Notice: Wiley Online Library will be unavailable on Saturday 30th July 2016 from 08:00-11:00 BST / 03:00-06:00 EST / 15:00-18:00 SGT for essential maintenance. Apologies for the inconvenience.
 The lithosphere-asthenosphere boundary (LAB) separating the rigid lid from the underlying weaker, convecting asthenosphere is a fundamental interface in mantle dynamics and plate tectonics. However, the exact depth and defining mechanism of the LAB interface remain poorly understood. The ocean plates are ideal for testing hypotheses regarding the nature of a plate since they make up 70% of Earth's surface area and have a relatively simple geological history. Seismically imaging the oceanic LAB at high resolution has proved challenging. Yet, several studies have recently increased resolution with provocative results. We summarize recent imaging of discontinuity structure beneath much of the Pacific using receiver functions from ocean floor borehole seismometers and land stations located at ocean-continent margins, SS precursors, and waveform modeling of multiple phases including multiple bounce S waves, ScS reverberations, and surface waves. Overall, there is much agreement among these different approaches about the reported depth of a negative discontinuity that occurs near the expected depth of the LAB. Some of the apparent discrepancies in depth are explained by the variation in sensitivity of seismic waves that sample structure at different wavelengths. Yet, when the results are considered together, no single age-depth relationship is illuminated. There are also puzzling discrepancies in where the discontinuity is detected, which again suggests greater complexity. Here we test the possibility that discrepant detection of a strong sharp discontinuity is caused by anisotropic structure. We stack SS waveforms with bounce points in the central Pacific into azimuthal bins. We use two methods, one that inverts for discontinuity structure based on subtle variations in the character of the SS waveform, and another that considers SS at higher frequency. We find azimuthal variation in the amplitude of the waveform, including a polarity reversal. We suggest that anisotropy is an important factor in imaging and constraining discontinuity structure of the oceanic plate, and must be carefully considered to constrain the age-depth dependence and defining mechanism of the oceanic lithosphere.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Oceanic plates form at mid-ocean spreading ridges, subsequently cooling, thickening, and subsiding as the plate ages and migrates away from the ridge axis. Yet, many questions remain even for this relatively simple realization of the tectonic plates. For instance, relatively old (>70 My) oceanic lithosphere does not replicate the subsidence predicted by simple cooling models [Stein and Stein, 1992]. Similarly, gravity lineations observed on young ocean seafloor are not explained by simple plate tectonic models [Harmon et al., 2009, 2011].
 A complete understanding of oceanic lithosphere requires imaging of the lithosphere-asthenosphere boundary (LAB) to illuminate the dynamics acting at the base of the plate. The LAB delineates the transition from the rigid lid to the underlying weaker asthenosphere. Simple geodynamic models place the LAB at the intersection of the geotherm and adiabat, the lithosphere characterized by conductive cooling and subadiabatic temperatures versus the convecting and adiabatic asthenosphere. These simple thermal models predict that the LAB increases in depth with age following subsidence patterns. Subsidence is well-predicted by half-space cooling models for seafloor <70 My old, and the empirical plate model provides a better fit for older seafloor where an additional mechanism such as convective instabilities or alteration from hot spots restricts subsidence [Parsons and Sclater, 1977; Stein andStein, 1992; Smith and Sandwell, 1997; Korenaga and Korenaga, 2008]. Thermal models predict gradual seismic velocity gradients in depth between lithosphere and asthenosphere. Plate model predictions are slightly sharper, but gradients are still quite gradual, >80 km thick for seafloor older than 25 My. The predicted LAB depth (adiabiat-geotherm) definition roughly corresponds to the depth of the lowest velocities (Figure 1).
 The general trend of plate cooling and thickening with age has been captured by surface waves, which offer the most comprehensive sampling of the oceanic lithosphere, and broadly image a progressively thicker and faster lid [Nishimura and Forsyth, 1989; Ritzwoller et al., 2004; Maggi et al., 2006a; Priestley and McKenzie, 2006; Nettles and Dziewonski, 2008; Lekic and Romanowicz, 2011b]. However, the long period of these waveforms means that sharper features (lateral or vertical) may not be resolvable if they exist. This is important because deviations from these thermal predictions either for lithospheric thickening with age and/or the sharpness of the seismic velocity gradient imply greater complexity, suggesting additional mechanisms are required to define the LAB.
