Seismic anisotropy and compositionally induced velocity anomalies in the lithosphere above mantle plumes: a petrological and microstructural study of mantle xenoliths from French Polynesia

www.elsevier.com/locate/epsl

Earth and Planetary Science Le

Seismic anisotropy and compositionally induced velocity

anomalies in the lithosphere above mantle plumes:

a petrological and microstructural study of mantle xenoliths

from French Polynesia

Andrea Tommasi*, Marguerite Godard, Guilhem Coromina,

Jean-Marie Dautria, Hans Barsczus

Laboratoire Tectonophysique, ISTEEM, CNRS/Universite Montpellier II, 34095 Montpellier Cedex 5, France

Received 22 April 2004; received in revised form 2 September 2004; accepted 9 September 2004

Available online 18 October 2004

Editor: E. Bard

Abstract

In addition to thermal erosion, plume/lithosphere interaction may induce significant changes in the lithosphere chemical

composition. To constrain the extent of this process in an oceanic environment and its consequences on the lithosphere

seismic properties, we investigated the relationship between petrological processes and microstructure in mantle xenoliths

from different hotspots tracks in South Pacific Superswell region: the Austral-Cook, Society, and Marquesas islands in

French Polynesia. Olivine forsterite contents in the studied spinel peridotites vary continuously from Fo91 to Fo83. Dunites

and wehrlites display the lowest forsterite contents. Their microstructure and high Ni contents preclude a cumulate origin,

suggesting that these rocks result from melt/rock reactions involving olivine precipitation and pyroxene dissolution. In

addition, lherzolites and wehrlites display evidence of late crystallization of clinopyroxene, which may result from a near-

solidus melt–freezing reaction. These data suggest that the lithosphere above a mantle plume undergoes a complex sequence

of magmatic processes that significantly change its composition. These compositional changes, particularly iron enrichment

in olivine, result in lower P- and S-waves velocities. Relative to normal lithospheric mantle, compositionally induced seismic

anomalies may attain �2.2% for S-waves and �1% for P-waves. Smaller negative anomalies for P-waves are due to a

higher sensitivity to modal composition. Conversely, crystal-preferred orientations (CPO) and seismic anisotropy are little

affected by these processes. Lherzolites and harzburgites, independent from composition, show high-temperature

porphyroclastic microstructures and strong olivine CPO. Dunites and wehrlites display annealing microstructures to which

is associated a progressive dispersion of the olivine CPO. Very weak, almost random olivine CPO is nevertheless rare, suggesting

0012-821X/$ - s

doi:10.1016/j.ep

* Correspon

E-mail addr

tters 227 (2004) 539–556

ee front matter D 2004 Elsevier B.V. All rights reserved.

sl.2004.09.019

ding author. Tel.: +33 467144912; fax: +33 467143603.

ess: [email protected] (A. Tommasi).

A. Tommasi et al. / Earth and Planetary Science Letters 227 (2004) 539–556540

that CPO destruction is restricted to domains of intense magma–rock interaction due to localized flow or accumulation of

magmas.

D 2004 Elsevier B.V. All rights reserved.

Keywords: upper mantle; lithosphere; mantle plumes; melt percolation; melt–rock reaction; seismic tomography; seismic anisotropy

1. Introduction

Seismic tomography and anisotropy measurements

are undoubtedly powerful tools to unravel the present-

day thermal structure and deformation of the mantle.

However, velocity heterogeneities may also reflect

variations in composition within the mantle. Fast

seismic velocity anomalies beneath cratonic domains,

for instance, are usually interpreted as due to both a

cold geotherm and a highly refractory lithospheric

mantle [1]. Recent seismic tomography experiments

show that the Ontong Java, Deccan, and Parana

mesozoic large igneous provinces (LIP) are underlain

by an abnormally slow upper mantle [2–5]. Unless

related to present-day mantle plumes that coincidently

impact below the three LIPs, these slow velocity

domains cannot be interpreted as thermal anomalies,

since the latter should have diffused since the

Mesozoic. Moreover, the upper mantle below the

Ontong Java plateau shows low seismic attenuation

[6]. Together, these observations suggest that the slow

seismic velocities below these basaltic provinces

characterize an abnormal mantle root, formed by

mantle rocks which composition has been modified

by plume activity.

Seismic anisotropy in the upper mantle results

essentially from orientation of olivine crystals during

plastic deformation. In the high-temperature convect-

ing mantle, olivine crystal-preferred orientations

(CPO) is continuously modified and seismic aniso-

tropy records present-day flow. On the other hand, the

low temperatures and resulting high viscosities that

prevail in the lithospheric mantle may freeze the

olivine CPO over very long time spans. Indeed, shear

wave splitting and Pn azimuthal anisotropy data in

ancient continental domains (e.g., [7–9]) display a

very good correlation with the major structures of

collisional belts of Archean or Neoproterozoic age.

Refraction experiments also show that fast P waves

propagation directions in the uppermost mantle

beneath oceans correlate with past seafloor spreading

directions, suggesting that olivine CPO formed at the

ridge may be frozen in the lithosphere for z100 My

[10]. However, shear wave splitting data in some very

young ocean islands, like Tahiti or La Reunion, detect

no anisotropy [11,12]. This bapparentQ isotropy has

been interpreted as resulting from a blocalQ destructionof the olivine CPO in the lithosphere by the plume

activity.

These observations suggest that plumes may

induce long-lived changes in the upper mantle seismic

properties. Possible candidates to produce seismic

velocity and anisotropy variations in the mantle above

a plume are changes in composition and texture of

mantle rocks due to partial melting and melt–rock

reactions. Direct analysis of mantle samples brought

to the surface by plume-related magmas allows one to

constrain the extent of these magmatic processes and

their effect on seismic properties. In this study, which

is part of the multidisplinary Polynesian Lithosphere

and Upper Mantle Experiment (PLUME) [13], we

investigate the relationship between petrological

processes, microstructure, and seismic properties in

a series of mantle xenoliths from different hotspots

tracks in South Pacific Superswell region: the Austral-

Cook, Society, and Marquesas islands in French

Polynesia. Two issues are essential to the interpreta-

tion of seismic data in terms of mantle structure and

deformation patterns: How strongly do compositional

heterogeneities produced by melt–rock interactions

contribute to the velocity anomalies observed in

seismic tomography? Do these processes modify the

olivine CPO and hence the seismic anisotropy

signature of the upper mantle?

2. Sampling

After optical analysis of a collection of 80 spinel-

bearing mantle xenoliths, we selected 20 samples for

Fig. 1. (a) Modal compositions of the studied peridotites plotted

onto the peridotite field of the olivine–orthopyroxene–clinopyrox

ene diagram. Modes were obtained by image analysis of photo

micrographs and of electron back-scattered compositional images o

thin sections. (b) Modal compositions of harzburgites and lherzo

lites plotted on a cpx/opx vs. olivine diagram. Compositions were

recalculated in wt.% for comparison with melting models: (1

polybaric melting after [46] with (a) or without (b) bexcess olivineQ

A. Tommasi et al. / Earth and Planetary Science Letters 227 (2004) 539–556 541

detailed microstructural investigation, electron

microprobe analyses, and crystallographic orientation

measurements: 4 lherzolites, 2 dunites and 2 wehrlites

sampled within basanite boulders from the Fataua

River valley in Tahiti (Society Islands), 4 harzbur-

gites, 1 dunite and 1 wehrlite sampled in alkali-basalts

outcropping on the western slopes of Mount Tanga in

Rapa, and 1 lherzolite sampled within tephritic flows

from the quarry of Mount Haramea in Tubuai (Austral

Islands), 4 dunites from alkali-basalts from Ua Huka

and 1 harzburgite from Fatu Hiva in the Marquesas

archipelago. Ages of extraction vary from ~9.5 Ma in

Tubuai [14] to b1 Ma in Tahiti [15]. These xenoliths

range between 2 and 15 cm in diameter and are

extremely fresh. All dunites and wehrlites as well as

most lherzolites and harzburgites display sharp con-

tacts with the enclosing basalt. Lherzolite samples

from Tahiti display mm-scale reaction rims at the

contact with the basalt, which were avoided in both

textural and chemical analyses. Proportions of the

various rock types (harzburgite, lherzolite, dunite, and

wehrlite) in the suite selected for detailed analysis are

representative of those observed for each sampling

site in the total collection, except for Tahiti where

dunites and wehrlites dominate.

modal composition is given by [46] for polybaric melting and was

fixed as 55% olivine, 28% opx, 15% cpx and 2% spinel for isobaric

melting. White symbols represent Marquesas samples, light gray

Society samples, and dark gray, Austral samples. Squares

harzburgites and lherzolites, circles: dunites, diamonds: wehrlites

Small gray circles represent Tahiti xenoliths analyzed by [48].

