The scale and geometry of chemical and isotopic heterogeneities in the source of plumes have important scientific implications on the nature, composition and origin of plumes and on the dynamics of mantle mixing over time. Here, we address these issues through the study of Marquesas Islands, one of the Archipelagoes in Polynesia. We present new Sr, Nd, Pb, Hf isotopes as well as trace element data on lavas from several Marquesas Islands and demonstrate that this archipelago consists of two adjacent and distinct rows of islands with significantly different isotopic compositions. For the entire 5.5 Ma construction period, the northern islands, hereafter called the Ua Huka group, has had systematically higher 87Sr/86Sr and lower 206Pb/204Pb ratios than the southern Fatu Hiva group at any given 143Nd/144Nd value. The shape and curvature of mixing arrays preclude the ambient depleted MORB mantle as one of the mixing end-members. We believe therefore that the entire isotopic heterogeneity originates in the plume itself. We suggest that the two Marquesas isotopic stripes originate from partial melting of two adjacent filaments contained in small plumes or “plumelets” that came from a large dome structure located deep in the mantle under Polynesia. Low-degree partial melting under Marquesas and other “weak” Polynesian hot spot chains (Pitcairn-Gambier, Austral-Cook, Society) sample small areas of the dome and preserve source heterogeneities. In contrast, more productive hot spots build up large islands such as Big Island in Hawaii or Réunion Island, and the higher degrees of melting blur the isotopic variability of the plume source.
 The Marquesas archipelago has also been the object of great interest for geophysical studies, particularly since McNutt and coworkers [McNutt, 1998; McNutt and Bonneville, 2000] discovered that French Polynesia was built on top of a large topographic height attributed to the so-called Polynesian Superswell. Its existence is usually thought to be linked to the presence of a large region of hot material rising from deep in the mantle [Davaille, 1999; Romanowicz and Gung, 2002; Courtillot et al., 2003; Montelli et al., 2006] but the way these observations relate to geochemical constraints remains unsolved.
 Here we reexamine the scale of isotopic variations using high-precision Sr, Nd, Hf and Pb isotopic measurements obtained on a large set of new samples that were collected and dated during the first detailed geological mapping of Marquesas Islands. We demonstrate the existence of isotopic stripes different from those suggested recently by Huang et al.  and we propose that they originate from small-scale heterogeneities located within elongated filaments in rising plumelets coming from a large dome structure. Further comparison with other ocean islands suggests that weak plume-derived islands such as Polynesia are ideal sites to determine the scale of plume heterogeneities while strong plume-derived islands such as Hawaii are the best choices to establish the average composition of plumes.
2. Geological Setting of the Marquesas Archipelago
 The ca. 350 km-long Marquesas Archipelago is located in northern French Polynesia. It includes eight main islands (Eiao, Nuku Hiva, Ua Huka, Ua Pou, Hiva Oa, Tahuata, Motane, Fatu Hiva; Figure 1) and a few islets (“motus”) and seamounts (Figure 1). Their age decreases toward the SE [Duncan and McDougall, 1974] from 5.5 Ma in Eiao to 0.6–0.35 Ma on a seamount south of Fatu Hiva (DH12 [Desonie et al., 1993]). The Marquesas Islands lie on 53–49 Ma old oceanic crust generated at the axis of the Pacific-Farallon ridge. This crust is anomalously shallow given its age [Crough and Jarrard, 1981] and the so-called “Marquesas swell” has been interpreted as resulting from uplift due to a rising mantle plume [Crough and Jarrard, 1981]. However, the crustal thickness reaches 15–20 km below the central part of the Archipelago [Filmer et al., 1993; Caress et al., 1995], and the swell could also result from the buoyancy of such a thick basaltic crust [McNutt and Bonneville, 2000]. The origin of the crustal thickening was attributed either to (i) Plio-Quaternary underplating of plume magmas just below the Moho [Caress et al., 1995; McNutt and Bonneville, 2000] or (ii) to the edification of the Archipelago over a small 50–45 Ma old oceanic plateau that formed near the axis of the Pacific-Farallon ridge [Gutscher et al., 1999].
 A final puzzling feature is that no active volcanoes are known at the southeastern end of the chain. 40K-40Ar ages younger than 1 Ma have only been obtained on a seamount south of Fatu Hiva (DH12, [Desonie et al., 1993]), and on the post-shield strombolian Teepoepo and Tahoatikikau cones on Ua Huka (Table 1 and Table S1 in the auxiliary material). Most authors consider that the Marquesas Fracture Zone (MFZ) [Pautot and Dupont, 1974] may have terminated the hot spot activity [McNutt et al., 1989; Brousse et al., 1990; Guille et al., 2002]. However, the MFZ is aseismic [Jordahl et al., 1995] and young volcanic rock has not been recovered from the adjacent Marquesas Fracture Zone Ridge (MFZ ridge; Figure 1). The seamount chain parallel to the MFZ ridge ca. 50 km to the north (Figure 1) has not yet been dredged and it could potentially be the present location of Marquesas hot spot activity.