 Seismically imaging the LAB using high resolution approaches allows us to test the viability of models that include melt, hydration, and anisotropy and to assess the effects of anomalous features such as hot spots and small-scale convection. Imaging the oceanic plate at high resolution has proven challenging due to the paucity of instruments on the ocean floor and overall higher noise of ocean island stations. The remoteness of the oceans, additional difficulty in the design and installation of ocean bottom seismometers beneath 4 km of ocean, and high noise levels in oceanic seismograms makes in situ measurements challenging and expensive. However, a number of recent studies using innovative techniques and exploring higher frequencies have increased imaging resolution of the LAB suggesting a relatively sharp discontinuity is present in many locations, in particular beneath the Pacific. The Pacific is the de facto focus site since it is surrounded by active margins that provide an abundance of events at ideal geometries for a variety of seismic methodologies.
 Here we summarize results and sensitivities of a multitude of seismic tools that have been used to image the shallow upper mantle of the Pacific Ocean. These methodologies include surface waves, receiver functions, SS precursors, and velocity inversions using waveform-modeling of phases such as multiple-S reflections, ScS, and surface waves.
2.1. Surface Waves
 Surface waves provide excellent resolution of velocity structure of the crust and upper mantle. Surface wave arrival times at various frequencies are used to invert for velocity-depth profiles, with longer periods (lower frequencies) sampling deeper structure. These waves resolve lateral variations at the scale of 100s of km regionally and 1000s of km globally [Nettles and Dziewonski, 2008].
 Surface waves also reveal important constraints on oceanic anisotropy. Global averages suggest a peak in radial anisotropy – alignment in the horizontal plane that results in horizontally polarized waves traveling faster than vertically polarized waves – at 120–150 km depth, increasing in depth with age [Nettles and Dziewonski, 2008; Lekic and Romanowicz, 2011a]. Similarly inversions for azimuthal anisotropy – horizontal alignment that results in azimuthally dependent velocities – find a correlation between the fast direction of anisotropy and absolute plate motion at 100–200 km depth, except in the vicinity of plumes [Maggi et al., 2006b]. The anisotropy is typically thought to reflect the region where the asthenosphere is deforming in response to plate tectonic forces, again suggesting that convection occurs in this region.
 Overall surface waves detect velocity minima and strong horizontal alignment at 120 to 150 km depth, increasing in depth with age, in agreement with the plate model definition of the LAB. Determining the exact thickness of the plate using surface waves alone is complicated by the smoothness of the velocity-depth profiles. The same is true for estimates of the sharpness of the boundary. The long periods and associated sensitivities of these waveforms means that discontinuities >40 km thick cannot be distinguished from sharper discontinuities. In addition, error in the depth to the base of the lid is ±20 km for regional studies [Li and Burke, 2006]. Therefore, the exact depth of the transition from rigid lid to convecting asthenosphere is somewhat ambiguous, especially if a sharp boundary does exist. Higher frequency techniques such as receiver functions, SS precursor methods, and waveform modeling of S waves, multiple bounce S waves, and ScS are required to provide more precise constraints for the sharpness and depth of the LAB.
2.2. Multiphase Velocity Inversions for Pacific Transects
 Waveform modeling of S waves, multiple bounce S waves, and surface waves have been used to invert for upper mantle seismic velocity structure beneath the central Pacific [Tan and Helmberger, 2007]. Similar techniques were also used in combination with ScS in a separate study, with partially coincident sensitivity [Gaherty et al., 1999]. Multiple bounce S waves reflect at the surface of the earth one or more times before reaching a station and provide sensitivity to shallow mantle velocity structure along a seismic corridor, as described in the work of Tan and Helmberger . ScS waves that reflect from the core also reverberate at strong velocity contrasts in the upper mantle, and give information on the depth and sharpness of mantle discontinuities (Figure 2).
 This joint approach reveals a 6% velocity decrease at a nearly constant depth of 63 [Gaherty et al., 1999]–66 km [Tan and Helmberger, 2007] beneath the Pacific seafloor (Figures 3 and 4). The discontinuity is <30 km thick [Gaherty et al., 1999]. Though the sensitivity of each seismic phase in these studies is slightly different (Figure 4), all share a large area of overlap in a relatively wide and elongated seismic corridor. Another more recent study using ScS reverberations finds more variability in the depth of a shallow discontinuity in transects running from Hawaii to 7 surrounding locations, all primarily in the western half of the Pacific [Bagley and Revenaugh, 2008]. This ScS reverberation study indicates a significant negative (5–14%) shear velocity discontinuity occurring at 72–112 km depth that is <30 km thick, and a lack of a negative contrast in this depth range for the transect from Hawaii to central America [Bagley and Revenaugh, 2008]. Overall, a clear age-dependence is not present in either the ScS reverberation or joint seismic approaches.
