Trans-Pacific upper mantle shear velocity structure

Click Here

JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 112, B08301, doi:10.1029/2006JB004853, 2007

for

Full Article

Trans-Pacific upper mantle shear velocity structure Ying Tan1 and Don V. Helmberger1 Received 10 November 2006; revised 21 March 2007; accepted 25 April 2007; published 1 August 2007.

[1] We use 50 Tonga-Fiji events recorded at the broadband TriNet array, southern

California, to develop a pure path upper mantle shear velocity model across the Pacific. At the epicentral distances of 70°–95°, multibounce S waves up to S5 are observed, including their triplicated branches, which become particularly clear after stacking. Since these S wave multiples turn at various depths, simultaneously modeling their differential traveltimes and waveforms provides strong constraints on the radial velocity structure. We parameterize the velocity model according to a priori information from the previous oceanic models, so that we can take a grid search approach, to fully investigate possible interdependencies among the model parameters. We construct synthetics with a reflectivity code and study both the SH and SV components. By modeling the whole recordings from events at different depths, with different mechanisms, we are able to separate shallow low-velocity zone (LVZ) features from deeper structure. Our preferred model (PAC06) contains a fast lid (Vsh = 4.78 km s 1, Vsv = 4.58 km s 1) with a thickness of 60 km. The underlying LVZ is prominent with the lowest velocities Vsh = 4.34 km s 1, and Vsv = 4.22 km s 1 occurring at a depth of 160 km. These velocities are below the estimates of solid-state LVZ, suggesting the presence of partial melt. The anisotropy (Vsv < Vsh) of PAC06 extends to 300 km depth, which is constrained by S5 turning at this depth. Besides the 406 km and 651 km discontinuities, PAC06 also has a small (1%) velocity jump at 516 km. We consider these main features of PAC06 to be well determined, since PAC06 explains a large data set from various events. Therefore it is ideally suited for comparing with mineralogical models. Citation: Tan, Y., and D. V. Helmberger (2007), Trans-Pacific upper mantle shear velocity structure, J. Geophys. Res., 112, B08301, doi:10.1029/2006JB004853.

1. Introduction [2] Seismic velocity structure provides the most important constraint on modeling the mineralogical composition and dynamics of the Earth’s interior. Of particular interest is the oceanic upper mantle structure, due to its large departure from a global average (e.g., PREM by Dziewonski and Anderson [1981]). The origin of the characteristic highvelocity lid, prominent low-velocity zone (LVZ) and the general lack of 220 km discontinuity has warranted a large number of investigations from the mineralogical point of view. The recent work by Stixrude and Lithgow-Bertelloni [2005] suggested a solid-state LVZ, which is the natural consequence of a thermal boundary layer and the effects of pressure and temperature on the elastic wave velocity of subsolidus mantle assemblages. The properties atypical of the mantle, such as partial melt and bound water, would be permissible, but not required. Their result agrees with Faul and Jackson [2005]. However, the quantitative comparison of their ‘‘null’’ hypothesis (elasticity, isotropy, absence of partial melt, homogeneity and thermodynamic 1 Seismological Laboratory, California Institute of Technology, Pasadena, California, USA.

Copyright 2007 by the American Geophysical Union. 0148-0227/07/2006JB004853$09.00

equilibrium) with seismological models revealed discrepancies in the negative velocity gradient, the absolute velocity value in the LVZ, and the high-velocity gradient below the LVZ. To reconcile these discrepancies forced them to examine variants of the ‘‘null’’ hypothesis and to include the effects of partial melt, attenuation, and anisotropy. [3] Such mineralogical studies have been matched in seismology. In particular, the last decade has seen the remarkable progress of seismic tomography, mostly driven by the advancement of theory and computational power, where global three-dimensional (3-D) models have spanned the entire depth range of the Earth’s mantle and achieved lateral resolution of less than 1000 km [Romanowicz, 2003]. However, many oceanic areas remain unsampled or poorly resolved in most high-resolution P velocity models. The other type of long-wavelength S velocity models based on surface waves generally have good coverage across the oceans, but most of them lack radial resolution. Hence the current generation of global models is not capable of resolving the hypotheses regarding the layering of the oceanic upper mantle. Instead, ‘‘pure path’’ regional models can provide significant information complementary to the tomographic models [Gaherty et al., 1999]. [4] Although there is increasing evidence for short length scale variations in lithospheric thickness across plate boundaries where subduction has occurred or is occurring [e.g.,

B08301

1 of 20

B08301

TAN AND HELMBERGER: PACIFIC UPPER MANTLE SHEAR STRUCTURE

B08301

Figure 1. Source-receiver geometry on a seafloor age map. The source mechanisms are from Harvard CMT solutions and color coded in depth. Blow up is the broadband TriNet array in southern California. The dashed line from Tonga-Fiji to the Hawaiian stations, HON and KIP, indicates the corridor studied by Gaherty et al. [1996]. Melbourne and Helmberger, 2001], structures away from plate boundaries display much smoother lateral variations of smaller scale, where simple corridors appear one-dimensional. Here we will investigate one such strip that runs nearly across the central Pacific Ocean connecting the Tonga-Fiji source region to TriNet, southern California (Figure 1). Cross sections from global tomographic models display laterally homogeneous features along this corridor [e.g., Ritsema et al. 1999], where the largest anomalies are associated with adjustments made to eliminate the 220 km discontinuity in the reference model PREM [Dziewonski and Anderson, 1981]. The plate history along the corridor is relatively simple, with most of the seafloor developed by the spreading of the ancient Pacific-Farallon ridge. The lithospheric age spans a wide range from approximately 125 Ma near the source region to 10 Ma approaching the California coastal line, where the lithosphere is about 50 km thick [Melbourne and Helmberger, 2001]. The pure surface wave studies [e.g., Nishimura and Forsyth, 1989; Ritzwoller et al., 2004; Maggi et al., 2006] indicated progressive increase of lithosphere velocity or thickness with increasing plate age, qualitatively mimicking isotherms of theoretical thermal cooling models. However, these results solely from surface wave data can be debated because of their poor vertical resolution. In contrast, Gaherty et al. [1999] suggested very little age dependence of the lithosphere thickness when they compared regional models of the old Pacific Ocean (100 – 125 Ma) and the younger Philippine Sea (15 – 50 Ma), where the transitions from the high-velocity lid to the

underlying LVZ (the G discontinuity) were mainly determined from ScS reverberation observations [Revenaugh and Jordan, 1991b]. They interpreted the G discontinuity mainly as a compositional boundary rather than a thermal boundary. [5] The whole corridor has been included in an early 2-D study from Tonga to Newfoundland [Graves and Helmberger, 1988], where multibounce S waves from a single event were modeled with a 2-D WKB code and the emphasis was given to the sharp transition across the Pacific – North America boundary. More recently, Gaherty et al. [1996] investigated the older half of the corridor from Tonga-Fiji to Hawaii (see Figure 1) with a combined data set of ScS reverberations and measured frequencydependent (10 – 45 mHz) phase delays for body and surface waves recorded on the island stations KIP and HON. Here we use a novel data set of broadband seismograms from the dense TriNet array in southern California. At the epicentral distances of 70°– 95°, multibounce S waves up to S5 sampling over the upper mantle triplication range are observed, including the guided waves from shallow events transversing the LVZ. Simultaneously modeling the moveouts of all these S wave multiples that have their turning points at different depths provides strong constraints on the radial velocity structure. [6] Because velocity anomalies associated with a single leg of a multibounce phase will be mapped into the whole path, we interpret our model PAC06 as an effective average along the corridor as discussed by Zhao and Helmberger [1993]. However, it explains record sections from a large