Table 1. Sr, Nd, Hf, and Pb Isotopic Compositions of Lavas From Ua Huka, Fatu Hiva, Hiva Oa, Motane, and Tahuataa
The average daily measurements of the Ames-Grenoble Hf standard ranged from 0.282153 to 0.282162 (23 measurements) and all data are normalized to 0.282160, the value recommended by Chauvel et al. . The average daily measurements for the Ames-Rennes Nd standard ranged from 0.511967 to 0.511980 (29 measurements) and all data are normalized to 0.511961, the value recommended by Chauvel and Blichert-Toft . The average 87Sr/86Sr value for the NBS 987 standard was 0.710263 (9 measurements). Based on repeated measurements of NBS 981, we estimate maximum errors on the Pb isotopic ratios at 100 ppm for 206Pb/204Pb, 150 ppm for 207Pb/204Pb and 200 ppm for 208Pb/204Pb. Sources of ages: unspiked 40K-40Ar ages measured on separated groundmass and reported in Table S1 and in Blais et al.  for Ua Huka. Based on field relationships, the age of UH66 is assumed to be similar to that of UH27, the age of MT12 similar to that of MT11 and the ages of TH04 and TH05 similar to that of TH14. Longitudes and latitudes are expressed in decimal degrees. The 87Sr/86Sr initial ratios were calculated using the Rb and Sr concentrations reported in Table S2. Dup stands for complete duplicate analysis.
Post shield phase
Post shield phase
Post shield phase
 Most previous authors have used the main N30–40°W trend of the Marquesas Archipelago, oblique to the current N65°W motion of the Pacific plate, to draw age-distance plots and these display considerable scatter [Brousse et al., 1990; Desonie et al., 1993; Guille et al., 2002]. Similar plots using the N65°W trend give better fits [Legendre et al., 2006], especially when considering only unspiked 40K-40Ar ages measured on separated groundmass [Charbit et al., 1998] because they are less scattered than conventional data on whole rocks.
 Marquesas basaltic lavas range from quartz tholeiites to basanites [Guille et al., 2002]. Generally speaking, alkalinity tends to increase through time in a given island, e.g., Nuku Hiva or Ua Huka. Shields are often tholeiitic while post-shield lavas are often alkaline, but there are many exceptions. For instance, alkali basalts and basanites make up most of the shield of Ua Pou [Legendre et al., 2005b]. Alkaline lavas also occur in the shields of Eiao [Caroff et al., 1995], Motane and Tahuata [Maury et al., 2012], and conversely tholeiites are present in the post-shield units of Nuku Hiva [Legendre et al., 2005a], Hiva Oa, Fatu Hiva and Tahuata [Maury et al., 2012].
 Here, we present new geochemical data obtained on five Marquesas Islands: Ua Huka, Fatu Hiva, Hiva Oa, Motane and Tahuata. The simplified geological maps shown in Figure S1 highlight their main features. The five islands have very different sizes with only 13 km2 for Motane and about 320 km2 for Hiva Oa. Motane represents the crescent-shaped remnant of the caldeira walls of a large shield volcano. Ua Huka, Tahuata and Fatu Hiva Islands consist in external shield volcanoes in the caldeira of which are nested post-shield inner volcanoes (Figure S1). There is usually little to no time gap between the end of edification of the shield and the start of the post-shield volcano for all these islands (Figure S1). Ua Huka shows, in addition, two much younger edifices attributed to a Quaternary rejuvenation of volcanism [Legendre et al., 2006]. Finally, Hiva Oa consists of three coalescent shields, the central one being partly overlain by post-shield lavas (Figure S1).
3. Analytical Techniques
 Before chemical analyses, all samples were crushed using an agate mortar. Major element data were obtained by Inductively Coupled Plasma-Atomic Emission Spectrometry (ICP-AES) at IUEM, Plouzané using the method of Cotten et al. . The international standards used for calibration were ACE, BEN, JB-2, PM-S and WS-E and the relative standard deviations are ±1% for SiO2, and ±2% for other major elements except P2O5 and MnO (absolute precision ±0.01%). Trace elements were measured on the same rock powders using an ICP-MS in Grenoble and the method described in detail by Chauvel et al. . Reproducibility of the measurements is typical better than 5% and the accuracy of the measured concentrations is better than 5%, as testified by repeated measurements of international rock standards (BHVO-2, BR, BEN, BR 24) (see Table S2).
 To avoid potential contamination of the samples during crushing, rock chips were used for the isotopic measurements. The chips were leached using 6N HCl for about one hour and the leachate was discarded. The leached chips were dissolved in ultraclean HF combined with HNO3. The chemical protocol is described by Chauvel et al.  but we also specify that Pb was isolated using a two columns procedure to insure proper purification of the element. Several blanks were measured during the course of the study. They were lower than 70 picograms for Pb, 50 picograms for Sr, 40 picograms for Nd and 40 picograms for Hf. These values are all insignificant relative to the amount of element isolated for each isotopic system.