3. Microstructures: deformation and magmatic

reactions

3.1. Modal composition and microstructures

The studied lherzolites and harzburgites are char-

acterized by relatively fertile compositions, with

clinopyronene contents ranging from 4–5 to 12

vol.%. Only two samples, harzburgite RPA18A and

lherzolite THTFA1A, display modal compositions

consistent with those predicted by partial melting

models (Fig. 1). The remaining are enriched in

clinopyronene or in olivine, the strongest enrichments

being observed in Tahiti lherzolites and in the

Marquesas harzburgite FTH101A.

Most lherzolites and harzburgites exhibit coarse-

grained porphyroclastic microstructures characteristic

of deformation by dislocation creep under high-

temperature, low-stress conditions (TN1100 8C).Foliation and lineation are usually marked by elonga-

tion of olivine porphyroclasts up to 5 mm long with

-

-

f

-

)

,

l

,

:

.

curvilinear grain boundaries and well-developed and

widely spaced (100) subgrain boundaries (Fig. 2a).

Interpenetrating olivine–olivine grain boundaries indi-

cate active grain boundary migration that, in some

samples (e.g., RPA12), gives rise to cm-scale olivine

grains (abnormal grain growth).

In a few samples (e.g., RPA18A), elongated lens-

shaped orthopyroxene crystals (Fig. 2a) mark the

deformation fabric. However, orthopyroxene is mostly

present as irregularly shaped crystals (0.5 to 4 mm),

which either display no shape-preferred orientation or

are elongated at high angle to the olivine fabric

(THTFA1A and THTFA4A, Fig. 2b). Orthopyroxene

(opx) often displays corroded grain shapes, with

corrosion embayments filled by olivine (Fig. 2d) or

clinopyroxene (Fig. 2e), and secondary crystallization

Fig. 2. Typical microstructures (a–c) and reactional features (d–f) in lherzolites and harzburgites. (a) High-temperature, coarse-grained

porphyroclastic microstructure (harzburgite RPA18A), foliation and lineation (X) are marked by elongation of olivine (ol) and enstatite

porphyroclasts (en), the former display well-developed and widely spaced (100) subgrain boundaries as well as grain boundary migration

features. (b) Cpx-rich lherzolite (THTFA4A) with characteristic interstitial diopside (di); black and white arrows mark irregularly shaped

enstatite (en) crystals elongated at high angle to the olivine fabric and a diopside-rich veinlet, respectively. (c) Porphyroclastic harzburgite

(RPA1B) displaying a bimodal grain size distribution, characterized by olivine (ol) and enstatite (en) porphyroclasts surrounded by a finer-

grained matrix composed by polygonal olivine and enstatite grains, interstitial diopside, and spinel. (d) Enstatite (en) grain with corrosion

embayments filled by olivine (ol). (e) Diopside (di) reaction rim around enstatite (en) (RPA1B). (f) Olivine with corrosion embayments filled by

diopside. All photomicrographs were taken under crossed polarizers, except (b).

A. Tommasi et al. / Earth and Planetary Science Letters 227 (2004) 539–556542

of clinopyroxene, olivine, or spinel along kinks or

fractures. Corrosion may explain the babnormalQorthopyroxene shape-preferred orientation observed

in the Tahiti lherzolite suite.

Clinopyroxene (cpx) generally occurs as isolated

grains or monophase aggregates with a clear

interstitial habit (Figs. 2f and 3) or within corrosion

embayments of orthopyroxene and olivine grains

(Fig. 2e–f). Typical opx–cpx–spinel aggregates are

less common. Tahiti lherzolites also show 1–2 mm

wide cpx-rich veinlets, clearly truncated by the basalt

at the xenolith margin (Fig. 3). These observations

suggest an origin by secondary crystallization for, at

least, part of the clinopyroxenes in the studied

peridotites. Together with orthopyroxene corrosion,

this could result in the high cpx/opx ratios that

distinguish Tahiti lherzolites THTFA4A and

THTFA5 (Fig. 1b).

Fig. 3. Electron back-scattered compositional image of the high-

temperature porphyroclastic lherzolite THTFA1A. Gray tonalities

are a function of the mean atomic number that increases from

enstatite (en, darkest gray) to spinel (sp, white); fractures and holes

are displayed in black. Diopside grains (light gray), which show

clear interstitial shapes, suggesting late crystallization from a melt,

tend to form discontinuous seams in two directions, one roughly

parallel and the other at 608 to the foliation.

A. Tommasi et al. / Earth and Planetary Science Letters 227 (2004) 539–556 543

Surprisingly, spinels do not mark the deformation

fabric, not even in peridotites that display a well-

developed olivine shape-preferred orientation. They

occur either as vermicular grains within opx–cpx–

spinel aggregates (Fig. 2b) or, more often, as

interstitial grains (isolated or in association with

clinopyroxene), which cusp shapes suggest a secon-

dary origin (Fig. 3). Locally, symplectitic intergrowths

of spinel and clinopyroxene are also observed.

Two samples, lherzolite TB110D and harzburgite

RPA1A, are distinguished by a bimodal grain size

distribution characterized by coexistence of olivine

and orthopyroxene porphyroclasts with a fine-grained

matrix (~0.5 mm) composed by polygonal olivine and

orthopyroxene grains, as well as interstitial clinopyr-

oxene and spinel. This matrix forms either cm-scale

bands roughly parallel to the elongation of orthopyr-

oxene porphyroclasts (TB110D) or discontinuous

mm-scale seams that crosscut the entire xenolith in

two roughly orthogonal directions (RPA1B, Fig. 2c).

Olivine porphyroclasts either display widely spaced

subgrains and sutured boundaries, indicative of active

grain boundary migration, or display polygonal

shapes. Matrix grains usually do not show any strain

features. Orthopyroxene porphyroclasts often exhibit

reaction rims (Fig. 2e) or secondary crystallization of

olivine, spinel or clinopyroxene along kinks or

fractures. A similar microstructure is also observed

in a cm-scale band in harzburgite FTH101A, which

otherwise displays a typical high-temperature por-

phyroclastic microstructure. The bimodal grain size

distribution suggests deformation under higher stress

(lower temperature?) conditions, in which dynamic

recrystallization lead to significant grain size refine-

ment. However, only lherzolite TB110D shows both a

shape-preferred orientation and kinks in orthopyrox-

ene porphyroclasts supporting this interpretation.

In contrast to lherzolites and harzburgites, dunites

and wehrlites do not display any clear deformation

microstructures; shape-preferred orientations marking

a macroscopic foliation or lineation are never

observed. Both dunites and wehrlites are equigranular

and show microstructures that indicate that static

recrystallization processes and, in particular, grain

boundary migration were very active. These micro-

structures range from highly lobated, interpenetrating

olivine grain boundaries, observed in most dunites

from Austral and Society islands (Fig. 4a), to

polygonal textures, characterized by straight grain

boundaries meeting at 1208 (Fig. 4b), which are

typical of dunites from Marquesas islands and of

wehrlites. Despite the important grain boundary

migration, grain sizes are usually smaller than in the

lherzolites. Olivine crystals are generally devoid of

internal deformation features. However, a few por-

phyroclasts still retain clear subgrain boundaries (Fig.