 The relatively constant depth of the negative discontinuity observed in these Pacific transect studies was used to argue for a frozen-in compositionally depleted layer related to melting processes at the ridge [Gaherty et al., 1999] given that the oceanic plate is expected to thicken with age. Shallow melting at the ridge continually removes hydrogen and iron, leaving behind a depleted lithospheric residue; the velocity contrast between fertile and depleted materials was argued to give rise to the observed discontinuity. The magnitude of the required drop, and thus the low velocity beneath the lid (VSV = 4.22 km/s) were also used to argue for partial melting in the lower layer [Tan and Helmberger, 2007]. A reexamination of the velocity drop associated with dehydration of mantle materials suggests that hydration may also be able to explain velocity drops of this magnitude [Karato, 2003, 2012].
2.3. Receiver Functions
 The waveforms used in receiver functions travel from an earthquake as a parent phase (P or S) before converting to a daughter phase (s and p, respectively) at a seismic velocity discontinuity located beneath a station (Figure 2). The parent phase is deconvolved from the daughter to make the receiver function, i.e., the Earth's impulse response. These waveforms are particularly useful for their ability to constrain sharp velocity contrasts, since conversions are only large for sharp velocity gradients with thicknesses less than a quarter wavelength [Richards, 1972; Bostock, 1999; Rychert et al., 2007, 2010]. Assuming upper mantle velocities, this corresponds to depths of 2–10 km for P waves (periods 1–5 s), and 8–14 km for S waves (periods 7–16 s). The Fresnel zone sensitivity radius of these waves at lithospheric depths is also small, <1 degree from the conversion point.
 Until recently receiver function imaging of oceanic lithosphere was limited to ocean islands due limitations of seismic station locations and high noise levels in instruments deployed on the ocean floor. Velocity decreases interpreted as the LAB at 40–140 km depth have been reported from stations on or buried near ocean islands [Rychert et al., 2010]. Beneath Hawaii both P-to-S (Ps) and S-to-P (Sp) have been used to image the LAB deepening from Kauai (50–60 km) to Oahu (65–90 km) to the Big Island (90–140 km) [Li et al., 2000; Collins et al., 2002; Li et al., 2004; Wölbern et al., 2006]. Depths have also been reported beneath Iceland (80 km), eastern and western Greenland (70 km and 100–120 km), Jan Mayan (40–60 km) [Kumar et al., 2007], Galapagos (70 km), Easter Island (50 km) [Heit et al., 2007], and islands in the Indian Ocean (80 km) [Kumar et al., 2007]. The global Ps average for ocean island stations is 70 ± 4 km [Rychert and Shearer, 2009]. However, ocean islands do not necessarily represent typical oceanic lithosphere, given their location on top of a hot spot. Although constraining the impact of a hot spot influence on the lithosphere is important for our understanding of the lithosphere and hot spot-lithosphere dynamics, the effects of plume-lithosphere interaction on LAB depth are not well known globally. It has been hypothesized that the plume may locally decrease [Li et al., 2004] or increase [Hall and Kincaid, 2003; Yamamoto and Morgan, 2009] lithospheric thickness. Therefore, the implications of these results for “normal” lithosphere are not clear, and we continue, focusing on receiver function results from other regions.
 Recently Ps and Sp receiver function imaging of the lithosphere-asthenosphere boundary was achieved using ocean bottom borehole seismometers [Kawakatsu et al., 2009]. Burying seismometers beneath the ocean floor, while expensive, can eliminate the noise from ocean currents and dramatically improve signal-to-noise ratios. Receiver functions using two borehole seismometers, one on the Philippine Sea Plate and one on the Pacific Plate were used to measure discontinuities at 55, 76, and 82 km depth beneath seafloor aged 25, 49, and 129 My [Kawakatsu et al., 2009]. The two younger measurements come from different groupings of conversion points recorded at the same borehole station on the Philippine Sea Plate. These results were used to argue for a positive depth-age trend of a LAB-associated discontinuity. The magnitude of the velocity discontinuities imaged was large (7–8% drop), consistent with the presence of melt in the asthenosphere. The authors suggested that aligned melt sheets [Holtzman et al., 2003; Takei and Holtzman, 2009] could explain the observation without the need for high degrees of partial melting which may not be stable in the mantle [Kawakatsu et al., 2009].
