Lower mantle heterogeneity beneath Eurasia imaged by parametric ...

Physics of the Earth and Planetary Interiors 108 Ž1998. 201–218

Lower mantle heterogeneity beneath Eurasia imaged by parametric migration of shear waves Susan L. Bilek ) , Thorne Lay Institute of Tectonics, UniÕersity of California, Santa Cruz, CA, USA Received 3 November 1997; accepted 2 April 1998

Abstract Lower mantle structure beneath central Eurasia is examined using a recently developed technique for migrating parametric representations of shear waves from teleseismic events. The migration method of Lay and Young wLay, T., Young, C.J., 1996. Imaging scattering structures in the lower mantle by migration of long period shear waves. J. Geophys. Res., 101: 20 023–20 040.x assumes isotropic scattering from discrete heterogeneities to account for S wave coda arrivals, and provides resolution at scale lengths of about 500 km. The migration is applied to 15 s period shear wave data from 21 western Pacific earthquakes recorded in Europe and the Middle East that exhibit extra arrivals between S and ScS on the transverse components. Migrations are performed using PREM as well as model SGLE wGaherty, J.B., Lay, T., 1992. Investigation of laterally heterogeneous shear velocity structure in DY beneath Eurasia. J. Geophys. Res., 97: 417–435.x, which has a discontinuity at 2605 km, 286 km above the core–mantle boundary. The migration images for the most coherent coda arrival ŽScd. resemble those expected for a lower mantle discontinuity model and simulations are performed for models with discontinuities at various depths. Quantitative correlation of the images for the data and synthetic migrations show the optimal average depth of the discontinuity to be 2605 km. Migrations of subsets of the data suggest that the discontinuity depth varies from 2605 km in the northern end of the study area to 2620 km in the southern end. An additional intermediate arrival ŽScd2. is observed in a limited portion of the dataset, and migrations indicate this phase is generated by localized discrete scatterers located at depths of 2750 to 2770 km under north–central Eurasia. These results are generally consistent with previous forward modeling studies, but the migration approach allows lateral variations to be modeled more systematically. The heterogeneous structure near the base of the mantle appears to be a manifestation of a complex boundary layer. q 1998 Elsevier Science B.V. All rights reserved. Keywords: Coda waves; Lower mantle; Shear waves; Discontinuity

1. Introduction

) Corresponding author. Earth Sciences Department, University of California, Earth and Marine Sciences Building, Room A232, Santa Cruz, CA 95064, USA. Fax: q1 408 469 3074; e-mail: [email protected]

The structure of the lower mantle has been characterized as very heterogeneous. Seismic tomography shows global patterns of large scale velocity heterogeneities, with lateral resolution of 2000–4000 km Že.g., Su et al., 1994; Li and Romanowicz, 1996;

0031-9201r98r$19.00 q 1998 Elsevier Science B.V. All rights reserved. PII S 0 0 3 1 - 9 2 0 1 Ž 9 8 . 0 0 1 0 5 - 8

202

S.L. Bilek, T. Lay r Physics of the Earth and Planetary Interiors 108 (1998) 201–218

Masters et al., 1996.. The consistency between midmantle regions in two recent global tomographic models ŽGrand et al., 1997; van der Hilst et al., 1997. indicates that the large scale heterogeneities are gradually becoming well defined by arrival time tomography. There is less consistency between models in the lowermost mantle, possibly due to strong chemical and thermal effects ŽLoper and Lay, 1995.. The main challenge in global arrival time tomography is obtaining large datasets with many crossing raypaths. In order to assess lateral heterogeneities at smaller wavelength, tomographic models have been developed for smaller regions such as beneath North and South America and the adjacent oceans ŽGrand, 1994. or the approximately 1008 = 1008 block encompassing Southeast Asia and Australia ŽWysession et al., 1994.. Until more data become available, the areas of the world where high resolution tomography is feasible will be limited. Because arrival time tomography methods do not yet resolve small scale deep mantle heterogeneities, other techniques are being developed to accomplish this goal. Most of these alternative techniques use datasets which sample localized regions and exploit a variety of phases found within the waveforms. Core reflected phases, phases diffracted by the core, and phases that enter the core have all been successfully used to image lower mantle structure and the core–mantle boundary Že.g., Vidale and Benz, 1992; Weber, 1993; Wysession et al., 1994; Garnero and Helmberger, 1993, 1996.. Analyses of 10–20 s period body waves augment the existing tomographic results, providing higher resolution images of heterogeneity with scale lengths between 500–2000 km Žsee Lay, 1995 for a review.. The method employed in this study images ; 500 km scale lateral variations in the lower mantle by using a parametric migration of travel time and amplitude measurements of 10–20 s period shear waves and their coda arrivals at distances of 708 to 828. Several possibilities have been proposed for the origin of the teleseismic S coda arrivals, such as SKS scattering, receiver reverberations, and source multipathing, but previous studies indicate that a lower mantle origin is a more likely explanation for the systematic arrivals observed between S and ScS ŽLay and Young, 1986; Lay, 1986.. Long period SH waves are used because they have simple waveforms

and the relatively low shear velocities in the lower mantle where their raypaths bottom separate arrivals more than for P waves ŽYoung and Lay, 1987.. Lay and Young Ž1996. developed the parametric migration technique to study the lower mantle shear velocity structure beneath Alaska, but coda migration techniques have been used earlier in a variety of areas Že.g., Lynnes and Lay, 1989; Revenaugh, 1995.. Scherbaum et al. Ž1997. exploit back azimuth constraints on scattered secondary arrivals for P waves to enhance lower mantle imaging, but this is only viable when array data are available. The most ambitious application of a complete diffraction tomography inversion for lower mantle structure was performed by Ying and Nataf Ž1998., who assumed a particular scattering parameterization Žcylindrical plume structures.. In general, we need methods that can image arbitrary scatterer geometries given the possible complexity of lower mantle structure.