2 of 20


B08301

B08301

Table 1. Source Parameters of the Tonga-Fiji Events Used in This Study From Harvard’s CMT Catalog Date Event ID

Year/Month/Day

Origin Time, UT

9026350 9530269 9533473 9563017 9564185 9685024 9611653 9648517 9658105 9658317 9743493 9792597 13980540 9927909 9942373 14106728 9982833 10100677 14132656 14170576

1997/10/14 2000/01/26 2000/02/25 2000/09/11 2000/09/14 2000/01/08 2001/01/09 2001/04/28 2001/05/26 2001/06/03 2002/01/03 2002/06/17 2003/07/27 2003/07/03 2003/09/02 2004/11/17 2004/02/12 2005/05/18 2005/03/19 2005/08/06

0953:32.7 1327:1.0 0144:5.2 1818:0.3 1500:2.3 1647:30.2 1649:37.9 0450:1.9 1057:31.0 0242:9.7 1017:49.6 2126:33.9 0204:15.6 0621:57.3 1828:10.9 2109:19.5 1347:35.9 1027:15.0 1734:45 0956:19.3

Location Latitude/Longitude 21.94/ 176.15 17.21/ 173.51 19.55/174.17 15.74/ 173.33 15.59/ 179.92 16.84/ 173.81 14.90/167.11 18.07/ 176.68 20.25/ 177.65 29.37/ 178.23 17.84/167.82 12.49/166.25 21.09/ 176.12 21.46/ 173.80 15.14/ 172.92 19.87/ 178.40 19.41/ 172.80 15.22/ 172.69 21.92/ 179.28 19.57/ 175.35

FM

Depth, km

Mw

Strike/Dip/Rake

166 70 17 125 15 162 115 367 414 200 20 44 216 16 15 629 12 12 610 218

7.7 6.2 7.1 6.3 6.2 7.2 7.0 6.8 6.3 7.1 6.6 6.6 6.6 5.9 6.4 6.5 5.7 6.2 6.3 6.6

257/17/ 30 213/16/ 66 315/74/169 331/10/ 129 154/77/176 79/8/ 13 183/44/160 106/16/161 134/68/ 176 227/41/138 359/30/97 163/38/81 294/22/ 3 192/24/83 210/28/171 77/8/117 213/26/96 144/33/48 97/16/160 266/33/ 24

D Correction, km 50 – – – – – – – – 40 – – 30 70 – – 60 – – –

D correction applies to the events that appear mislocated as discussed in section 3.3. The correction is simply an average distance shift for all the stations from a particular event.

number of events at different depths, spanning a wide distance range of 70°– 95°. Moreover, the good agreement between PAC06 and PA5 developed for the older segment [Gaherty et al., 1996], as well as the excellent fits of PAC06 to the waveform data from the Hawaiian earthquake sampling the younger segment, indicates a relatively simple, uniform structure along the corridor, which proves particularly useful for mineralogical investigations.

2. Data [7] Since early 1999, when the southern California TriNet array was expanded into a dense broadband network with over 120 stations, hundreds of large events (M > 5.5) that occurred in the Tonga-Fiji seismic zone have been recorded. Among them, we selected 50 events (Figure 1), whose seismic sources are relatively simple, and whose multibounce phases are rich. We assumed the event locations and origin times from Harvard centroid moment tensor (CMT) solutions (see Table 1 for the events discussed here)

[Dziewonski et al., 1981], and the source depths were adjusted based on the relative timing between direct and depth phases, if necessary. The exact event locations and origin times are not critical, since we model differential traveltimes between S multiples, and the effects of event location or origin time become second order [Grand and Helmberger, 1984b]. [8] Because of the small aperture of TriNet, a single event provides limited lateral coverage. However, this problem is largely compensated by studying a number of events spanning a wide distance range of 70° – 95° (Figure 2). Moreover, modeling multiple events of different mechanisms, at different depths, provides an effective way to eliminate contaminations of source effects on the velocity structure. Particularly within the studied distance range, both SSS and S4 become triplicated by the 410 km and 660 km discontinuities. A sketch of their triplication patterns is displayed in Figure 3. Note that different portions of the triplication patterns can be observed for SSS and S4 on the same record section.

Figure 2. Schematic raypaths of multibounce S waves at epicentral distances of 70° – 95° used in this study. Particularly for SSS and SSSS that are triplicated in this distance range, only the early arriving branches are shown. 3 of 20

B08301


Figure 3. Triplication plot of SSS and S4 from PAC06sh with a 10 km deep source. The small triplication patterns on the ‘‘cd’’ branches are due to the 516 km discontinuity in PAC06. The dashed lines were added to indicate the positions of diffractions. [9] We rotated the original horizontal component (N-S and E-W) data into radial (SV) and tangential (SH) motions, although we generally found the SV waves on the vertical component to be the most useful for modeling purposes. An example of the tangential records from a shallow event (9533473) is displayed as a distance record section in Figure 4, where the various S wave multiples are labeled. The dense array produces coherent seismograms, which warrants a simple stacking procedure and facilitates identification of triplications or phase interference. At this distance range (82° – 87°), SSS is triplicated due to the 660 km discontinuity, resulting in a forward branch (S3ef) turning below the discontinuity and a shallower backward branch (S3cd). S4 contains more complexity, in that the triplications are interfering with the shallow guided waves (Love waves). In fact, the apparent interference displayed in Figure 4 is mainly due to the guided waves that are most sensitive to shallow depths, since the amplitude of S4 is much smaller due to the radiation pattern effect. By examining various events of different depths and mechanisms, it becomes possible to separate Love waves from S4 triplications. However, modeling such guided waves at their natural frequencies (>10 s) provides powerful constraints on the shallow part of the velocity model. The whole group becomes simpler and the so-called ‘‘G phase’’ when filtered to longer periods (>30 s), which is normally used in tomographic studies.