 Stations located at the continent-ocean boundary were used to estimate thickness of subducting oceanic lithosphere [Kumar and Kawakatsu, 2011]. Sp receiver functions were used to image the base of the Philippine and Pacific Plates using waveforms recorded by stations located on Japan, the Aleutians, and the North American west coast. The thickness of these plates was calculated by subtracting estimates of the depth to the top of the slab from the Sp measurements of its base. The top of the slab was estimated two ways: 1) using Sp and 2) from previously published active source experiments (landward of the trench) and bathymetry (seaward of trench) (http://www.ngdc.noaa.gov/mgg/global/global.html). Sp receiver functions were stacked in rectangular conversion point bins with dimensions that ranging from 2–7°. The ages of the bins varied from ∼10–130 My and the discontinuity depth varied from 45 km to 85 km. These depths were also used to argue for a positive age depth trend of a discontinuity at the LAB.
2.4. SS Precursors
 The SS seismic phase is an S wave that reflects once at the surface of the Earth before arriving at a station (Figure 2). Underside reflections from deeper boundaries form small amplitude arrivals that arrive several 10s to 100s of seconds before the main SS phase. The SS precursors are particularly powerful imaging tools owing to the sensitivity to discontinuity structure at the center of the raypath (bounce point), providing coverage for regions where oceanic station installations are sparse or non-existent. They have the ability to place tight constraints on discontinuity depth (±5 km). SS precursors may be used to constrain velocity gradients, and are sensitive to sharp velocity gradients, <15–30 km thick.
 A methodology has recently been developed that uses subtle variations in the character of the SS waveform to invert for structure as shallow as the Moho, SS Lithospheric Interface Profiling (SSLIP) [Rychert and Shearer, 2010]. This has been made possible in part by the large number of events in the IRIS FARM database. SSLIP resolves a discontinuity at 25–130 km depth over a large swath of the Pacific (Figure 4). Discontinuity results are well-correlated with Voigt-averaged shear velocities at 100 km depth [Nettles and Dziewonski, 2008]. The result shows an increase in depth to the discontinuity with seafloor age [Parsons and Sclater, 1977] and distance from the ridge along a flow line [Adam and Vidal, 2010]. All of these indicate a discontinuity that follows the subsidence of the oceanic plate, and its depth is thermally defined. Long period SS is only sensitive to velocity gradients <20–30 km thick, and more gradual gradients reported in the SSLIP model may be caused by variation in depth to the discontinuity in the sensitivity region of the waveform.
 A discontinuity with a weak depth-age dependence was also imaged using SS waveforms in the depth range of 40–100 km at several locations beneath the Pacific, with two populations near 55 and 75 km depth [Schmerr, 2012]. The major difference between the methodology of Rychert and Shearer  was that in the work of Schmerr  data were hand-picked to eliminate poorly formed arrivals, and considered acceleration instead of displacement, which allowed access to higher frequencies and giving access to precursor waveforms in close proximity to the SS phase. The higher frequency SS approach imaged a sharp discontinuity primarily beneath regions with active surface volcanism, including the South Pacific Superswell, Socorro Hot spot, Louisville Hot spot, and near Hawaii [Schmerr, 2012]. A similar feature is also observed beneath most subduction zones in the circum-Pacific, but given the long-period waveform of SS and juxtaposition of multiple lithospheres, it is difficult to ascertain the exact depth of the LAB in these regions. The result was used to argue for a sharp (<20 km thick), thermally perturbed LAB beneath hot spots and other volcanic locales, and for a broad velocity gradient across the LAB in regions without a sharp interface.
3. Discussion of Previous Results
 To summarize, a sharp (<30 km thick) 6% velocity drop at relatively constant depth (60 ± 20 km) is observed across a large transect of the central Pacific using a combination of ScS, multiple S bounces, and surface waves [Gaherty et al., 1999; Tan and Helmberger, 2007]. A sharp (<20 km thick) ≥5% velocity drop showing a weak age dependence (∼55–75 km) primarily beneath anomalous, hot spot regions is imaged using high frequency SS [Schmerr, 2012] but not detected elsewhere. A sharp (<30 km thick) 5–14% velocity drop at 72–112 km depth is observed using ScS reverberations beneath a large swaths of the western Pacific, but a corresponding boundary is not observed in the eastern Pacific [Bagley and Revenaugh, 2008]. A sharp (<15 km thick) 7–8% velocity drop [Kawakatsu et al., 2009] in the 44–85 km depth range [Kawakatsu et al., 2009; Kumar and Kawakatsu, 2011] is imaged using receiver function imaging of subducting oceanic lithosphere and also on the old Pacific and younger Philippine Sea Plate. A discontinuity at 25–130 km depth is imaged beneath large portions of the Pacific, increasing in depth with age. The discontinuity is sharp beneath young seafloor, but not necessarily everywhere [Rychert and Shearer, 2011].
