Mantle P-wave velocity structure beneath the Hawaiian hotspot

Earth and Planetary Science Letters 303 (2011) 267–280

Contents lists available at ScienceDirect

Earth and Planetary Science Letters j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / e p s l

Mantle P-wave velocity structure beneath the Hawaiian hotspot Cecily J. Wolfe a,⁎, Sean C. Solomon b, Gabi Laske c, John A. Collins d, Robert S. Detrick d, John A. Orcutt c, David Bercovici e, Erik H. Hauri b a

Hawaii Institute of Geophysics and Planetology, University of Hawaii at Manoa, Honolulu, HI 96822, USA Department of Terrestrial Magnetism, Carnegie Institution of Washington, Washington, DC 20015, USA Cecil H. and Ida M. Green Institute of Geophysics and Planetary Physics, Scripps Institution of Oceanography, University of California, San Diego, CA 92093, USA d Department of Geology and Geophysics, Woods Hole Oceanographic Institution, Woods Hole, MA 02543, USA e Department of Geology and Geophysics, Yale University, New Haven, CT 06520, USA b c

a r t i c l e

i n f o

Article history: Received 14 July 2010 Received in revised form 14 December 2010 Accepted 6 January 2011 Available online 5 February 2011 Editor: L. Stixrude Keywords: Hawaii hotspot mantle plume finite-frequency tomography

a b s t r a c t Three-dimensional images of P-wave velocity structure beneath the Hawaiian Islands, obtained from a network of seafloor and land seismometers, show an upper-mantle low-velocity anomaly that is elongated in the direction of the island chain and surrounded by a high-velocity anomaly in the shallow upper mantle that is parabolic in map view. Low velocities continue downward to the mantle transition zone between 410 and 660 km depth and extend into the topmost lower mantle, although the resolution of lower mantle structure from this data set is limited. Comparisons of inversions with separate data sets at different frequencies suggest that contamination by water reverberations is not markedly biasing the P-wave imaging of mantle structure. Many aspects of the P-wave images are consistent with independent tomographic images of S-wave velocity in the region, but there are some differences in upper mantle structure between P-wave and S-wave velocities. Inversions without station terms show a southwestward shift in the location of lowest P-wave velocities in the uppermost mantle relative to the pattern for shear waves, and inversions with station terms show differences between P-wave and S-wave velocity heterogeneity in the shallow upper mantle beneath and immediately east of the island of Hawaii. Nonetheless, the combined data sets are in general agreement with the hypothesis that the Hawaiian hotspot is the result of an upwelling, high-temperature plume. The broad upper-mantle low-velocity region beneath the Hawaiian Islands may reflect the diverging “pancake” at the top of the upwelling zone; the surrounding region of high velocities could represent a downwelling curtain; and the low-velocity anomalies southeast of Hawaii in the transition zone and topmost lower mantle are consistent with predictions of plume tilt. © 2011 Elsevier B.V. All rights reserved.

1. Introduction The Hawaiian–Emperor seamount and island chain is one of the major volcanic features on Earth. Spanning thousands of kilometers across the northern Pacific, the Hawaiian chain records ageprogressive volcanism for N75 My (Duncan and Keller, 2004) and exhibits a broad, ~ 1000-km-wide region of elevated topography, known as the Hawaiian Swell (Crough, 1978; Dietz and Menard, 1953; Wessel, 1993), that surrounds the locus of current volcanism. Located far from any plate boundary in the interior of the Pacific plate, the Hawaiian hotspot has been suggested to be the product of a mantle plume: a localized upwelling of hot, buoyant material from Earth's deep mantle (Morgan, 1971). Estimated to transport a larger buoyancy flux than any other present-day plume (Bourdon et al., 2006; Davies, 1988; Sleep, 1990) and erupting magmas that differ

⁎ Corresponding author. Fax: +1 808 956 3188. E-mail address: [email protected] (C.J. Wolfe). 0012-821X/$ – see front matter © 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.epsl.2011.01.004

chemically and isotopically from mid-ocean ridge basalts (e.g., Hart et al., 1992; Hauri, 1996; Hofmann, 1997; Kurz et al., 1983; Sobolev et al., 2005), the Hawaiian hotspot has been the focus of a number of geochemical and geodynamical studies that have attempted to define its origin. Yet, a debate has long raged about whether Hawaii's volcanism is plume-related or is instead the consequence of more shallow processes (e.g., Norton, 2007; Turcotte and Oxburgh, 1973) because of a lack of crucial geophysical data. Given that Hawaii has been proposed by many to be the archetypical example of a mantle plume, understanding its formation would resolve the debate over whether mantle plumes exist as a mode of mantle convection in the Earth, a topic that has important implications for the transfer of mass and energy between the upper and lower mantle as well as the thermal and compositional evolution of the Earth (e.g., Davies, 1999; Lay et al., 2008). Seismological imaging of mantle structure provides a means for directly testing ideas for the origin of a hotspot (Nataf, 2000; Nolet et al., 2007). The presence of a narrow, vertically continuous zone of low seismic velocities in the underlying mantle indicative of higher-

268

C.J. Wolfe et al. / Earth and Planetary Science Letters 303 (2011) 267–280

than-normal temperatures, anomalous mantle composition, or partial melt beneath a hotspot is supportive of the mantle plume hypothesis (Boschi et al., 2007; Courtillot et al., 2003; Ito and van Keken, 2007). Global seismic tomography suggests that low seismic velocities may extend beneath the Hawaiian hotspot from the upper to the lower mantle (e.g., Li et al., 2008; Montelli et al., 2004, 2006; Nolet et al., 2007), but there is controversy about how well, and to what depth, narrow plumes may be resolved by global studies (Boschi et al., 2006; van der Hilst and de Hoop, 2005). Moreover, global models of mantle structure are limited in their resolution around Hawaii by the sparseness of wave paths, given the lack of seismometers on the ocean floor and the great distances between the islands and most circum-Pacific earthquake sources. Early seismological investigations conducted with stations on the Hawaiian Islands were limited in geographic coverage and numbers of instruments. Mantle seismic structural studies conducted with island stations have included body-wave tomography (Tilmann et al., 2001; Wolfe et al., 2002), surface wave dispersion (Woods and Okal, 1996), shear-wave splitting (Walker et al., 2001, 2003), and receiver function analyses (Li et al., 2000; Li et al., 2004). Early investigations with seafloor sensors addressed surface wave imaging (Laske et al., 1999, 2007), mantle discontinuity structure from receiver functions, and anisotropy from shear-wave splitting (Collins et al., 2002), but the networks deployed for these studies had few instruments. Given the limitations of past data, fundamental characteristics of mantle seismic structure in the vicinity of Hawaii had not been fully resolved. Results of some prior regional seismic studies nonetheless support the mantle plume hypothesis at Hawaii. P-to-s conversions at the mantle transition zone observed with receiver functions (Collins et al., 2002; Li et al., 2000; Shen et al., 2003) and SS-wave precursors (Schmerr and Garnero, 2006) indicate an area of thinned transition zone around Hawaii, consistent with higher-than-normal temperatures, although the horizontal scale of such an anomaly has not been well defined by such studies. Lower than normal velocities in the lower lithosphere and asthenosphere beneath the Hawaiian Swell are indicated by surface-wave imaging conducted with a network of seafloor instruments that spanned regions on and off the swell (Laske et al., 1999, 2007), consistent with heating from below. In this study, we present results from P-wave tomographic imaging of the mantle beneath the Hawaiian hotspot from the Plume-Lithosphere Undersea Melt Experiment (PLUME), a deployment of land and ocean-bottom seismometers (OBSs) that provided dense wave-path coverage beneath the Hawaiian region. This paper is complementary to the S-wave tomographic images presented earlier from the same experiment (Wolfe et al., 2009).

