Hot spot volcanoes are commonly constructed in a characteristic sequence of stages. After the volumetrically dominant shield stage, a protracted period of quiescence ends with a final stage of activity: rejuvenated volcanism. The mechanism responsible for generating rejuvenated volcanism is not generally agreed upon. New data obtained for samples 200 m down-section in a deeply incised canyon on Savai‘i (Samoa) are unusually enriched isotopically and indicate a relatively voluminous rejuvenated stage compared to other intraplate volcanoes. Using a modified flexural model originally proposed for Hawai‘i, we suggest that the location of Samoa near the Tonga Trench terminus causes plate flexure resulting in upward flow of the shallow mantle driving partial melting. In particular, subduction-related plate bending in the Samoan region may cause a larger flexural amplitude than generated by volcanic loading in Hawai‘i. The larger amplitude may explain the larger volume of rejuvenated melt in Samoa, constrained by our new data. Moreover, we argue that Sr-Nd-Pb-Os-He-Ne isotopes in Samoan rejuvenated lavas are all consistent with sampling of a lithospheric component that is characterized by a metasomatic imprint from the Pacific Plate's earlier passage over the Rarotonga hot spot. Furthermore, temperature estimates for the melts suggest a drop in temperature during the predicted shallower melting due to flexural uplift, compared to the conditions during shield volcanism. Thus, flexural bending and metasomatism of the Samoan lithosphere may have generated the voluminous and geochemically distinct Samoan rejuvenated lavas, implying the lithosphere may play an important role during this stage in non- Hawaiian hot spots.
 Following the volumetrically dominant shield stage volcanism, a period of quiescence of at least 0.5 Ma ends with eruptions of the rejuvenated stage, the final stage of volcanism at hot spots [e.g., Clague and Dalrymple, 1987; Garcia et al., 2010]. In Hawai‘i, the archetype for hot spot volcanoes, the rejuvenated stage is erupted after a 0.5–2 million year long hiatus in volcanic activity, and it is geochemically distinct from earlier, shield-stage volcanism [e.g.,Garcia et al., 2010]. Temporal geochemical variations are clearly observed in samples from vertical sections such as canyons and drill cores [e.g., Stolper et al., 1996; DePaolo et al., 2001a; Garcia et al., 2007, 2010], and have helped constrain the comparatively small total volumes for the rejuvenated volcanic stage. Despite their small volumes, rejuvenated lavas have a surprisingly large impact on our understanding of the origin of volcanism at hot spots.
 Hot spots are thought to be generated by buoyantly upwelling mantle plumes that melt beneath a moving plate, producing age-progressive chains of volcanoes that extend away from a central region of melting, or ‘hot spot’ [Morgan, 1971]. In Hawai‘i, volcanism consists of multiple stages [e.g., Clague and Dalrymple, 1987] where the first three stages of each volcano—pre-shield, shield, and post-shield—are thought to be caused by the volcano passing over and sampling different parts of a mantle plume [e.g.,Hauri, 1996; Kurz et al., 1996; Lassiter et al., 1996; Frey and Rhodes, 1993; DePaolo et al., 2001b; Blichert-Toft et al., 2003; Abouchami et al., 2005; Fekiacova et al., 2007]. However, the fourth and final stage of volcanism—rejuvenated volcanism—has always posed a problem for the mantle plume origin of hot spots, due to the preceding hiatus in activity and the distinct composition of the lavas. A hiatus between shield and rejuvenated volcanic stages implies that the volcano is transported ∼35–140 km (Pacific Plate motion of ∼70 km/Ma) away from the location where the last shield-building eruptions occurred. Rejuvenated volcanism is difficult to reconcile with a plume-origin for this final stage of hot spot activity: the plume is far removed from the locus of volcanism when rejuvenated volcanism initiates. Therefore, other (non-plume) mechanisms, including plate flexure, have been proposed to explain the origin of rejuvenated lavas [e.g.,ten Brink and Brocher, 1987; Bianco et al., 2005; Konter et al., 2009]. Most models for rejuvenated volcanism have been based on observations of Hawaiian volcanoes, but other ocean islands, including Samoa, exhibit rejuvenated volcanism and may provide insight into potential causes.
 Rejuvenated volcanism is a feature of volcanic evolution at a number of intraplate volcanic chains, including Kerguelen, Canary Islands, Madeira, Mauritius, Marquesas, Society Islands, and Fieberling-Guadalupe Seamounts, but the Samoan hot spot hosts some of the most extensive rejuvenated volcanism in the world (Figures 1 and 2) [Macdonald et al., 1983; Wright and White, 1987; Woodhead, 1992; White and Duncan, 1996; Ielsch et al., 1998; Paul et al., 2005; Workman et al., 2004; Konter et al., 2009; Garcia et al., 2010; White, 2010]. The largest island in Samoa, Savai‘i (similar in size and age to the Hawaiian islands of O‘ahu and Kaua‘i; Figure 2) is known for its substantial volume of rejuvenated (or post-erosional) volcanism [Stearns, 1944; Kear and Wood, 1959; Natland, 1980]. In Samoa, this stage is so extensive that it has almost completely “resurfaced” the island with a veneer of young lavas [Workman et al., 2004]. The resurfacing is so complete that subearially exposed shield stage lavas were only recently reported in the deeply incised canyon of the Vanu river [Konter et al., 2010]. Hawkins and Natland  pointed out that the eruption of the youngest stage stretches to the islands of Upolu and Tutuila, but they are not as far along in their volcanic construction, so the island of Savai‘i provides the best estimate of erupted volumes of rejuvenated volcanism. Moreover, the vertical exposure afforded by the Vanu river canyon (Figure 3) provides a unique opportunity to estimate the volume of erupted rejuvenated lavas in Samoa. The geochemical compositions of several lavas sampled over the bottom ∼50 m of the canyon walls help constrain the thickness of the rejuvenated lava veneer on Savai‘i, since shield-stage lavas are geochemically distinct and should occur below the lowest rejuvenated lavas. The thickness of this veneer of rejuvenated volcanism in Samoa is important for evaluating the proposed models for the origin of this stage.