 Where sensitivity is overlapping results for the depth of the negative discontinuity agree within the estimated errors for the most part (Figure 4). We plot areas where sensitivity is overlapping, but depth to a discontinuity does not agree, and avoiding possible complications from subduction zones (Figure 5) and present a table summary (Table 1). However, many apparent disagreements could be explained by multiple discontinuities in depth, velocity gradients with different character (sharpness), and/or lateral variations in depth. Long period SS has the potential to resolve more gradual velocity gradients, and those that vary in depth more, and is sensitive to a slightly larger lateral area than short period SS. Transect studies and ScS have sensitivity over the long sections that the waves transverse, in comparison to the circular SS bins. One notable depth disagreement occurs beneath region 5, the disagreement between the ScS result (107–108 ± 5–7 km depth) and the Pacific transect studies (60 ± 20 km depth) which have very similar areas of sensitivity.
The depths of negative velocity discontinuities (km) reported by studies with overlapping Fresnel zones are presented for regions with apparent discrepancies (1–8, Figure 5). The studies are labeled as follows: ScS [Bagley and Revenaugh, 2008], SSlong [Rychert and Shearer, 2011], SSshort [Schmerr, 2012], and transects [Gaherty et al., 1999; Tan and Helmberger, 2007]. These apparent discrepancies are plotted in Figure 5. Depths reported here represent regional summaries, which do not perfectly reflect discrepancies in individual measurements, e.g., region 8 where individual measurements are in some cases in agreement.
90 ± 8
72 ± 8
112 ± 7
86 ± 8
108 ± 7
65 ± 11
107 ± 8
65 ± 11
60 ± 20
65 ± 7
108 ± 7
60 ± 20
58 ± 13
 Although there is much agreement and sensitivity arguments can explain discrepancies between overlapping and adjacent results where results are reported, no singular dynamic model has been illuminated to explain the evolution of the oceanic lithosphere (Figure 6), either how it thickens with age or the mechanism that defines it. Does a sharp negative discontinuity exist everywhere? How does this relate to the thickening of the plate? What is the role of small-scale dynamics? Is the sharp constant depth boundary or the age dependent boundary related to a variation in viscosity that defines the plate?
 Indeed, greater complexity is similarly suggested by discrepancies in the presence/absence of the observed boundary (Figure 5). Existence discrepancies are determined by comparing the region where SS bounce point coverage is high (>500 waveforms for long period SS) and yet no discontinuity is resolved (long or short period SS), including the one “non-observation” from ScS [Bagley and Revenaugh, 2008], with the regions where a strong discontinuity is positively identified by any of the other studies, and broadly accounting for Fresnel zones. Again, sensitivity arguments may provide some explanation, including different sensitivity to velocity gradient thickness (short period SS is sensitive only to the sharpest velocity gradients, <20 km), variation in discontinuity depth (long period SS is sensitive to some wide variations in comparison to others), or even multiple discontinuities in depth (precluded owing to simple long period SS parameterization). However, the large, sharp velocity drop proposed by Pacific transect studies should also be imaged by other techniques, in particular short period SS. One possibility is that structure is more complicated, and anisotropy may play a role.
3.1. Anisotropy Hypothesis Test – Method
 We propose that the shallow discontinuity at relatively constant depth may have a strong anisotropic component. To test the possibility we stack long and short period SS data with bounce points approximating a Pacific transect study (blue rectangle in Figure 5) [Tan and Helmberger, 2007] into azimuthal bins. In this region bounce point coverage is high, and a strong negative constant depth discontinuity has been proposed [Gaherty et al., 1999; Tan and Helmberger, 2007] but is not resolved or intermittently resolved by SS [Rychert and Shearer, 2011; Schmerr, 2012].
 For the long period SS stacks we use shallow events (<75 km depth) from epicentral distances 90–140° in the IRIS FARM database 1990–2007. We Hilbert transform the transverse component, low-pass filter at 0.1 Hz, normalize to unit amplitude, flip polarity of the negative pulses, align on peak amplitude in an 80 s window around the theoretical SS arrival, weight based on signal-to-noise, and stack. Signal-to-noise ratios are calculated using a window 285 s to 60 s before the SS peak, and waveforms with signal-to-noise ratios <4 are rejected. 4056 waveforms fit our selection criteria with bounce points approximating the study region of a Pacific transect study (blue rectangle in Figure 5) [Tan and Helmberger, 2007]. We divide these into azimuthal bins, one roughly parallel to (25 ± 30°) (1074 waveforms) and one roughly perpendicular (295 ± 30°) (972 waveforms) to the transects [Gaherty et al., 1999; Tan and Helmberger, 2007].