2. Data The PLUME experiment (Laske et al., 2009) consisted of two yearlong deployments of a dense, large-aperture network comprising three-dozen broadband OBSs and the concurrent operation of 10 portable broadband seismometers on the Hawaiian Islands (Fig. 1). All instruments operated continuously to record teleseismic and local earthquakes. Additional information on the experiment has been provided by Laske et al. (2009), Leahy et al. (2010), and the supporting online information of Wolfe et al. (2009). For the work presented in this paper, we analyzed direct P waves on OBS vertical components. We did not observe any well-separated PKP phases that yielded adequate signal-to-noise for obtaining measurements on either the ocean-bottom or land seismometers. On the first PLUME deployment, several OBSs did not return vertical data. For three instruments we used P waves observed on the radial component as a substitute at moderate and high frequencies (but not at low frequencies, owing to the greater contaminating influence of water reverberations on radial-component waveforms).

Fig. 1. (A) Locations of seismometers used in this P-wave tomography study. Land stations are indicated by blue triangles and OBSs by red (first deployment, 2005–2006) or yellow (second deployment, 2006–2007) circles. Only stations that successfully recorded either vertical-component data or two horizontal components are shown. Bathymetry is taken from an updated version of Smith and Sandwell (1997). (B) Corrected mean station delays (0.05–0.1 Hz), adjusted to vertical incidence. Early arrivals are shown by blue circles and late by red triangles, with symbol size scaled linearly to the magnitude of the delay (see scale at lower left).

We measured the relative P-wave arrival times from large (moment magnitude Mw ≥ 5.5) teleseismic earthquakes in multiple frequency bands, approximately 0.05–0.1 Hz, 0.08–0.12 Hz, 0.1–0.2 Hz, and 0.4–1 Hz, with the multichannel cross-correlation method of VanDecar and Crosson (1990). Because high background noise on the seafloor around Hawaii (e.g., Stephen et al., 2003; Webb, 1998) reduces the number of observable waves at frequencies above 0.1 Hz, we obtained the largest number of picks (2148 measurements) and the best azimuthal distribution of earthquakes in the frequency band ~0.05– 0.1 Hz (Fig. 2). In the frequency band 0.08–0.12 Hz we obtained 1580 measurements from 51 earthquakes, and we augmented this data set with 220 measurements at a few key back azimuths from eight earthquakes at 0.05–0.1 Hz. The number of measurable phases is greatly reduced at higher frequencies, and it was not possible to construct a high-frequency data set having a sufficient azimuthal distribution of sources for good tomographic resolution with measurements only at 0.4–1 Hz. We therefore constructed a “combined higher-frequency” data set, with 814 measurements, by combining moderate-frequency, 0.1–0.2 Hz, and high-frequency, 0.4–1 Hz, data. The window lengths of the 0.1–0.2-Hz data were generally taken as less than 10 s and centered near the strongest-amplitude P-wave arrival with 1-s tapering at the edges, in order to minimize any possible influence of later-arriving water reverberations.

C.J. Wolfe et al. / Earth and Planetary Science Letters 303 (2011) 267–280

A

-> uth

20

E AST

WEST


NORTH

40 60

80 100

120 140 an gul ar d i sta nce

SOUTH

B

-> uth

20

40 60

80 100

120 140 an gul ar d i sta nce

SOUTH

E AST

WEST


NORTH

269

The analysis of multiple frequency bands was undertaken to assess the possible effect of water reverberations on P-wave delay times, which should exhibit a strong frequency dependence. Blackman et al. (1995) discussed how the amplitude of the first water-reverberated Pw wave is stronger on hydrophone or differential pressure gauge data owing to constructive interference at the seafloor and weaker (one-third the amplitude of the primary P wave) on the vertical-component seismometer owing to destructive interference at the seafloor. Contamination by water reverberations is thus likely to be less important for the vertical seismometer data analyzed in our study. For the ~4.5–5.5-km water depths of OBSs in the PLUME experiment, the Pw reverberation arrives 6–7 s, and the smaller amplitude Pww reverberation (with negative polarity) arrives 12–14 s, after the P wave. These reverberations can be identified in plots of 0.4–1-Hz waveforms (Figs. S1 and S2, supporting online material), with highly variable amplitudes (both strong and weak) owing to the effects of focusing, defocusing, multipathing, or scattering by topography. Such focusing and defocusing effects are likely to be less important at lower frequencies. On the 0.1–0.2-Hz waveforms with ~7-s period (equivalent to the time delay of the first-arriving water reverberation), we did not notice any strong and systematic reverberations later in the OBS waveforms compared with land seismometer records (Figs. S3 and S4). On OBS deployments, water reverberations are best identified in highfrequency data, but at low frequencies for which the wave period exceeds the separation of the P and Pw phases, direct and reverberated phases overlap and become indistinguishable. Water reverberations thus have the potential to bias delay-time measurements to an unknown degree. Our combined higher-frequency data set (0.1–0.2 Hz, and 0.4–1 Hz with shorter window lengths) was included in large part to minimize bias by water reverberations, but it suffers from greatly reduced wave-path coverage. Lower-frequency data sets (0.05–0.1 Hz or 0.08–0.12 Hz with 10–20-s window lengths) have more optimal wave-path coverage but may contain greater bias. Comparison of higher- and lower-frequency data sets and images from their separate inversions is therefore important to assess potential bias, as it is expected that strong bias would produce differences among images. Wave-path coverage for the 0.05–0.1-Hz data set is indicated in Fig. 3. Direct P phases provide crossing wave paths down to 600–700 km depth, but at depths greater than ~1000 km direct P-wave paths diverge outward. Wave-path coverage is sparser in the combined higherfrequency data set (Fig. S5), given the smaller number of measurements (about one-third). We note that the resolution of lower mantle structure is more limited in this direct P-wave study than in the S-wave study of Wolfe et al. (2009), where SKS phases provided additional crossing wave paths down to 1500 km depth that complemented the direct S-wave data. The seismic waveforms were corrected for the differing instrument responses, and the measured relative arrival times were corrected for estimated variations in station elevation and crustal thickness, as described in the supporting online material of Wolfe et al. (2009). Corrected mean P-wave station delays of about ±1 s reflect upper mantle heterogeneity and indicate relatively low velocities beneath and to the west of the main Hawaiian Islands and high velocities to the east of Hawaii and at distant stations around the margins of the swell (Figs. 1B and S6). The pattern of corrected mean P-wave station delays in Fig. 1B correlates well (with a correlation coefficient of 0.77) with S-wave station delays reported by Wolfe et al. (2009), although S-wave station delays are larger by factors of 2–3. One important difference in patterns is that P-wave arrivals are not as delayed relative to other stations as S-wave arrivals on the island of Hawaii in the band 0.05–0.1 Hz.