Fig. 1. Schematic drawing of the coda scattering concept. Ža. Raypath for a direct S arrival and the scattering ellipsoid for an observed coda arrival, Scd. The scattering surface is defined by the differential time Scd–S and a reference background velocity model. Žb. Three source–receiver combinations and scattering ellipsoids for specific coda arrivals at each station. Intersections of all three ellipsoids, denoted by stars, reveal possible locations of common scatterers. Identifying kinematically viable scattering locations is the key concept for the parametric migration.


The parametric coda migration technique used here does not assume an a priori scatterer geometry, and a formal diffraction tomography method is not pursued. Instead, coda arrivals are attributed to isotropic scattering from deep velocity heterogeneity. The SH waveforms can generally be characterized as a sequence of discrete arrivals with measurable relative arrival times and amplitudes. A spike-train parameterization simplifies the signals and eliminates instrument and source complexity effects, while preserving the general characteristics of the main arrivals. For a large number of stations, time shifts are measured between the S coda arrivals and a reference phase, either S or ScS. Assuming S to S scattering, each differential time defines a scattering ellipsoid surrounding the direct S wave path ŽFig. 1a.. Multiple source–receiver combinations for several events provide a large number of ellipsoids which intersect in areas where common scattering can account for the coda arrivals ŽFig. 1b.. A scattering surface or reflector is indicated by a smooth, continuous surface of intersections, whereas a point scatterer should yield a very small area with a large number of intersections ŽLay and Young, 1996.. Because only one or two extra arrivals are typically

203

observed in the S coda, images with complex discrete scattering surfaces or convoluted structures are less probable, as every station should see multiple scattered arrivals in that case. This procedure is applied to study deep mantle structure beneath Eurasia.

2. Data The data are transverse component waveform records from 21 magnitude 5.6–6.1 events in the western Pacific between 1967 and 1984 ŽTable 1., recorded by 45 long-period WWSSN stations in Europe and the Middle East ŽFig. 2.. Fig. 3 shows representative waveforms, and many additional data are shown in Gaherty and Lay Ž1992.. Event depths range from 115 km to 566 km. All of the data have clear S and ScS arrivals. Many observations between 708 to 828 display an additional arrival, which is commonly attributed to reflection from a discontinuity a few hundred kilometers above the core–mantle boundary ŽCMB.. At wide angles of incidence, a discontinuity produces a triplication, with two arrivals generated; Sbc, involving energy reflected from

Table 1 Event information Date

Origin time, UT

Latitude, deg

Longitude, deg

Depth Žkm.

Region

7r4r74 8r13r67 12r1r67 2r28r68 3r31r69 9r5r70 1r29r71 5r27r72 1r31r73 9r10r73 7r10r76 12r12r7 6r21r78 9r2r78 8r16r79 3r31r80 11r27r8 7r3r83 7r24r83 10r8r83 4r23r84

2342:12.9 2006:52.3 1357:03.4 1208:03.4 1925:27.0 0752:27.2 2158:03.2 0406:49.6 2055:54.2 0743:32.2 1137:14.0 0108:51.1 1110:38.7 0157:34.2 2131:24.9 0732:32.4 1721:44.3 0249:28.2 2307:31.8 0745:26.3 2140:34.2

43.10 35.43 49.45 32.95 38.49 52.28 51.69 54.97 28.22 42.48 47.31 28.04 48.27 24.81 41.85 35.49 42.93 20.19 53.91 44.21 47.44

142.58 135.49 154.40 137.85 134.52 151.49 150.97 156.33 139.30 131.05 145.75 139.67 148.66 121.87 130.86 135.52 131.19 122.41 158.36 130.74 146.73

157 367 144 348 397 560 515 397 508 552 402 503 380 115 566 362 525 221 190 551 399

Sea of Okhotsk Sea of Japan Sea of Okhotsk Izu-Bonin Sea of Japan Sea of Okhotsk Sea of Okhotsk Kamchatka Izu-Bonin Sea of Japan Sea of Okhotsk Izu-Bonin Sea of Okhotsk Taiwan Sea of Japan Sea of Japan Sea of Japan Philippine Tr. Kamchatka Sea of Japan Sea of Okhotsk

204


Fig. 2. Source–receiver geometry for this study shown in Mercator projection. Circles indicate the 21 western Pacific events and triangles indicate the 45 long period WWSSN stations in Europe and the Middle East.

the discontinuity, and Scd, involving energy refracted from below the discontinuity. Separate Scd and Sbc arrivals are not distinguishable in our longperiod data, and the extra arrival, which we call Scd, could be a single scattered phase rather than a composite of triplication arrivals. An intermittently observed additional arrival will be referred to as Scd2. Fig. 3 shows that the Scd and Scd2 arrivals appear to be simple extra pulses in the waveforms, and we treat them as such. Lay and Young Ž1996. discuss possible complexities if these arrivals are actually triplication phases, noting that minor bias in the final interpretations may be incurred, but these are intrinsic given the resolution of the long-period data and are eliminated by the simulation approach used below. Because the Scd arrival is most stably observed in the 708–828 distance range, we only consider source–receiver combinations in this range.