 The character of the SS waveform is modeled using a reference phase convolved with a lithospheric operator (Figure 2) [Shearer, 1996; Rychert and Shearer, 2011]. We invert the 100 s preceding the SS waveform peak for the best fitting seismic velocity discontinuity using a grid search over discontinuity depth (15–140 km), amplitude (total S velocity change, positive or negative), and sharpness.
 For the short period SS stacks we also use shallow events (<75 km depth) from the IRIS-Data Management Center. We use events with magnitudes Mw ≥ 5.8 occurring from 1990 to 2008 that sampled beneath the Pacific at an epicentral distance of 90–170°. The station response is deconvolved from each record and the horizontal components of motion were rotated into the source-receiver azimuth to obtain transverse component seismograms. The resulting seismograms are differentiated twice to obtain acceleration, and band-pass filtered with corners at 0.1 Hz and 0.02 Hz. The maximum SS amplitude of each record is then normalized to one. Each record is weighted by a signal-to-noise ratio determined by comparing the maximum enveloped amplitude in a ±15 s window around the theoretical SS arrival time to a 100-s noise window beginning 200 s before the theoretical S or Sdiff arrival time. Seismograms with a signal-to-noise ratio <3.0 are eliminated from the data set. The records from each event are visually inspected to determine the polarity of the SS pulse and eliminate poorly formed SS arrivals. Seismograms with epicentral distances between 115° and 140° are excluded from the data set to avoid interference with the phases ScS410ScS and ScS660ScS. The final data set consists of 30,423 seismograms that sample across the Pacific plate [Schmerr, 2012], though we only use a smaller subset of this data that sample within the Tan and Helmberger  corridor and also a 2000 km wide zone around Hawaii for this analysis [Schmerr, 2012] (Figure 5). The data are aligned on the theoretical arrival time for a precursor at 80 km depth and stacked into 30 degrees azimuthal bins. Finer azimuthal binning is permitted in this case since short period SS does not require as many waveforms to resolve discontinuities, though bins with <50 records are under sampled owing to the non-uniform distribution of sources and receivers around the Pacific. Stacking algorithms and methodology are described in Schmerr and Garnero . Stacked energy above the 95% confidence bound is evaluated using a bootstrap-resampling algorithm [Efron and Tibshirani, 1986].
3.2. Anisotropy Hypothesis Test – Result
 In the long period SS inversions, the only resolvable feature in the 2 azimuth bins is a 5.8% velocity increase (positive contrast) at 70 ± 10 km depth in the 295° bin (Figure 7), i.e., opposite polarity and azimuth to the Pacific transect studies. Long period SS do not resolve a discontinuity parallel to the direction of the Pacific transects, but this is likely due to limitations of the parameterization which cannot resolve the two discontinuities in depth that likely exist in this region, i.e., one sharp at 60 ± 20 km depth, and a deeper more gradual velocity drop (Figure 3) [Gaherty et al., 1999; Tan and Helmberger, 2007].
 In the short period SS stacks a polarity reversal is resolved. A negative discontinuity is imaged at ∼70 km depth only for azimuths near 45° (same polarity and azimuth as Pacific transect studies) (Figure 8). In the remainder of the azimuth bins a positive polarity discontinuity or no discontinuity is resolved in the 70 ± 20 km depth range.
 These SS results at long and short period agree with each other (positive polarity discontinuities at azimuths roughly perpendicular to the Pacific transects) and also with Pacific transect studies [Gaherty et al., 1999; Tan and Helmberger, 2007] (negative discontinuities at azimuths roughly parallel to the Pacific transects). The observed polarity change with back-azimuth suggests an anisotropic boundary. The simplest explanation involves one anisotropic layer and one isotropic layer, though many more complicated scenarios may exist. For simplicity we explore the possibility of either shallow or deep anisotropy.
3.3. Anisotropy Hypothesis Test – Discussion, Deep versus Shallow
 One explanation involves shallow azimuthal anisotropy, typically ascribed to frozen–in lattice preferred orientation (LPO), reflecting ancient plate motions. Fossil spreading direction in this region is ∼265° [Müller et al., 2008], which does not explain our observations, producing negative polarities near ∼175° rather than the observed ∼445° (Figure 8). Although, fossil spreading direction does rotate to 305° in the southwestern most corner of the transect, our result is not likely dominated by such a small region. Plume related complications could play some role in obliterating fossil alignment [Collins et al., 2012], although this scenario is not likely over the large area as sampled by our waveforms.