Fig. 2. Distribution of earthquakes that yielded measurements of P-wave relative arrival times (0.05–0.1 Hz). Azimuthal equidistant projection centered on 20°N, 158°W. (A) Distribution for the first deployment. (B) Distribution for the second deployment.

270

C.J. Wolfe et al. / Earth and Planetary Science Letters 303 (2011) 267–280

beneath the Hawaiian Islands also tended to be delayed, suggesting the presence of low velocities in the transition zone and lower mantle. In the P-wave data set, similar patterns are seen (Fig. S8), though not as commonly as in the S-wave data set. Late P-wave arrivals are observed for wave paths that cross the transition zone and uppermost lower mantle southeast of Hawaii from an earthquake in South America at 120° back azimuth, similar to the delay patterns of SKS phases from earthquakes in this region. An earthquake at 170° back azimuth also sampled this mantle region and yielded strong delays for wave paths crossing beneath Hawaii. 3. Inversion results

Fig. 3. Wave-path coverage (0.05–0.1-Hz data) at intervals of depth between 0 and 900 km.

After correcting for the travel times predicted by the IASPEI91 (Kennett and Engdahl, 1991) one-dimensional velocity model and removing the mean station delay, P waves display azimuthal delays of ±0.8 s and are 3 times smaller in magnitude than the observed S-wave delays; the standard deviation of S-waves delays is 0.92 s and that of P-waves is 0.29 s. Because the measurement noise levels are similar for both datasets (both for the 0.05–0.1 Hz band), the signal-to-noise ratio of the P-wave delay-time data set will be lower (by one-third) than that for the S-wave data set. As discussed below, the azimuthal patterns of relative P-wave delays display both similarities to and differences from those of the S-wave data set. With the data from the first PLUME OBS deployment, Wolfe et al. (2009) observed that S phases having wave paths that passed beneath Hawaii tended to be delayed by up to several seconds. But the P-wave data from the first deployment do not exhibit the same pattern (Fig. S7). P and S delay-time patterns both show positive delays for paths that cross beneath Hawaii at some back azimuths. In contrast to the S-wave delays, however, at westerly back azimuths, a region of early P-wave delays is seen on and around the island of Hawaii. From southeasterly back azimuths, a region of late P-wave arrivals is observed on and around the island of Hawaii. These different P-wave patterns are also seen on the more limited combined higher-frequency data set, and they thus are unlikely to result from contamination by water reverberations. With data from the second PLUME OBS deployment, Wolfe et al. (2009) observed that S and SKS phases having wave paths that crossed

The relative arrival times were inverted for heterogeneity in P-wave velocity VP, damped earthquake relocations, and origin time terms using a finite-frequency methodology (Dahlen et al., 2000; Hung et al., 2004; Montelli et al., 2004). We employed the same model parameterization and technical approach as that described by Wolfe et al. (2009), and the reader is referred to the supporting online material for that paper for further information. For the combined higher-frequency data set, the Gaussian wavelet used for calculating the finite-frequency kernel was varied according to the frequency band of measurement to allow consistent merging of data of differing sensitivity. One strategy for mitigating the effect of streaking of shallow heterogeneity deeper into a model is to include station terms to partially absorb shallow structure, but this approach comes at the expense of decreased resolution at shallow depths. Moreover, station terms may not successfully absorb structure at 100–200 km depth if velocity variations are strong. We therefore performed inversions both with (Fig. 4) and without (Fig. 5) station terms. Station terms have a noticeable influence on shallow structure, but they affect less on deep structure, where low velocities beneath and southeast of Hawaii are imaged in the transition zone and uppermost lower mantle in inversions with all of the different data sets (Fig. 6). A model from the 0.08–0.12-Hz data set is shown in Fig. S9, and models from the combined higherfrequency data set are displayed in Figs. 7 and S10. Comparison of inversions with the separate 0.05–0.1 Hz, 0.08– 0.12 Hz, and combined high-frequency data sets suggests that contamination by water reverberations is not biasing the imaging of upper mantle structure to any major extent. The combined higher-frequency P-wave images do not reveal any novel patterns but rather appear as degraded versions of the 0.05–0.1-Hz images, likely because of poor resolution given the sparse wave-path coverage. We performed a two-step inversion to assess the vertical extent of upper mantle structure, which is not well constrained by P-wave data alone. In the first step, we inverted for a model in which all structure was “squeezed” between 50 and 250 km depth. In the second step, we corrected the observations for the effects of the squeezed model and inverted for an additional residual model. The two-step model (Fig. S11) continues to recover a low-velocity anomaly in the transition zone and lower mantle (although amplitude is slightly decreased), but much of the heterogeneity at 300 km depth can be squeezed to shallower depths, similar to what was observed with the inversions of S-wave delay times (Wolfe et al., 2009). However, in the S-wave inversions, the extremely large amplitude (±8%) of shallow structure in a model squeezed between 50 and 250 km depth became inconsistent with the amplitude of variations constrained by surface wave observations, suggesting that such a squeezed model was unlikely to be a good representation of regional structure. We also performed a combined two-step inversion to assess whether the lower mantle structure can be squeezed to more shallow depths (Fig. 8). The first step squeezed all structure above 700 km, and a second step was an inversion for a residual model. The results demonstrate that some structure in the topmost lower mantle can be squeezed more shallowly into the transition zone and above, although smaller-amplitude heterogeneity remains in the lower mantle.

C.J. Wolfe et al. / Earth and Planetary Science Letters 303 (2011) 267–280

271

Fig. 4. Mantle P-wave velocity heterogeneity at selected depths from 100 to 1200 km for a model with station terms (0.05–0.1-Hz data). The scale of heterogeneity is indicated in the upper right corner of each panel, and the station terms are indicated as scaled symbols in the top left panel.

4. Resolution tests We conducted a range of resolution tests, in which we constructed a synthetic model of velocity heterogeneity, calculated synthetic arrivaltime data using finite-frequency theory, added normally distributed random errors (with a mean of zero and standard deviation of 0.1 s), and re-inverted synthetic data for three-dimensional structure using the same smoothing constraints as with the actual data. The use of the linear finite-frequency theory of Dahlen et al. (2000) in calculating the synthetic data takes into account the first-order effects of wavefront healing, by which diffraction acts to fill in or heal irregularities in the wavefront (e.g., Hung et al., 2001; Nolet and Dahlen, 2000; Wielandt,

1987). The Fresnel zones of these P waves increase in horizontal dimension with depth, and as a consequence of wavefront healing objects much smaller than the width of the Fresnel zone will not appreciably influence the observed travel times. One type of test used as input a low-velocity anomaly in the form of a vertical cylinder for which the normalized velocity contrast varied with horizontal distance r from the cylinder's axis as a Gaussian of half-width w [δVP/VP ~ (δVP/VP)0exp(−r2/w2)] from a maximum anomaly (δVP/VP)0 at its center. Such a simple, vertically coherent structure is easier to resolve at small horizontal scales than more complex structures. The resolution of the 0.05–0.1-Hz P-wave data is worse than the resolution of the S-wave data set described by Wolfe et al. (2009), because