4a), suggesting an early deformation incompletely

erased by static recrystallization. Spinel occurs as

small polygonal grains (~0.5 mm) at triple junctions

or as inclusions in olivine, as a result of the active

migration of olivine grain boundaries. Most dunites

display small amounts (b5%) of interstitial clinopyr-

oxene, but orthopyroxene is never observed. In the

wehrlites, clinopyroxene content ranges from 31 to 37

vol.%. It is either interstitial, forming semi-continuous

seams (Fig. 4c), or poikiloblastic, enclosing corroded

olivine grains (Fig. 4d). This suggests that the

wehrlites formed at the expense of dunites by

secondary crystallization of clinopyroxene. In addi-

tion, the annealing microstructure, which is better

developed in the Marquesas dunites and wehrlites,

suggests that these rocks have been submitted to high

temperatures, probably under static conditions.

Fig. 4. Photomicrographs of typical microstructures in dunites (a–b, crossed polarizers) and wehrlites (c–e, plane-polarized light). (a) Dunite

RPA6 displaying an equigranular texture; arrow points a porphyroclast that still retains clear subgrain boundaries. (b) Dunite UAH289A

displaying a polygonal texture. (c) Interstitial clinopyroxene (Fe-rich diopside, light gray) forming semi-continuous seams around dunitic lenses

in wehrlite THTFA5A/3A. (d) Poikiloblastic clinopyroxenes (light gray) enclosing corroded olivine grains in wehrlite RPA9.

A. Tommasi et al. / Earth and Planetary Science Letters 227 (2004) 539–556544

3.2. Crystal-preferred orientations

Olivine and pyroxenes crystallographic orienta-

tions (CPO) were determined by indexation of

electron back-scattered diffraction (EBSD) patterns.

All samples were analyzed manually (i.e., indexing of

every crystal was verified by the operator) to avoid

errors due to the pseudo-hexagonal symmetry of

olivine in the [100] direction. Measurements were

done in a grain by grain basis along 3-mm-spacing

profiles parallel to the long axis of the thin section.

Lherzolites and harzburgites, independently from

composition or microstructure, show very strong and

homogeneous olivine CPO, characterized by a strong

alignment of [100] axes close to the lineation (marked

by the olivine shape-preferred orientation) and a girdle

distribution of [010] and [001] normal to it, with

weaker maxima normal and parallel to the foliation,

respectively (Fig. 5). This olivine CPO suggests

deformation by dislocation creep with dominant

activation of the high-temperature (010)[100] and

(001)[100] slip systems. Lherzolite TB110D is the

single exception to this pattern. It displays a much

weaker olivine CPO characterized by a concentration

of [100] axes parallel to the lineation and of [001]

axes normal to the foliation, which, in this sample, are

clearly marked by the elongation of orthopyroxene

porphyroclasts and by the preferred orientation of

fine-grained domains. Coarse- and fine-grained

domains display similar olivine CPO. This suggests

that the bimodal texture and the weak CPO result from

deformation by dislocation creep under high-stress

conditions, in which dynamic recrystallization lead to

significant grain size refinement.

Orthopyroxene CPO are coherent with the olivine

ones, but much weaker (see online supplementary Fig.

1). [001] axes concentrate close to the lineation and

[100] and [010] are distributed in a girdle normal to it,

suggesting activation of {hk0}[001] systems. Yet, in

most samples, there is a small obliquity (~158)between the olivine and orthopyroxene CPOs. This

suggests that olivine and orthopyroxene underwent

Fig. 5. Olivine crystal-preferred orientations (CPO) in the lherzolites

and harzburgites. Lower hemisphere equal-area projection, n

measurements, contours at 1 multiple of a uniform distribution

intervals. Full line marks the orientation of the foliation (XY plane);

lineation (X direction) is horizontal. Mean forsterite content (Mg#)

of olivine in each sample is also indicated.

A. Tommasi et al. / Earth and Planetary Science Letters 227 (2004) 539–556 545

the same deformation, but harder orthopyroxene

grains accommodated smaller strains than olivine

ones. Clinopyroxene usually displays almost random

CPO, except in harzburgite FTH101A that is charac-

terized by parallelism of orthopyroxene and clinopyr-

oxene CPOs and in lherzolite TB110D, which

displays a clinopyroxene CPO oblique to both the

olivine and orthopyroxene ones.

Dunites and wehrlites display weak olivine CPO,

which are often characterized by higher concentra-

tions of the [010] axes relative to [100] and [001] (Fig.

6). CPO intensity is related to microstructure. Dunites

with lobate grain boundaries retain an olivine CPO

similar to, although much weaker, than those observed

in the lherzolites and harzburgites. Polygonal dunites

and wehrlites display even weaker, almost isotropic,

olivine CPO. This suggests that static recrystallization

contributed to the dispersion of the olivine CPO.

Clinopyroxene CPO in wehrlites also relates to the

microstructure (Background Dataset Fig. 2). Poikilo-

blastic diopside in wehrlite RPA9 displays an almost

random orientation, whereas interstitial diopside in

wehrlite THTFA5A/3A displays a weak, but well-

organized CPO in good agreement with the olivine

one, suggesting a common deformation.

4. Mineral chemistry: evidence for extensive melt/

rock interaction

Olivine in lherzolites and harzburgites is charac-

terized by a wide range of compositions with Mg#

(Mg#=Mg/(Mg+Fe)) ranging from 86 to 91 and Ni

contents from 2400 to 3400 ppm (Fig. 7a). These

variations are correlated to the olivine content;

harzburgites usually display higher Mg# as expected

for partial melting. Peridotites with bimodal textures

display variations in olivine composition at the sample

scale: olivine from fine-grained bands in harzburgites

RPA1B and FTH101A displays lower Ni contents and

Mg# than the porphyroclasts (Fig. 7a). Fe-rich olivine

compositions of lherzolites overlap with those of the

dunites and wehrlites (Mg#=83–86 and Ni=1600–

2900 ppm). Core–rim compositional gradients are

restricted to one sample, lherzolite THTFA4A, which

shows a clear decrease in Mg# and Ni content from

core (Mg#=89; Ni=3000 ppm) to rim (Mg#=83–87;

Ni=1700–2400 ppm).

A. Tommasi et al. / Earth and Planetary Science Letters 227 (2004) 539–556546

Orthopyroxene (opx) is enstatite with Mg# ranging

from 87 to 91 and variable TiO2 contents (0.02–0.19

wt.%). Clinopyroxene is Cr-rich diopside (Cr2O3=0.7–

1.1 wt.%) with Mg# ranging from 84 to 93, the lowest

Mg# (84–87) values being measured in wehrlites.

Clinopyroxenes display also variable TiO2 contents

(0.02–0.94 wt.% in lherzolites and harzburgites, 0.83–

1.33 wt.% in wehrlites). In contrast to olivine, pyrox-

enes are homogeneous at both grain and sample scales

except for wehrlite RPA9, whose small interstitial

grains display the lowest Mg# (83) and Cr2O3 values

(0.46 wt.%) and highest TiO2 content (2.2 wt.%)

observed in the studied xenolith suite (Fig. 7b–c).

Al2O3, Na2O and CaO contents in lherzolites and

harzburgites distinguish the different archipelagoes.