 The canyon of the Vanu river on Savai‘i is a steep amphitheater-shaped canyon with a vertical relief of ∼500 m. This canyon resembles the large erosional canyons of Kohala Volcano (Hawai‘i), which started forming potentially due to waterfall erosion during the shield stage as evidenced by late shield stage lavas flowing into the canyon [Lamb et al., 2007]. In Samoa, the canyon west of the Vanu river seems to have experienced significant refilling by younger lavas based on the canyon morphology, but the eastern canyon is more steeply incised and has alluvium exposed on the canyon floor, resting on minor younger lava fill [Kear and Wood, 1959]. Such canyons afford direct access to the vertical stratigraphy of different volcanic stages. One major difference with the canyons of Kohala is that the mouth of the Samoan canyons lies near 400 m elevation, while those in Hawai‘i are at sea level and the canyon walls form seacliffs. In the most recent eruption, a major canyon was filled with lava backing up against the offshore reef [Anderson, 1910]. Given that the erosional profile of the Vanu river flattens near 400 m, its canyon may have been filled in a similar fashion. Since this is the deepest canyon on Savai‘i, combined with the high eruption rate observed in Samoa [Anderson, 1910], it appears qualitatively that larger volumes of lava are erupted in Samoa compared to Hawaii, where major canyons erode down to sea level. Despite a potential filling of the lower parts of the Vanu river canyon, the ∼500 m nearly vertical upper canyon wall provides constraints on the thickness of the rejuvenated lavas on Savai‘i. We show that this exposure suggests a much thicker sequence of rejuvenated lavas that covers a much larger portion (∼99%) of the volcano surface area compared to the very small volumes of rejuvenated volcanism that cover less than half of the surface of a Hawaiian volcano [Garcia et al., 2010].
 We present new major element and Pb isotope data that demonstrates a minimum thickness for the rejuvenated lavas of ∼200 m. An upper bound on this thickness—approximately 500 m, which is the depth of the canyon—is provided by the presence of shield-stage trachytes as cobbles in the Vanu river that drains this canyon [Workman et al., 2004]. We calculate a minimum thickness of 200 m, which we argue constitutes a volume of erupted rejuvenated lava that is many times that observed for rejuvenated volcanism in Hawai‘i [Garcia et al., 2010]. The unusually large volumes of rejuvenated lavas in Samoa greatly extend the available set of observations for this volcanic stage among hot spots globally. We use both geochemical compositions and melting temperature estimates to critically evaluate proposed models for the origin of rejuvenated volcanism in Samoa.
 The exceptional volumes and unusual geochemical compositions of Samoan rejuvenated volcanism can help constrain its origins. We argue that the large volumes of Samoan rejuvenated lavas are the result of tectonic bending associated with Samoa's location near the Tonga trench. The tectonic regime near a trench corner provides a mechanism for extensive lithospheric flexure, and we use shallow upwelling rates from modeling to calculate related partial melting. We will show that the resulting melts may have inherited a significant contribution from metasomatized lithosphere beneath Samoa that we argue was enriched during the plate's prior passage over the Rarotonga hot spot (part of the Cook-Austral hot spot chain) ∼15 million years ago. Therefore, the model we present below suggests that the unique tectonic regime permits melting of metasomatized lithosphere and provides a common mechanism linking the tectonic setting with the unusually large volumes of isotopically enriched rejuvenated lavas in Samoa.
2. Background on the Samoan Hot Spot
2.1. Samoa as an Age Progressive Volcanic Chain
 The Samoan volcanoes are located in a linear chain ∼100 km northeast of the northern terminus of the Tonga trench (Figure 1). While the eastern seamount, Vailulu‘u, is considered the youngest volcano [Hart et al., 2000; Konter et al., 2004; Staudigel et al., 2006; Sims et al., 2008], the western-most subaerial volcano, Savai‘i, has been active since at least 5.29 Ma and was last active in 1905–1911 [e.g.,Sapper, 1906; Anderson, 1910; Natland and Turner, 1985; Koppers et al., 2008]. However, construction of the volcanic shields in Samoa follows an age progression that is anchored at the easternmost volcano, Vailulu‘u Seamount, and continues 1750 km to west to Alexa Bank, which has lavas dated at 23–24 Ma [e.g., Duncan, 1985; Natland and Turner, 1985; Hart et al., 2004; Koppers et al., 2008, 2011; Németh and Cronin, 2009; McDougall, 2010]. On the eastern young (0–2 Ma) side of the chain, the islands and seamounts east of Tutuila Island form volcanic edifices that constitute various parts of the shield-building stage of Samoan volcanism. Moving west, the islands of Tutuila, Upolu and Savai‘i each host rejuvenated stage volcanism, while volcanoes even further to the west have only been sampled on their lower submarine flanks such that no extensive set of rejuvenated lavas are available. The subaerial rejuvenated lava cover increases from less than one quarter cover on Tutuila, to nearly half cover on Upolu, and a near-complete “resurfacing” of the island of Savai‘i with rejuvenated lavas (Figures 1 and 2). Moreover, on Savai‘i the only shield exposure is present in the canyon of the Vanu river [Konter et al., 2010], constraining the total erupted volume of rejuvenated lavas.