Gaherty and Lay Ž1992. measured differential travel times between the direct S and ScS phases and the intermediate Scd and Scd2 arrivals, as well as for corresponding surface reflected arrivals ŽsS, sScS, sScd., from the digitized WWSSN records. We use these previously measured times directly, as Lay and Young Ž1996. found that more sophisticated methods for measuring the differential travel times gave only minor changes for a comparable dataset beneath Alaska. We did return to the digitized records to measure the relative amplitudes of the Žs.S, Žs.Scd, Žs.Scd2 and Žs.ScS phases for incorporation into the parametric migration. In total, 120 Scd, 25 sScd, 42 Scd2 and 5 sScd2 arrivals were measured from these records. The measurements essentially characterize the SH waveforms as spiketrains with three or four arrivals with relative arrival times accurate to about "1 s and relative amplitudes accurate to 20%. Our


205

large numbers of intersections or few intersections. Areas with many intersections indicate possible common scatterers for many of the paths. Map projections at each of the grid depths provide a visual representation of the distribution of intersections throughout the grid volume. The hits are typically weighed by the amplitude ratio of the scattered and reference phases to emphasize stronger arrivals. The resulting migration images intrinsically contain artifacts caused by incomplete raypath coverage, lack of destructive interference by noise Žas occurs in true migration., and errors in the reference model. Simulations allow us to account for the primary artifacts caused by geometry of the data distribution, and we adjust our reference model to match ScS–S times.

3. Method validation

Fig. 3. Examples of transverse component data used in this study. Gaherty and Lay Ž1992. show many more waveforms. The S, ScS, and coda arrivals, Scd and Scd2, are identified.

migration processing is not intended to detect the readily observed arrivals but to explore the significance of amplitude and travel time fluctuations in the arrivals with respect to possible lower mantle structures. As the geometry of the dataset requires a lower mantle origin for the frequently observed Scd phase, a grid of possible scattering locations was defined within the lower mantle beneath Eurasia, extending from 2200 km to the CMB, with 50–100 km depth spacing and about 18 horizontal spacing. The PREM 1 s SH velocity model ŽDziewonski and Anderson, 1981. was initially used for migrations, but any model can be considered, and we also use other structures. We do not make any a priori assumption that the intermediate arrivals are caused by discontinuities, but treat them as scattered phases. A scattering ellipsoid intersection is recorded as a hit if the measured and theoretical ScdŽ2. –S or ScS–ScdŽ2. times for a particular grid location match within the "1 s tolerance level. Summing the ellipsoid intersections, or hits, at each grid point reveals areas with

Prior to using this migration method on real data, we applied it to simulations with specified scatterers to test the resolution for our particular source–receiver geometry. A synthetic dataset with the same ray coverage as the actual dataset was generated, with a single point scatterer located at 608N, 878E, and 2500 km depth. Isotropic scattering was assumed, producing intermediate arrivals ŽS ) . in the synthetic dataset. The result produced by migration of the noise-free synthetic data is shown in Fig. 4. The largest symbol, representing 193 intersections Žthe total number of source–receiver combinations., is located at the true location because the velocity model used for the migration is correct. Placing the point scatterer at other grid locations produces similar images with the point scatterer being accurately imaged in each case. This test indicates the type of image expected for a point scatterer, with some streaking of the image caused by the particular raypath geometry. This streaking is a function of both the limited, nonisotropic ray coverage and the "1 s tolerance assumed for the timing of the arrivals. Similar results were found by Lay and Young Ž1996. beneath Alaska, although there is better resolution in the present case. Of course, if multiple arrivals are present, one can image multiple scatterers by considering all coda arrivals at each station without needing to associate them a priori.

206


Fig. 4. Migration images produced for a simulation involving a single point scatterer. Circles indicate the number and location of ellipsoid intersections at points in the lower mantle grid. The largest symbol represents a hit count of 193 Žall possible source–receiver combinations. and the smallest represents a hit count of three. A point scatterer was inserted at 608N, 878E at a depth of 2500 km. This location is well resolved but there is some smearing along the predominately east–west raypaths.

It is unlikely that most deep mantle scattering is caused by isolated point scatterers. Another possibility is that the extra arrivals are produced by localized regions of strong velocity gradient, such as discontinuities, plumes, or slabs. As the core–mantle boundary is a spherical boundary with a strong velocity gradient, the ScS phase can be viewed as a scattered arrival for the special case of a specular reflection, as the smoothness of the boundary gives rise to only one arrival at each station rather than multiple arrivals from the finite surface. The ScS arrivals in the real dataset were treated as scattered arrivals and their lag times measured with respect to the direct S

phase. These ScS–S differential times then define scattering ellipsoids, and a migration image is formed, with no amplitude weighing. As seen in Fig. 5, the migration images for a phase reflected from a spherical boundary are quite different from those for a point scatterer. Instead of one grid point with a large number of hits, the reflector is manifested in a coherent, or spatially concentrated, grouping of the largest hit counts at the apex of the scattering ellipsoids, with many shallower artifacts. In this case, the most coherent scattering surface is near a depth of 2850 km, shallower than the actual CMB depth of 2891 km. This is because PREM overpredicts the


207

Fig. 5. Results of migration of ScS relative to S using the actual ScS–S differential times and PREM as the reference velocity model. In this case, we know that the correct interpretation is reflection from a surface, but there is a complex image formed throughout the volume. The most coherent pattern is imaged at a depth of 2850 km rather than at the CMB depth of 2891 km. This is because PREM overpredicts the ScS–S times on average in this region. The largest symbol represents 26 hit counts, with no amplitude weighing.

