM and DWKM synthetics for Ritsema s tomographic model. Most tomographic models do not have sharp variations due to damping in the inversion and thus do not explain the rapid variations in traveltimes and waveforms as shown in Fig. 8.

9 Structure of the African superplume Figure 8. Great circle paths for event sampling the ALVS model broken down into three panels, with green lines representing segments of Sdiff ray paths within various depth ranges. The upper plot shows the delayed S path sampling from depths of 19 to 2 km, the middle panel from 2 to 25 km, and the bottom panel from 25 to 289 km. The large thickness of the ALVS model accommodates the large delays in Sdiff without strong velocity reductions in D. conserving the required traveltimes. If we force the entire anomaly into D (the lowermost patch), we would need to drop the shearwave velocity by about 6 per cent relative to PREM. To avoid introducing waveform distortions for the profile associated with the middle of the ALVS (99), one needs to taper the velocity decrease from about 2 per cent at the top to about 9 per cent at the CMB, as proposed by Wen (1). With an average velocity reduction of only 6 per cent in D,ittakes about 25 of epicentral distance to produce the observed delays of 15 s at the South Africa Array (see the bottom of Fig. 8). However, the vertical traveltime anomaly for this D structure is only 2 s, which is too small to explain about 1 observations of an SKS traveltime jump associated with crossing the boundary, with average delays of 6 s(ni & Helmberger 3b). Thus, the combination of near vertical delays of 6 s and the diffracted paths of.6 s per degree require a large-scale structure. However, the LVZ structure beneath the Indian Ocean is poorly known and there are some uncertainties about its shape and lateral velocity distribution. Such uncertainties could affect SKS delays which are currently limited to the tip of South Africa. But the data presented here are consistent with a tall structure which is more in agreement with modern tomographic models (Ritsema et al. 1999). An S velocity reduction of up to 9 per cent confined in D greatly delays ScS even to large epicentral distances (Wen 1). Indeed, Wen (1) explains the second pulses in the data set for event as ScS caused by a ULVZ. However, this model would also predict a strong ScS on the radial component, and would have a strong effect on SVdiff. Note that for PREM-like models, the S and ScS are asymptotic in timing beyond 93, and because they have similar amplitude but opposite signs, the SVdiff amplitude damps out very rapidly with distances beyond the shadow zone in most data sets. Although SKS is apparent for all the events listed in Table 1, none displayed clear SVdiff arrivals, even after stacking (Fig. 9). In other global regions of relatively slow D such as the mid-pacific SVdiff is apparent even to those distances (112 1 )ascompared with SKS (Ritsema et al. 1997). Thus, we suggest that the second pulses in the SH component for events and 9794 are not ScS and a ULVZ is not needed to explain the data. To further enhance the evidence for a tall ALVS, we analysed SS traveltimes for event 99, sampling the mid-section of the ALVS, employing a cross-correlation method (Ni & Helmberger 3b). We found that SS traveltimes exhibit systematic azimuthal variations (solid circles in the bottom panel of Fig. 1) with arrivals at the northern stations (larger azimuth) delayed up to 5 s. Next, we measured the traveltime variation with the same method for the SEM synthetics for the Ritsema et al. (1999) model and found that SEM synthetics (open triangles) show a similar trend. Since SS ray paths for event 99 only sample a depth of around 17 km (top panel, Fig. 1), any thin model (<3 km) with a very slow basal layer cannot explain the observed SS traveltime variations. Lastly, the issue of multipathing caused by source structure could be a problem; in particular slabs are expected to produce such effects, and P-waveform multipathing due to slab structure