3. Waveform modeling [10] Our modeling effort involves matching the differential traveltimes between multibounce S waves and simulating their waveforms, particularly the waveforms of the higher-order multiples, S3 and S4, plus shallow guided

B08301

waves, which are triplicated or interfering in the upper mantle. Since the different multiples have their own turning depths, simultaneously modeling their traveltimes provides unique estimates of velocity at various depths. Moreover, finer-scale structures can be inferred by simulating their subtle triplication or interference patterns. [11] To model these recordings, we take a ‘‘trial and error’’ approach which has proven effective in previous similar studies [e.g., Grand and Helmberger, 1984b; Melbourne and Helmberger, 2002]. In addition, we parameterize the velocity model as a composite of a high-velocity lid, a negative gradient zone, a positive gradient zone and the transition zone, according to a priori information from previous oceanic models [e.g., Grand and Helmberger, 1984a; Gaherty et al., 1996]. This simple parameterization enables us to efficiently investigate possible interdependencies among the thicknesses, average velocities and gradients of the layers in a grid search manner, and hence to resolve the main characteristics of the model. [12] We construct three-component full synthetics with an isotropic reflectivity code [Zhu and Rivera, 2002], while we use generalized ray theory [Helmberger, 1983] to help identify individual arrivals in cases where ambiguity exists. We model the SH (tangential) and SV (vertical) components respectively, since the higher-order S wave multiples, such as S3, S4 and S5 are significantly split over the two components. The isotropic nature of the synthetics restricts our ability to speculate the geometry of anisotropy, although the resolved Vsh > Vsv is consistent with transverse isotropy reported by previous investigators [e.g., Gaherty et al., 1996; Ekstro¨m and Dziewonski, 1998]. However, the unified methodology for both the SH and SV components enables us to address the depth extent of anisotropy and its relationship to the upper mantle layering. [13] We started with the attenuation model developed by Ding and Grand [1993] for the East Pacific Rise (EPR) using multibounce S waves. This model has proved effective in a later study by Melbourne and Helmberger [2002] for EPR events recorded at TriNet. However, we found that nearly doubled Q values in the upper 200 km of the model are required to explain the progressive amplitude decay of the S multiples (see Table 2). This adjustment is reasonable considering that Ding and Grand’s [1993] Q model was derived from one of the hottest structures on Earth and is probably near the lower bound. [14] We display the waveform comparison between the data shown in Figure 4 and the synthetics from our best SH model PAC06sh assuming these Q values in Figure 5. We align both the data and synthetics on SS to minimize the source region complexity, since SS has a take off angle close to those of the higher-order multiples, while the direct S wave leaves the source at a much steeper angle (see Figure 2). The whole record section is nicely reproduced with the simple 1-D model (Figure 5), particularly the waveforms of the guided waves overriding the S4 triplications. There is a slight tendency for the coastal stations to be matched better than the inland ones, such as hec, gla, dan, mtp and nee. Moreover, the time lags of the observed ‘‘G + S4’’ waves relative to the synthetics are systematically increasing with station distance (Figure 5c). This ‘‘unmodeled’’ feature is consistent with the lithospheric thickness change across the Pacific – North America boundary reported by Melbourne

4 of 20

B08301


B08301

Figure 4. (left) Velocity record section of the tangential component from a shallow event (9533473). The original records are desampled down to 1 Hz for display purpose. S wave multiples are labeled. (right) Selected records after a simple running stack process, where the record at each station has been replaced by the average of the nearest records within 10 km. The number following a station name indicates how many records have been stacked for the particular station. and Helmberger [2001]. Similar sharp transitions are also seen in recent tomographic models [e.g., Tanimoto and Prindle Sheldrake 2002]. The major discrepancy of the comparison in Figure 5 is the time lags of the synthetic S phases (5 s) relative to the observations. We attribute this advancement of the observed S waves to the fast slabs in the source region, mainly because the same problem does not exist for the deeper events. [15] In contrast, we display in Figures 6 and 7 the waveform fits for the same event 9533473 shown in Figure 5 with the preliminary Earth reference model (PREMsh) [Dziewonski and Anderson, 1981] and the path average of a typical tomographic model S20RTS by Ritsema et al. [1999]. Because S20RTS was developed mainly with Rayleigh wave dispersion data, we extrapolated the SH counterpart of the path averaged S20RTS assuming the same anisotropy as in the reference model PREM. The comparison between PREM, the path average of S20RTS and our preferred model PAC06sh is given in Figure 8.

Table 2. Model PAC06 Depth, km

5 of 20

0.2 0.2 6 6 66 66 163 303 406 406 516 516 651 651 736 900

Vsh, km s 1.00 3.20 3.20 4.78 4.78 4.51 4.34 4.64 4.86 5.02 5.22 5.28 5.54 6.01 6.22 6.33

1

Vsv, km s 1.00 3.20 3.20 4.58 4.58 4.31 4.22 4.64 4.86 5.02 5.22 5.28 5.54 6.01 6.22 6.33

1

Qs 20 20 20 160 160 100 100 120 160 180 180 180 180 310 310 310

B08301


B08301

Figure 5. (a) Waveform comparison between the SH data (black) from event 9533473 and the synthetics with PAC06sh (red). Time axis is relative to the event origin time. (b) Source parameters of the event from Harvard’s CMT catalog. (c) Time delays (gray circles) of the observed S multiples relative to the synthetics (tobs tsyn) measured by waveform cross correlation. Black circles are the residuals after the synthetics and data are aligned on SS. Moreover, to correct the crustal effect in PREMsh and S20RTSsh, we have replaced their thick crust (21 km) with the much thinner one (6 km) in our model PAC06sh. As a consequence, the underlying mantle layers in the two models have been thickened to conserve the overall travel time. The synthetics from PREMsh and

S20RTSsh with their original thick crust are compared with data in the auxiliary material (Figures S29 – S30).1 As displayed in Figure 6, PREM failed to predict both the 1 Auxiliary materials are available in the HTML. doi:10.1029/ 2006JB004853.

6 of 20

B08301


B08301

Figure 6. Selected waveform fits for event 9533473 as shown in Figure 5 with the preliminary Earth reference model (PREMsh). Both the data (black) and synthetics (red) are aligned on SS. interference pattern of the guided waves and their traveltimes. The tomographic model (S20RTS) containing a welldeveloped LVZ produces the two interfering branches; however, their traveltimes plus their separations are apparently off. These comparisons suggest strong dependence of the guided waves on the shallow part of the model, which we will explore in section 3.1. [16] Although there is some redundancy, the selected 50 events contain extensive sampling of SSS and S4 triplications (Figure 3), plus their interference with the guided waves from shallow events. Therefore they provide the most stringent constraint for different depth ranges of the model. For example, as the guided waves traversing the LVZ provides a means of estimating the shallow part of the model, the triplications of SSS or S4 mainly constrain the transition zone. In the following, we will give typical examples of the events that we use to refine the velocity structure at different depths, where we will also address

representative sensitivity tests in an attempt to quantify the model resolution. Our discussion will be restricted to the SH (tangential) component alone, since the adequate SV observations are rather limited. 3.1. Lid and LVZ [17] The velocity in the shallow part of the model, such as the lid or the low-velocity zone (LVZ) has the strongest effect on the differential traveltimes of the multibounce S waves, since the higher-order multiples systematically spend more time traveling at these depths. Moreover, the shallow radial structure largely controls the waveforms of the guided waves traversing the LVZ. However, its influence becomes significantly less on the triplication patterns of SSS or S4. In particular, it can hardly change the time separations between the triplicated branches, except for a small overall shift. This allows us to separate the effect of the shallow part from that of the underlying