observed ScS–S differential times by about 1 s in this region. Hence PREM is not a perfect reference model for migration. There are also blurring effects caused by strong lateral variations in ScS–S anomalies in these data ŽLay et al., 1997.. The artifacts at shallower depths can have large hit counts, but the spatial pattern is complex. Any complex distribution of scatterers should give rise to multiple coda arrivals at each station. A simple reflector gives rise to a single extra arrival in the data rather than multiple discrete arrivals, so it is the uniformity of the image at a particular depth that favors the interpretation of the simple coherent reflector, in this case representing the CMB scatterer. The complex smearing of the image can actually be used to an advantage as shown

below, as it results from the specific path sampling of the reflector. This migration of the ScS data indicates that PREM may bias the Scd migrations. A model which matches the ScS–S differential times more closely should be superior. Because we found that our initial migrations indicated the existence of a lower mantle reflector, as previously proposed, we adopt a reference discontinuity model that was specifically developed for this region. Gaherty and Lay Ž1992. derived model SGLE as the best average model to explain the differential time observations in this area. SGLE is identical to PREM at shallow depths, however, SGLE has a 2.75% shear velocity increase at a depth of 2605 km ŽFig. 6. and has higher velocities in DY

208


Fig. 6. Reference shear velocity models PREM and SGLE. Both models are identical above a depth of 2300 km. SGLE has a 2.75% shear wave velocity discontinuity at depth 2605 km.

than PREM. This predicts shorter ScS–S differential times than PREM by about 1 s. We now consider results for this reference model.

4. SGLE migrations Using SGLE as the reference velocity model for the observed ScS–S differential time migration gives the results in Fig. 7. Larger hits are recorded at 2891 km, and the image at a depth of 2850 km is not as coherent as in the migration using PREM ŽFig. 5.. Lateral variations appear to be responsible for the still imperfect focusing of the image at 2891 km, with subregions suggesting shallower or deeper CMB depths. Simulations with synthetic ScS–S times are similar to Fig. 7, but with greater coherence at 2891 km. Since SGLE gives a better baseline for the ScS–S times than PREM, we perform migrations of Scd using SGLE as the reference model. To assess

the effects of lateral variations in the S times, we consider both Scd–S and ScS–Scd migrations. Fig. 8 shows the migration of the Scd arrivals relative to the S phase. The hit counts are weighed by each ScdrS amplitude ratio. This scaling gives weight to more robust Scd arrivals and reduces the importance of those Scd arrivals that were small or questionable measurements. For each depth section, the circles indicate the amplitude weighed sum of scattering ellipsoid intersections with the lower mantle grid. The largest symbol results from 23 intersections at a particular grid point, while the smallest symbols involve three intersections. The most coherent scattering structure is located near 2600–2620 km, where the maximum hit count is 17, with a very coherent patch in northern Eurasia. Areas with high hit counts, but lower amplitudes are found beneath central Eurasia with the best coherence at 2620 km. The northernmost area coincides with the region where clear evidence for both P and S wave discontinuities has been previously presented Že.g., Lay and Helmberger, 1983; Weber and Davis, 1990; Weber, 1993.. Migration of the observed ScS–Scd differential times is shown in Fig. 9, with the ScdrScS amplitude ratios being used as weights. As before, the circles indicate location and number of amplitude weighed hits, with the largest number of hits being 26 and the smallest number being three. Similar to the Scd–S case, the 2580–2620 km depth sections shows the most coherent overall structures. The ScS–Scd migrations are thought to be more stable because the turning depth of the ScS phase is at the CMB, closer to the depth at which Scd is generated than is the case for the S wave bottoming points. The amplitude weighing again reduces the coherence and strength of the southern section of the imaged reflector at this depth ŽFig. 10., similar to the Scd–S migrations, but the overall coherence near 2600 km is better for the ScS–Scd case. This can be understood as the result of the larger Scd–S fluctuations than ScS–Scd fluctuations relative to model SGLE in these data ŽLay et al., 1997.. The overall results of these migrations are in excellent agreement with those of Gaherty and Lay Ž1992., who determined a discontinuity depth in this area at 2605 km. Fig. 11 is a Mercator projection of the study area with the grid point hit counts for the ScS–Scd


209

Fig. 7. Results of migration of ScS relative to S using the actual ScS–S differential times and SGLE as the reference velocity model. In this case, the most coherent surface appears to be deeper than in Fig. 5, although the image at the CMB is not perfectly imaged. The largest symbol represents a hit count of 20, with no amplitude weighing.

migration at 2600 km depth. Also plotted are the ScS bottoming points for low amplitude and non-observations of the Scd arrival as designated by Gaherty and Lay Ž1992.. Diamonds indicate no Scd arrival observed for that particular station–event combination, while squares indicate a very weak or questionable Scd arrival. All but one of the non-observations occur in areas where ray coverage is sparse and there are few grid intersections of scattering ellipsoids. Most of the low amplitude Scd observations occur in similar regions. Note that areas of sparse coverage with low hit counts may actually have a strong reflector, and the migration image must be evaluated by considering the hit counts as well as the amplitude weighed values. The somewhat weaker features

in the southern areas do correspond to low amplitude Scd arrivals with large hit counts. There are a few low amplitude observations and even one non-observation located in the central area where the number of grid intersections and relative amplitudes are high. This complexity may arise from interference effects. One further test of our method assesses whether the scattering images are a true product of coherence of the data or merely a coincidence of the source–receiver geometry. We use the observed differential times but shuffle the latitude and longitude of the stations as well as the depths of the events. Two migrations were performed for this test, one using the ScS–Scd times and the other using Scd–S times.