has been modelled (Kendal & Thomson 1993). However, the diffracted P waves observed by the South Africa Array typically do not show much complexity (Wen 1). Fig. 11 displays a comparison of observed and predicted P and S waveforms for event 9794 with our preferred model. The simplicity of the P waveform would seem to eliminate this explanation, since the ray paths for P and S are very similar in the source region. In conclusion, the ALVS must have relatively sharp shear-velocity variations across nearly vertical walls to produce the strong azimuthal patterns discussed above, while the P waves show very little variation (Fig. 11). Both the sharp boundary and the anomalous P/S ratio support arguments for a thermochemical structure with broad dimensions, as suggested by Anderson (2). The top of the ALVS could be quite complicated, with some small-scale upwellings causing the scattered SS traveltime observations (Fig. 11) and the smallscale SKS scatter presented in the work of Ni & Helmberger (3b). Such features would be consistent with chemical heterogeneity in geodynamic modelling (Tackley ) and broad superplume-like

10 292 S. Ni, D. V. Helmberger and J. Tromp SHdiff SKS SKKS SVdiff Figure 9. Stacked SHdiff, SKS and SVdiff waveforms for event (heavy dashed, solid and light dashed lines, respectively). Only stations with multipathed SHdiff are stacked. The two SHdiff pulses are modelled as 3-D multipathing effects. SVdiff is very weak as compared with SKS, arguing that there is no substantial low-velocity gradient in the lowermost mantle (Ritsema et al. 1997). Consequently, a ULVZ as proposed by Wen (1) seems to be absent in this region, because it would produce strong SVdiff. upwelling in laboratory stimulations (Davaille 1999; Gonnermann et al. 2). ACKNOWLEDGMENTS The data used in this study were provided by the IRIS DMS. Special thanks go to the Kaapvaal project scientists for producing this extraordinary data set ( This project is supported by NSF under grants EAR-885 and EAR Contribution number 938, Seismological laboratory, Division of Geological and Planetary Science, California Institute of Technology. REFERENCES Figure 1. Top: SS ray paths for epicentral distances of 126.Atthese distances the rays sample the structure about 1 km above the CMB. Bottom: observed SS traveltime variations (event 99, solid circles) and SEM predictions (open triangles). The 5 s delay in the SS traveltime requires that the ALVS must exceed 1 km in height. Anderson, D.L., 2. The case for irreversible chemical stratification of the mantle, Int. Geol. Rev., 44(2), Davaille, A., Simultaneous generation of hotspots and superswells by convection in a heterogenous planetary mantle, Nature, 2(6763), Dziewonski, A.M. & Anderson, D.L., Preliminary Reference Earth Model, Phys Earth planet. Inter., 25(4), Garnero, E.J. & Helmberger, D.V., Further structural constraints and uncertainties of a thin laterally varying ultralow-velocity layer at the base of the mantle, J. geophys. Res. Solid Earth, 13(B6), Gilbert, F. & Helmberger, D.V., Generalized ray theory for a layered sphere, Geophys. J. R. astr. Soc., 27, Gonnermann, H.M., Manga, M. & Jellinek, A.M., 2. Dynamics and longevity of an initially stratified mantle, Geophys. Res. Lett., 29(1) doi:1.129/2gl Grand, S.P., 2. Mantle shear-wave tomography and the fate of subducted slabs, Phil. Trans. R. Soc. Lond., A, 36(18), 5 1. Helmberger, D. & Ni, S., 5. Approximate 3D Body-Wave synthetics for Tomographic Models, Bull. seism. Soc. Am., 95, 1,, doi: /14. Kendal, J.M. & Thomson, C.J., Seismic modeling of subduction zones with inhomogeneity and anisotropy, 1. Teleseismic P-wave-front tracking, Geophys. J. Int., 112(1),