High Al2O3 contents (5–5.2 wt.% in opx and 6.2–6.85

wt.% in cpx) in Tahiti lherzolites stand out against the

low Al2O3 values of the Australs and Marquesas

peridotites (2.1–3.1 wt.% in opx and 2.5–4.2 wt.% in

cpx). The lower CaO and higher Na2O content of

clinopyroxenes in Tahiti lherzolites suggest that these

peridotites were equilibrated at higher pressures than

the other studied xenoliths. Higher AlIV/AlVI ratios in

clinopyroxenes from wehrlites (2.5–4.5) relative to

Rapa harzburgites (1.5–2.9) and to Tahiti lherzolites

(0.8–1.2) suggest that wehrlites were equilibrated at

lower pressures. Thermometers based on Al–Ca

distribution in pyroxenes indicate also higher equili-

bration temperatures for Tahiti samples (1040–1080

8C [16] and 1010–1035 8C [17]) than Australs

peridotites (890–950 8C [16], 910–980 8C [17]).

Spinels show a wide range of compositions (Fig.

7d) consistent with the variable degree of fertility of

the studied peridotites. Variations in spinel composi-

tion also distinguish the different archipelagoes. Tahiti

lherzolites display the lowest Cr# (15–26) and highest

Mg# (64–72), while Marquesas harzburgite and

dunites display higher Cr# (50–57) and lower Mg#

(30–45). The most striking feature in the studied

spinels is the strong variation in TiO2 content (0.04–

9.45 wt.%). The most enriched sample is Marquesas

harzburgite FTH101A and the high Mg# Rapa

Fig. 6. Olivine crystal-preferred orientations in the dunites and

wehrlites. Lower hemisphere equal-area projection, n measure-

ments, contours at 1 multiple of a uniform distribution intervals. No

shape-preferred orientation marking a foliation or lineation is

observed in these samples.

A. Tommasi et al. / Earth and Planetary Science Letters 227 (2004) 539–556 547

harzburgites display the lowest values. The studied

xenoliths (with exception of FTH101A) show an

inverse correlation between TiO2 in spinel and olivine

Mg#, the higher spinel TiO2 content and lower olivine

Mg# being found in dunites and wehrlites (Fig. 7d).

Dunites with Fe-rich olivines are common in

oceanic environments (e.g., Hawaii [19], French

Polynesia, and La Reunion [20]). Because of (1) their

systematic association with wehrlites and (2) their

high Fe-content, they were first interpreted as

cumulates [19]. However, olivines in these dunites

and wehrlites differ from those crystallizing from a

basalt by their high Ni content at a given Mg# (Fig.

7a). Basalts in equilibrium with peridotites at depth

become increasingly orthopyroxene-undersaturated as

pressure decreases [21]. Berger and Vannier [20] and

Kelemen et al. [21] proposed therefore that dunites

formed at low pressures (b1.5 GPa) by reaction

between peridotites and an olivine-saturated basaltic

melt leading to dissolution of pyroxene and precip-

itation of olivine. Microstructural evidence of early

high-temperature deformation followed by static

recrystallization (Fig. 4a) also favors a reactional

origin for the studied dunites.

The interstitial habit, high AlIV/AlVI ratios, and

high TiO2 content of clinopyroxene in wehrlites

suggests that it is the product of a late, low-pressure

crystallization of a percolating basaltic melt within the

dunites. Similar impregnation features are observed at

the Moho Transition Zone (MTZ) in ophiolites [22–

24]. Variations in clinopyroxene composition may be

related to changes in melt composition, the higher Ti

Fig. 7. Olivine Ni content (a), clinopyroxene Mg# (b), orthopyrox-

ene Mg# (c), and spinel TiO2 content (d) as a function of the olivine

Mg# in the studied xenolith suite. Symbols are shown in inset.

Variations in olivine composition within samples are distinguished

by L (large grains) and S (small grains) when dependent on grain

size, and by (c) and (r) for core to rim variations. Interstitial cpx

compositions in RPA9 are highlighted by (i). For comparison, a

compilation of olivine compositions previously measured in Society

and Austral xenoliths is shown in (a) [20,49–53]. Evolution of Ni

content and Mg# in olivine during fractional crystallization in a

closed system is calculated by subtracting iteratively (in increments

of 0.1 wt.%) olivine in equilibrium with the evolving melt using

olivine/liquid partition coefficient values from [54] for FeO/MgO

and from [55] for Ni. Initial melt composition is Mg#=74 (after

[56]) and Ni=360 ppm. Diamonds mark crystallized fractions by 1%

increments, numbers in square brackets label 5% increments. Black

diamonds distinguish the fractional crystallization trend from the

equilibrium crystallization trend (white diamonds).

A. Tommasi et al. / Earth and Planetary Science Letters 227 (2004) 539–556548

interstitial clinopyroxene in RPA9 resulting from late

crystallization of a more differentiated melt. Dunites

and wehrlites thus formed by melt/rock reactions

involving olivine precipitation, followed by melt

consumption through clinopyroxene precipitation.

Fe-rich compositions of harzburgites and lherzo-

lites, as well as Ti-enrichment in pyroxenes and spinel

(Fig. 7) and low Cr2O3 contents in clinopyroxene,

suggest that these rocks also underwent chemical

reequilibration, at depth, with percolating Fe–Ti-rich

melts. This interaction is more developed in Tahiti

lherzolites and in the fine-grained domains of RPA1B

and FTH101A. Thus, melt–rock interaction probably

played an important role in the formation of the

bimodal texture in these samples. Although Fe–Mg

interdiffusion rates in olivine are fast [18], homoge-

neous Mg# at cm-scale in lherzolites and harzburgites

imply melt–rock interaction times N104 years, which

suggest that Fe-enrichment is not associated with fast

magma flow through dikes. Incomplete melt–rock Fe–

Mg reequilibration, as evidenced by core–rim com-

positional gradients, is recorded only in sample

THTFA1A and may be related to xenolith extraction.

The high cpx/opx ratios, the corrosion features in

orthopyroxene, and the often interstitial habit of

clinopyroxene suggest that, in the lherzolites, reaction

with basaltic melts also modified modal compositions

through precipitation of clinopyroxene. Secondary

crystallization of clinopyroxene has been described

in orogenic and ophiolitic massifs and interpreted as a

melt–freezing reaction at the lithosphere/astheno-

sphere interface [25]. In the studied xenoliths,

secondary crystallization of clinopyroxene is related

to the equilibration temperature, the most affected

being the high-temperature Tahiti lherzolites.

5. Seismic properties of the lithosphere above a

mantle plume

To constrain the effect of the compositional and

textural changes induced by melt–rock interactions on

the mantle seismic properties, the three-dimensional

distributions of seismic velocities in each sample were

estimated by averaging the individual grain elastic

constants tensors as a function of the crystallographic

orientations and modal composition [26]. In the

present calculations, we used single-crystal elastic

constants tensors of olivine, enstatite, and diopside at

ambient conditions [27–29] and Voigt–Reuss–Hill

averages. In addition, the dependence of olivine

elastic constants and density on the forsterite content

[30] was explicitly taken into account.

5.1. Seismic anisotropy

Lherzolites and harzburgites display compres-

sional (P) waves velocity distributions and shear

(S) waves anisotropy patterns (Fig. 8) typical of

upper mantle rocks deformed under high-temperature

conditions. P-waves are the fastest when propagating

parallel to the lineation (maximum concentration of

olivine [100] axes) and the slowest when propagat-

ing normal to the foliation (parallel to the maximum

concentration of olivine [010] axes). Except for

lherzolite TB110D, all lherzolites and harzburgites

are highly anisotropic. P-waves azimuthal anisotropy

is 9–11% and S-waves polarization anisotropy may

attain 5–7%. The highest anisotropies are displayed

by the Mg-rich harzburgite RPA18A, which displays

the least modified chemical and mineralogical

composition (Fig. 9). Fast S-waves are polarized

parallel to the maximum concentration of olivine

[100] axes, i.e., the lineation. Polarization anisotropy

is minimum for S-waves propagating at low angles

to the lineation and maximum for those propagating

at high angles to the lineation.