210


Fig. 8. Migration results for the observed Scd–S differential times, with ScdrS amplitude ratios used to weigh the data. Symbols indicate the number of weighed hits at each grid point, with the largest hit count representing 23 ellipsoid intersections. The most coherent structure occurs near depths of 2600–2620 km.

The results of these migrations reveal that while a few isolated grid points have moderate hit counts, the coherence of the scattering surface near 2600 km depth is much weaker when using the incorrect station locations and event depths. Thus, we infer that the data migrations do contain coherent structure.

5. Synthetic migrations The migrations of the Scd phase are clearly more compatible with a scattering surface than with a point scatter or small number of point scatterers. We discuss above how a scattering surface will produce extensive streaking artifacts in the images which are

results of the geometry. While we can qualitatively interpret the existence of a reflector within the depth range 2580 km to 2620 km, a more quantitative method of choosing the optimal depth is desired. In order to determine this depth in a more systematic fashion, we perform simulations for modified versions of the SGLE discontinuity model, allowing us to account for predictable streaking effects. These models, shown in Fig. 12, are identical to SGLE at shallow depths, but vary in both the depth and size of velocity increase in the depth range of 2400–2680 km. Seven models were created; the shallow discontinuity models have 100 km increments in reflector depth, as the migration images do not support the likelihood of a simple reflector between depths 2400 km–2580 km. Velocity discontinuities with depth


211

Fig. 9. Migration results for the observed ScS–Scd differential times, with ScdrScS amplitude ratios used to weigh each datum. Symbols indicate the number of weighed hits at each grid point, with the largest hit count representing 26 ellipsoid intersections. The images at 2600 km and 2620 km depths are the most coherent, but the results for depth 2600 km have higher hit counts.

Fig. 10. Migration results for observed ScS–Scd differential times at 2600 km depth showing hit counts on the left and amplitude weighed ŽScdrScS. hit counts on the right.

212 S.L. Bilek, T. Lay r Physics of the Earth and Planetary Interiors 108 (1998) 201–218 Fig. 11. Mercator projection of the study area showing hit counts from the unweighed ScS–Scd migration at depth 2600 km. Also plotted are the ScS turning points of rays that have weak Scd arrivals Žsquares. and no Scd observations Ždiamonds., as determined by Gaherty and Lay Ž1992.. Three regions of dense ray sampling are defined as well.


Fig. 12. Discontinuity models used for the synthetic data migrations. The models mimic SGLE except for size and depth of the shear wave velocity discontinuity. The numbers for each model indicate the depth of the discontinuity.

increments of 20 km from 2580 km to 2680 were constructed to fine tune the reflector model. Travel times were calculated for each of these models, using the actual source–receiver combinations of the data and assuming that the Scd phase is the forward triplication branch involving energy turning just below the discontinuity. While the reflected branch ŽSbc. could be used to more precisely image the boundary, it is likely that actual observations ŽScS–Scd and Scd–S differential times. involve a combination of Scd and Sbc arrivals. Synthetic migrations were performed for each model using the same procedure as in the data migrations. Synthetic amplitudes were not calculated and no relative amplitude weighing was made, thus the synthetic migration results are compared to unweighed hit counts for the actual data. The images for simulations with different discontinuity depths vary at every point in the grid volume. Correlation coefficients are calculated between the data and synthetic migration results for each trial reflector depth. For each model, the individual grid

213

depths were 2200 km to 2500 km in 100 km increments and 2580 km to 2700 km in 20 km increments. Correlations for each grid depth were computed and averaged to find an overall value for each model. To avoid bias toward the deeper models with finer depth increments in the grid we averaged only the correlations for grid layers down to the actual reflector depth. For example, to calculate the average correlation coefficient for the model with a discontinuity at 2640 km, the individual coefficients of the 2200–2500 km, 2600 km and 2640 km grid depths were summed and averaged. Fig. 13 shows the correlation coefficients as a function of model discontinuity depth for both the ScS–Scd and Scd–S cases. In both cases, model SGLE, with the velocity discontinuity at 2605 km depth, has the highest average correlation coefficients. The ScS–Scd case has a correlation of 0.79, and the Scd–S case has a correlation coefficient of 0.67 for model SGLE. This indicates that the complex images are actually well accounted for by a simple discontinuity model. The Scd–S case does not have a sharp peak like the ScS–Scd case; instead it has a broader peak of high correlation for models with discontinuity depths of 2580–2605. As noted above, this may reflect the large lateral variations in direct S times. In order to explore small scale lateral variations in the depth of the discontinuity, the lower mantle grid was dissected into three discrete areas defined by the most coherent patches with high hit counts ŽFig. 11.. The source–receiver geometries were then examined and only those that had a portion of their raypath traveling within the latitude and longitude boundaries of the respective areas were chosen for migration. Migrations were performed for the three different patches separately, using the real data as well as the synthetic datasets generated for the previous migrations. The results of the migrations were as expected; the high hit count areas were reduced from the full migration and generally confined to the area around the assigned box. Since both data and synthetic cases were considered, correlations were calculated for these results. Fig. 13 shows the average correlation coefficients as a function of discontinuity depth for each area for both the ScS–Scd and Scd–S measurements.

214


Fig. 13. Correlation coefficients between the data and synthetic migrations. Average correlation coefficients as a function of model discontinuity depth for both the ScS–Scd and Scd–S migrations averaged over the entire lower mantle grid.

For the ScS–Scd cases, the northern and central areas show peaks at 2605 km, although in the central patch, it is a broad peak extending to 2620 km. The southern area indicates a sharp peak at 2620 km. The Scd–S cases are similar, with the northern area having a sharp peak at 2605 km, a broad peak around 2620 km in the central area, and a sharper peak at 2620 km in the south. These results indicate minor topography on the boundary, but, of course, lateral variations in velocity structure could account for the apparent topography.