272

C.J. Wolfe et al. / Earth and Planetary Science Letters 303 (2011) 267–280

Fig. 5. P-wave velocity heterogeneity at the same depths as shown in Fig. 4 but for a solution without station terms. See Fig. 4 for further information.

of the lack of deep crossing phases analogous to SKS and because the wavelengths of 0.05–0.1-Hz P waves are twice as large and the effects of wavefront healing will be correspondingly greater. Figure S12 shows a resolution test for a cylinder extending to 2000 km depth with 100-km half-width, demonstrating that such a short-wavelength feature is resolvable only at depths shallower than 400 km, likely a result of wavefront healing. When the half-width of such a cylinder is increased to 200 km, structure is resolved deeper, to about 800 km depth (Fig. S13). Similarly, a vertical cylinder with 200-km half-width and limited vertical extent (400–800 km depth) is also resolvable (Fig. S14). Figs. S15 and S16 show tests with a cylinder of limited vertical extent between 700 and 2000 km depth centered southeast of Hawaii, with half-widths of 200 km

and 300 km, respectively. Only the longer-wavelength lower mantle structure is resolvable, and then only down to about 1000 km depth. We used “checkerboard” tests (models with rectangular blocks in map view that are alternately 1% higher or lower in velocity than average) to gain additional information on how resolution may vary with depth if structure is more complex. In one checkerboard test (Figs. S17 and S18), we adopted as an input model a single-layer, vertically coherent horizontal checkerboard (alternating every 4°) above 700 km depth, and the output shows that such structure is well resolved with little vertical streaking into the lower mantle. In another checkerboard test (Figs. S19 and S20), the input model was a single-layer, vertically coherent horizontal checkerboard (alternating every 7°) between 700 and

C.J. Wolfe et al. / Earth and Planetary Science Letters 303 (2011) 267–280

273

Fig. 6. Additional representations of the model shown in Fig. 5. Top panels are plots in map view, showing the two traces of the vertical cross sections (with scale given at upper right). Middle panel is a northwest-to-southeast vertical cross section. Bottom panel is a southwest-to-northeast vertical cross section.

2000 km depth. The results show that complex long-wavelength lower mantle structure is marginally resolvable and only in the topmost portions of the lower mantle, and that the recovered amplitude is one-fourth the input amplitude. Finally, we show a test with two different layers of checkerboard structure (Figs. S21 and S22), which demonstrates that complex lower mantle structure remains marginally resolvable in the presence of complex upper mantle structure. These P-wave tests demonstrate that inversion solutions have better resolution in the upper mantle and transition zone than in the lower mantle. There is higher resolution of simple lower mantle structures, such as a vertical cylinder, than of complex lower mantle structure, such as a checkerboard. These tests indicate that structures similar to the simple, low-velocity anomaly southeast of Hawaii may be detectable

with our data, although the resolution will be less than optimal. Given the better resolution in the transition zone, it is possible that much of the P-wave structure at greater depth could be generated by transition-zone structure beneath Hawaii rather than a lower mantle anomaly, as also indicated by the two-step inversion in Fig. 8. 5. Sampling of the Pacific large low-shear-wave-velocity province Our inversions are based on the premise that structural variations occur only within the model volume beneath the region of Hawaii. The limited spatial aperture of the PLUME network means that Fresnel zones from different stations to a common source will converge and overlap as the source is approached, causing the differential arrival times to lose

274

C.J. Wolfe et al. / Earth and Planetary Science Letters 303 (2011) 267–280

Fig. 7. Upper mantle P-wave velocity heterogeneity at 100, 300, and 400 km depth for the combined high-frequency data set. Left column shows model from an inversion with station terms, and right column shows model from an inversion without station terms. See Fig. 4 for further information.

direct sensitivity to heterogeneity outside the model region. From an examination of differential travel-time sensitivity kernels of direct S waves at stations that are 1° apart, for instance, Marquering et al. (1999) showed that sensitivity is primarily to structure in the near-receiver vicinity. Wolfe et al. (2009) examined the possible aliasing effects of lower mantle structure on SKS phases due to strong, sharp heterogeneity in the region just outside of the model volume, and we present an analogous assessment for P phases. Large low-shear-wave-velocity provinces (LLSVPs) are known to exist in the lowermost 500–1000 km of the mantle beneath Africa and the Pacific (e.g., Garnero et al., 2007; Grand, 2002; Ishii and Tromp, 1999). Whereas global tomographic models are typically smooth and thus do not resolve sharp features,

forward modeling of body waveforms and travel times has suggested that, at least in some locations, the margins of LLSVPs may exhibit large changes in velocity over short lateral distances (e.g., Bréger and Romanowicz, 1998; Garnero et al., 2007; Ni et al., 2002). As the PLUME network is located over the northern margin of the Pacific LLSVP, an important question is whether the observed delay patterns could be generated by the sharp edges of this anomaly. As discussed by Garnero et al. (2007), global models of S-wave structure of LLSVPs are quite consistent, but there is less consistency among models of P-wave structure. Moreover, the fractional amplitudes of LLSVPs in global P-wave models are 3–5 times smaller than for S-wave models, suggesting that any biasing effect from the Pacific LLSVP on P-wave

C.J. Wolfe et al. / Earth and Planetary Science Letters 303 (2011) 267–280

275

Fig. 8. Mantle P-wave velocity heterogeneity for a model from a two-step inversion, with structure at the first step squeezed above 700 km depth (0.05–0.1-Hz data). See Fig. 4 for further information.

delays should be correspondingly smaller. Figs. S22 and S23 show P-wave path segments in the lowermost 1000 km of the mantle superimposed on an updated (S. P. Grand, personal communication, 2009) version of an earlier global shear-wave velocity model (Grand, 2002). It can be seen from the figure that some paths do cross the edge of the LLSVP, but, as discussed in the next section, the sampling of this feature by P-wave paths differs from that for S-wave paths. 6. Comparison of P- and S-wave models Our results improve the fidelity of imaging of VP in the mantle beneath the Hawaiian hotspot region and provide an important

complement to models of S-wave velocity VS. The resolving power of the PLUME P-wave data set is not as high as with the S-wave data set, however, given the smaller delay times and therefore lower signalto-noise of the measurements, sparser wave-path coverage in the topmost lower mantle, and larger wavelengths of P waves at the frequency band (0.05–0.1 Hz) where coverage is best. The P-wave images in this study display some similarities to the S-wave images of Wolfe et al. (2009) (Figs. 9 and 10). For both wave types, in inversions without station terms (Fig. 9), there is a broad low-velocity volume in the upper mantle that is surrounded by a high-velocity anomaly that is parabolic in map view. In both cases, low velocities are imaged in the transition zone and the

276

C.J. Wolfe et al. / Earth and Planetary Science Letters 303 (2011) 267–280

Fig. 9. Comparison between P-wave models and the S-wave model of Wolfe et al. (2009), for inversions without station terms. (Top row) S-wave model. (Second row) P-wave model (0.05–0.1-Hz data). (Third row) P-wave model (0.08–0.12-Hz data). (Fourth row) P-wave model (combined higher-frequency data).