 Hf and Nd isotopic measurements were performed at ENS Lyon using the Nu instrument, normalizing ratios 179Hf/177Hf = 0.7325 and 146Nd/144Nd = 0.7219, and a daily bias correction based on the average measurement of Ames-Grenoble Hf and Ames-Rennes Nd standards (see footnote of Table 1). The same instrument was used for the Pb isotopes. Mass fractionation was corrected using the Tl addition technique [White et al., 2000] and standard bracketing with the NBS 981 standard which was measured every third sample. Standard values are from Galer and Abouchami . Sr isotopic ratios were measured on a TIMS in Brest in static mode. Values measured for the NBS 987 standard during the course of the study are given in the footnote of Table 1.
 K-Ar ages were obtained at LSCE (Laboratoire des Sciences du Climat et de l'environnement) in Gif-sur-Yvette. The samples were crushed, sieved to 0.25–0.125 mm size fraction and ultrasonically washed in acetic acid. Potassium and argon were measured on the microcrystalline groundmass, after removal of phenocrysts using heavy liquids and magnetic separations. This process removes at least some potential sources of systematic error due to the presence of excess 40Ar in olivine and feldspar phenocrysts [Laughlin et al., 1994]. The K content of the separated groundmass was measured by ICP-AES in Brest using the same method as for major elements while Ar analyses were performed in Gif-sur-Yvette using the procedures of Guillou et al. . Ages given in Table S1 were calculated using the constants recommended by Steiger and Jäger .
 Major and trace element data are provided in Table S2. Basalts belonging to the shield and post-shield phases are either tholeiitic or alkalic with SiO2 contents ranging from 45 to 48% and MgO between 4 and 12%. In contrast, the rejuvenated phase of volcanic activity in Ua Huka consists of basanites with much lower SiO2 (41–45%) and generally higher MgO contents. In Ua Huka, four differentiated rocks belonging to the post-shield phase were also analyzed. Trace elements for all samples are plotted as spidergrams in Figure 2. All basaltic lavas display patterns enriched in incompatible elements as usually observed in other ocean islands and no marked differences exist between the various islands. However, four samples (TH04 and TH13 from Tahuata, MT04 from Motane and HV76 from Hiva Oa) display positive Pb anomalies, a feature that might be explained by the presence of post-magmatic sulfides that occur sporadically in basaltic samples from these islands [Maury et al., 2012]. In Ua Huka, the benmoreite, the two trachytes and the phonolite present very distinct trace element patterns (Figure 2) due to the effect of intense fractional crystallization.
 The new isotopic data obtained on five different Marquesas Islands are given in Table 1 and compared with published Nd, Sr, Hf and Pb isotopic ratios in Figure 3. Data from the five analyzed islands cover the entire range of known isotopic variation for the Marquesas in 87Sr/86Sr versus 143Nd/144Nd space: inclusion of samples from the shield-building phase of Ua Huka extend it to lower Nd and higher Sr (Figure 3a). In addition, each individual island defines a large field covering at least half of the total Marquesas variation. In comparison to Nd and Sr, Pb isotopic ratios are relatively constant (see Figures 3b and 3d) with 206Pb/204Pb ratios ranging only from 18.8 to 19.7 and 207Pb/204Pb from 15.55 to 15.63. The Hf isotopic compositions, the first reported for Marquesas Islands, vary between 0.28288 and 0.28306, correlate with Nd isotopic ratios, and define in Hf-Nd isotopic space an array located above the Austral-Cook Islands in the middle of the oceanic island array (see Figure 3c).
 The most interesting feature displayed by the isotopic data is the existence within the Marquesas Archipelago of two parallel arrays in Nd versus Sr isotopic space (Figure 4). Two groups of islands can be distinguished, one to the north, the other to the south of a N65°W line that follows the direction of the Pacific plate motion (see Figure 1) and passes through Hiva Oa where it separates the Taaoa shield from the other volcanoes (see map shown in Figure S1). The northern group of islands, shown by various shades of blue dots in Figure 4, includes Eiao, Hatutu, Nuku Hiva, Ua Huka, Fatu Huku and the young lavas from the eastern side of Hiva Oa Island (Puamau, Ootua and Temetiu volcanoes). The southern group, shown by various shades of orange dots in Figure 4, includes Ua Pou, Tahuata, the old Hiva Oa lavas (from the Taaoa volcano located on the western side of the island), Motane, Fatu Hiva and Motu Nao. Like Chauvel et al.  who used the name of the youngest island for each of the three trends recognized along the Austral-Cook Island chain, we call the northern group the “Ua Huka group” and the southern group the “Fatu Hiva group.” The distinction between the two groups is not related to the age of the islands: volcanism in the Ua Huka group occurs over a time period from 5.52 Ma at Eiao to 0.76 Ma at Ua Huka (Table 2) while it extends from 4.0 Ma at Ua Pou to 0.35 Ma (seamount DH12 south of Fatu Hiva) in the Fatu Hiva group.