 The isotopic and trace element composition of Samoan shield and rejuvenated lavas are extreme in the context of global oceanic volcanic rocks: The isotopic composition of Samoan lavas is defined by highly radiogenic Sr isotope ratios, relatively unradiogenic Nd isotopes, and intermediate Pb isotope compositions [Hofmann and White, 1982; Natland and Turner, 1985; Wright and White, 1987; Workman et al., 2004; Jackson et al., 2007, 2010a]. The Sr isotope ratios of several submarine lavas from Savai‘i are the highest documented for oceanic intraplate volcanism, and these data therefore define one extreme end-member composition (EM2) [Zindler and Hart, 1986] in the total compositional range of oceanic volcanic lavas [Jackson et al., 2007]. By contrast, the rejuvenated lavas are distinct and trend away from the field defined by Samoan shield lavas toward a component with moderately elevated 208Pb/204Pb and relatively low 206Pb/204Pb, moderately high 87Sr/86Sr and low 143Nd/144Nd [Natland and Turner, 1985; Wright and White, 1987; Workman et al., 2004]. Moreover, these lavas also show high Ba/(Sm, Nb, Th) ratios compared to shield stage lavas (Figure 5) [Hart et al., 2004; Workman et al., 2004; Jackson et al., 2010a]. Such compositions are different from rejuvenated lavas in Hawai‘i, where the isotopic composition trend toward the compositions of mid-ocean ridge basalts, which exhibit low87Sr/86Sr and 206Pb/204Pb isotope ratios and high 143Nd/144Nd. While Hawaiian rejuvenated lavas do have similar high Ba/(Sm, Nb, Th) [Garcia et al., 2010], the distinct isotopic compositions of the rejuvenated lavas in both island chains imply that different source materials are involved in the generation of the melts. We will examine these compositions in terms of the potential processes and sources that may were involved in generating the observed geochemical signatures in Samoan rejuvenated lavas.
 We will demonstrate a further role of the lithosphere in modifying volcanism at Savai‘i is expressed in the distinct isotope geochemistry of its rejuvenated lavas: Instead of trending toward a depleted mantle component like Hawaiian rejuvenated lavas, Samoan rejuvenated lavas sample an enriched EM1-like component that forms an isotopically distinct group from the EM2 component found in Samoan shield lavas (EM1, EM2: enriched mantle 1 and 2 [Zindler and Hart, 1986]). Below we argue that flexural forces from the nearby trench cause shallow melting of an EM1 rich component that may be related to the Pacific Plate's passage over the EM1 hot spot of Rarotonga (Figure 1) [e.g., Chauvel et al., 1997; Konter et al., 2008; Jackson et al., 2010a].
3. Samples and Methods
3.1. Sample Description
 Samples were collected during a field campaign in 2009 on the island of Savai‘i that focused on sampling the shield stage of Savai‘i and the oldest rejuvenated lavas, exposed in the canyon of the Vanu river. This canyon was mapped as shield stage volcanics (Fagaloa Series) by Kear and Wood , but the new data show that the amount of exposed shield lavas was overestimated [Konter et al., 2010]. Rejuvenated lavas that we recovered are used here to examine the minimum total erupted volumes of rejuvenated lavas at Savai‘i. Collection sites are located within the upper canyon of the Vanu river, encompassing the lower ∼50 m of a 200 m vertical waterfall section (Sinaloa Falls). The samples we collected from the waterfall section are all olivine-phyric basanites and picrites. Groundmass phases consist of olivine, plagioclase, spinel and clinopyroxene. Olivine phenocrysts occur both as anhedral and euhedral crystals, and the overall Mg number of the olivines spans a wide range that indicates that only some of the olivine phenocrysts are in equilibrium with the Mg number of their whole rock composition (see Text S1 and Figure S1 in theauxiliary material). Rare small phenocrysts of clinopyroxone or plagioclase are anhedral. These observations are in agreement with the picritic major element composition of the samples, suggesting accumulation of olivine and clinopyroxene. The geochemical composition of these samples suggests that they are rejuvenated-stage lavas.
3.2. Analytical Techniques
 All samples were crushed avoiding contact with any metal, and the freshest chips were picked with a binocular microscope. Major element analyses were performed on unleached whole-rock powders by X-ray fluorescence (XRF) at the GeoAnalytical Lab at Washington State University. The technique employed is described byJohnson et al. . Isotopic analysis of Pb was performed at the University of Texas at El Paso, using standard HBr-HCl-based chemistry [Hanan and Schilling, 1989] to separate and purify Pb in a class 100 environment. Mass spectrometric analysis was carried out on the Nu Plasma high resolution multicollector inductively coupled plasma mass spectrometer (HR MC-ICP-MS) at the University of Texas at El Paso. To monitor machine performance on the MC-ICP-MS, standards were run systematically after every two samples. The Pb isotopic analyses were carried out using a combination of bracketing samples with SRM 981 Pb (correcting to values ofTodt et al. ) and monitoring by Tl doping (SRM 997 Tl and 205Tl/203Tl = 2.3889 [White et al., 2000; Hanan et al., 2004, 2008]) to correct for mass fractionation and machine bias [White et al., 2000]. The standard SRM 981 Pb (20 ppb solution, N = 70) achieves a long-term (October 2010–August 2011) average of206Pb/204Pb = 16.943 ± 3, 207Pb/204Pb = 15.500 ± 3, and 208Pb/204Pb = 36.726 ± 8 (±2σ). The reported uncertainties for each sample represent in-run precision. Total procedural blanks were less then 50 pg for Pb. Rock standard BCR2 is well within error of the average reported byWeiss et al. , where the average renormalized to Todt et al.  SRM 981 values corresponds to: 206Pb/204Pb = 18.7451 ± 11, 207Pb/204Pb = 15.6192 ± 9, and 208Pb/204Pb = 38.6987 ± 24. All compositional data are presented in Table S1 in the auxiliary material.