11 Structure of the African superplume 293 SEM (----P),S Data (----P),S Figure 11. Comparison of SEM P-wave (dotted) and S-wave (solid) synthetics for event 9794 with observations. The P and S waves are aligned relative to PREM predictions for both data and synthetics. The data range in distance from 113 to 118 and are plotted in a fan profile as a function of azimuth. The data have been low-pass filtered at 15 s to match the synthetics. The ALVS model has PREM compressional velocities, which agrees with the observations. However, the S waves exhibit a strong delay when crossing the boundary. Komatitsch, D., Ritsema, J. & Tromp, J., 2. The spectral-element method, Beowulf computing, and global seismology, Science, 298(5599), Luo, S.N., Ni, S.D. & Helmberger, D.V., 1. Evidence for a sharp lateral variation of velocity at the core- mantle boundary from multipathed PKPab, Earth planet. Sci. Lett., 189(3 4), Masters, G., Johnson, S., Laske, G. & Bolton, H., A shear-velocity model of the mantle, Phil. Trans. R. Soc. Lond., A, 354(1711), Ni, S. & Helmberger, D.V., 1. Probing an ultra-low velocity zone at the core mantle boundary with P and S waves, Geophys. Res. Lett., 28(12), 5 8. Ni, S.D. & Helmberger, D.V., 3a. Ridge-like lower mantle structure beneath South Africa, J. geophys. Res. Solid Earth, 18(B2) doi:1.129/1jb1545. Ni, S.D. & Helmberger, D.V., 3b. Seismological constraints on the South African superplume; could be the oldest distinct structure on Earth, Earth planet. Sci. Lett., (1 2), Ni, S.D., Ding, X.M., Helmberger, D.V. & Gurnis, M., Low-velocity structure beneath Africa from forward modeling, Earth planet. Sci. Lett., 17(4), Ni, S.D., Ding, X. & Helmberger, D.V.,. Constructing synthetics from deep earth tomographic models, Geophys. J. Int., 1, Ni, S.D., Tan, E., Gurnis, M. & Helmberger, D., 2. Sharp sides to the African superplume, Science, 296(5574), Ni, S, Cormier, V.F. & Helmberger, D.V., 3. Construction of synthetics for 2D structures: analytical vs. numerical, Bull. seism. Soc. Am., 93(6), Ritsema, J., Garnero, E. & Lay, T., A strongly negative shear velocity gradient and lateral variability in the lowermost mantle beneath the Pacific, J. geophys. Res., 12, Ritsema, J., Ni, S., Helmberger, D.V. & Crotwell, H.P., Evidence for strong shear velocity reductions and velocity gradients in the lower mantle beneath Africa, Geophys. Res. Lett., 25(23), 4 4. Ritsema, J., van Heijst, H.J. & Woodhouse, J.H., Complex shear wave velocity structure imaged beneath Africa and Iceland, Science, 286(5446), Russell, S.A., Lay, T. & Garnero, E.J., Small-scale lateral shear velocity and anisotropy heterogeneity near the core-mantle boundary beneath the central Pacific imaged using broadband ScS waves, J. geophys. Res. Solid Earth, 14(B6), Su, W.J., Woodward, R.L. & Dziewonski, A.M., Degree-12 model of shear velocity heterogeneity in the mantle, J. Geophys. Res. Solid Earth, 99(B4), Tackley, P.,. Mantle convection and plate tectonics: toward an integrated physical and chemical theory, Science, 288, 2 7. Tkalcic, H. & Romanowicz, B., 2. Short scale heterogeneity in the lowermost mantle: insights from PcP P and ScS S data, Earth planet. Sci. Lett, (1), Tromp, J. & Komatitsch, D.,. Spectral-element simulations of wave propagation in a laterally homogeneous Earth model, in Proceedings of

12 294 S. Ni, D. V. Helmberger and J. Tromp the Erice 1999 School of Geophysics, eds Boschi, E., Ekström, G. & Morelli, A., pp Tsuboi, S., Komatitsch, D., Ji, C. & Tromp, J., 3. Broadband modeling of the 2 Denali fault earthquake on the Earth Simulator, Phys. Earth planet. Inter., 139(3 4), Wen, L.X., 1. 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(1 2), Wen, L.X. & Helmberger, D.V., Ultra-low velocity zones near the core mantle boundary from broadband PKP precursors, Science, 279(5357), Widiyantoro, S., Kennett, B.L.N. & van der Hilst, R.D., Seismic tomography with P and S data reveals lateral variations in the rigidity of deep slabs, Earth planet. Sci. Lett., 173(1 2), Wysession, M.E., Fischer, K.M., Al-eqabi, G.I., Shore, P.J. & Gurari, I., 1. Using MOMA broadband array ScS S data to image smaller-scale structures at the base of the mantle, Geophys. Res. Lett., 28(5), Zhao, D.P., Hasegawa, A. & Kanamori, H., Deep-structure of Japan subduction zone as derived from local, regional, and teleseismic events, J. Geophys. Res. Solid Earth, 99(B11),