Table 2. Main Characteristics (Volume, Age, and Location) of the Marquesas Islandsa
Emerged Surface (km2)
Edifice Volume (km3)
Groundmass/Whole Rock K-Ar Ages
Distance to the MFZR Along N65W Trend (km)
Submerged edifice volumes above the −3,000 m isobaths were calculated with the GMT software and emerged volumes from topographic maps. The corresponding total volume of volcanic edifices is estimated at 42,500 km3. This value includes 2,800 km3 of seamounts scattered throughout the archipelago (Figure 1). When available, unspiked 40K-40Ar ages measured on separated groundmass [Charbit et al., 1998] were preferred to conventional data on whole rocks (Hatutu and Fatu Huku from Brousse et al.  shown in italics) because of possible excess of radiogenic Ar into olivine and pyroxene phenocrysts [Laughlin et al., 1994]. Sources of unspiked ages on groundmass: Table S1; Caroff et al.  for Eiao; Maury et al.  for Nuku Hiva; Legendre et al.  and Blais et al.  for Ua Huka; Legendre et al. [2005b] for Ua Pou and Maury et al.  for Hiva Oa, Tahuata, Motane and Fatu Hiva. Two distinct volcanic events are recognized in Ua Huka (3.24–2.43 Ma and 1.15–0.76 Ma). Ages younger than 1 Ma only occur for Ua Huka and a seamount south of Fatu Hiva (DH12, 0.60–0.35 Ma [Desonie et al., 1993]). MFZR: Marquesas Fracture Zone Ridge.
 Both groups define a very large range of isotopic compositions (for example, 87Sr/86Sr varies from 0.703 to 0.706 in the two groups) but the Ua Huka group has systematically higher Nd isotopic compositions at any given Sr isotopic composition. The distinction between the two groups is less clear in other isotopic spaces but in general the Fatu Hiva group has higher 206Pb/204Pb ratios than the Ua Huka group at any given 87Sr/86Sr or 143Nd/144Nd (Figures 5a and 5b). Finally, the maximum width on the seafloor of the Ua Huka group is about 120 km while that of the Fatu Hiva group is smaller at about 60 km (Figure 1).
5.1. Is There a Relationship Between Timing of Eruptions, Types of Lavas and Isotopic Compositions?
 In some oceanic islands, a clear relationship exists between timing of lava eruption and chemical and isotopic compositions. For example, in Hawaii, extensive studies [e.g., Garcia et al., 2010; Hanano et al., 2010] demonstrated that the shield volcanic phase consists almost exclusively of tholeiites whose Sr isotopic compositions are systematically more radiogenic than those of the post-shield alkali lavas. In contrast, in Kerguelen, the opposite is observed [Doucet et al., 2002]. In the Marquesas, no clear relationship exists. For example, in Ua Huka, the shield-building phase consists of tholeiites with radiogenic Sr isotopes associated with low Nd isotopic ratios and it is followed by an alkaline post-shield phase with similar isotopic compositions (see Table 1). It is only the rejuvenated basanitic volcanism that has distinctly lower Sr isotopic compositions. In Tahuata and Motane, tholeiites and alkali basalts occur both in shield and post-shield phases (Table 1) while in Fatu Hiva and Hiva Oa only tholeiites exist. However, in all these islands, large Sr isotopic variability is observed. Our new results together with previously published data show therefore that no clear systematic exists and they do not support the claim that Marquesas shield lavas have lower 87Sr/86Sr ratios than post-shield lavas [Castillo et al., 2007]. Our new trace element data also show that fractionation between highly incompatible and mildly incompatible trace elements is primarily controlled by degree of partial melting. Figure 6 clearly shows that when La/Yb is plotted as a function of La concentration, the alkali series defines a field with systematically higher La/Yb than the tholeiites, as expected if alkali basalts result from lower degrees of partial melting than tholeiites. In contrast, if the lavas are sorted as shield versus post-shield, no clear distinction exists and the only lavas with clearly higher La/Yb ratios are the basanites forming the Ua Huka rejuvenated volcanic phase. Finally, simple relationships between trace elements and isotopic compositions are not really obvious. Willbold and Stracke , who compiled worldwide data for ocean island basalts, showed that the Ce/Pb ratio of EM-type lavas is usually lower than the canonical “mantle” value of 25 [Hofmann et al., 1986] and that their Nb/U ratio is generally scattered and not significantly different from the value of 47 suggested by Hofmann et al. . They also mention that EM basalts have elevated Th/U ratios and lower Nb/Th ratios than HIMU basalts. Using our new data, we examine the relationship between these key trace element ratios and 87Sr/86Sr, a good proxy for the intensity of the EM signal present in the various lavas. When taken as a whole, our new data define a scattered field for Ce/Pb versus 87Sr/86Sr, but if samples with positive Pb anomalies (see Figure 2) are ignored, lavas define a broad negative trend between Ce/Pb ≈ 40 at low 87Sr/86Sr and Ce/Pb ≈ 20 at high 87Sr/86Sr, covering the entire range reported for ocean island basalts (Figure 7). No clear relationship appears between 87Sr/86Sr and Nb/U or Th/U (Figure 7), but our data confirm the high Th/U suggested by Willbold and Stracke , with an average value of 4.8. In contrast, when Nb/Th is plotted as a function of 87Sr/86Sr (Figure 7), a distinct negative trend is observed between Nb/Th ≈ 13 at low 87Sr/86Sr and Nb/Th ≈ 8 at high 87Sr/86Sr; this relationship confirms the presence in the Marquesas plume source of low Nb sedimentary material as already suggested by numerous authors [Dupuy et al., 1987; Dostal et al., 1998; Willbold and Stracke, 2006]. In summary, the relationships between isotopic compositions as exemplified with 87Sr/86Sr, and trace element ratios do not provide new and unambiguous constraints on the composition of the depleted and enriched end-members present in the source of Marquesas basalts but they confirm previous interpretations about the presence of crustal material in the source of lavas with elevated Sr isotopic compositions.