 The major element compositions, presence of anhedral phenocrysts and occurrence of groundmass olivine indicate that the samples we report on here are alkalic lavas that likely accumulated some olivine and clinopyroxene. The difference in major element compositions between samples is small (e.g., Figure 4), but they have different modal contents in phenocrysts and differences in phenocryst grain sizes. The samples have Pb isotope compositions ranging from 206Pb/204Pb = 18.801–18.808, 207Pb/204Pb = 15.591–15.599, and 208Pb/204Pb = 38.981–39.003 (Figure 7). This narrow range makes it difficult to distinguish these samples from each other based on their Pb isotope compositions. Compared to other rejuvenated lavas in Samoa, these samples are more picritic in composition, showing the lowest SiO2 and total alkali content found in rejuvenated lavas from Savai‘i, although rejuvenated lavas with lower SiO2 are found on Upolu (Figure 4). The isotopic compositions fall within the characteristic range of rejuvenated lavas on Savai‘i, and are distinct from Samoan shield lavas (Figure 7). Therefore, our samples from the Vanu river canyon enhance aereal cover of the rejuvenated lavas (the only available data in Upolu and Tutuila) with an estimate of minimum thickness. On Savai‘i the thickness is ∼200 m (the height of the waterfall), and below we use these constraints with the geochemical signatures of these lavas to explore models for the origin of rejuvenated volcanism.
5.1. Volume Constraints on Samoan Rejuvenated Lavas
 The stages of volcanism on Savai‘i were crudely mapped by Kear and Wood based on erosional profiles, soil thickness and rock descriptions. However, geochemical compositions are needed to distinguish different stage lavas. Therefore the volume of the rejuvenated lavas on the mature volcano of Savai‘i has not been previously constrained. Owing to the presence of shield-stage cobbles in the Vanu river [Konter et al., 2010], we infer that the boundary between shield stage and rejuvenated lavas is exposed in the deeply incised canyon that is drained by the river. This information can be used to place constraints on the thickness and total volume of the rejuvenated lavas. We identified rejuvenated lavas 200 m down-section in the wall of a >500 m deep canyon on Savai‘i, based on their rock types (picrite/basanite) and Pb isotope compositions compared to shield lavas (Table S1 in theauxiliary material and Figures 2, 4, and 7). The samples were collected from the lower ∼50 m of the 200 m exposure; although we do not have a detailed sample record of the entire 200 m section, it is unlikely that a change occurred up-section (i.e., back to the shield stage). Therefore, the discovery of rejuvenated lavas at the base of the 200 m stratigraphy of the waterfall indicates that 200 m is the minimum thickness for the rejuvenated lavas. Additionally, the presence of a shield-stage cobble originating in the canyon indicates that the 500 m depth of the canyon represents a maximum thickness for the rejuvenated stage. Unfortunately, no systematically collected well cores [Garcia et al., 2010] or scientific drill cores [Stolper et al., 1996; DePaolo et al., 2001a; Garcia et al., 2007] are available to estimate the paleotopography and the thickness of the rejuvenated lavas in detail. Therefore, while the paleo-topography could have an important effect on the thickness distribution of the rejuvenated lavas, it is not well-constrained on Savai‘i. The only other location where a minimum thickness can be estimated is a 70 m rejuvenated lava escarpment near the northern coast line (Figure 2) [Workman et al., 2004], suggesting the thickness of the rejuvenated lavas remains great across the island. We therefore assume that the subaerial part of Savai‘i volcano is covered by a minimum 200 m thickness estimated in the canyon, despite known offshore occurrence of rejuvenated lavas [Jackson et al., 2007, 2010a]. The observed eruption volume of the 1905 eruption lends credibility to this approach as well. The flows filled a 10 mile canyon with a lava series up to 130 m thick [Anderson, 1910], and the rare occurrence of deeply eroded canyons compared to Kaua‘i, suggests that rejuvenated volcanism on Savai‘i can easily smooth out an eroded shield like Tutuila or Upolu. Assuming a thickness of 200 m results in a total volume estimate of rejuvenated lavas on Savai‘i of ∼0.75 × 103 km3. This is ∼1.8% of the total volume of Savai‘i, which at ∼42 × 103 km3 is similar to the total volume of similarly aged Hawaiian volcanoes (Figure 2) [Robinson and Eakins, 2006]. This likely still represents a conservative estimate, since it uses the minimum thickness of 200 m and it assumes that offshore rejuvenated eruptions (reported in Jackson et al. [2010a]) are volumetrically unimportant. Alternatively, if the veneer is 200 m near the summit of Savai‘i and tapers off, disappearing at the coastline (an underestimate since a 70 m escarpment is observed near the north coast), a minimum estimate of 0.091 × 103 km3 is obtained (0.22% of the total volume of Savai‘i). Thus, even the most conservative estimate of the Samoan rejuvenated lava volume is nearly double the 0.1% relative volume estimated for the largest rejuvenated stage outpouring in Hawai‘i (0.058 × 103 km3 on Kaua‘i [Garcia et al., 2010]), and this is qualitatively consistent with the areal extent (Figure 2) of the rejuvenated lavas on Kaua‘i (half the island) versus Savai‘i (nearly the entire island). However, using our preferred estimate for Samoa (1.8%) suggests that the erupted volume of rejuvenated lavas on Savai‘i is more than an order of magnitude larger than that for Kaua‘i.