 Instead, shallow anisotropic alignment near the direction of absolute plate motion (APM), is more consistent with our observations. Average APM is about ∼295° in this region, and alignment at ∼315° provides the closest match to the observed polarity flip. SS-precursor polarities from synthetic seismograms, reflected from an anisotropic shallow layer (hexagonal symmetry with fast axes oriented at 315°) and an isotropic, average velocity deeper layer [Keith and Crampin, 1977] produce the polarity flip observed in short period SS stacks, and in agreement with transect polarities and long period SS inversions (Figures 8 and 9). One exception is the short period SS bin centered at 115°, which has >600 waveforms and yet does not resolve the predicted positive polarity discontinuity.
 The amplitudes of many individual short period SS stacks are not resolved sufficiently for interpretation, and so we focus on polarity rather than absolute amplitude. However, the amplitudes of long period SS waveform stack (∼5.8% increase at azimuth ∼295°) and Pacific transects (∼5.7% decrease at azimuth ∼25°), provide some constraints on velocity structure. These ∼6% velocity drops were calculated assuming isotropic discontinuities. However, comparison to synthetic amplitude predictions for an azimuthally anisotropic lithosphere and isotropic asthenosphere (Figure 8) suggests a ∼2–5% isotropic velocity drop combined with mild to moderate anisotropic alignment (∼30–60%) provides the best fit to observed amplitudes (Figure 9b), in other words, a small velocity increase (∼1–3%) for waveforms traveling at ∼295° azimuth and a velocity decrease of (∼5.7%) for waveforms traveling at ∼25–45° azimuth. This asymmetric velocity model, i.e., greater velocity decrease than increase is an effect of the large predicted positive SS precursor/SS amplitude ratios that are caused both by incidence angle and splitting of the main SS phase. This is demonstrated by amplitude ratios created at a discontinuity between an anisotropic lithosphere and an isotropic asthenosphere with average velocity (symmetric velocity model), where positive amplitudes are much bigger than negative amplitudes (Figure 8).
 Alternatively, an anisotropic deeper layer could explain our results (Figures 8 and 9). Typically, mantle flow driven by plate motion is expected to produce LPO in the direction of APM [Blackman and Kendall, 2002]. However, neither deep layer alignment in the direction of APM or fossil spreading explain the observed polarity flip, producing negative polarities at ∼115° or ∼85°, respectively, instead of the observed 25–45°. Instead, anisotropic fast directions perpendicular to APM in the deeper layer are required to match the observed polarity flip. Synthetic seismogram amplitudes of SS precursors created at a boundary between a deeper hexagonal anisotropic layer assuming shape-preferred orientation (SPO), with slow velocities in the APM direction and fast velocities in the APM perpendicular plane also match the observed polarity flip (Figure 8). Melt in shear bands may produce anisotropic fast directions oriented perpendicular to the direction of shear [Holtzman et al., 2003; Holtzman and Kendall, 2010]. In reality the shear bands may be aligned sub-vertically [Holtzman et al., 2003; Holtzman and Kendall, 2010] though we do not model this complication here. More dynamic testing of the predicted effects of melting on subsidence and evolution of the lithosphere evolution are required.
3.4. Anisotropy Hypothesis Test – Detailed Analysis of Shallow Alignment
 Our data cannot distinguish between shallow versus deep anisotropy, and in reality the result may be complicated by a combination of the two. In neither case does the required direction of alignment agree with simple expectations from either plate driven mantle flow (deep) [Conrad et al., 2007] or fossil orientation (either shallow or deep). Thin layers of melt may explain the deep case, i.e., fast alignment perpendicular to mantle flow, although whether or not this is a viable mechanism for such a large area of the seafloor is unknown. Here we further investigate the possibility that alignment in the shallow layer could explain the observations.