Dunites and wehrlites display weak anisotropies

for both P- and S-waves. Anisotropy intensity does

not depend on olivine composition (Fig. 9). There is,

however, a relation between anisotropy and micro-

structure (Fig. 10). The lowest anisotropies are

displayed by dunites and wehrlites with well-devel-

oped polygonal textures. Although weaker, P-waves

velocity distributions and S-waves anisotropy patterns

of dunites and wehrlites are similar to those displayed

by the lherzolites. However, since these rocks do not

show any shape-preferred orientation marking a

foliation or a lineation, seismic anisotropy may only

be related to flow in the upper mantle by assuming

that olivine [100] and [010] axes align, respectively,

parallel to the flow direction and normal to the flow

plane, as usually in peridotites deformed under high-

temperature conditions.

We interpret the reduction in seismic anisotropy in

the dunites and wehrlites relative to lherzolites and

Fig. 9. P-wave (a) and maximum S-wave (b) anisotropies as a

function of the mean forsterite content of olivine in each sample

Symbols as in Fig. 7.

Fig. 8. Modeled three-dimensional compressional waves velocity

and shear wave anisotropy (intensity and polarization direction o

the fast wave) distributions for lherzolites and harzburgites. Voigt–

Reuss–Hill averages calculated from crystallographic orientation

data and elastic constants tensors for olivine, enstatite, and diopside

at ambient conditions [27–29]. Modal compositions used in the

calculations, as well as the forsterite content of olivine are indicated

on the right of each plot. The variation of elastic constants and

density of olivine associated with changes in forsterite content [30

was explicitly taken into account in the calculation. Lowe

hemisphere equal-area projections, contours for P-waves velocities

and S-wave anisotropy at 0.1 km/s and 1% intervals, respectively

Full line marks the foliation (XY plane); lineation (X direction) is

horizontal.

A. Tommasi et al. / Earth and Planetary Science Letters 227 (2004) 539–556 549

.

harzburgites as a result of the dispersion of the

olivine CPO due to crystallization of non-oriented

olivine neoblasts during dunite-forming melt–rock

reactions. Static recrystallization may have contrib-

uted to the dispersion of the olivine CPO and,

hence, to the decrease in anisotropy, by favoring

growth of the new, undeformed neoblasts at the

expenses of the old grains. Thus, melt–rock reac-

tions may locally erase the seismic anisotropy

signature of the lithospheric mantle. They may also

modify the seismic velocities.

f

]

r

.

Fig. 11. Density (a), P-wave (b) and S-wave (c) velocities as a

function of the mean forsterite content of olivine in each sample

Full lines show the theoretical variation of density, as well as of P-

and S-waves velocities as a function of olivine forsterite conten

(calculated for a 100% olivine isotropic aggregate). For each

sample, we represented the bisotropicQ P- and S-waves velocities

(darker symbols, calculated for an isotropic aggregate with similar

modal and mineralogical compositions) as well as the maximum and

minimum P-wave velocities (b) or the mean fast and slow S-wave

velocities (c) calculated for the actual anisotropic samples (lighter

symbols). Symbols as in Fig. 7.

Fig. 10. Modeled three-dimensional compressional waves velocity

and shear wave anisotropy (intensity and polarization direction of the

fast wave) distributions for dunites and wehrlites. Mode and forsterite

content of olivine are indicated on the right of each plot. Lower

hemisphere equal-area projections, contours for P-waves velocities

and S-wave anisotropy at 0.1 km/s and 1% intervals, respectively.

A. Tommasi et al. / Earth and Planetary Science Letters 227 (2004) 539–556550

.

t

5.2. Compositional seismic velocity anomalies

Analysis of the variation of bisotropicQ P- and S-

waves velocities as a function of the mean olivine

forsterite content in each sample (Fig. 11) shows that

A. Tommasi et al. / Earth and Planetary Science Letters 227 (2004) 539–556 551

S- and P-waves velocities decrease with Fe-enrich-

ment in olivine. However, P-wave velocities are also

highly sensitive to the modal composition of mantle

rocks; a decrease in enstatite content from 25% to 0%

(i.e., the transformation of a harzburgite in a dunite by

progressive dissolution of pyroxenes and crystalliza-

tion of olivine) results in an increase of P-waves

velocities equivalent to a variation in olivine compo-

sition from Fo91 to Fo85. As a consequence, dunites

display bisotropicQ P-wave velocities similar or

slightly higher than the ones shown by the lherzolites.

Lower P-wave, but similar S-wave velocities in

wehrlites relatively to dunites with similar olivine

Mg# also point to a stronger sensitivity of P-waves to

modal composition.

If we consider harzburgite RPA18A, which dis-

plays the least modified mineral compositions and

strongest CPO, as representative of the bnormalQoceanic lithospheric mantle, S-wave negative velocity

anomalies display a clear anticorrelation with olivine

forsterite content (Fig. 12). Maximum anomalies

(�2.25%) are associated with the wehrlites. However,

even the small compositional changes observed in the

lherzolites do result in significant negative seismic

Fig. 12. P-wave (a) and S-wave (b) velocity anomalies relatively to

RPA18A as a function of the mean forsterite content of olivine in

each sample. Symbols as in Fig. 7.

anomalies (up to �1.25%) for shear waves. P-wave

velocity anomalies show a more complicated pattern,

since the effects of olivine forsterite content and

modal composition may either add or subtract

depending on the physico-chemical conditions of the

magma–rock interaction. Maximum negative anoma-

lies (�1%) are displayed by Fe-rich wehrlite

THTFA3A and pyroxene-rich lherzolite THTFA1A.

Thus, Fe-enrichment in olivine and pyroxene crystal-

lization do result in a decrease of P-wave velocities,

as shown by the dunites and wehrlites and also,

although less clearly, by the lherzolites and harzbur-

gites. However, if Fe-enrichment in olivine is

accompanied by dunite-forming reactions, i.e., dis-

solution of pyroxenes and crystallization of olivine,

the final result is most often an increase in P-wave

velocities. Influence of enstatite content on P-wave

velocities should nevertheless decrease with increas-

ing depth, since bulk modulus pressure-derivatives

are higher in enstatite than in olivine, leading to

similar P-wave velocities in both minerals at ~200 km

depth [31].

6. Discussion

The studied xenolith suite samples an oceanic

lithosphere modified, to variable extent, by melt

percolation associated with mantle plumes. Interaction

with basaltic melts results in changes in both the

chemical (mainly Fe-enrichment) and the mineralog-

ical compositions of mantle. Clinopyroxene-rich

lherzolites are produced by a near-solidus melt–

freezing reaction occurring at the boundary of a

partial melting domain. This reaction preserves the

pre-existing deformation microstructures and CPO.

On the other hand, as they infiltrate the lithosphere,

melts formed at depth equilibrate with peridotites by

dissolving orthopyroxene. They become, as a result,

mineralogically non-reactive. However, Fe–Mg

exchanges (as well as trace element exchanges, such

Ti in spinel and pyroxenes) between melt and

minerals may still occur as long as percolation

continues. Complete dunitification takes place only

at low pressure, when the melt becomes olivine

saturated, and for high melt/rock ratios. Crystalliza-

tion of new, non-oriented olivine neoblasts during

these dunite-forming reactions leads to dispersion of

A. Tommasi et al. / Earth and Planetary Science Letters 227 (2004) 539–556552

pre-existing olivine CPO. Late melt crystallization in

dunites produces wehrlites. Predominance of olivine

or pyroxene crystallizing reactions, as well as their

extent, depends both on the composition and volume

of the magma and on pressure and temperature

conditions [25].

6.1. Seismic signature of melt–percolation processes

in the mantle

Although reactions change their composition,

crystallographic-preferred orientations and, hence,

seismic anisotropy of lherzolites and harzburgites

are little affected by the percolation of magmas. Most

harzburgites and lherzolites display coarse porphyro-

clastic microstructures and strong olivine CPO,

characteristic of deformation under high-temperature,

low-stress, asthenospheric conditions, which were

frozen in the lithospheric mantle by progressive

cooling of the plate [32]. Weakening of the olivine

CPO leading to almost isotropic properties is only

observed in dunites and wehrlites, suggesting that it is

restricted to domains of intense magma–rock inter-

action due to enhanced percolation or to accumulation

of magmas.