5.2. Does Samoa's Tectonic Setting Explain the Large Volumes of Rejuvenated Lavas?
 A combination of modeling and observations near the Tonga Trench terminus can be shown to be in agreement with the idea that Samoa's unique tectonic setting may have an impact on the mantle melting that produces the Samoan Islands. The Tonga Trench south of the Samoan Islands is part of a tectonically complex zone consisting of subduction along the Tonga-Kermadec subduction zone and back-arc spreading in the Lau Basin, which together accommodate the very fast convergence of the Pacific and Australian Plates (26–27 cm/a [Hart et al., 2004]). While the Pacific Plate is subducted south of the terminus of the Tonga Trench, the Pacific plate including the Samoan Islands north of the terminus continues further west along a seismically observed tear in the plate that is expressed as the trace of the Vitiaz Lineament [e.g., Millen and Hamburger, 1998; Hart et al., 2004]. The Samoan Islands are located just ∼100 km north of the intersection of the Tonga Trench with the Vitiaz Lineament, and their proximity to this complex plate boundary requires an assessment of the role of the nearby trench on proposed models for the origin of rejuvenated lavas.
 Owing to plate flexure outboard of the trench, which extends north of the northern terminus, the Pacific Plate is upwarped in the Samoan region [Levitt and Sandwell, 1995; Govers and Wortel, 2005]. The resulting upward mantle flow under the elastic part of the plate may cause small amounts of decompression melting, in a manner not unlike that proposed by Bianco et al.  for plate flexure due to volcanic loading in Hawai‘i. Bathymetric and gravity data across the trench have been used to model flexure of the Pacific Plate in the Samoan region. The data from three profiles suggest a flexural amplitude of ∼250–1000 m, resulting from tectonic forces from the Tonga trench [Levitt and Sandwell, 1995]. Govers and Wortel assumed visco-elastic rheology and employed a finite element model of the trench-transform intersection to evaluate the near-instantaneous response of the lithosphere and underlying mantle to a high density, subducting slab. The results suggest significant flexure-driven uplift (∼1000 m) of the lithosphere near Savai‘i (Figure 6), and the shallow upward velocities (∼1–4 mm/year) associated with bending extend into the asthenosphere. While tectonic forces associated with the nearby Tonga trench have been suggested to crack the plate parallel to the Vitiaz Lineament [Natland, 1980], we suggest that plate upwarping drives decompression melting in the Samoan region. Below, we assess whether this melt-generating mechanism is greater in magnitude, but qualitatively similar to, the flexure-driven melting in Hawaiian rejuvenated lavas.
5.3. Rejuvenated Melt Production Due to Flexure in Samoa and Hawai‘i
 Melt production due to the flexural response of the Pacific Plate under the growing load of a new Hawaiian volcano has been modeled by Bianco et al.  with a 3D finite element model that includes an evaluation of melt production. A full 3D numerical treatment of the complicated tectonic region around Samoa goes beyond the scope of this work. However, we have calculated a 1D melt production example (a parcel of mantle flowing along the top of the asthenosphere) and compare the cases of Samoa and Hawai‘i using this approach (Figure 8). Our calculation is based on the work of Phipps Morgan . We estimate melt production per unit time (dF/dt in Ma−1) due to decompression:
where F is the melt fraction, P is pressure and t is time. We follow Bianco et al.  to estimate dP/dt, using the vertical (upwelling) velocity:
This formulation only considers lithostatic decompression, using the density (ρ), acceleration of gravity (g), and upwelling velocity (vz). The upwelling velocity used here is derived from a cross section through the results of Govers and Wortel  (see Figure 6), which corresponds to the lithospheric uplift velocity, providing the maximum velocity for the asthenosphere directly underneath. Evidence that their model produces reasonable results comes from the flexural amplitude (up to ∼1000 m) calculated by Levitt and Sandwell  for the Tonga Trench. Next, we use Phipps Morgan's  formulation to estimate the second term, dF/dP:
and we use his estimates for its parameters (∂Tm/∂z = 4°C/km for the solidus gradient with depth; αgT/cp = 0.4°C/km for the temperature gradient for adiabatic expansion; TΔS/cp = 550°C for the enthalpy of fusion; and ∂Tm/∂F = 250°C for the solidus depletion gradient). The calculated productivity per 1 Ma at a single location per 1 Ma (dF/dt; Figure 8) shows a nonlinear profile across the flexural bulge (Figure 8). The sum of these melt fraction increments starting from the current hot spot (Vailulu‘u) following the chain is the cumulative melt fraction that would have been produced through time (bottom panel). For the Hawaiian model, a scaled (100 m uplift in 1 Ma) flexural profile from Bianco et al.  and its cumulative melt fraction predict that rejuvenated melt production in Hawai‘i is nearly ten times smaller, in agreement with a flexural origin for Samoan rejuvenated volcanism. A different feature of the model that is consistent with observations is that only the Samoan islands with significant flexurally induced melt production host rejuvenated lavas.