6. Scd2 arrival We now consider the second intermediate arrival, Scd2, which appears more intermittently than Scd. This phase is observed after the Scd arrival and before the ScS arrival for events located in the Japan and Kurile arcs. Gaherty and Lay Ž1992. modified model SGLE to incorporate two 2.3% velocity discontinuities, placed at depths of 2591 km and 2731 km in order to fit the second intermediate arrival observed in the data. This model can match the waveform features, but does not address the spatial extent of the structure. We use the parametric migration method to explore the spatial extent and depth of the structure causing the additional arrival.

For this set of migrations, we use SGLE as the reference velocity model, as there is always an Scd arrival when Scd2 is observed. There is nothing inherent in SGLE that would produce a second intermediate arrival. We treat the Scd2 arrival as a scattered arrival and determine the grid intersections of the ellipsoids with a lag time that matches the Scd2 arrival times. Similar to the other data migrations, we use both the ScS–Scd2 and Scd2–S differential times in separate migrations. Fig. 14 shows the migration results for ScS–Scd2, with the circles again representing the amplitude ratio weighed number of ellipsoid intersections at each grid point. These images are similar to unweighed hit count images. The ScS–Scd2 image is more coherent and stronger than the Scd2–S image Žnot shown., again the expected result given the proximity of the CMB to the scatterer image and the known lateral fluctuation of the S times. Thus we place greater weight on the results shown in Fig. 14, particularly the coherent localized feature near 2750 km. While the higher amplitude observations of Scd2 appear to be very localized, there are relatively widespread observations of the corresponding arrival. A second patch, slightly to the south of the main feature, has peak coherency at 2770 km. The relatively small size of the most coherent ScS–Scd2 patch in Fig. 14 is significant. This fea-


215

Fig. 14. Migration results using ScS as a reference phase for the Scd2 arrival. Scd2 amplitudes are weighed by ScS amplitudes. Symbols represent weighed hit counts at lower mantle grid points, with the largest indicating 12 intersections. This migration results in a strong yet localized scatterer at a depth of 2750 km.

ture is much smaller than the reflector surface imaged in the Scd migrations ŽFig. 9., indicating that the deeper scatterer is spatially concentrated. A few scenarios could account for such a scatterer. Weber Ž1993. suggests that strong lateral gradients in a single discontinuity depth can give rise to double arrivals at some stations. It is possible that this area is a depression of the velocity discontinuity boundary that is imaged elsewhere at 2605 km. In this case, the second arrival could be energy reflected from the sides or bottom of this valley. This is difficult to reconcile with the relatively strong image formed for the Scd arrival in the same area in Fig. 9. A more likely possibility for the origin of this arrival given the migration features is a localized scatterer beneath the 2605 km discontinuity, perhaps similar to the discrete heterogeneities proposed by Hadden

and Buchbinder Ž1987.. The small size of the scattering image resembles the single scatterer in Fig. 4, although spread over a larger region. These results are consistent with the findings of Gaherty and Lay Ž1992., who felt the heterogeneity must be limited spatially in order to produce the intermittent nature of the Scd2 arrival, but they did not locate it. The secondary Scd2 scatterer near 2770 km appears to have similar dimensions, but a weaker scattering coefficient.

7. Discussion A large number of studies indicate that the lowermost mantle is quite complex globally and not characterized by any single class of velocity structure

216


ŽLay and Helmberger, 1983; Weber and Kornig, ¨ 1992; Garnero et al., 1993; Nataf and Houard, 1993; Vidale and Benz, 1992, 1993; Kendall and Shearer, 1994; Kruger et al., 1995; Kendall and Nangini, ¨ 1996; Revenaugh and Jordan, 1991.. Several studies have concentrated on structure beneath Eurasia, finding evidence for a deep mantle discontinuity ŽLay and Helmberger, 1983; Weber and Davis, 1990; Houard and Nataf, 1992, 1993; Weber, 1993; Thomas and Weber, 1997; Scherbaum et al., 1997.. The migration results here support the previous work and suggest that the lower mantle complexity takes the form of mildly varying depth of the shear velocity discontinuity throughout the area as well as the presence of small discrete scatterers within the DY layer. Array based methods that measure slownesses and azimuths provide alternate tools for high resolution examination of the lower mantle. A number of studies have demonstrated the benefits of using arrays in many different areas of the world, such as detecting signals not observed with conventional networks ŽKohler et al., 1997. and developing new lower mantle imaging methods ŽKruger ¨ et al., 1993, 1995.. Weber and Davis Ž1990. use array data to find evidence for a reflector located 290 km above the CMB beneath Eurasia with velocity increases of 3% for P waves and 2% for S waves, while Weber Ž1993. locates a P wave reflector at 2612 km under the Nansen basin and 2605 Ž"10–20. km beneath northern Siberia with similar velocity increases as Weber and Davis Ž1990.. While our dataset provides only sparse coverage in the Nansen basin area, the compatibility between our results and the previous work in the Kara Sea and northern Siberia is encouraging, with the discontinuity depth found using array data agreeing with ours to within the error estimates. The possibility of topography on the reflector is examined by Thomas and Weber Ž1997., who compare PdP observations and non–observations with synthetics for 2-D models to suggest a DY discontinuity beneath northern Siberia 400–1200 km in lateral extent with 10–100 km of topography. In this region, we do not observe significant topography changes of this magnitude, but the large Fresnel zones of our data could conceivably average over rough topography. Scherbaum et al. Ž1997. apply the double beam stacking method to source and receiver