On the other hand, the iron-enrichment results in

an increase in density and a decrease in seismic

velocities for both P and S-waves. P-wave velocities

are also highly sensitive to variations in pyroxene

content, in particular enstatite. Therefore, olivine-

crystallization reactions may result in higher P-wave

velocities, while pyroxene crystallization decreases

them. Relative to normal mantle, seismic anomalies

associated with a Fe-rich wehrlite may attain �0.75%

and �2.25% for compressional and shear waves,

respectively. This S-wave negative seismic anomaly,

in particular, is equivalent to the one produced by a

200–100-K temperature anomaly in the mantle, the

lower values corresponding to more effective viscoe-

lastic relaxation processes in the upper mantle due to

higher temperature, lower grain sizes, and lower

frequencies [33,34]. Even the smaller compositional

changes observed in the lherzolites produce signifi-

cant seismic anomalies: up to �1% and �1.25% for

P- and S-waves, respectively.

Melt–rock interactions above mantle plumes can

both weaken seismic anisotropy and modify seismic

velocities. The resulting seismic anomalies may be

preserved in the lithospheric mantle for very long time

spans. Yet, before the present observations are used to

interpret seismic tomography or seismic anisotropy

data, some fundamental questions concerning the

extent and spatial distribution of these percolation-

induced compositional and textural variations should

be addressed. What is the spatial distribution of the

dunites and wehrlites in the lithospheric mantle? Do

they correspond to a large-scale modification of the

uppermost lithospheric mantle above a mantle plume

or are they reactional rocks produced by a localized

magma flow in the lithospheric mantle? Do the

compositional changes observed in the lherzolites

result from a diffuse percolation that may affect the

large domains of the lithosphere above the plume?

The spatial distribution and structural relationships

between different mantle lithologies (lherzolites,

harzburgites, and dunites) may be inferred from the

study of peridotite massifs representative of tectoni-

cally uplifted oceanic (ophiolites) and continental

(orogenic peridotites) mantle sections. These studies

highlight that melt transfer and melt–rock interactions

take place by mechanisms as varied as melt flow in

lithospheric centimetric to metric-scale vein conduits

and wall–rock reactions [35], melt extraction from

mantle sources via channeled porous flow [36], or

propagation of km-scale melting and percolation

fronts associated with thermal erosion of lithospheric

mantle [37].

Peridotite massifs are generally characterized by a

predominance of lherzolites or harzburgites. Dunites

are observed as irregular lenses or, most commonly,

tabular bodies, a few tens of centimeters to a few

hundreds meters thick. Within ophiolites, dunite

bodies are particularly common in the uppermost

mantle section, where they are interlayered with

gabbros and websterites [22]. Predominance of Fe-

rich dunites, wehrlites, and websterites is observed

only in km-scale ultramafic massifs interpreted as

representative of the shallow mantle beneath mag-

matic arcs, such as the Urals and Kohistan ultramafics

[25]. These observations suggest that formation of

dunites is essentially related to focused melt flow,

either in veins or via channeled porous flow, or to melt

accumulations at permeability barriers such as the

Moho discontinuity or the base of the lithosphere [25].

Thus, we regard as unlikely the development of large-

scale (few tens of kilometers) dunitic bodies in the

A. Tommasi et al. / Earth and Planetary Science Letters 227 (2004) 539–556 553

upper mantle and suggest therefore that dunitization

above a mantle plume cannot erase the lithospheric

seismic anisotropy over the length scales sampled in

seismological studies.

On the other hand, compositional changes

observed in the lherzolites and harzburgites, like Fe-

enrichment in olivine or pyroxene enrichment due to

melt–freezing reactions, might be related to pervasive

melt flow. Both processes are therefore good candi-

dates to form large-scale compositional anomalies in

the upper mantle. Melt–freezing reactions are thought

to result from infiltration of melt down a thermal

gradient from near- to subsolidus conditions [25].

They occur at the margins of asthenospheric melting

domains in oceanic ridges [23] or at the base of a

lithospheric plate brought into contact with a partially

molten, upwelling asthenosphere due to plume activ-

ity or mantle delamination processes [37]. The

progression of pervasive melt flow is limited by the

melt solidus, i.e., an isotherm that depends on the melt

composition, varying from 1150 8C for basaltic

magmas to 800 8C for volatile-rich metasomatic fluids

[38]. Thus, large-scale melt-induced compositional

changes should start at base of the lithosphere and

progress upwards as the plume heats the lithosphere.

6.2. Seismic anisotropy above mantle plumes

The present data suggests that, as far as it only

implies thermal and petrological processes, plume–

lithosphere interaction preserves the lithospheric

seismic anisotropy rather than erases it. This predic-

tion is in good agreement with shear wave splitting

measurements in plume-related islands in the Pacific,

Atlantic, and Indian oceans that usually show fast-

shear waves polarized at small angles to the absolute

plate motion in the hotspot reference frame and delay

times of 1 to 1.5 s [11,12,39–41].

Recent geochemical and petrophysical studies in

the Ronda peridotite massif (southern Spain) also

show that partial melting and melt transport associated

with an basthenospherizationQ process may change the

microstructure and the chemical composition of litho-

spheric mantle [37], but preserve the pre-existing

olivine CPO and seismic anisotropy [42]. Thus,

thermo-chemical erosion of the lithospheric mantle

may produce contrasting signatures for seismic veloc-

ities and anisotropy. The hot basthenospherizedQ

lithosphere will be imaged by seismic tomography as

a shallow low velocity anomaly suggesting a thinned

lithosphere, but seismic anisotropy measurements will

reflect the pre-existing lithospheric structure and hence

detect no variation between thinned and bnormalQdomains [42]. Such an apparent paradox between

seismic tomography and anisotropy data sets charac-

terizes for instance the Yellowstone hotspot wake in

the western US [43].

The apparent isotropy observed in Tahiti, Azores,

and La Reunion [11,41] remains an unsolved ques-

tion. Lherzolite xenoliths from Tahiti display indeed

very strong olivine CPO. Absence of shear wave

splitting may yet result from non-coherent olivine

CPO at the length scales sampled by SKS waves (50

km), from a vertical alignment of olivine [100] axes,

i.e., from vertical flow directions in the lithosphere

and asthenosphere (Fig. 8), or from a destructive

interference between the lithospheric and astheno-

spheric contributions. None of these hypotheses is

fully satisfactory. An upwelling plume may produce

vertical flow or small-scale variations in flow direc-

tion in the asthenosphere, but the lithosphere should

retain its olivine CPO unless it is mechanically

eroded. Moreover, seismic anisotropy data in most

oceanic island stations are better explained by olivine

CPO developed in response to a constant velocity

gradient between the plate and the deep mantle,

leading to similar orientations of the lithospheric and

asthenospheric anisotropies [32].

6.3. Compositionally induced seismic velocity

anomalies

The compositionally induced seismic anomalies

calculated for the lherzolites are equivalent to those

observed within the lithospheric mantle below the

Deccan and the Parana mesozoic large igneous

provinces: �1.5% and �2%, respectively, the higher

anomalies being observed in S-wave models [2–4].

They are however much smaller than those observed

using Rayleigh waves beneath the Ontong Java plateau

(�200 to �300 m.s�1, i.e., �5% to �7%; [5]). Such

high velocity anomalies can only be accounted for by

an extreme iron enrichment of the lithospheric mantle,

resulting in olivines with Mg#=76–78, values never

observed in upper mantle rocks at the Earth surface. A

thermal origin is still more unlikely, since it would

A. Tommasi et al. / Earth and Planetary Science Letters 227 (2004) 539–556554

imply a temperature anomaly z250 K [33,34]. Such a

temperature anomaly would induce both partial melt-

ing and high attenuation and neither of these phenom-

ena is observed below Ontong Java today [6].