 One caveat is that the parameters for the dF/dP estimate (particularly dTm/dF) are not well characterized, and therefore the calculated curves may have to be scaled appropriately. In general only rough estimates are available for dry peridotite [Phipps Morgan, 2001], while even less is known for other lithologies, including metasomatized peridotite. As a result, for the comparison of Samoa to Hawai‘i equal values for dF/dP are used at both localities. The difference in upwelling velocity due to flexure therefore controls the difference in melt generation in this calculation. This is probably a reasonable assumption, considering the observed larger volumes of Samoan rejuvenated lavas relative to Hawai‘i can be produced by melting to a larger degree, and/or melting a larger source volume. The dTm/dF term can change the most significantly, especially showing an increase due to very low degree melting [Phipps Morgan, 2001], which logically predicts lower melt production (dF/dP). Therefore, other reasonable values for dTm/dF only amplify the difference between Samoa and Hawai‘i. Alternatively, melting of a larger source volume requires a larger volume of mantle near its solidus and efficient melt segregation. However, in the flexural calculation, the region with vertical upwelling velocities is larger in Samoa (Figure 8), and therefore a flexural origin would still be consistent with the available data. Therefore, we suggest that flexural upwarping of the Pacific Plate is indeed quantitatively consistent with more rejuvenated-stage melt production in Samoa. Moreover, this process generates large quantities of rejuvenated volcanism in Samoa that can erupt along established conduits of the island volcanoes as they are rafted past the northern terminus of the trench.
5.4. Can the Hawaiian Flexural Model Be Modified to Explain Samoan Rejuvenation?
 Similarly, the mechanism responsible for generating rejuvenated lavas is still not well known, and the eruptive hiatus that separates rejuvenated volcanism from shield stage volcanism (e.g., 0.5–2 Ma in Hawai‘i [Garcia et al., 2010]) is a major characteristic that needs an explanation. This hiatus is not a natural result of any model for the internal structure of a mantle plume. Some models for rejuvenated volcanism place the rejuvenated mantle sources in the plume, but vary the source geometry. For example, suggested geometries for the rejuvenated mantle component include concentric zoning of the plume, filament structures or small blobs within the plume [e.g., DePaolo et al., 2001b; Abouchami et al., 2005; Marske et al., 2007]. However, these models fail to explain why a new phase of activity initiates after a period of protracted quiescence, when the mantle plume is hundreds of km away. As a result, plume-lithosphere interaction has developed as an alternative for explaining the initiation of rejuvenated volcanism. This class of models is divided into three broad categories: 1.) Conductive melting of the lithosphere driven by lithospheric heating from the plume [Gurriet, 1987], 2.) Decompression of plume material by lateral spreading of the plume head underneath the lithosphere [Ribe and Christensen, 1999; Sherrod et al., 2003; Paul et al., 2005], or 3.) Decompression due to plate flexure around a newly constructed volcanic shield [ten Brink and Brocher, 1987; Bianco et al., 2005].
 The Samoan volcanoes can contribute to the evaluation of these models, given their exceptional volumes of rejuvenated lavas and their unusual geochemical composition (Figure 1) [Hawkins and Natland, 1975; Wright and White, 1987; Workman et al., 2004; Koppers et al., 2008]. It is unlikely that conductive heating or plume spreading can generate more melt in Samoa compared to Hawai‘i, since the estimated buoyancy flux for the Samoan plume source is ∼6 times smaller and the potential temperature is lower than that for Hawai‘i [Courtillot et al., 2003]. Instead of plume spreading or conductive heating, we consider the flexural model the most appropriate for generating rejuvenated lavas in Samoa. The proximity to the Tonga Trench may lead to more substantial flexure-driven melting compared to the model for Hawai‘i [Bianco et al., 2005].
 The large volumes of rejuvenated lavas in Samoa (up to 10 times more than in Hawai‘i) provide a unique opportunity to assess the potential role of lithospheric flexure in rejuvenated volcanism. In a plate flexure model as applied in Hawai‘i, the load of a large volcanic edifice on an elastic plate can cause the plate to upwarp at a distance of 200–300 km, and the resulting decompression may generate melts [ten Brink and Brocher, 1987; Bianco et al., 2005]. Unlike Hawai‘i, which is located far from any plate boundary, Samoa is located next to the northern terminus of the Tonga Trench, and we suggest that melting under subducting and bending lithosphere causes greater flexure-driven melting due to the greater flexural amplitude compared to Hawai‘i. The plate flexure in Samoa (from ∼250–1000 m [Levitt and Sandwell, 1995; Govers and Wortel, 2005]) can create greater uplift and decompression melting than the magnitude of plate flexure predicted for Hawai‘i (∼100 m [e.g., Bianco et al., 2005]). This observation makes decompression melting due to flexural uplift an attractive explanation for explaining the greater volumes of rejuvenated volcanism in Samoa. Therefore, compared to Hawai‘i, larger volumes of rejuvenated melt are predicted (and observed) in Samoa.