5.2. What Do the Isotopic Stripes Tell Us About the Plume Source?
 In Figure 4 we show that the relationship between Sr and Nd isotopic compositions in Marquesas lavas defines two parallel isotopic bands that are related to the position of the islands along the chain. The data previously published and compiled in the GEOROC database gave a hint [Chauvel et al., 2009; Huang et al., 2011] but our new data on Ua Huka, Hiva Oa, Tahuata, Fatu Hiva and Motane make the distinction extremely clear. The most striking feature comes from data for Hiva Oa where we observe that the 2.55 Ma old Taaoa volcano lavas, which are located on the western side of the island, belong to the Fatu Hiva group while the younger lavas (2.27 to 1.44 Ma) have very different isotopic compositions and belong clearly to the Ua Huka group (see Figure 4 and Tables S1 and S3 for high-precision ages and chemical compositions). The clear difference between the two groups within Hiva Oa shows that different mantle sources are sampled at distances less than 10 km and with a time difference of only 0.3 Ma (Table 1 and Figure S1).
 The difference between the two isotopic stripes observed along the Marquesas Archipelago can be evaluated in a quantitative way by calculating the statistical difference between the two arrays. The thick blue and red curves shown in Figure 4 represent the best fit to the fields defined by samples in each group. The only assumptions used to calculate the regression curves are the Sr isotopic compositions, which were fixed at 0.70275 and 0.70650 for the depleted and enriched end-members. In Figure 4, we show the two calculated regressions curves (thick dark blue and red curves) as well as the associated errors shown as thin blue and orange curves (see Text S1 for more explanations). The Ua Huka group defines an array between two end-members: a depleted one (A) with a 143Nd/144Nd ratio of 0.51311 ± 13 at 87Sr/86Sr = 0.70275 and an enriched end-member (B) with a 143Nd/144Nd ratio of 0.51270 ± 6 at 87Sr/86Sr = 0.7065. The mixing hyperbola has a curvature controlled by the ratio R of Sr/Nd in component A over Sr/Nd in component B of 2.2 (max: 5.7, min: 0.88) (see Table 3). In contrast the Fatu Hiva group defines a regression curve between a depleted end-member (C) with a 143Nd/144Nd ratio of 0.51295 ± 5 at 87Sr/86Sr = 0.70275 and an enriched end-member (D) with a 143Nd/144Nd ratio of 0.51253 ± 18 at 87Sr/86Sr = 0.7065 (see Figure 4 and Table 3). Here, the R ratio is equal to 0.73 (2.0, 0.27). Statistically, the two arrays have curvatures that are close to straight lines but the mixing end-members are significantly different.
Table 3. Calculated Compositions for the Four Mixing End-Members
 In isotopic spaces other than Sr versus Nd, the distinction between the two groups is less clear. Our new Hf isotopic data do not show a clear distinction between the two groups when combined with Nd isotopic compositions (see Figure 3 where data for Ua Huka define a trend that is indistinguishable from the field defined by the Fatu Hiva group islands – Tahuata, Fatu Hiva and Motane). This suggests that the process that created the Sr-Nd isotopic difference between the two groups did not decouple the Hf and Nd isotopic systems. Indeed, if Sr isotopes are plotted as a function of Hf isotopic compositions, the distinction highlighted in Figure 4 is reproduced. Finally, in Pb isotopic spaces, straight lines such as those reported for Hawaiian rocks are absent when all published data are compiled (Figures 3b and 3d). However, when only our high-precision Pb data are used, two trends with significantly different slopes can be distinguished (Figure 8). In each group, samples with high 206Pb/204Pb also have high 208Pb/204Pb but, for 207Pb/204Pb, the arrays differ significantly since the slope of the Ua Huka group is positive while that of the Fatu Hiva group is slightly negative. In addition, the two arrays intersect at a 206Pb/204Pb of about 19.15. When 206Pb/204Pb is plotted versus 87Sr/86Sr or 143Nd/144Nd and all data from the literature are used, the two groups are also distinct: the Fatu Hiva group is always located at higher 206Pb/204Pb than the Ua Huka group (Figures 5a and 5b). The array defined by the Fatu Hiva group has a strong curvature that constrains the composition of two end-members, if interpreted as a mixing array. The enriched end-member (D) with elevated 87Sr/86Sr has a 206Pb/204Pb of 18.95 ± 0.68 while the depleted end-member (C) with low 87Sr/86Sr has a 206Pb/204Pb of 19.91 ± 0.73 (Table 3). Using the relationship between 143Nd/144Nd and 206Pb/204Pb ratios (see Figure 5b), we obtain similar values for the Pb isotopic compositions of the two end-members (C) and (D). Interestingly, the 206Pb/204Pb ratio close to 20 obtained for the depleted end-member of the Fatu Hiva group eliminates the possibility that this component is ambient depleted mantle, as sampled along mid-ocean ridges. We suggest therefore that the entire isotopic range defined by the lavas comes from isotopic heterogeneities located in the plume itself.