The close spatial correlation between 1% and 2%

slow velocity anomalies within the lithospheric

mantle and the location of Paleogene volcanics and

magmatic underplating (part of the North Atlantic

igneous province) observed in a recent seismic

tomography study on the British Isles [44] suggests

that slow velocities may result from compositional

changes in the lithospheric mantle composition

similar to those described in this study. Finally, the

lower than average seismic velocities observed in the

mantle beneath the Bushveld province in the Kapvaal

craton also suggest local Fe-enrichment of the

cratonic lithosphere by this 2.05-Ga-old magmatic

event [45].

7. Conclusion

The microstructural and petrological analysis of a

series of mantle xenoliths from different archipelagoes

within the South Pacific Superswell suggests that the

oceanic lithosphere above a mantle plume undergoes a

complex sequence of magmatic processes that induces

significant changes in its chemical and modal

composition. These compositional changes, particu-

larly secondary crystallization of pyroxenes and iron

enrichment in olivine, result in lower seismic veloc-

ities for P- and S-waves. Relative to normal litho-

spheric mantle, compositionally induced seismic

anomalies may attain �2.2% for S-waves and �1%

for P-waves. Melt–freezing reactions and iron-enrich-

ment of olivine associated with diffuse flow of

magmas in the lithospheric mantle may thus be good

candidates to form large-scale compositional anoma-

lies in the upper mantle.

On the other hand, crystal-preferred orientations

(CPO) and hence seismic anisotropy are little affected

by these processes. Lherzolites and harzburgites,

independently from Fe-content, show high-temper-

ature porphyroclastic microstructures and strong

olivine CPO. Dunites and wehrlites display annealing

microstructures to which is associated a progressive

weakening of the olivine CPO. However, very weak,

almost random olivine CPO is rare, suggesting that

CPO destruction is restricted to domains of intense

magma–rock interaction due to localized flow or

accumulation of magmas.

Acknowledgements

Emilie Roulleau and Annelise Jourdan performed

some ESBD and microprobe analyses as part of their

Master and graduate research projects, respectively.

Thoughtful reviews by S. Karato, I. Jackson, G. Hirth,

and an anonymous referee as well as frequent

discussions with A. Vauchez greatly improved the

ms. We thank C. Nevado for the high-quality polished

thin sections for EBSD measurements and C. Merlet

and J.-M. Peris for help with electron microprobe

analyses. The Laboratoire de Tectonophysique’s

EBSD system was funded by the CNRS/INSU,

Universite of Montpellier II, and NSF project

bAnatomy of an Archean cratonQ. This work is part

of the PLUME project funded by the French

bMinistere de la RechercheQ program bAction Con-

certee Incitative Jeunes ChercheursQ.

Appendix A. Supplementary material

Supplementary data associated with this article can

be found, in the online version, at doi:10.1016/

j.epsl.2004.09.019.

References

[1] T.H. Jordan, Structure and formation of the continental

tectosphere, J. Petrol. (1988) 11–37.

[2] B.L.N. Kennett, S. Widiyantoro, A low seismic wavespeed

anomaly beneath northwestern India: a seismic signature of

the Decan plume, Earth Planet. Sci. Lett. 165 (1999)

145–155.

[3] J.C. VanDecar, D.E. James, M. Assumpcao, Seismic evidence

for coherent flow of the crust and upper mantle below south

America since the breakup of Gondwana, Nature 378 (1995)

25–31.

[4] M. Schimmel, M. Assumpcao, J.C. VanDecar, Upper mantle

seismic velocity structure beneath SE Brazil from P- and S-

wave travel time inversions, J. Geophys. Res. 108 (2003).

DOI:10.1029/2001JB000187.

[5] W.P. Richardson, E.A. Okal, S. van der Lee, Rayleigh-wave

tomography of the Ontong-Java plateau, Phys. Earth Planet.

Inter. 118 (2000) 29–51.

A. Tommasi et al. / Earth and Planetary Science Letters 227 (2004) 539–556 555

[6] B.M. Gomer, E.A. Okal, Multiple Scs probing of the

Ontong-Java plateau, Phys. Earth Planet. Inter. 138 (2003)

317–331.

[7] P.G. Silver, W.W. Chan, Shear wave splitting and subcon-

tinental mantle deformation, J. Geophys. Res. 96 (1991)

16429–16454.

[8] M. Heintz, A. Vauchez, M. Assumpcao, G. Barruol, M.

Egydio-Silva, Shear wave splitting in SE Brazil: an effect of

active or fossil upper mantle flow, or both? Earth Planet. Sci.

Lett. 211 (2003) 79–95.

[9] J.M. Kendall, S. Sol, C.J. Thomson, D.J. White, I. Asudeh,

C.S. Snell, F.H. Sutherland, Seismic heterogeneity and

anisotropy in the Western Superior Province, Canada: insights

into the evolution of an Archean craton, in: C.M. Fowler, C.J.

Ebinger, C.J. Hawkesworth (Eds.), The Early Earth: Physical,

Chemical, and Biological Development, Geol. Soc. London

Spec. Publ., 199 (2002) 27–44 London.

[10] H.H. Hess, Seismic anisotropy of the uppermost mantle under

oceans, Nature 203 (1964) 629–631.

[11] G. Barruol, R. Hoffman, Upper mantle anisotropy beneath

the GEOSCOPE stations, J. Geophys. Res. 104 (1999)

10757–10773.

[12] R.M. Russo, E.A. Okal, Shear-wave splitting and upper mantle

deformation in French Polynesia: evidence for small-scale

heterogeneity related to the Society hotspot, J. Geophys. Res.

103 (1998) 15089–15107.

[13] G. Barruol, D. Bosch, V. Clouard, E. Debayle, M.P. Doin, F.

Fontaine, M. Godard, F. Masson, A. Tommasi and C.

Thoraval, PLUME investigates the South Pacific Superswell,

Eos Trans. Am. Geophys. Union 83 (2002) 511, 514.

[14] R.C. Maury, M. El Azzouzi, H. Bellon, J.-M. Liotard, G.

Guille, H. Barczus, C. Chauvel, C. Diraison, C. Dupuy, P.

Vidal, R. Brousse, Geologie et petrologie de l’ıle de Tubuai

(Australes, Polynesie Francaise), C. R. Acad. Sci. Paris II 318

(1994) 1341–1347.

[15] R.A. Duncan, M.R. Fisk, W.M. White, R.L. Nielsen, Tahiti:

geochemical evolution of a French Polynesian volcano, J.

Geophys. Res. 99 (1994) 24341–24357.

[16] G.P. Brey, T. Koehler, Geothermobarometry in four-phase

lherzolites: II. New thermobarometers, and practical assess-

ment of existing thermobarometers, J. Petrol. 31 (1990)

1353–1378.

[17] P.R.A. Wells, Pyroxene thermometry in simple and complex

systems, Contrib. Mineral. Petrol. 62 (1977) 129–139.

[18] G.A. Gaetani, E.B. Watson, Open system behavior of

olivine-hosted melt inclusions, Earth Planet. Sci. Lett. 183

(2000) 27–41.

[19] G. Sen, A petrologic model for the constitution of the upper

mantle and crust of the Koolau Shield, Oahu, Hawaii, and

Hawaiian magmatism, Earth Planet. Sci. Lett. 62 (1983)

215–228.

[20] E.T. Berger, M. Vannier, Les dunites en enclaves dans les

basaltes alcalins des ıles oceaniques: approche petrologique,

Bull. Mineral. 107 (1984) 649–663.

[21] P.B. Kelemen, D.B. Joyce, J.D. Webster, J.R. Holloway,

Reaction between ultramafic rock and fractionating basaltic

magma: II. Experimental investigation of reaction between

olivine tholeiite and harzburgite at 1150–1050 8C and 5 kb, J.