arrays to examine the lower mantle beneath the Arctic and northern Siberia, finding a P wave discontinuity 293 km above the CMB in northern Siberia. Most array studies use P wave data as, unfortunately, array data for S waves are quite limited. The parametric migration technique ŽLay and Young, 1996. was developed to use sparse shear wave datasets to examine lower mantle structure. This technique avoids the restrictive constraints needed for full inversions, such as assumption of the scatterer geometry Žas used in Ying and Nataf, 1998., and simplifies the three dimensional imaging. As tomography and waveform modeling studies better constrain deep mantle structure, it may become possible to formulate useful diffraction tomography for lowermost mantle structure. It is still unclear what causes the velocity discontinuity and how DY relates to dynamics of the lower mantle region ŽLay, 1995; Wysession et al., 1998.. Possibilities for the origin of the discontinuity include a decrease in temperatures caused by subducted slabs ponding at the base of the mantle, mineralogical phase changes at the high CMB P–T conditions, perovskite reacting with the liquid iron of the outer core to generate reaction products ŽKnittle and Jeanloz, 1989, 1991., or residual material from core formation. Most likely, it is a combination of factors in the boundary layer above the core that creates the velocity heterogeneity. The results of this contribute to the evidence for both large and small scale heterogeneities in the boundary layer, possibly of thermal or chemical origin. However, resolving the precise nature of these structures is a major challenge intimately linked to issues such as mantle plume formation, the fate of subducting slabs, and lower mantle mineralogy and thermo-chemical processes.

8. Conclusions The migration method used here provides a systematic approach to mapping lateral variations in the lower mantle. Lay and Young Ž1996. first used the technique to examine heterogeneities beneath Alaska, favoring a lower mantle reflector. Similar conclusions are drawn here regarding a lower mantle discontinuity beneath Eurasia. This study further sug-


gests the presence of internal structure in the DY layer below the main shear velocity discontinuity. This study advanced the methodology in Lay and Young Ž1996. by incorporating correlation with simulations to account for smearing artifacts in the images produced by non-uniform ray coverage, which enhances the resolution for reflectors with mild depth undulations. The optimal correlations found are with simulations for models with discontinuities at depths near 2600 " 20 km. Separating the well sampled regions of the lower mantle into three distinct groups suggests mild variations in the depth of the discontinuity, from 2605 km in the northern and central areas, deepening to 2620 km in the south. The observation of a second intermediate arrival located between the Scd and ScS phases in 25% of the data indicates the presence of small, localized scatterers at depths of 2750 and 2770 km beneath Central Eurasia. While neither the main discontinuity nor the localized scatterers within DY are yet understood in terms of precise boundary layer processes, continued efforts to resolve the strength and scale lengths of DY structures are paramount to making progress in quantifying dynamics in DY . Acknowledgements This research was supported by NSF Grant EAR9418643. Contribution No. 342 of the Institute of Tectonics. We made extensive use of the GMT software provided by Wessel and Smith Ž1991.. We wish to thank S. Schwartz, R. Hartog, M. Hagerty, and S. Russell for helpful comments on the manuscript, as well as H.-C. Nataf and an anonymous reviewer for careful reviews. References Dziewonski, A.M., Anderson, D.L., 1981. Preliminary reference Earth model. Phys. Earth Planet. Inter. 25, 297–356. Gaherty, J.B., Lay, T., 1992. Investigation of laterally heterogeY neous shear velocity structure in D beneath Eurasia. J. Geophys. Res. 97, 417–435. Garnero, E.J., Helmberger, D.V., 1993. Travel times of S and SKS: implications for three-dimensional lower mantle structure beneath the central Pacific. J. Geophys. Res. 98, 8225– 8241.

217

Garnero, E.J., Helmberger, D.V., 1996. Seismic detection of a thin laterally varying boundary layer at the base of the mantle beneath the central Pacific. Geophys. Res. Lett. 23, 977–980. Garnero, E.J., Helmberger, D.V., Grand, S., 1993. Preliminary evidence for a lower mantle shear wave velocity discontinuity beneath the central Pacific. Phys. Earth Planet. Inter. 79, 335–347. Grand, S.P., 1994. Mantle shear structure beneath the Americas and surrounding oceans. J. Geophys. Res. 99, 11591–11622. Grand, S.P., van der Hilst, R.D., Widiyantoro, S., 1997. Global seismic tomography: a snapshot of convection in the Earth. GSA Today 7, 1–7. Hadden, R.A.W., Buchbinder, G.G.R., 1987. S wave scattering by 3-D heterogeneities at the base of the mantle. Geophys. Res. Lett. 14, 891–894. Houard, S., Nataf, H.-C., 1992. Further evidence for the Lay discontinuity beneath Northern Siberia and the North Atlantic from short period P waves recorded in France. Phys. Earth Planet. Inter. 72, 264–275. Houard, S., Nataf, H.-C., 1993. Laterally varying reflector at the Y top of D beneath northern Siberia. Geophys. J. Int. 115, 168–182. Y Kendall, J.-M., Nangini, C., 1996. Lateral variations in D below the Caribbean. Geophys. Res. Lett. 23, 399–402. Y Kendall, J.-M., Shearer, P.M., 1994. Lateral variations in D thickness from long period shear wave data. J. Geophys. Res. 99, 11575–11590. Knittle, E., Jeanloz, R., 1989. Simulating the core–mantle boundary: an experimental study of high-pressure reactions between silicates and liquid iron. Geophys. Res. Lett. 16, 609–612. Knittle, E., Jeanloz, R., 1991. Earth’s core–mantle boundary: results of experiments at high pressures and temperatures. Science 251, 1438–1443. Kohler, M.D., Vidale, J.E., Davis, P.M., 1997. Complex scattering Y within D observed on the very dense Los Angeles Region Seismic Experiment passive array. Geophys. Res. Lett. 24, 1855–1859. Kruger, F., Weber, M., Scherbaum, F., Schlittenhardt, J., 1993. ¨ Double beam analysis of anomalies in the core–mantle boundary region. Geophys. Res. Lett. 20, 1475–1478. Kruger, F., Weber, M., Scherbaum, F., Schlittenhardt, J., 1995. ¨ Evidence for normal and inhomogeneous lowermost mantle and core–mantle boundary structure under the Arctic and northern Canada. Geophys. J. Int. 122, 637–657. Lay, T., 1986. Evidence of a lower mantle shear wave discontinuity in S and sS phases. Geophys. Res. Lett. 13, 1493–1496. Lay, T., 1995. Seismology of the lower mantle and core–mantle boundary. Rev. Geophys. 33, 325–328. Lay, T., Helmberger, D.V., 1983. A lower mantle S wave triplicaY tion and the shear velocity structure of D . Geophys. J. R. Astron. Soc. 75, 799–838. Lay, T., Young, C.J., 1986. The effect of SKS scattering on Y models of the shear velocity structure of the D region. J. Geophys. 59, 11–15. Lay, T., Young, C.J., 1996. Imaging scattering structures in the lower mantle by migration of long period shear waves. J. Geophys. Res. 101, 20023–20040.