 In contrast to the Fatu Hiva group, the Ua Huka group defines far more diffuse arrays in Figures 5a and 5b suggesting that the isotopic heterogeneities cannot be simply explained by binary mixing when Pb isotopic compositions are considered. Nevertheless the mixing arrays point to a poorly defined 206Pb/204Pb ratio higher than 19.5 for the depleted end-member (A) and the enriched end-member (B) has a better constrained 206Pb/204Pb ratio at 19.00 ± 0.63 (Table 3).
 The 206Pb/204Pb ratio of about 19 for the two enriched end-members (B and D) calculated using the observed correlations with Sr and Nd isotopic compositions in Figure 5 are entirely consistent with the tendencies shown by Pb isotopes measured on our samples alone (Figure 8). However, it is not the case for the depleted end-member of the Ua Huka group (end-member A), which has a value lower than 18.8 when our samples are considered alone (Figure 8) while it has a very poorly constrained value, but higher than 19.5, when all data are used (Figure 5). Acquisition of high-precision Pb isotopic data on samples from Eiao and Nuku Hiva might help solve the discrepancy.
 Finally, the curvature of the arrays in Figures 5a and 5b constrains the difference in Nd/Pb and Sr/Pb elemental ratios of the depleted and enriched components: for the Fatu Hiva group, the depleted end-member, which has high 206Pb/204Pb, has Sr/Pb and Nd/Pb ratios that are about 2.5 times those of the enriched end-member (Table 3). For the Ua Huka group, the contrast between the two end-members is less well constrained but the relationships between 87Sr/86Sr, 143Nd/144Nd and 206Pb/204Pb suggest that both Sr/Pb and Nd/Pb in the depleted component are significantly higher than in the enriched component (Table 3).
 Such differences in Sr/Pb and Nd/Pb ratios between sources should translate into differences in the lavas themselves and this is indeed seen when Sr/Pb and Nd/Pb are plotted as a function of 87Sr/86Sr (Figure S2) even though melting and fractional crystallization blurred the source ratios due to the difference in incompatibility of the three elements.
5.3. Origin of the Isotopic Stripes
 Since the two isotopic arrays are evident in small geographical areas (about 120 km for the Ua Huka trend and 60 km for the Fatu Hiva trend, see Figure 1) and located very close to one another in the archipelago, it is possible that the two trends originate through shallow processes. Contamination of the ascending magmas by the oceanic plate and overlying sediments could be a cause of isotopic changes, especially in the intermediate and evolved liquids [Caroff et al., 1995, 1999; Legendre et al., 2005a, 2005b]. In Figure 9, we demonstrate that the very large range of Sr isotopic compositions (0.7028 to 0.7061) is unrelated to the amount of MgO, a good proxy for fractional crystallization and potential interactions with the surrounding crust. We are therefore convinced that the range of isotopic compositions displayed by the Marquesas volcanics is not due to near-surface processes but is characteristic of the plume source.
Abouchami et al.  attributed the existence of the two Pb isotopic stripes in Hawaii to heterogeneities in the Hawaiian plume. More recently, the observation was extended to other isotopic systems and back in time to about 5 Ma [Tanaka et al., 2008; Weis, 2010] but the origin of the asymmetry remains strongly debated. Very recently, Weis  and Weis et al.  associated the existence of the two Hawaiian trends with material from the Pacific “Large Low Shear Velocity Province” (LLSVP) located at the core-mantle boundary. They further suggested that the isotopic similarities between Hawaii and Pitcairn were not fortuitous and that the two plumes were sampling the same source material. Huang et al.  made a different parallel: they compared the Hawaiian isotopic stripes with similar features along Samoan and Marquesas islands to conclude that Hawaii was sampling the northern side of the Pacific LLSVP while Samoa and Marquesas were sampling its southern side. Like Huang et al. , it might be argued that the location of the Marquesas Archipelago above the LLSVP could explain the two isotopic bands.
 Here we adopt a different approach because we believe that the main characteristics of the Marquesas isotopic stripes differ from those recognized for Hawaii and Samoa. We need to explain the coexistence of two different sources that melted next to each other below the Marquesas Archipelago for an extended period of time of over 5 Ma. First of all, in contrast to the sharp distinction in 208Pb*/206Pb* versus 143Nd/144Nd displayed by the Kea and Loa trends in Hawaii [Huang et al., 2011; Weis et al., 2011] and the Malu and Via trends in Samoa [Huang et al., 2011], there is no significant difference in 208Pb*/206Pb* between the Ua Huka and the Fatu Hiva groups in the Marquesas (Figure 10). Indeed, the ranges of 208Pb*/206Pb* in the two groups are basically identical and the small difference seen in Figure 10 comes from the slightly higher 143Nd/144Nd ratios of the Ua Huka group. The real distinction between the two Marquesas groups comes primarily from the relationship between Nd and Sr isotopic compositions, as shown in Figure 4. The second very important difference is the strength and persistence of the plumes. The Hawaiian plume has been active for at least 85/81 Ma [Regelous et al., 2003] and led to the production of very large volumes of lavas. Within the past million years, 213,000 km3 of lavas erupted to form the Big Island of Hawaii [Robinson and Eakins, 2006], a figure that can be compared to the 42,500 km3 of lavas produced over 5.5 Ma, the total length of volcanic activity documented in the Marquesas. In other words, while Hawaii results from the activity of a strong and long-lived plume that probably came from very deep in the mantle [Sleep, 1990], the Marquesas Archipelago results from a short-lived and weak plume. We believe the same model cannot apply to these two very different island chains.