7 of 20

B08301


B08301

Figure 7. Selected waveform fits for event 9533473 as shown in Figure 5 with the path average of a typical tomographic model S20RTS by Ritsema et al. [1999]. transition zone when modeling the records. The transition zone, however, mainly controls the triplications. [18] The shallow events (depth < 50 km) are the most suitable for determining the lid and LVZ structure, since they generate the strongest guided waves. Moreover, their seismograms are relatively simple without obvious depth phases. Figure 9 displays a composite record section of four shallow events together with overlain synthetics from our preferred SH model PAC06sh. The four events have different source mechanisms, however, their records are well explained with a single model, suggesting the source effects on the velocity structure are largely eliminated. The radial structure of the lid and LVZ shapes the waveforms of the guided waves. Such influences, however, are rather difficult to quantify or predict, since they generally involve both timing and phase changes in the complex waveform pockets. Therefore we have parameterized the model as a composite of three simple layers (the lid, negative gradient

zone and positive gradient zone) and conducted extensive ‘‘grid searches’’ for the basic parameters, such as thickness, average velocity and gradient of each layer. Our preferred model PAC06sh produces the best waveform fits to the observations. The basic characteristics of PAC06sh include a 60 km thick high velocity lid (Vsh = 4.78 km), a prominent LVZ with the lowest velocity of 4.34 km s 1 at a depth of 160 km, and a high gradient (0.002 s 1) zone underneath. [19] There are some trade-offs among certain model parameters, of which those between the lid and LVZ structure appear the most severe. Figure 10 displays the sensitivity of the complex ‘‘G + S4’’ waveforms as shown in Figure 9 to the lid and LVZ structure, as well as their tradeoffs. We have discarded the traces from the two closer events in Figure 10 for display purpose, while the results for the whole record section are given in the auxiliary material (Figures S1 – S4). In particular, Figure 10a displays the sensitivity of the ‘‘G + S4’’ waveforms to the lid thickness.

8 of 20

B08301


B08301

cannot be determined since we are dealing with surface waves. We present the sensitivity tests for the high gradient zone (HGZ) beneath the LVZ in the auxiliary material (Figure S5), which suggest very little trade-off between the velocities of the HGZ and the shallower part.

Figure 8. Comparison of our preferred SH model PAC06sh with PREM (Vsh > Vsv) and the path average of the tomographic model S20RTS by Ritsema et al. [1999]. Since S20RTS was developed mainly with Rayleigh wave data, we have extrapolated its SH counterpart assuming exactly the same anisotropy as in PREM.

The lid thickness change in the perturbed models has been offset by a velocity adjustment, so that the traveltimes for deeper phases, such as S, SS, and SSS waves are conserved. However, the interference pattern of the two surface wave branches are affected significantly, due to their differential sensitivity to different depths. Particularly, the backward branch, that is the most sensitive to the very top of the lid, gets accelerated by the faster velocity there or delayed by the slower velocity. The similar trend is observed as we fix the lid thickness, but vary the size of the G discontinuity between the lid and LVZ (Figure 10b). Figure 10c displays possible trade-offs between the lid and LVZ structure by comparing the waveform fits from PAC06sh to the perturbed models, where the lid thickness trades off with the G discontinuity. The models shown, model 1 with a 80 km thick lid, and model 2 with a 40 km thick lid, are considered two ‘‘end-members’’ outlining the uncertainty of the lid and LVZ structure of PAC06sh, beyond which waveform misfit increases significantly. These results from the grid search indicate the robustness of the shadow structure in PAC06sh, although the detailed sharpness of the G discontinuity

3.2. Transition Zone [20] The transition zone comprising the two major mantle discontinuities plays the most significant role in shaping the triplication patterns, since the triplicated branches travel nearly the same paths except in the vicinity of the discontinuities. Therefore modeling the separation, the cross over and the relative amplitudes of the triplicated branches provides tight constraints on the transition zone [e.g., Grand and Helmberger, 1984b; Melbourne and Helmberger, 1998; Wang et al., 2006]. In this study, we observe the triplications of the higher-order multiples SSS and S4 (Figure 3). Among them, those of S4 are mainly from intermediate and deep events, where the shallow guided waves are well suppressed. We started with a trial structure for the transition zone from the tectonic North America model (TNA) [Grand and Helmberger, 1984b] (Figure 11), and used a large number of the observed triplications to examine the details. We let the TNA model stand unless modifications were required to better explain our observations, since the transition zone structure of TNA is particularly well constrained with good sampling of all the triplicated branches due to both the 410 km and 660 km discontinuities. The resulting transition zone structure in our preferred model PAC06 turned out to be very similar to that of TNA, and the major difference is that PAC06 has a small velocity jump (DVs 1%) at approximately 516 km. Although we did not observe clear triplications due to such a small discontinuity, it is needed mainly to explain the timing and amplitudes of the triplicated ‘‘cd’’ branches (Figure 3), turning within the transition zone. [21] Our model PAC06 inherited a 406 km discontinuity from TNA. The velocity jump, however, is smaller (4%), mainly due to the generally larger velocities in the shallower part of the model (see Figure 11), which are required to explain the differential traveltimes between the S multiples and the guided waves. We have conducted various tests to examine the exact depth, velocity jump and sharpness of the discontinuity. However, these fine-scale features cannot be well resolved with our data set. In particular the synthetics constructed with the perturbations produce nearly equal waveform fits to most of the observations. This is because our data set does not contain sampling of the 410 km triplications as well as those addressed by Melbourne and Helmberger [1998] or Song et al. [2004]. [22] However, the radial structure within the transition zone is well determined, particularly, the average velocity gradient and the 1% velocity jump at approximately 516 km. The constraint is mainly from modeling the relative behavior of the ‘‘cd’’ and ‘‘ef’’ branches. Figure 12 gives such an example, where we compare selected SSS waveform fits with our preferred model PAC06 and the perturbed ones. The data shown were selected from the shallow event collection to avoid complications caused by the depth phases. As the distance increases, SSScd systematically moves backward relative to the forward branch SSSef. Although there are small perturbations on the indi-

9 of 20

B08301


B08301

Figure 9. Waveform comparison between data (black) and the synthetics (red) from our preferred SH model PAC06sh for four shallow events. Both the data and synthetics are aligned on SS. The traces for the two thrust events 10100677 and 9743493 have been multiplied by a factor of ‘‘ 1’’ for display purposes. vidual observations, the stability of the triplication pattern indicates little significant lateral variation along the path. Our model PAC06 accurately produces both the differential times and relative amplitudes between the triplicated branches SSSef and SSScd. The perturbed model ‘‘A’’ with a similar gradient, but no 516 km velocity jump, significantly overpredicts the amplitudes of the backward branch SSScd. In contrast, the larger 516 km discontinuity in the perturbed model ‘‘B’’ exaggerates the triplication in SSScd,

and causes SSScd to fade away beyond 82°. We also included two models ‘‘C’’ and ‘‘D’’ with different velocity gradients in Figure 12 to display model sensitivity. As the larger gradient in model ‘‘C’’ tends to suppress the ‘‘cd’’ branch, model ‘‘D’’ with the smaller gradient produces the magnified and advanced SSScd. We have tested the depth resolution of this small discontinuity. Varying its depth mainly changes the position of the triplication on the ‘‘cd’’ branch, as well as the travel times near the crossover

10 of 20

B08301


Figure 10. (a) Waveform fits with our preferred SH model PAC06sh (left) compared to those from the perturbed models with a different lid thickness. The model comparison is displayed to the right where PAC06 is in black. (b) Waveform fits with our preferred SH model PAC06sh (left) compared to those from the perturbed models with a different-sized G discontinuity between the lid and LVZ. (c) Waveform fits with our preferred SH model PAC06sh (left) compared to those from the perturbed models where the lid thickness trades off with the G discontinuity. In particular, model 1 has a 80 km thick lid, whereas model 2 has a thinner lid of 40 km thick.