C.J. Wolfe et al. / Earth and Planetary Science Letters 303 (2011) 267–280

topmost lower mantle. The P-wave structure does not extend as deeply into the lower mantle as the S-wave structure, however, and squeezing tests indicate that much of this signal could alternatively be produced within the transition zone, consistent with the differing wave path coverage and weaker resolution at depth of the P-wave data set. Future studies of transition-zone structure, e.g., from receiver function analyses of PLUME data, may place better constraints on transition-zone structure that will help to resolve such ambiguity. Joint inversions of surface-wave and body-wave observations or including PLUME data in inversions of global seismic data will also help to improve the resolution of structures and their variations with depth. Because Poisson's ratio is not constant with depth in the onedimensional IASPEI91 velocity model, predicted teleseismic wave paths from a common set of earthquakes to a common set of stations can be different for P and S waves. In particular, the sampling of the lowermost mantle by P phases in the near-receiver region (Figs. S22 and S23) is different from the sampling by SKS phases in the study by Wolfe et al. (2009). Given these path differences, any similarities in P-wave and SKS-wave delay patterns – or in models derived from such data – should not be caused by outside aliasing from lowermost mantle heterogeneity. It is unlikely that the sharp edges of the Pacific LLSVP are generating the transition-zone and lower mantle structures seen in both P- and S-wave models. A more plausible explanation given the similarity of the VP and VS images is that P- and S-wave delays reflect common isotropic heterogeneity that these wave paths sample within the interior of our model region. Although there are strong similarities in the P-wave and S-wave models, there are also notable differences in upper mantle structure in the region around the island of Hawaii. In the P-wave models from inversions without station terms, the upper mantle velocities are lowest to the southwest of the Hawaiian Island chain, whereas in a comparable S-wave model the lowest local velocities directly underlie the island chain (Fig. 9). The amplitude of the high-velocity upper mantle region to the northeast of Hawaii is also relatively larger in the P-wave images than in the S-wave images (although it should be noted that our tomography resolves only relative velocity variations across the model region, so an arbitrary baseline shift can be added to the models at any given depth). The differences between shallow upper mantle P-wave and S-wave structure beneath and just east of the island of Hawaii are more marked when comparing images with station terms (Fig. 10), which tend to emphasize the azimuthal patterns of relative delay times. Compared with the S-wave model, low velocities are not evident beneath the island of Hawaii in the P-wave model and velocities east of Hawaii are relatively stronger in amplitude. On the basis of the inversion of the higher-frequency data set (bottom row of Fig. 10), we do not believe that these P- and S-wave model differences reflect contamination by water reverberations; nor do they appear to be caused by the larger wavelengths and lower resolution of the 0.05–0.1-Hz P-wave data. Some of the differences may be generated by delay patterns on stations on the island of Hawaii. For example, as described above, the azimuthal patterns of Hawaii Island P-wave delays show early arrivals from earthquakes to the west, whereas this pattern is not present in S-wave data. It may be that the strong region of high velocities immediately to the east of Hawaii Island in the VP models hampers imaging of low velocities beneath Hawaii, particularly if there is a sharp boundary between contrasting features. Higher-frequency data with shorter seismic wavelengths should be better at delineating any sharp features. The changes in the pattern of VP anomalies with frequency shown in Fig. 9 support the idea that the position of the boundary to the highvelocity volume to the east of Hawaii may pose a challenge to tomographic imaging. At 100 km depth, for instance, the boundary position is farther to the southwest at lowest frequency (0.05–0.1 Hz) but is more northeastward at higher frequencies. We reiterate that at the lowest frequencies where wave-path coverage is best (0.05–0.1 Hz),

277

the wavelengths of P waves are about twice as large as the S waves used in the VS model. 7. Discussion Our three-dimensional VP and VS images can be compared with the predictions of simple geodynamic models of thermal plumes. In inversions without station terms, the broad, low-velocity anomaly at shallow (b 200 km) mantle depths (see left column of Fig. 9) remains in agreement with the prediction from geodynamic models of a “pancake” of high-temperature material as the plume impacts and spreads laterally beneath the rigid lithosphere (Davies, 1988; Ribe and Christensen, 1999; Sleep, 1990). Some geodynamic models (e.g., Moore et al., 1998; Ribe and Christensen, 1994) also predict a parabola-shaped curtain of cold, downwelling material that isolates plume flow from the surrounding mantle, consistent with the surrounding region of high upper-mantle velocities imaged in all models, although variations outside the boundaries of the highvelocity volume are constrained only to the southeast, where station coverage extends beyond its edge. Mean station delay patterns display asymmetry around the island of Hawaii (Figs. 1 and S6), with late arrivals in the western offshore region between Hawaii and Maui (where bathymetry is correspondingly shallower), and early arrivals to the east (where the seafloor is deeper). Images of shallow structure (Fig. 9) also display such asymmetry. These coupled characteristics of seismic velocity and seafloor topography likely reflect three-dimensional complexities in uppermost mantle flow and melting, as suggested by some geodynamic models (Moore et al., 1998). At greater depths, P-wave models reveal low velocities within the mantle transition zone and in the topmost lower mantle, suggesting a deep source region for the Hawaiian plume. The southeastern location of the center of low velocities at 900 km depth is consistent with our earlier S-wave models and with geodynamic arguments that mantle convection in general, and the fast-moving Pacific plate in particular, shear and tilt the plume conduit in the lower and upper mantle (Richards and Griffiths, 1988; Steinberger and O'Connell, 1998; Steinberger et al., 2004). Nonetheless, some of the imaged features at Hawaii are more complex than previously recognized and may motivate future revisions to geodynamic models. In the upper mantle, a near vertical, narrow plume conduit, or “stem,” beneath the island of Hawaii is not seen in P-wave models that include station terms and becomes apparent only in the S-wave models (see Fig. 10). Inversions of PLUME P-wave and S-wave observations without station terms bear more resemblance to each other, and these models likely emphasize the signal of the upper mantle plume “pancake.” Yet the upper mantle seismic signal of this feature appears stronger and may extend deeper than has been predicted by some purely thermal geodynamic models. For example, in the Ribe and Christensen (1999) model, the pancake (defined by a temperature anomaly of ~ 200°) extends only over the depth range 80–160 km. From Table 20.2 of Karato (2008), at 115 km depth such a temperature anomaly might produce a ~3.5% reduction in S-wave velocity and a ~2% reduction in P-wave velocity. However, the delays accumulated over such an 80 km-thick-anomaly would be only 0.6 s for S waves and 0.2 s for P waves, values that are much smaller than the PLUME observations of the corrected mean station delays of ± 2 s for S-waves and ± 1 s for P waves (Fig. S6). Additionally, our simplified estimate of delays is for a uniform 200° temperature anomaly in the pancake layer, whereas the geodynamic model has a variable vertical structure that would further reduce the predicted values. This discrepancy may reflect both greater seismic velocity variations and/or a thicker plume pancake than predicted by some geodynamic models. Laske et al. (Asymmetric shallow mantle structure beneath the Hawaiian Swell — Evidence from Rayleigh waves recorded by the PLUME network, manuscript in preparation,

278

C.J. Wolfe et al. / Earth and Planetary Science Letters 303 (2011) 267–280

Fig. 10. Comparison of P-wave models (second through fourth rows) and an S-wave model (top row), for inversions with station terms. See Fig. 9 for further information.