5.5. What is the Difference With Previous Lithosphere-Based Models for Samoan Rejuvenated Volcanism?
Hawkins and Natland  proposed that tectonic stresses in the Samoan lithosphere played an important role in generating the unusually large volumes of rejuvenated lavas along nearly 300 km of volcanic chain, from Tutuila to Savai‘i. Bending of the lithosphere southward, toward the tearing and subducting plate, was proposed to create tensional faults that provide pathways for melt [Hawkins and Natland, 1975; Natland, 1980; Natland and Turner, 1985] (inset in Figure 6). However, 3D numerical (finite element) modeling of the lithospheric response to the Pacific Plate tearing and subducting suggests that plate bending generates low strain rates under the Samoan chain (Figure 6) [Govers and Wortel, 2005]. Strain rates are only high well within 100 km of the subduction-transform intersection at the northern terminus of the Tonga Trench (i.e., ∼100 km south of Savai‘i Island). In contrast to the localized high strain rates, theGovers and Wortel  model does show that the Samoan lithosphere flexurally bends up around the northern terminus of the Tonga trench, with upward velocities extending into the top ∼80 km of the asthenosphere. In fact, the vertical velocity field of Govers and Wortel  correlates very well with the location of rejuvenated lava eruption with Savai‘i, Upolu, and Tutuila, which are located near the maximum in vertical velocities (Figures 6 and 8). Similarly, upward mantle flow caused by plate flexure under the flexural bulge of Hawai‘i has been modeled to generate rejuvenated lavas [Bianco et al., 2005]. It is important to note that the flexural amplitude of the Pacific Plate entering the Tonga Trench is at up to 10 times as large as the flexural amplitude in Hawai‘i that is only related to flexural loading instead of bending into a subduction zone [Levitt and Sandwell, 1995; Govers and Wortel, 2005]. Additionally, in the Hawaiian flexure model, the region in which lithospheric uplift and upwelling occurs is <200 km across. By comparison, the region affected by flexural uplift near the Tonga Trench spans 300 km along the Samoan hot spot. From these observations, we suggest that upward mantle flow beneath Samoa caused by lithospheric bending generates enhanced volumes of melt through decompression melting.
6. Plate-Flexure Induced Melts of the Lithospere in Samoa: Evidence From Isotopes
 In the following sections we assess whether traditional heavy radiogenic isotopes (Sr-Nd-Pb), noble gas isotopes and Os isotopes for the Samoan rejuvenated lavas are all in agreement with melt generation from lithospheric flexure. Since the predicted flexural melts are all shallowly based (section 5.3), it is important to consider the geologic history of the Pacific Plate. Plate reconstructions show that the region of the Pacific plate now occupied by Samoa passed over the Rarotonga hot spot 15 million years ago (inset in Figure 1) [e.g., Chauvel et al., 1997; Wessel and Kroenke, 2008; Konter et al., 2008; Jackson et al., 2010a]. Given the similar isotopic and trace element composition to the Rarotonga lavas, we will argue that the Samoan rejuvenated mantle source may contain a Rarotonga hot spot signature hosted in the shallow Pacific mantle instead of being hosted entirely in the upwelling Samoan mantle. During the shield stage, such a component would likely be overwhelmed by melting of the Samoan mantle source, since the predicted amount of melting due to flexure is very low during the shield stage (Figure 8 and section 5.3). With limited decompression melting due to flexure, conductive melting due to heat transfer from hot plume material is another potential cause of melting of the rejuvenated source material. However, this has been modeled in Hawai‘i by Gurriet , and would cause a delay in melting of ∼2–10 Ma, suggesting melt production from heat conduction into the lithosphere during the shield stage should be minimal as well. Moreover, heat conduction and the delay in melting it causes [Gurriet, 1987] is unlikely to be the cause for rejuvenated volcanism by itself, since the likely hotter mantle in Hawai‘i [Courtillot et al., 2003; Herzberg and Gazel, 2009] is producing smaller volumes of rejuvenated melt than Samoa. Therefore, the lithosphere-driven hypothesis is our preferred model, and we will assess below whether the geochemical predictions associated with this model (e.g., Rarotonga influence) can be confirmed.
6.1. Evidence From the Isotopes of Incompatible Trace Elements (Sr-Nd-Pb-He-Ne) for Involvement of EM1 (Rarotonga) Metasomatized Lithosphere in Samoan Rejuvenated Lavas
 The rejuvenated lavas are distinct from the shield lavas in Samoa, with a rejuvenated composition that is moderately elevated in 208Pb/204Pb for its low 206Pb/204Pb (high Δ8/4 [Hart, 1984]), and intermediate in 87Sr/86Sr and 143Nd/144Nd compared to the shield (Figure 9). Although individual samples are scattered, the rejuvenated field is elongated away from the shield compositions, and it is nearly parallel to the East Pacific Rise (EPR) MORB field in Pb isotope space (Figure 9). The rejuvenated lavas have Pb isotopic compositions that overlap with Rarotonga lavas. However, this overlap does not exist for Pb-Sr isotope compositions. Instead the rejuvenated lavas trend toward EM1-like isotope compositions in Rarotonga. In more detail, the scatter of the Samoan rejuvenated lava compositional field implies that the source likely contains at least three components. Since the compositions of the Savai‘i rejuvenated lavas are all contained between the Samoan shield lavas, the EPR MORB, and the most EM1-rich sample from Rarotonga, we explore whether there is evidence for a mixture of a Rarotonga component and a lithospheric component in the Samoan rejuvenated lavas. In this model, the Samoan component may represent either the admixture of the Samoan shield stage lavas with the lithosphere (in a similar way to the Rarotonga metasomatizing component), or Samoan plume mantle that was not completely depleted during the shield stage.