11 of 20

B08301

B08301


Figure 11. Comparison of PAC06 with the earlier pure path upper mantle models, TNA, SNA [Grand and Helmberger, 1984b], and ATL [Grand and Helmberger, 1984a], developed by modeling similar multibounce S wave data sets. point (see Figure 3). Although the depth of approximately 516 km provides the best fits to the observations, its uncertainty is relatively large. Particularly, a change of ±20 km produces little significant increase in waveform misfits. [23] PAC06 contains a slightly shallower 651 km discontinuity and a faster high gradient zone right below the discontinuity compared to TNA (see Figure 11). These features were well determined with our data. Synthetic experiments showed significant advancement of the ‘‘ef’’ branches, particularly of SSS and S4 by the elevated discontinuity and faster high gradient zone, which however, were required by our observations. A similar issue plus the mineralogical implication has been addressed in a recent study by Wang et al. [2006]. 3.3. Summary: SH Component [24] In sections 3.1 and 3.2, we concentrated on the derivation of the model, where we showed complete waveform comparisons only for a small fraction of the studied events. Here we summarize the overall fit of the differential traveltimes between the S multiples for all the events in Figure 13, where the differences of the differential traveltimes, DTSS S, DTSSS SS and DTSSSS SS between the syn-

B08301

thetics and data are displayed. To calculate these timing differences, we have measured the time lags between the data and synthetics for S, SS, SSS and S4, respectively, by waveform cross correlation. In cases for which SSS or S4 is triplicated, the emphasis has been given to the early branch. The alignment of synthetics and data on SS has mostly eliminated possible origin time problems. Moreover, we have also applied event mislocation corrections (see Table 1) for the few events that displayed relatively large, systematic discrepancies in the differential time measurements. In particular, the observed higher-order multiples from these events are increasingly faster or slower than the synthetic predictions, which can be easily explained by the events’ mislocation error. The data points in Figure 13 are color coded by the depths of the events, and the overlain waveform comparisons for representative events at different depths, with different mechanisms are given in the auxiliary material (Figures S6– S28). [25] The most apparent anomaly in Figure 13 occurs for DTSS S, where large negative values of DTSS S up to 7 s are observed. This implies that the observed direct S waves are faster than the model would predict. We attribute the anomaly to the fast slab effects (see the earlier discussion of Figure 5). Although this warrants further investigation, there are two main lines of evidence that favor our argument. First, nearly all the problematic events are shallow, sampling particular distance ranges. Moreover, many of these events are located along the complex New Hebrides subduction slab. Secondly, the upper mantle structure is not an effective modifier on the differential traveltimes between S and the higher-order multiples. Another noteworthy feature in Figure 13 is the linear trend for DTS4 SS, where DTS4 SS systematically increases from the westernmost coastal stations to the easternmost inland ones. Such a tendency becomes particularly obvious when the guided waves are involved for the shallow events, due to the sharp transition from fast oceanic lithosphere to thick continental crust across the Pacific – North America boundary. 3.4. SV Component [26] The above discussion was restricted to the SH (transverse) component, and here we address the SV wave observations. We will focus on the vertical component, since it generally produces better signal-to-noise ratios than the radial component. The greatest difficulty in modeling the SV waves is caused by the SV P wave coupling. The problem can be overwhelming, especially when the SVcoupled PL waves are developed, and mainly controlled by the crustal waveguide for mixed continental-oceanic paths [e.g., Helmberger and Engen, 1974]. Although we widely observe the energy leakage of SV waves into P waves in this study, their coupling effect is not significant, which implies that the shallow oceanic structure is more uniform. For an example, we display the record section of the vertical component from a shallow event in Figure 14, where the multibounce SV waves are easily recognized and labeled. Among them, S4 is the strongest due to the constructive interference of the triplicated branches at the sampled distance range (see Figure 3). On the SV records, even S5 can be clearly observed, since the Rayleigh waves are much slower. As we will address later in this section, these S5 waves that turn near a depth of 300 km provide strong

12 of 20

B08301


B08301

Figure 12. (a) Selected SSS waveform fits from four shallow events (10100677, 9982833, 9564185, and 9533473 from the bottom to top) with our preferred model PAC06sh and the perturbed models. (b) Model perturbations that are particularly concentrated in the transition zone, where the triplicated branch SSScd is the most sensitive. Both the data (black) and synthetics (red) are aligned on SSSef, and the normalized amplitudes by the maxima on the records are plotted. Note the different behavior of the synthetic SSS from the perturbed models, in particular, the amplitude contrast between SSSef and SSScd. constraint on the depth extent of the observed seismic anisotropy. [27] Many investigators have reported radial anisotropy (or transverse isotropy) from surface wave studies in the upper mantle beneath the Pacific Ocean [e.g., Ekstro¨m and Dziewonski, 1998]. However, it is not appropriate for us to contrast the SH and SV observations directly, since the multibounce SH and SV waves contain different phase shifts from the bounce points and the station-side receiver func-

tions. One way to bypass the problem is to compare the SV observations with the synthetics from our well-constrained SH model (Figure 14). Since there is some contamination of SS, we have chosen to align the synthetics and data on S3 instead. The alignment resulted in a large shift of over 20 s for the synthetics relative to the data, which is not seen for the SH component. Moreover, the observed higher-order multiples S4 and S5 are progressively slower than the synthetic predictions. In particular, S5 are delayed by over

13 of 20

B08301


B08301

tinuity, in parameterizing the SV model. These features, however, are not well resolved with the SV data. Figure 15 displays the waveform fits for the same observations as shown in Figure 14, with our preferred SV model PAC06sv, which accurately reproduces both the waveforms and traveltimes of the SV wave multiples. The comparison between the SV and SH model is given in Figure 16c. [29] PAC06sv features slower velocities than the SH model down to a depth of approximately 300 km (Figure 16c). Although there is some trade off between the lid and LVZ velocities for the SV model, the depth extent of anisotropy (Vsv < Vsh) is well determined, and the major constraint is attributed to the S5 turning at a depth of approximately 300 km. For an example, we display in Figure 16 the waveform comparisons with two trial models, where the anisotropy is confined to the top 170 km (model I), essentially in the lid and LVZ, or extends to the 406 km discontinuity (model II). The lid and LVZ velocities have also been adjusted accordingly, so that the traveltimes of the lower-order multiples, such as S, SS, SSS and S4, that turn below the 406 km discontinuity, are conserved. However, since S5 travels nearly flat at a depth of 300 km (Figure 16d), it is rather sensitive to small velocity perturbations there. While the synthetic S5 gets accelerated by the faster velocities at that depth in model I, it is significantly delayed by the slower velocities in model II. In fact, such discrepancies between S5 and the lower-order multiples would be more severe if the lid and LVZ velocities remained unchanged. Moreover, they cannot be reconciled by perturbing the lid or LVZ velocities alone. Our experiments with various perturbed models suggest significant anisotropy (DVsh sv > 1%) extending below 220 km depth.