C.J. Wolfe et al. / Earth and Planetary Science Letters 303 (2011) 267–280

2011) report peak-to-peak VS anomalies of ~6% from surface wave tomography, suggesting that greater seismic velocity variations contribute to some of the discrepancy, though a greater thickness remains necessary. Whereas partial melt may contribute to the increased amplitude of low-velocity anomalies in some locations, the Ribe and Christensen (1999) model predicts that melting is centered beneath the island of Hawaii (with a secondary zone of melting downstream around Kauai), rather than being broadly distributed throughout the pancake. Thus melting in this geodynamic model does not explain the observed broad patterns of mean station delays, although melt may alternatively be more widely distributed and other types of variations (as discussed below) may also contribute to the seismic anomalies. Future geodynamic models should explore whether plume models can account for the body-wave observations documented here, and in particular the conditions for formation of a pancaking region of greater vertical extent than in the Ribe and Christensen (1999) model, i.e., to account for greater accumulated delays. The large signal produced by the high-velocity volume surrounding Hawaii, best constrained to the east of the island of Hawaii, and the asymmetry in uppermost mantle structure and its coupling to seafloor topography also merit further geodynamic study. Such types of features, likely signatures of a cold curtain of downwelling of mechanically eroded lithosphere and three-dimensional complexities in the flow pattern, respectively, have been predicted in a general sense by some prior models (e.g., Moore et al., 1998), but the detailed features now resolvable in seismic velocity models of the region provide an opportunity to refine our understanding of these processes. A number of factors, of course, can lead to velocity heterogeneity in the mantle, including variations in temperature (with contributions from both anharmonic and anelastic components, e.g., Karato, 1993), water content (Karato, 2003), melt (e.g., Hammond and Humphreys, 2000), grain size (Faul and Jackson, 2005), and bulk composition. The last factor includes depletion by melt extraction (Jordan, 1979), although recent studies suggest that the effects of depletion on P- and S-wave velocities and VP/VS ratios may be minor (Afonso et al., 2010; Schutt and Lesher, 2006). [Note that the suggested minor effect of depletion in increasing velocities may allow geodynamic models containing a depleted, buoyant, and more viscous “swell root” (Phipps Morgan et al., 1995) to remain compatible with the broad uppermantle low-velocity region imaged beneath the Hawaiian Islands.] More detailed analysis and better integration with other approaches are needed to explore the joint interpretation of our P- and S-wave models and to assess possible contributions of the various processes, including a study of which factors may give rise to the differences seen in P- and S-wave images (Figs. 9 and 10). Anisotropy may be a particularly important influence in producing differences between VP and VS structures. Because temperature and melt will produce larger delays for S waves than for P waves, anisotropy may be a more important source of bias in P-wave images than in S-wave images. Additionally, as discussed above, there may also be issues regarding the different resolution in the P- and S-wave imaging that could contribute to dissimilarity in these structures. 8. Conclusions Three-dimensional images of P-wave velocity structure beneath the Hawaiian Islands, obtained from the ~ 2-year-long PLUME deployment of seafloor and land seismometers, show an uppermantle low-velocity anomaly that is elongated in the direction of the island chain and surrounded by a high-velocity anomaly that is parabolic in map view. Low velocities continue downward to the mantle transition zone between 410 and 660 km depth and extend into the topmost lower mantle, although the resolution of lower mantle structure is limited. Comparison of inversions with the separate data sets at different frequencies suggests that contamina-

279

tion by water reverberations is not markedly biasing the P-wave imaging of mantle structure. The VP models provide an important complement to the VS models of Wolfe et al. (2009). P-wave and S-wave velocity images show similarities in transition-zone and topmost lower mantle structure that support the inference that both types of wave paths sample common isotropic structure within the interior of our model region and are not strongly affected by aliasing from lowermost mantle heterogeneity. However, inversions without station terms show a southwestward offset in the position of low VP velocities in the uppermost mantle relative to the VS structure, and inversions with station terms show notable differences between VP and VS structure in the shallow upper mantle beneath and immediately east of the island of Hawaii. Both VP and VS images display features that are consistent with many geodynamic predictions of thermal plumes from the lower mantle, such as the broad upper mantle low-velocity region beneath the Hawaiian Islands that may reflect the diverging “pancake” zone at the top of the upwelling zone; the surrounding parabola-shaped area of high velocities that may represent a downwelling curtain; and the location southeast of Hawaii of low-velocity anomalies in the transition zone and topmost lower mantle that is consistent with predictions of plume tilt by the background mantle flow field. The seismic structure revealed by the PLUME project should provide a basis for sharpening further the dynamical models for this archetypical hotspot region. Acknowledgements This project was supported by the U.S. National Science Foundation. We thank Garrett Ito, Maxim Ballmer, Shun Karato, and Greg Hirth for helpful discussions, Guust Nolet for input regarding methodology, and Steve Grand for a copy of his global model. We acknowledge Brandon Schmandt and an anonymous reviewer for their constructive comments. We thank the crews of the research vessels Melville, Ka'imikai-OKanaloa, and Kilo Moana, the Jason remotely operated vehicle team, and the Woods Hole Oceanographic Institution and Scripps Institution of Oceanography Ocean Bottom Seismograph Instrument Pool teams for their work on the ocean-bottom deployments. We acknowledge Peter Burkett and Steven Golden of the Carnegie Institution of Washington's portable seismology laboratory for their work on the land deployment, as well as Brian Savage, Linda Warren, and Dayanthie Weeraratne for their help in the field. We are grateful to Paul Okubo and Steve Brantley at the U.S. Geological Survey Hawaiian Volcano Observatory for logistical assistance with the land experiment, and we thank Glenn Shepherd, Ted Jordan, David Grooms, and Clare Horan for also providing support. We acknowledge those who helped host temporary stations on the Hawaiian Islands: Ruth Mizuba, Bob Carroll, Martie and Don Nitsche at Bougainvillea Bed & Breakfast, Marc Rice at Hawai'i Preparatory Academy, Marshall Mock at Kaua'i Community College, Kelly Kim at NASA Kokee Park Geophysical Observatory, Ed Bartholomew at Lahainaluna High School, James Millar at Molokai Ranch, Jack Spruance at Pu'u O Hoku Ranch, and Mike Weber at Brigham Young University. Appendix A. Supplementary data Supplementary materials related to this article can be found online at doi:10.1016/j.epsl.2011.01.004. References Afonso, J.C., Ranalli, J., Fernàndez, M., Griffin, W., O'Reilly, S.Y., Faul, U., 2010. On the Vp/Vs–Mg# correlation in mantle peridotites: implications for the identification of thermal and compositional anomalies in the upper mantle. Earth Planet. Sci. Lett. 289, 606–618. Blackman, D.K., Orcutt, J.A., Forsyth, D.W., 1995. Recording teleseismic earthquakes using ocean-bottom seismographs at mid-ocean ridges. Bull. Seismol. Soc. Am. 85, 1648–1664. Boschi, L., Becker, T.W., Soldati, G., Dziewonski, A.M., 2006. On the relevance of Born theory in global seismic tomography. Geophys. Res. Lett. 33, L06302 doi:10.1029/ 2005GL025063.