 Many authors relate the superswell below French Polynesia, together with its numerous hot spots, to a very large thermochemical dome in the lower mantle. Cadio et al.  proposed that this rising structure culminated at about 670 km depth and several little plumelets originated from this large hot plume. This structure is entirely consistent with Courtillot et al.  who relate French Polynesia to “secondary plumes” from a large structure at the base of the upper mantle. We believe for the following reasons that this model is the most consistent with the geochemistry and age pattern of French Polynesia. The Polynesian island chains are numerous (Austral-Cook, Society, Marquesas and Pitcairn-Gambier) but restricted to a small geographical area (3.6 106 km2), a tricky issue if plumes come from very deep in the mantle. In addition, the age pattern is also extremely complex with, for example, concomitant volcanic activity in three different locations in the Austral-Cook chain [Turner and Jarrard, 1982; Chauvel et al., 1997] and near-synchronous eruptions over 300 km apart not only in the Marquesas (e.g., Ua Huka and Fatu Hiva; see Figure 11a) but also in the Austral-Cook chain [Chauvel et al., 1997]. Moreover, the total volume of erupted lavas is very low (7,700 km3 per Ma in the Marquesas, 11,000 km3 per Ma in the Society, 6,600 km3 per Ma in the Austral-Cook and only 3,000 km3 per Ma in the Pitcairn-Gambier chain), a feature consistent with weak hot spots. In Polynesia, alkali basalts and basanites are extremely common and dominate over tholeiites, which suggests low-degree partial melting [Caroff et al., 1997]. Finally, the sources are extremely heterogeneous, as shown by the large range of isotopic compositions. It is in this context that the two isotopic stripes in the Marquesas Archipelago must be examined. We believe that the very large isotopic range of each Marquesas island group originates from very low degree melts of an extremely heterogeneous plume. Small plumelets from a large dome at the base of the upper mantle, as suggested by Courtillot et al.  and Cadio et al. , would sporadically ascend through the asthenosphere, each carrying the geochemical diversity of its source. Such activity explains why volcanism occurs at the same time in several locations (see Figure 11a) in the Marquesas Archipelago and why isotopically distinct magmas erupt synchronously within short distances (see for example, Ua Huka, Ua Pou, Fatu Huku and Hiva Oa between 2.5 and 3 Ma in Figures 11b and 11c).
5.4. Isotopic Heterogeneities and Plume Strength
 One of the most puzzling features of Polynesia is its remarkable isotopic heterogeneity: almost the entire range of Sr, Nd, Hf and Pb isotopic compositions known in ocean island basalts exists within the four island chains – Austral-Cook, Society, Pitcairn-Gambier and Marquesas (Figure 3). The question to be asked is how small islands, dispersed over a large area, can come from such diverse sources? Here we suggest that the range of isotopic heterogeneity is related to the strength of the plume. In Figure 12, we plot the isotopic variability of the island chains as a function of the volume of erupted lavas normalized per Ma of volcanic activity. The amplitude of change of Sr, Nd, Hf and the three Pb isotopic ratios for each island chain is normalized to the total range reported for OIB in the literature. To calculate the amplitude, we used data from this study and the GEOROC database and estimated the range simply by calculating the difference between the highest and the lowest values for each isotopic system. We believe that this is better than calculating averages and standard deviations because histograms of measured values do not define Gaussian curves with normal distributions. To calculate the volume of magmas erupted per Ma, we used literature data for the size of the islands or chains (sum of the emerged and immerged parts) and divided the total volume by the known duration of volcanic activity. This method overestimates the volumes emitted per Ma, as only the uppermost 3% to 10% of each edifice is generally sampled (during fieldwork on the islands, dredging or drilling) and dated, but it allows comparison between islands and linear chains. When volumes of volcanic edifices (islands and seamounts) were not available, we calculated them using GMT software and a reference isobath corresponding to the top of the oceanic crust in the bathymetric chart [Smith and Sandwell, 1997]. Further details including data and references for each island or island chain, are provided in Table 2 and in the caption of Figure 12.