4. Results and Discussions

Figure 13. Differences of the S wave multiples’ differential times, particularly DTSS S, DTSSS SS, and DTSSSS SS, between the data and synthetics. A positive value implies that the separation between the multiples on the data is smaller than that on the synthetics, whereas a negative value suggests the opposite. The data points from different events are color coded by the events’ depths. 10 s relative to the synthetics. All these discrepancies suggest that the SV model needs to be slower than the SH counterpart, especially in the shallow part, such as the lid or LVZ, since the higher-order multiples increasingly spend more time there. [28] Therefore we have concentrated on the shallow part in deriving the SV model. Because of the lack of good triplication data in the SV observations, the same structure below the 406 km discontinuity as in the SH model is assumed. We also retain the characteristic features in the SH model, such as the lid thickness and the large ‘‘G’’ discon-

[30] In this study, we concentrated on resolving the upper mantle layering by constructing simple models with as few parameters as possible. In particular, if a linear gradient between two depths proved effective in explaining most observations, we did not add any other posterior constraints. Our preferred 1-D model, PAC06 (Table 2), however, provides excellent fits to both the waveforms and traveltimes of the observed S wave multiples from a large number of events at different depths and with different mechanisms. [31] We interpret PAC06 as an effective path average for the corridor, since the sampled lithospheric age spans a wide range from approximately 125 Ma near the source region (Tonga-Fiji) to 10 Ma approaching the California coastal line (see Figure 17b). The thinning of lithosphere, or the decreasing of seismic velocity with the crustal age getting younger has been reported from pure surface wave studies by many researchers [e.g., Nishimura and Forsyth, 1989; Ritzwoller et al., 2004; Maggi et al., 2006]. To address such lateral variation is out of the scope of this paper. However, we will briefly discuss two pieces of evidence here, which suggest little systematic variation along the path. [32] Figure 17a displays the comparison between the model PAC06 from this study and the model PA5 developed by Gaherty et al. [1996] for the older half of the corridor from Fiji-Tonga to Hawaii. In their study, Gaherty et al. [1996] mainly utilized relatively long period (>25 s)

14 of 20

B08301


B08301

Figure 14. Selected waveform fits for the vertical component of a shallow event 9792597 (see Table 1) with our preferred SH model PAC06sh.The data are shown in black with synthetics in red. We chose to align the synthetics and data on ‘‘S3’’ since there is some contamination on ‘‘SS.’’ S wave multiples up to S5 that can be easily recognized are labeled. Note that the synthetics have been shifted by 20 s for such an alignment. However, that is about the amount the observed direct S waves are advanced. Moreover, the higher-order multiples S4 and S5 are delayed on the data relative to the synthetics. body and surface waves plus the ScS reverberations [e.g., Revenaugh and Jordan 1991b] for determining the discontinuities. Although PA5 samples relatively old oceanic lithosphere (100– 125 Ma), while our studied path covers a wider range from 125 Ma to 10 Ma at the receiver side, the two models display remarkable agreement (Figure 17a) despite small detailed differences. Moreover, PA5 does a reasonably good job of explaining our data as presented in the auxiliary material (Figure S31). On the other hand, Figure 17c compares the synthetic prediction of PAC06sh

to the data from the recent Hawaii earthquake. The shown comparison for the coastal station VES is typical for the over a hundred stations of TriNet. This data set from the Hawaii event samples the younger half of the path, however, the waveforms and the separations between S and ‘‘G + SS’’ are nicely reproduced from PAC06. In particular, the separations between S and ‘‘G + SS’’ are rather sensitive to the velocities in the lid or LVZ. For example, a decrease of lid velocity by less than 4% from our experiments would have narrowed the separation between S and ‘‘G + SS’’ by over 15 s. In

15 of 20

B08301


B08301

Figure 15. Selected waveform fits for the vertical component of the same event as shown in Figure 14 with our preferred SV model PAC06sv. The data are shown in black with synthetics in red. We aligned the synthetics and data on ‘‘S3,’’ which resulted in an average shift of 2 s for the synthetics relative to the data. Note the nice fits for all the labeled S wave multiples. Particularly, the poorly fit phases following direct S are the shear coupled PL waves. summary, these comparisons between PAC06 and the half path model (PA5) or observations suggest very little systematic lateral variation in the lid or LVZ, and the whole path appears uniform. [33] Another supporting evidence comes from our study of multiScS waves. For about 1/3 of our studied events, that are intermediate depth and thrust, the multibounce ScS waves are rich and strong. Figure 18 displays an example, where we include the multiScS waves in the waveform comparison for an intermediate-depth event (9648517). As it might be expected from lower mantle studies [e.g.,

Garnero and Helmberger, 1993; Ritsema et al., 1997; Breger et al., 2001], these multibounce ScS waves (ScSn and sScSn) are consistently delayed by over 5 s relative to PAC06. However, if we align the synthetics and data on a particular multiple ScSn, we obtain remarkable fits for their first-order reverberations, e.g., S660 S, S660+S from the mantle discontinuities as discussed by Revenaugh and Jordan [1989]. Note that these ‘‘precursors’’ (ScSnSd S) or ‘‘peglegs’’ (ScSnSd+S) each comprise a family of n or n + 1 rays that travel different paths, but have the same traveltime, amplitude, and phase, assuming a 1-D model. Consequently,