280

C.J. Wolfe et al. / Earth and Planetary Science Letters 303 (2011) 267–280

Boschi, L., Becker, T.W., Steinberger, B., 2007. Mantle plumes: dynamic models and seismic images. Geochem. Geophys. Geosyst. 8, Q10006 doi:10.1029/2007GC001733. Bourdon, B., Ribe, N.M., Stracke, A., Saal, A.E., Turner, S.P., 2006. Insights into the dynamics of mantle plumes from uranium-series geochemistry. Nature 444, 713–717. Bréger, L., Romanowicz, B., 1998. Three-dimensional structure at the base of the mantle beneath the central Pacific. Science 282, 718–720. Collins, J.A., Vernon, F.L., Orcutt, J.A., Stephen, R.L., 2002. Upper mantle structure beneath the Hawaiian swell: constraints from the ocean seismic network pilot experiment. Geophys. Res. Lett. 29, 1522 doi:10.1029/2001GL013302. Courtillot, V., Davaille, A., Besse, J., Stock, J., 2003. Three distinct types of hotspots in the Earth's mantle. Earth Planet. Sci. Lett. 205, 295–308. Crough, S.T., 1978. Thermal origin of mid-plate hot-spot swells. Geophys. J. Roy. Astron. Soc. 55, 451–469. Dahlen, F.A., Hung, S.-H., Nolet, G., 2000. Frechet kernels for finite-frequency traveltimes — I. Theory. Geophys. J. Int. 141, 157–174. Davies, G.F., 1988. Ocean bathymetry and mantle convection, 1. Large-scale flow and hotspots. J. Geophys. Res. 93, 10467–10480. Davies, G.F., 1999. Dynamic Earth: Plates, Plumes and Mantle Convection. Cambridge University Press, Cambridge. 458 pp. Dietz, R.S., Menard, H.W., 1953. Hawaiian swell, deep, and arch, and subsidence of the Hawaiian Islands. J. Geol. 61, 99–113. Duncan, R.A., Keller, R.A., 2004. Radiometric ages for basement rocks from the Emperor Seamounts, ODP Leg 197. Geochem. Geophys. Geosyst. 5, Q08L03 doi:10.1029/ 2004GC000704. Faul, U.H., Jackson, I., 2005. The seismological signature of temperature and grain size variations in the upper mantle. Earth Planet. Sci. Lett. 234, 119–134. Garnero, E.J., Lay, T., McNamara, A., 2007. Implications of lower-mantle structural heterogeneity for the existence and nature of whole-mantle plumes. In: Foulger, G.R., Jurdy, D.M. (Eds.), Plates, Plumes and Planetary Processes. Spec. Paper, 430. Geological Society of America, Boulder, Colo, pp. 79–101. Grand, S.P., 2002. Mantle shear-wave tomography and the fate of subducted slabs. Phil. Trans. Roy. Soc. Lond. A. 360, 2475–2491. Hammond, W.C., Humphreys, E.D., 2000. Upper mantle seismic wave velocity: effects of realistic partial melt geometries. J. Geophys. Res. 105, 10975–10986. Hart, S.R., Hauri, E.H., Oschmann, L.A., Whitehead, J.A., 1992. Mantle plumes and entrainment: isotopic evidence. Science 256, 517–520. Hauri, E.H., 1996. Major-element variability in the Hawaiian mantle plume. Nature 382, 415–419. Hofmann, A.W., 1997. Mantle geochemistry: the message from oceanic volcanism. Nature 385, 219–228. Hung, S.-H., Dahlen, F.A., Nolet, G., 2001. Wavefront healing: a banana-doughnut perspective. Geophys. J. Int. 146, 289–312. Hung, S.-H., Shen, Y., Chiao, L.-Y., 2004. Imaging seismic velocity structure beneath the Iceland hot spot: a finite frequency approach. J. Geophys. Res. 109, B08305 doi:10.1029/2003JB002889. Ishii, M., Tromp, J., 1999. Normal-mode and free-air gravity constraints on lateral variations in velocity and density of Earth's mantle. Science 285, 1231–1236. Ito, G., van Keken, P.E., 2007. Hotspots and melting anomalies. In: Bercovici, D. (Ed.), Mantle Dynamics. : Treatise on Geophysics, vol. 7. Elsevier, Amsterdam, pp. 371–435. Jordan, T.H., 1979. Mineralogies, densities, and seismic velocities of garnet lherzolites and their geophysical significance. In: Boyd, F.R., Meyer, H.O.A. (Eds.), The Mantle Sample: Inclusions in Kimberlites and other Volcanics. American Geophysical Union, Washington, D.C., pp. 1–14. Karato, S., 1993. Importance of anelasticity in the interpretation of seismic tomography. Geophys. Res. Lett. 20, 1623–1626. Karato, S., 2003. Mapping water content in the upper mantle. In: Eiler, J.M. (Ed.), Inside the Subduction Factory. Geophys. Monograph 138. American Geophysical Union, Washington, D.C., pp. 135–152. Karato, S., 2008. Deformation of Earth Materials: an Introduction to the Rheology of Solid Earth. Cambridge University Press, Cambridge. 463 pp. Kennett, B.L.N., Engdahl, E.R., 1991. Traveltimes for global earthquake location and phase identification. Geophys. J. Int. 105, 429–465. Kurz, M.D., Jenkins, W.J., Hart, S.R., Clague, D., 1983. Helium isotopic variations in volcanic rocks from Loihi Seamount and the island of Hawaii. Earth Planet. Sci. Lett. 66, 388–406. Laske, G., Phipps Morgan, J., Orcutt, J.A., 1999. First results from the Hawaiian SWELL pilot experiment. Geophys. Res. Lett. 26, 3397–3400. Laske, G., Phipps Morgan, J., Orcutt, J.A., 2007. The Hawaiian SWELL pilot experiment — evidence for lithosphere rejuvenation from ocean bottom surface wave data. In: Foulger, G.R., Jurdy, D.M. (Eds.), Plates, Plumes, and Planetary Processes. Spec. Paper, 430. Geological Society of America, Boulder, Colo, pp. 209–233. Laske, G., Collins, J.A., Wolfe, C.J., Solomon, S.C., Detrick, R.S., Orcutt, J.A., Bercovici, D., Hauri, E.H., 2009. Probing the Hawaiian hotspot with new broadband ocean bottom instruments. Eos Trans. Amer. Geophys. Un. 90, 362–636. Lay, T., Hernlund, J., Buffett, B.A., 2008. Core-mantle boundary heat flow. Nat. Geosci. 1, 25–32. Leahy, G.M., Collins, J.A., Wolfe, C.J., Laske, G., Solomon, S.C., 2010. Underplating of the Hawaiian Swell: evidence from teleseismic receiver functions. Geophys. J. Int. 183, 313–329. Li, X., Kind, R., Priestley, K., Sobolev, S.V., Tilmann, F., Yuan, X., Weber, M., 2000. Mapping the Hawaiian plume conduit with converted seismic waves. Nature 405, 938–941. Li, X., Kind, R., Yuan, X., Wölbern, I., Hanka, W., 2004. Rejuvenation of the lithosphere by the Hawaiian plume. Nature 427, 827–829. Li, C., van der Hilst, R.D., Engdahl, E.R., Burdick, S., 2008. A new global model for P wave speed variations in Earth's mantle. Geochem. Geophys. Geosyst. 9, Q05018 doi:10.1029/2007GC001806.