218


Lay, T., Garnero, E.J., Young, C.J., Gaherty, J.B., 1997. Scale lengths of shear velocity heterogeneity at the base of the mantle from S wave differential travel times. J. Geophys. Res. 102, 9887–9909. Li, X.D., Romanowicz, B., 1996. Global mantle shear velocity model developed using nonlinear asymptotic coupling theory. J. Geophys. Res. 101, 22245–22272. Loper, D.E., Lay, T., 1995. The core–mantle boundary region. J. Geophys. Res. 100, 6397–6420. Lynnes, C.S., Lay, T., 1989. Inversion of P coda for isotropic scatterers at the Yucca Flat test site. Bull. Seismol. Soc. Am. 79, 790–804. Masters, T.J., Johnson, S., Laske, G., Bolton, H., 1996. A shear velocity model of the mantle. R. Soc. London Philos. Trans. 354, 1385–1411. Nataf, H.-C., Houard, S., 1993. Seismic discontinuity at the top of Y D : a worldwide feature?. Geophys. Res. Lett. 20, 2371–2374. Revenaugh, J., 1995. A scattered-wave image of subduction beneath the Transverse Ranges. Science 268, 1888–1892. Revenaugh, J., Jordan, T.H., 1991. Mantle layering from ScS reverberations: 4. The lower mantle and core–mantle boundary region. J. Geophys. Res. 96, 19811–19824. Scherbaum, F., Kruger, F., Weber, M., 1997. Double beam imag¨ ing: mapping lower mantle heterogeneities using combinations of source and receiver arrays. J. Geophys. Res. 102, 507–522. Su, W.-J., Woodward, R., Dziewonski, A.M., 1994. Degree 12 model of shear velocity heterogeneity in the mantle. J. Geophys. Res. 99, 6945–6980. Thomas, Ch., Weber, M., 1997. P velocity heterogeneities in the lower mantle determined with the German Regional Seismic Network: improvement of previous models and results of 2D modelling. Phys. Earth Planet. Inter. 101, 105–117.

van der Hilst, R.D., Widiyantoro, S., Engdahl, E.R., 1997. Evidence for deep mantle circulation from global tomography. Nature 286, 578–584. Vidale, J.E., Benz, H.M., 1992. A sharp and flat section of the core–mantle boundary. Nature 359, 627–629. Vidale, J.E., Benz, H.M., 1993. Seismological mapping of the fine structure near the base of the Earth’s mantle. Nature 361, 529–532. Weber, M., 1993. P- and S- wave reflections from anomalies in the lowermost mantle. Geophys. J. Int. 115, 183–210. Weber, M., Davis, J.P., 1990. Evidence of a laterally variable lower mantle structure from P and S waves. Geophys. J. Int. 102, 231–255. Weber, M., Kornig, M., 1992. A search for anomalies in the ¨ lowermost mantle using seismic bulletins. Phys. Earth Planet. Inter. 73, 1–28. Wessel, P., Smith, W.H.F., 1991. Free software helps map and display data. EOS Trans. AGU 72 Ž441., 445–446. Wysession, M.E., Bartko, L., Wilson, J.B., 1994. Mapping the lowermost mantle using core-reflected shear waves. J. Geophys. Res. 99, 13667–13684. Wysession, M.E., Lay, T., Revenaugh, J., Williams, Q., Garnero, Y E.J., Jeanloz, R., Kellogg, L.H, 1998. The D discontinuity and its implications. In: Gurnis, M., Buffet, B., Knittle, E., Wysession, M.E. ŽEds.., AGU Monograph on the Core–Mantle Boundary, in press. Ying, J., Nataf, H.-C., 1998. Detection of mantle plumes in the lower mantle by diffraction tomography: Hawaii. Earth Planet. Sci. Lett., in press. Young, C.J., Lay, T., 1987. The core–mantle boundary. Annu. Rev. Earth Planet. Sci. 15, 25–46.