16 of 20

B08301


B08301

Figure 16. Waveform fits for the same event as shown in Figure 15 with two perturbed SV models: (a) model I and (b) model II. Note the traveltimes of the lower-order multiples up to S4, that turn below the 406 km discontinuity are conserved, however the synthetic S5 is advanced with perturbed model I, while slowed by model II. (c) Comparison of the perturbed models with PAC06. (d) Raypaths of the first legs of S4 and S5. The two dashed lines indicate the 406 km and 651 km discontinuities. Note that the raypath of S 5 is nearly at a depth of ~300 km; hence it is rather sensitive to small velocity perturbations there. the strength of these precursors or peglegs relative to their parent phases (ScSn’s) gets stronger for the higher-order ‘‘n’’, due to the increasing redundancy, which is exactly what we observe in the data (Figure 18b). Similar results are obtained for the other studied events, although they sample different portions along the path. Presently we do not know how much lateral variation is needed to decrease the redundancy in these multiScS reverberations, and such efforts will be addressed in the future study. [34] The derived anisotropy in PAC06 represents an effective average over detailed variability along the path. We are not able to resolve the geometry of anisotropy, or its true amount, due to the isotropic nature of the synthetics as well as the limited azimuthal sampling. However, the anisotropy of Vsh > Vsv and its magnitude in PAC06 are consistent with transverse isotropy reported by previous investigators [e.g., Gaherty et al., 1996; Ekstro¨m and Dziewonski, 1998]. The accuracy and uniqueness of map-

ping the actual anisotropy to transverse isotropy using data from a single corridor have been discussed in detail by Gaherty et al. [1996]. PAC06 contains anisotropy down to a depth of 300 km, beyond the lid and LVZ, which is consistent with the results from previous studies [e.g., Nishimura and Forsyth, 1989; Ekstro¨m and Dziewonski, 1998]. The anisotropy of Vsh > Vsv in the lid can be attributed to the lattice preferred orientation (LPO) in olivine when the lid was formed [e.g., Nicolas and Christensen 1987], since the studied path is consistently oblique to the ancient spreading direction. The anisotropy in and below the LVZ could be caused by the preferred orientation of meltfilled pockets [e.g., Schlue and Knopoff 1977] or shearing in the asthenosphere due to mantle flow or plate motions [e.g., Zhang and Karato, 1995; Montagner and Tanimoto, 1991]. [35] Figure 11 compares PAC06 to the earlier pure path models developed with similar multibounce S wave data, particularly, TNA, SNA for the North America shield

17 of 20

B08301


B08301

Figure 17. (a) Comparison of PA5 developed by Gaherty et al. [1996] for the corridor from Tonga-Fiji to Hawaii with our preferred model PAC06. (b) Comparison of the source receiver geometry for PAC06, PA5, and the recent Hawaii event (star). (c) Selected transverse component waveform fits with PAC06sh at the station VES for the Hawaii event 12261875 (25 October 2006, 1707:49.200 UT). The source parameters are from Harvard’s CMT solutions. The data are shown in black with synthetics in red. [Grand and Helmberger, 1984b] and ATL featuring the old Atlantic (70– 150 Ma) [Grand and Helmberger, 1984a]. Except for SNA, the other three models all contain a welldeveloped LVZ. Moreover, the high gradient zone (HGZ) beneath the LVZs appear unique for the oceanic paths, in contrast with the continental model SNA. The steep gradient in the HGZ is intriguing, since it cannot be easily explained with a subadiabatic thermal gradient or decreasing partial melt with depth. One way to match the fast velocity increase below the LVZ is to progressively increase the grain size by about one order of magnitude from approximately millimeters to approximately centimeters [Faul and Jackson, 2005]. A possible explanation for such increase in grain size with depth is a change in deformation mechanism, such as the transition from dislocation to diffusion creep suggested by Karato and Wu [1993]. A depth-dependent increase in the amount of pyroxenite could also produce an enhanced velocity gradient in the HGZ, though a plausible dynamical picture is absent [e.g., Stixrude and Lithgow-Bertelloni, 2005]. It is noteworthy in Figure 11 that the velocity

minimum in the LVZ migrates upward below the ridges (TNA) and downward beneath the old oceans (ATL). The value also tends to increase with age. Similar features have been observed in a number of surface wave studies [e.g., Nishimura and Forsyth, 1989; Zhang and Lay, 1999]. This might suggest that possible flows in the LVZ are not confined within a certain depth range. [36] Compared to the other models, PAC06 contains a velocity jump at 516 km (Figure 11). This is in good agreement with global stacks of SS precursors [Shearer, 1990, 1993] or ScS reverberations [Revenaugh and Jordan, 1991a]. The small velocity jump at the discontinuity is consistent with the b g transition in (Mg, Fe)2SiO4 of the mineralogical models [e.g., Rigden et al., 1991].

5. Conclusion [37] In this study, we developed a pure path shear velocity model PAC06 along the corridor from Tonga-Fiji to California. The model contains a fast lid (Vsh = 4.78 km s 1, Vsv = 4.58 km s 1) 60 km thick. The underlying low-

18 of 20

B08301


B08301

Figure 18. (a) Extended waveform comparison for an intermediate-depth (367 km) event (9648517) at three stations. See Figure S22 of the auxiliary material for the comparison of the whole record section of the multibounce S waves. The data are shown in black, while the synthetics from PAC06 in red. Labeled are the multibounce S waves as well as the multibounce ScS waves. The data and synthetics are aligned on SS. Note that the observed multiScS waves are apparently delayed. (b) Waveform comparison of the multibounce ScS waves (ScS2, ScS3, and ScS4), where the data and synthetic traces are the stacked records from the whole record section. Both the data and synthetics are aligned on ScS3 and the time axis is relative to the event’s origin time. The windowed data and synthetics are aligned up on each individual ScS multiple ScSn in the right, where their amplitudes are plotted as normalized. The first-order reflections of the ScS multiples, such as the precursors, S660 S, S410 S, sS660 S and the peglegs, S660+S, sS410+S, and sS660+S are easily recognized and labeled. velocity zone (LVZ) is prominent with the lowest velocities Vsh = 4.34 km s 1, and Vsv = 4.22 km s 1. The anisotropy (Vsv < Vsh) of PAC06 extends to a depth of 300 km. Besides the 406 km and 651 km discontinuities, PAC06 also has a small (1%) velocity jump at 516 km. We consider

these main features of PAC06 to be well determined, since PAC06 explains a large data set from various events. Therefore it is ideally suited for comparing with mineralogical models.

19 of 20

B08301


[38] Acknowledgments. The authors would like to thank James Gaherty and an anonymous reviewer for their constructive comments and suggestions, which have helped to improve the manuscript. We also thank Lupei Zhu for using his FK code. This work is supported by National Science Foundation under the contracts EAR-0636012 and EAR-0456433. Caltech Seismological Laboratory contribution 9168.