Figure 12 shows a negative correlation between the volume of lava emitted per Ma and the range of isotopic compositions. Unfortunately, we could only plot few islands or island chains because of the limited set of available volumes for volcanic edifices. The Big Island of Hawaii, with its huge volcanic activity, has the smallest range, with normalized amplitude values below 0.2 while the Pitcairn-Gambier and Austral-Cook chains in Polynesia have the lowest magma productions and the largest amplitudes, up to almost 1 for 207Pb/204Pb. Both the Marquesas and Society chains have low magma productions and high amplitudes. Interestingly, the six isotopic ratios considered here do not vary similarly in all island chains. Pb isotopes are the most variable in lavas of Reunion Island and the Austral-Cook and Canary chains where HIMU sources were identified. In contrast, Nd and Sr are the most variable in Pitcairn-Gambier, Marquesas, Societies and Hawaii as can be expected of magmas from enriched EM-type sources. Where data are available, the Hf isotopic compositions are almost always the least variable of all isotopic systems but this could be due to the much smaller data set for this isotopic system. Finally, the relative variability of the six isotopic systems not only demonstrates the complex nature of the plume sources but also demonstrates that when extreme compositions such as HIMU, EM I or EM II are present in an island chain, they usually coexist with very different compositions to produce the large observed variability. In contrast, where magma volumes are large, the isotopic variability is small. This suggests that the degree of partial melting controls the isotopic variability, an observation consistent with suggestions made by Chabaux and Allègre  using U-Th-Ra disequilibria data. When the degree of melting is low, the source heterogeneity is transmitted to the lavas. In contrast, when high-degree melting leads to a large amount of magma, the source volume is larger, the heterogeneities are better mixed, and the diversity is blurred. In other words, small and isotopically heterogeneous islands such as the Marquesas, the Austral-Cook and Pitcairn-Gambier map the diversity of source compositions whereas large and isotopically more homogeneous islands such as Big Island (Hawaii) and Réunion Island are ideal sites to estimate the average composition of a plume source.
 The relationship between isotopic variability and magma production rate shown in Figure 12 can be used to constrain the size of heterogeneities in the plume source. We focus on the Marquesas Archipelago where we have abundant and precise isotopic data from accurately dated lavas. Our data from Ua Huka, Fatu Hiva and Tahuata show clearly that lavas erupted within <50,000 years have diverse isotopic compositions (e.g., UH34 and UH40 in Ua Huka, FH01 and FH18 in Fatu Hiva and TH05 and TH08 in Tahuata, Table 1). This suggests that the source that melted within this very short time interval was isotopically heterogeneous. Using the overall magma production rate of 42,500 km3 over the 5.5 Ma of volcanic activity in the Marquesas, we estimate an average magma production of ≈400 km3 in 50,000 years. Given the large proportion of basanites and alkali basalts in the Marquesas, we also estimate that the degree of partial melting is generally quite low, probably less than 5%. This indicates that over the period of 50,000 years, 400 km3 of lava were produced from 8,000 km3 of the mantle source. This small source volume can be represented as a sphere with a ca. 10 km radius. Such small-scale heterogeneity in a mantle plume is not surprising given the considerable intraisland variations documented here as well as in previous studies of the Marquesas Archipelago.
 New isotopic (Sr, Nd, Pb, Hf) analyses of Marquesas lavas combined with literature data define two parallel rows of islands and seamounts with distinct and highly variable Nd and Sr isotopic compositions: the Ua Huka group in the northeast, and the Fatu Hiva group in the southwest. The two groups coexisted for the entire length of volcanic activity in the archipelago. The distinction between the two groups is less clear in other isotopic systems but in general the Fatu Hiva group has higher 206Pb/204Pb ratios than the Ua Huka group at any given 87Sr/86Sr value. Best fit hyperbolas for each island group demonstrate that the material that melted does not include ambient depleted mantle and was entirely contained in the plume source. The two end-members with depleted characteristics (low 87Sr/86Sr and high 143Nd/144Nd) have 206Pb/204Pb ratios around 20 while the enriched end-members have more subdued values around 19. We attribute these isotopic stripes to melting of two adjacent filaments within the raising plume as suggested for Hawaii by Farnetani and Hofmann [2009, 2010].
 We suggest that the very large isotopic ranges of each Marquesas group originates from sampling by very low-degree melts of small portions (<10,000 km3) of an extremely heterogeneous source in a large dome structure located under Polynesia. Similarly, the other Polynesian chains (Pitcairn-Gambier, Austral-Cook, Society) represent small volumes of magmas with unusually large isotopic variability. The same melting mechanism of other parts of the same dome could explain them as well. The contrast between the large isotopic variability observed in the surface expression of weak plumes (Polynesia) and the relative isotopic homogeneity of islands derived from strong plumes such as Hawaii or Réunion Island suggests that the factor controlling isotopic variability is the degree of partial melting. When partial melting degree is very low, source heterogeneity is preserved. In contrast, when high degrees of partial melting lead to high magma production rate, the diversity is blurred.
 We dedicate this study to the memory of Robert Brousse (1929–2010), pioneer of geological and petrologic studies in French Polynesia. He provided several samples used in this work, especially from Hiva Oa. Help from Sarah Bureau, Philippe Telouk and Philippe Nonnotte during analytical work was greatly appreciated. We thank Dominique Weis, an anonymous reviewer and the editor Joel Baker for their very constructive comments that helped us improve the overall content of the manuscript. This study was funded by CNRS (INSU and UMR 5025, 5275, 6538, 6518, 1572), BRGM, CEA-DASE and the ANR project “M &Ms.”