 We tested whether shallow alignment in the APM direction could have developed in the last ∼40 My since plate motion direction changed, based on the Hawaii volcano line. Channel flow modeling after Podolefsky et al.  indicates that the transition from diffusion to dislocation creep is grain size dependent, and begins shallower than 70 km depth in 60 My old lithosphere [Behn et al., 2009]. Strain accumulation at depths shallower than 70 km could be large enough (>1) to produce a significant lattice-preferred orientation (LPO) in the APM direction beneath 60 My old lithosphere as it ages to 100 My for a plate model (80 km thick plate) with mantle potential temperature 1450°C [Podolefsky et al., 2004]. A thicker plate (90–100 km) or cooler temperatures (1350°) increases the depth at which strain accumulation is >1 by ∼10 km, within the error bars of our model (±20 km). In addition, accounting for the original fossil alignment (assuming it existed) via a modeling approach that includes directionality [Becker, 2006; Becker et al., 2008], would likely enhance LPO, since the fossil direction (∼265°) is close to the required direction of alignment (∼295°). The LPO would be frozen-in to the base of the plate today, and the observed discontinuity would be owing to variation from shallow frozen-in alignment to a deeper less-aligned asthenosphere (Figure 9). Lithosphere-asthenosphere decoupling is implied, and a lack of LPO in APM direction at depth today could be caused by complicated plume flow in the low viscosity asthenosphere [Morgan et al., 1995].
 A model of strong shallow anisotropy, either due to fossil or APM alignment does not agree with the general view of anisotropy beneath the oceans. Radial anisotropy peaks beneath the oceans at ∼125 km for 25–100 My old oceanic lithosphere in surface wave models [Nettles and Dziewonski, 2008] (blue line in Figure 3b), and strong anisotropy is typically expected in the asthenospheric layer, below the plate. However, global averages do not hold in this region. Surface wave velocities averaged over the sensitivity regions of the Pacific transects [Tan and Helmberger, 2007] are reduced at deeper depths (Figure 3b). Radial anisotropy in what is typically thought to represent asthenospheric depths ∼100–150 km is reduced in comparison to the rest of the global oceanic averages (pink compared to red, blue, and green lines in Figure 3b). Indeed, other global model also finds increased radial anisotropy at shallow depth in comparison to surrounding regions [Lekic and Romanowicz, 2011b]. Azimuthal anisotropy is also reduced at deeper depths (∼120 km) in comparison to surrounding regions [Becker et al., 2003]. One possible cause for reduced radial anisotropy beneath the Pacific transects at 100–150 km depth in comparison to global averages could be that upwelling from the Hawaiian plume interferes with alignment from shearing in the direction of absolute plate motion. This type of deviation from plate driven flow has been suggested by global azimuthal anisotropy studies [Maggi et al., 2006b] and geodynamic modeling of Poiseuille flow [Höink and Lenardic, 2010].
3.5. Anisotropy Hypothesis Test – Implications
 A model of shallow anisotropy in the direction of APM implies that plate motion related strain is recorded strongly in the base of the plate. In other words, alignment just below the plate was once strong and plate driven, and has subsequently been frozen-in. Flow in the lower viscosity asthenosphere may occur in a less azimuthally coherent way, and strong decoupling between the layers is implied. Complicated deeper mantle flow could be caused by the plume influence in the region or it could be a general phenomenon such as the onset of small scale convection often invoked to explain the deviation from half-space cooling. A low viscosity asthenosphere may be caused by hydration [Hirth and Kohlstedt, 1996]. Indeed, hydration may enable grain boundary sliding at ∼70 km depth beneath the oceans, easily explaining the mild velocity drop inherent to this model [Karato, 2012].
 In the deeper anisotropy model, the melt bands might be created in regions where plume related heat/melting sharpens the boundary. This type of phenomenon would not be expected ubiquitously in the asthenosphere given that asthenospheric azimuthal anisotropy is not generally perpendicular to plate motions [Maggi et al., 2006a]. However, intermittent observations of sharp boundaries could easily be explained by plume related heating and melt in shear bands.
 The case for an anisotropic discontinuity and a better understanding of the location (shallow or deep) of anisotropy and also defining mechanism will be more complete with additional measurements of SS azimuthal dependence over the entirety of the Pacific. Although bounce point coverage is very high beneath the Pacific corridor studied here, azimuthal sampling across the Pacific is still insufficient for a basin-wide analysis. However, the global database of SS precursors is steadily increasing with the deployment of new arrays and ongoing seismicity; future large data sets will allow us to achieve anisotropic structure with higher resolution across the Pacific and other ocean basins.
 We thank Hitoshi Kawakatsu for helpful comments on an earlier version of the manuscript and Prakash Kumar for providing his receiver function results and locations. We thank Meredith Nettles for providing her surface wave velocity model. We acknowledge funding from the Natural Environment Research Council – UK (NE/G013438/1) [CAR] and NASA postdoctoral program [NS]. Data are obtained from the IRIS Data Management Center. Figures made using GMT.