References Breger, L., B. Romanowicz, and C. Ng (2001), The Pacific plume as seen by S, ScS and SKS, Geophys. Res. Lett., 28, 1107 – 1110. Ding, X. Y., and S. P. Grand (1993), Upper mantle q structure beneath the East Pacific Rise, J. Geophys. Res., 98(B2), 1973 – 1985. Dziewonski, A. M., and D. L. Anderson (1981), Preliminary reference Earth model, 25, 297 – 356, 1981. Dziewonski, A. M., T. A. Chou, and J. H. Woodhouse (1981), Determination of earthquake source parameters form waveform data for studies of global and regional seismicity, J. Geophys. Res., 86, 2825 – 2952. Ekstro¨m, G., and A. M. Dziewonski (1998), The unique anisotropy of the Pacific upper mantle, Nature, 394, 168 – 172. Faul, U. H., and I. Jackson (2005), The seismological signature of temperature and grain size variations in the upper mantle, Earth Planet. Sci. Lett., 234(1 – 2), 119 – 134, doi:10.1016/j.epsl.2005.02.008. Gaherty, J. B., T. H. Jordan, and L. S. Gee (1996), Seismic structure of the upper mantle in a central Pacific corridor, J. Geophys. Res., 101, 22,291 – 22,309. Gaherty, J. B., M. Kato, and T. H. Jordan (1999), Seismological structure of the upper mantle: A regional comparison of seismic layering, Phys. Earth Planet. Inter., 110, 21 – 41, doi:10.1016/S0031-9201(98)00132-0. Garnero, E. J., and D. Helmberger (1993), Travel times of S and SKS: Implications for three-dimensional lower mantle structure beneath the central Pacific, J. Geophys. Res., 98, 8225 – 8241. Grand, S. P., and D. V. Helmberger (1984a), Upper mantle shear structure beneath the North Atlantic Ocean, J. Geophys. Res., 89, 11,465 – 11,475. Grand, S. P., and D. V. Helmberger (1984b), Upper mantle shear structure of North America, Geophys. J. R. Astron. Soc., 76, 399 – 438. Graves, R. W., and D. V. Helmberger (1988), Upper mantle cross section from Tonga to Newfoundland, J. Geophys. Res., 93, 4701 – 4711. Helmberger, D. V. (1983), Theory and application of synthetic seismograms, in Earthquakes: Observation, Theory and Interpretation, pp. 174 – 222, Soc. Ital. di Fis., Bolgna. Helmberger, D. V., and G. R. Engen (1974), Upper mantle shear structure, J. Geophys. Res., 79, 4017 – 4028. Karato, S.-I., and P. Wu (1993), Rheology of the upper mantle: A synthesis, Science, 260, 771 – 778. Maggi, A., E. Debayle, K. Priestley, and G. Barruol (2006), Multimode surface waveform tomography of the Pacific Ocean: A closer look at the lithospheric cooling signature, Geophys. J. Int., 166, 1384 – 1397, doi:10.1111/j.1365-246X.2006.03037.x. Melbourne, T., and D. V. Helmberger (1998), Fine structure of the 410 km discontinuity, J. Geophys. Res., 103, 10,091 – 10,102. Melbourne, T., and D. V. Helmberger (2001), Mantle control of plate boundary deformation, Geophys. Res. Lett., 28, 4003 – 4006. Melbourne, T. I., and D. V. Helmberger (2002), Whole mantle shear structure beneath the East Pacific Rise, J. Geophys. Res., 107(B9), 2204, doi:10.1029/2001JB000332. Montagner, J.-P., and T. Tanimoto (1991), Global upper mantle tomography of seismic velocitites and anisotropy, J. Geophys. Res., 96, 20,337 – 20,351. Nicolas, A., and N. I. Christensen (1987), Formation of anisotropy in upper mantle peridotites: A review, in Composition, Structure, and Dynamics of

B08301

Lithosphere-Asthenosphere System, Geodyn. Ser., vol. 16, edited by K. Fuchs and C. Froidevaux, pp. 111 – 123, AGU, Washington, D. C. Nishimura, C. E., and D. W. Forsyth (1989), The anisotropic structure of the upper mantle in the Pacific, Geophys. J. R. Astron. Soc., 96, 203 – 229. Revenaugh, J., and T. H. Jordan (1989), A study of mantle layering beneath the western Pacific, J. Geophys. Res., 94, 5787 – 5813. Revenaugh, J., and T. H. Jordan (1991a), Mantle layering from SCS reverberations: 2. The transition zone, J. Geophys. Res., 96, 19,763 – 19,780. Revenaugh, J., and T. H. Jordan (1991b), Mantle layering from SCS reverberations: 3. The upper mantle, J. Geophys. Res., 96, 19,781 – 19,810. Rigden, S. M., G. D. Gwanmesia, J. D. Fitzgerald, I. Jackson, and R. C. Lieberman (1991), Spinel elasticity and the seismic structure of the transition zone of the mantle, Nature, 345, 143 – 145. Ritsema, J., E. J. Garnero, and T. Lay (1997), A strongly negative shear velocity gradient and lateral variability in the lowermost mantle beneath the Pacific, J. Geophys. Res., 102, 20,395 – 20,411. Ritsema, J., H. J. van Heijst, and J. H. Woodhouse (1999), Complex shear velocity structure imaged beneath Africa and Iceland, Science, 286, 1925 – 1928. Ritzwoller, M. H., N. M. Shapiro, and S.-J. Zhong (2004), Cooling history of the Pacific lithosphere, Earth Planet. Sci. Lett., 226, 69 – 84. Romanowicz, B. (2003), Global mantle tomography: Progress status in the past 10 years, Annu. Rev. Earth Planet. Sci., 31, 303 – 328, doi:10.1146/ annurev.earth.31.091602.113555. Schlue, J. W., and L. Knopoff (1977), Shear-wave polarization anisotropy in the Pacific basin, Geophys. J., 49, 145 – 165. Shearer, P. M. (1990), Seismic imaging of upper mantle structure with new evidence for a 520-km discontinuity, Nature, 344, 121 – 126. Shearer, P. M. (1993), Global mapping of upper mantle reflectors from long-period SS precursors, Geophys. J. Int., 115, 878 – 904. Song, T.-R. A., D. V. Helmberger, and S. P. Grand (2004), Low-velocity zone atop the 410-km seismic discontinuity in the northwestern United States, Nature, 427, 530 – 533. Stixrude, L., and C. Lithgow-Bertelloni (2005), Mineralogy and elasticity of the oceanic upper mantle: Origin of the low-velocity zone, J. Geophys. Res., 110, B03204, doi:10.1029/2004JB002965. Tanimoto, T., and K. Prindle Sheldrake (2002), Three-dimensional S-wave velocity structure in southern California, Geophys. Res. Lett., 29(8), 1223, doi:10.1029/2001GL013486. Wang, Y., L. Wen, D. Weidner, and Y. He (2006), SH velocity and compositional models near the 660-km discontinuity beneath South America and northeast Asia, J. Geophys. Res., 111, B07305, doi:10.1029/ 2005JB003849. Zhang, S., and S. Karato (1995), Lattice preferred orientation of olivine aggregates deformed in simple shear, Nature, 375, 774 – 777. Zhang, Y.-S., and T. Lay (1999), Evolution of oceanic upper mantle structure, Phys. Earth Planet. Inter., 114, 71 – 80. Zhao, L. S., and D. Helmberger (1993), Upper mantle compressional velocity structure beneath the northwest Atlantic Ocean, J. Geophys. Res., 98, 14,185 – 14,196. Zhu, L., and L. A. Rivera (2002), A note on the dynamic and static displacements from a point source in multi-layered media, Geophys. J. Int., 148, 619 – 627, doi:10.1046/j.1365-246X.2002.01610.x. D. V. Helmberger and Y. Tan, Seismological Laboratory 252-21, Division of Geological Planetary Sciences, California Institute of Technology, Pasadena, CA 91125, USA. ([email protected])

20 of 20