Petrol. 31 (1990) 99–134.

[22] F. Boudier, A. Nicolas, Nature of the Moho transition zone in

the Oman ophiolite, J. Petrol. 36 (1995) 777–796.

[23] M. Godard, D. Jousselin, J.-L. Bodinier, Relationships

between geochemistry and structure beneath a palaeo-spread-

ing centre: a study of the mantle section in the Oman

Ophiolite, Earth Planet. Sci. Lett. 180 (2000) 133–148.

[24] K.T. Koga, P.B. Kelemen, N. Shimuzu, Petrogenesis of the

crust–mantle transition zone and the origin of lower crustal

wehrlite in the Oman ophiolite, Geochem. Geophys. Geosyst.

2 (2001). DOI:10.1029/2000GC000132.

[25] J.L. Bodinier, M. Godard, Orogenic, Ophiolitic, and Abyssal

Peridotites, in: K. Turekian, H. Holland (Eds.), Treatise on

Geochemistry Volume 2: Geochemistry of the Mantle and

Core, 2003, pp. 103–176.

[26] D. Mainprice, M. Humbert, Methods of calculating petro-

physical properties from lattice preferred orientation data,

Surv. Geophys. 15 (1994) 575–592.

[27] M.D. Collins, J.M. Brown, Elasticity of an upper mantle

clinopyroxene, Phys. Chem. Miner. 26 (1998) 7–13.

[28] T.S. Duffy, M.T. Vaughan, Elasticity of enstatite and its

relationships to crystal structure, J. Geophys. Res. 93 (1988)

383–391.

[29] E.H. Abramson, M. Brown, L.J. Slutsky, J. Zaug, The elastic

constants of San Carlos olivine up to 17 GPa, J. Geophys. Res.

102 (1997) 12252–12263.

[30] J.D. Bass, Elasticity of minerals, glasses, and melts, in: T.

Ahrens (Ed.), Mineral Physics and Crystallography: A

Handbook of Physical Constants, AGU Reference Shelf,

vol. 2, American Geophysical Union, Washington, 1995,

pp. 45–62.

[31] L.M. Flesch, B. Li, R.C. Liebermann, Sound velocities of

polycrystalline MgSiO3 orthopyroxene to 10 GPa at room

temperature, Am. Mineral. 83 (1998) 444–450.

[32] A. Tommasi, Forward modeling of the development of seismic

anisotropy in the upper mantle, Earth Planet. Sci. Lett. 160

(1998) 1–13.

[33] S. Karato, Importance of anelasticity in the interpretation

of seismic tomography, Geophys. Res. Lett. 20 (1993)

1623–1626.

[34] I. Jackson, J.D. FitzGerald, U.H. Faul, B.H. Tan, Grain-size

sensitive seismic wave attenuation in polycrystalline olivine, J.

Geophys. Res. 107 (2002) 2075–2360.

[35] J.L. Bodinier, G. Vasseur, J. Vernieres, C. Dupuy, J. Fabries,

Mechanisms of mantle metasomatism; geochemical evidence

from the Lherz orogenic peridotite, J. Petrol. 31 (1990)

597–628.

[36] P.B. Kelemen, N. Shimizu, V.J.M. Salters, Extraction of mid-

ocean-ridge basalt from the upwelling mantle by focused flow

of melt in dunite channels, Nature 375 (1995) 747–753.

[37] X. Lenoir, C. Garrido, J.-L. Bodinier, J.-M. Dautria, F. Gervilla,

The recrystallization front of the Ronda peridotite: evidence for

melting and thermal erosion of lithospheric mantle beneath the

Alboran basin, J. Petrol. 42 (2001) 141–158.

[38] D. McKenzie, Some remarks on the movement of small melt

fractions in themantle, Earth Planet. Sci. Lett. 95 (1989) 53–72.

A. Tommasi et al. / Earth and Planetary Science Letters 227 (2004) 539–556556

[39] C.J. Wolfe, F.L. Vernon III, Shear-wave splitting at central

Tien Shan: evidence for rapid variation of anisotropic patterns,

Geophys. Res. Lett. 25 (1998) 1217–1220.

[40] C.J. Wolfe, S.C. Solomon, P.G. Silver, J.C. VanDecar, R.M.

Russo, Inversion of body-wave delay times for mantle

structure beneath the Hawaiian Islands; results from the

PELENET experiment, Earth Planet. Sci. Lett. 198 (2002)

129–145.

[41] F. Fontaine, G. Barruol, D. Reymond, E. Debayle, A.

Tommasi, Seismic evidence of the upper mantle anisotropy

beneath French Polynesia: PLUME preliminary results, Geo-

phys. Res. Abstract 5 (2003) 8203.

[42] A. Vauchez, C. Garrido, Seismic properties of an asthenosph-

erized lithospheric mantle: constraints from lattice preferred

orientations in peridotite from the Ronda massif, Earth Planet.

Sci. Lett. 192 (2001) 235–249.

[43] D. Schutt, E.D. Humphreys, K. Dueker, Anisotropy of the

Yellowstone hotspot wake, eastern Snake River Plain, Idaho,

Pure Appl. Geophys. 151 (1998) 443–462.

[44] S.J. Arrowsmith, J.-M. Kendall, J.C. Vandecar, N.J. White,

D.C. Booth, Seismic images reveal plume–lithosphere inter-

action beneath the British Isles, Geophys. Res. Abstr. 5 (2003)

646.

[45] D. James, M. Fouch, J. VanDecar, S. Van der Lee, Tecto-

spheric structure beneath southern Africa, Geophys. Res. Lett.

28 (2001) 2485–2488.

[46] Y. Niu, C.H. Langmuir, R.J. Kinzler, The origin of abyssal

peridotites: a new perspective, Earth Planet. Sci. Lett. 152

(1997) 251–265.

[47] M.J. Walter, T.W. Sisson, D.C. Presnall, A mass proportion

method for calculating melting reactions and application to

melting of model upper mantle lherzolite, Earth Planet. Sci.

Lett. 135 (1995) 77–90.

[48] R.J. Tracy, Petrology and genetic significance of an ultramafic

xenolith suite from Tahiti, Earth Planet. Sci. Lett. 48 (1980)

80–96.

[49] E.T. Berger, R. Brousse, Dunite nodules from Tubuai (Austral

Islands, South Pacific), Cah. Indo-Pac. 2 (1980) 207–218.

[50] E.T. Berger, Petrographic heterogeneities of the South Pacific

mantle below the Austral archipelago: evidence and interpre-

tation using ultramafic xenolith investigations, Bull. Soc.

Geol. Fr. 1 (1985) 207–216.

[51] E.T. Berger, B.C. Weber, Chemical and thermal exchanges

between basaltic plumes and the lithospheric mantle in the

Society islands: mineralogic and geothermometric evidence

from inclusions from the mouth of the Papenoo volcano,

Tahiti, C. R. Acad. Sci., Paris 308 (1989) 1049–1054.

[52] E.H. Hauri, N. Shimizu, J.J. Dieu, S.R. Hart, Evidence for

hotspot-related carbonatite metasomatism in the oceanic upper

mantle, Nature 365 (1993) 221–227.

[53] Q. Qi, B.L. Beard, Y. Jin, L.A. Taylor, Petrology and

geochemistry of Al-augite and Cr-diopside group mantle

xenoliths from Tahiti, Society islands, Int. Geol. Rev. 36

(1994) 152–178.

[54] C. Herzberg, M.J. O’Hara, Plume-associated ultramafic

magmas of Phanerozoic age, J. Petrol. 43 (2002) 1857–1883.

[55] D. Elthon, Partitioning of Ni between olivine and high-MgO

basaltic liquids, Lunar Planet. Sci. XVIII (1987) 258–259.

[56] R.J. Kinzler, T.L. Grove, Primary magmas of mid-ocean

ridge basalts: 2. Applications, J. Geophys. Res. 97 (1992)

6907–6926.