Marquering, H., Dahlen, F.A., Nolet, G., 1999. Three-dimensional sensitivity kernels for finite-frequency traveltimes: the banana-doughnut paradox. Geophys. J. Int. 137, 805–815. Montelli, R., Nolet, G., Dahlen, F.A., Masters, G., Engdahl, E.R., Hung, S.-H., 2004. Finitefrequency tomography reveals a variety of plumes in the mantle. Science 303, 338–343. Montelli, R., Nolet, G., Dahlen, F.A., Masters, G., 2006. A catalog of deep mantle plumes: new results from finite-frequency tomography. Geochem. Geophys. Geosyst. 7, Q11007 doi:10.1029/2006GC001248. Moore, W.B., Schubert, G., Tackley, P.J., 1998. Three-dimensional simulations of plumelithosphere interaction at the Hawaiian Swell. Science 279, 1008–1011. Morgan, W.J., 1971. Convection plumes in the lower mantle. Nature 230, 42–43. Nataf, H.-C., 2000. Seismic imaging of mantle plumes. Annu. Rev. Earth Planet. Sci. 28, 391–417. Ni, S., Tan, E., Gurnis, M., Helmberger, D., 2002. Sharp sides to the African superplume. Science 296, 1850–1852. Nolet, G., Dahlen, F.A., 2000. Wave front healing and the evolution of seismic delay times. J. Geophys. Res. 105, 19043–19054. Nolet, G., Allen, R., Zhao, D., 2007. Mantle plume tomography. Chem. Geol. 241, 248–263. Norton, I.O., 2007. Speculations on Cretaceous tectonic history of the northwest Pacific and a tectonic origin for the Hawaii hotspot. In: Foulger, G.R., Jurdy, D.M. (Eds.), Plates, Plumes, and Planetary Processes. Spec. Paper, 430. Geological Society of America, Boulder, Colo, pp. 451–470. Phipps Morgan, J., Morgan, W.J., Price, E., 1995. Hotspot melting generates both hotspot volcanism and a hotspot swell? J. Geophys. Res. 100, 8045–8062. Ribe, N.M., Christensen, U.R., 1994. Three-dimensional modeling of plume-lithosphere interaction. J. Geophys. Res. 99, 669–682. Ribe, N.M., Christensen, U.R., 1999. The dynamical origin of Hawaiian volcanism. Earth Planet. Sci. Lett. 171, 517–531. Richards, M.A., Griffiths, R.W., 1988. Deflection of plumes by mantle shear flow: experimental results and a simple theory. Geophys. J. Int. 94, 367–376. Schmerr, N., Garnero, E., 2006. Investigation of upper mantle discontinuity structure beneath the central Pacific using SS precursors. J. Geophys. Res. 111, B08305 doi:10.1029/2005JB004197. Schutt, D.L., Lesher, C.E., 2006. Effects of melt depletion on the density and seismic velocity of garnet and spinel lherzolite. J. Geophys. Res. 111, B05401 doi:10.1029/ 2003JB002950. Shen, Y., Wolfe, C.J., Solomon, S.C., 2003. Seismic evidence for a lower mantle discontinuity beneath Hawaii and Iceland. Earth Planet. Sci. Lett. 214, 143–151. Sleep, N.H., 1990. Hotspots and mantle plumes: some phenomenology. J. Geophys. Res. 95, 6715–6736. Smith, W.H.F., Sandwell, D.T., 1997. Global seafloor topography from satellite altimetry and ship depth soundings. Science 277, 1956–1962. Sobolev, A.V., Hofmann, A.W., Sobolev, S.V., Nikogosian, I.K., 2005. An olivine-free mantle source for Hawaiian shield basalts. Nature 434, 590–597. Steinberger, B., O'Connell, R.J., 1998. Advection of plumes in mantle flow: implications on hotspot motion, mantle viscosity and plume distribution. Geophys. J. Int. 132, 412–434. Steinberger, B., Sutherland, R., O'Connell, R.J., 2004. Prediction of Emperor-Hawaii seamount locations from a revised model of plate motion and mantle flow. Nature 430, 167–173. Stephen, R.A., Spiess, F.N., Collins, J.A., Hildebrand, J.A., Orcutt, J.A., Peal, K.R., Vernon, F.L., Wooding, F.B., 2003. Ocean seismic network pilot experiment. Geochem. Geophys. Geosyst. 4, 1092 doi:10.1029/2002GC000485. Tilmann, F.J., Benz, H.M., Priestley, K.F., Okubo, P.G., 2001. P-wave velocity structure of the uppermost mantle beneath Hawaii from travel time tomography. Geophys. J. Int. 146, 594–606. Turcotte, D.L., Oxburgh, E.R., 1973. Mid-plate tectonics. Nature 244, 337–399. van der Hilst, R.D., de Hoop, M.V., 2005. Banana-doughnut kernels and mantle tomography. Geophys. J. Int. 163, 956–961. VanDecar, J.C., Crosson, R.S., 1990. Determination of teleseismic relative phase arrival times using multi-channel cross correlation and least squares. Bull. Seismol. Soc. Am. 80, 150–169. Walker, K.T., Bokelmann, G.H.R., Klemperer, S.L., 2001. Shear-wave splitting to test mantle deformation models around Hawaii. Geophys. Res. Lett. 28, 4319–4322. Walker, K.T., Bokelmann, G.H.R., Klemperer, S.L., 2003. Reply to “Shear-wave splitting to test mantle deformation models around Hawaii” by Vinnik et al. Geophys. Res. Lett. 30, 1676 doi:10.1029/2002GL016712. Webb, S.C., 1998. Broadband seismology and noise under the ocean. Rev. Geophys. 36, 105–142. Wessel, P., 1993. Observational constraints on models of the Hawaiian hot spot swell. J. Geophys. Res. 98, 16095–16104. Wielandt, E., 1987. On the validity of the ray approximation for interpreting delay times. In: Nolet, G. (Ed.), Seismic Tomography. D. Reidel, Boston, Mass, pp. 85–98. Wolfe, C.J., Solomon, S.C., Silver, P.G., VanDecar, J.C., Russo, R.M., 2002. Inversion of body wave delay times for mantle structure beneath the Hawaiian Islands: results from the PELENET experiment. Earth Planet. Sci. Lett. 198, 129–145. Wolfe, C.J., Solomon, S.C., Laske, G., Collins, J.A., Detrick, R.S., Orcutt, J.A., Bercovici, D., Hauri, E.H., 2009. Mantle shear-wave velocity structure beneath the Hawaiian hotspot. Science 326, 1388–1390. Woods, M.T., Okal, E.A., 1996. Rayleigh wave dispersion along the Hawaiian Swell: a test of lithospheric thinning by thermal rejuvenation at a hotspot. Geophys. J. Int. 125, 325–339.