 Since the elongated 206Pb/204Pb–208Pb/204Pb array of rejuvenated lavas stretches from the Samoan shield lavas to a composition between an extreme Rarotonga composition and EPR MORB, it appears that an initial mixture between Rarotonga and MORB may be involved. Therefore, if a Rarotonga component is truly present in the Samoan source, it likely mixed with a Pacific MORB component, prior to exposure to Samoan shield compositions. One way to create such a Rarotonga-lithosphere mixture would be metasomatic overprinting of lithospheric compositions with Rarotonga melts when the Pacific Plate passed over the Rarotonga hot spot.Pilet et al. recently showed that ocean island alkalic basalts (including nephelinites and basanites) can be generated from melting of metasomatized mantle, and we here extend this idea to Samoan rejuvenated volcanism by using an “easy-to-melt by flexure” component hosted in the lithosphere. Since our best estimate for a non-contaminated lithospheric composition is EPR MORB, a range of potential lithospheric compositions needs to be assessed to describe the potential metasomatic process. To a first order, we can assume a composition for a Rarotonga component based on the most extreme EM1-type Rarotonga sample ofNakamura and Tatsumoto , and mix such a melt with either the “depleted” or “enriched” depleted (solid) mantle composition of Workman and Hart . These compositions require only about 0.5% of the Rarotonga melt component to explain the Pb isotope composition of the rejuvenated lavas, which plot halfway between EPR MORB and Rarotonga (Figure 9), while a less extreme Rarotonga composition would require a slightly larger melt contribution. If we adopt this 0.5% mixture (composed of Rarotonga plus DMM) as one (solid) end-member and a Samoan shield melt composition (dredge 115 from Savai‘i [Jackson et al., 2007]) as the other (melt) component, then only about 1–2% contribution from the Samoan shield source is required to explain the overall isotopic compositions of the Samoan rejuvenated lavas. Overall, these calculations are consistent with our hypothesis that Samoan rejuvenated lavas contain a metasomatic Rarotonga component.
 If the Samoan rejuvenated lavas are melts of MORB mantle metasomatized by the Rarotonga hot spot, they would also be expected to host noble gas isotopic signatures associated with an EM1 metasomatic component. Based on He and Ar isotope systematics, Burnard et al. argued for multiple pulses of metasomatism in Samoan xenoliths, confirming earlier reports of extensive metasomatic overprinting of the Sr-Nd-Pb isotopic compositions of Samoan xenoliths from the lithospheric mantle [Hauri et al., 1993; Hauri and Hart, 1994]. Other clues about the noble gas isotopic composition of shallow mantle beneath Samoa come from analyses of noble gases in Samoan peridotite mantle xenoliths hosted in Savai‘i rejuvenated lavas [Poreda and Farley, 1992]. In a plot of He versus Ne isotopes (Figure S2 in the auxiliary material), the Samoan xenoliths trend from a component similar to MORB (but with slightly less radiogenic He) toward a component similar to the EM1 mantle found in Pitcairn lavas [Honda and Woodhead, 2005]. Since Ne isotopes have not been measured in Rarotonga lavas, we assume that the He-Ne signature for EM1 in Rarotonga lavas is similar to the EM1 lavas from Pitcairn, given that Rarotonga forms a trend that partly overlaps with Pitcairn in Pb-Sr-Nd isotope space [Woodhead and McCulloch, 1989]. A subset of the Samoan mantle xenoliths indeed trend toward the EM1 composition we might expect based on observed Pitcairn compositions. The Samoan xenoliths erupted during the rejuvenated stage plot between MORB, EM1 and EM2 (Samoan shield lavas [Jackson et al., 2009]). Therefore, the He-Ne isotope observations in the rejuvenated lavas, similar to Sr-Nd-Pb isotope compositions, appear to be consistent with our proposed model for a metasomatic Rarotonga component in Samoan lithosphere.
 Combining all the results, the isotopic signatures of Sr, Nd, Pb, He and Ne define a unique geochemical composition for Samoan rejuvenated lavas that can be explained by metasomatic fluids from the Rarotonga hot spot. We have shown with a mixing calculation that a combination of an EM1 Rarotonga component, the MORB mantle, and the EM2 shield component can explain the Samoan rejuvenated lava composition. Our model assumes that the EM1 component can be contributed by the geochemical signature of Rarotonga (e.g., Rarotonga: 206Pb/204Pb = 18.256–18.975, 207Pb/204Pb = 15.481–15.564, 208Pb/204Pb = 38.598–38.994, 87Sr/86Sr = 0.70412–0.70444, 143Nd/144Nd = 0.512629–0.512750 [Nakamura and Tatsumoto, 1988; Schiano et al., 2001]). In our model, this metasomatic component is sampled by melting due to lithospheric flexure, where upward mantle flow and melting is concentrated near the bottom of the plate [Govers and Wortel, 2005]. According to our model, all the isotope systems described above would be affected by metasomatism, and we therefore need to evaluate evidence for lithospheric contributions using an isotope system not affected by metasomatism.
6.2. Evidence From Os Isotopes for Lithosphere Involvement in the Genesis of Samoan Rejuvenated Lavas
 In order to evaluate the potential contribution from the Pacific lithosphere, we here discuss the available Os isotope data for Samoa. Since the parent element Re is incompatible while Os is compatible during partial melting of the mantle, melts will have an increased Re/Os ratio while mantle peridotites will have a greatly diminished Re/Os ratio [Shirey and Walker, 1998; Bizimis et al., 2007]. Consequently radiogenic ingrowth of 187Os is slow in peridotites, yielding low 187Os/188Os ratios over time. Modification of these low ratios by metasomatic effects is expected to be minimal, given the relatively high Os concentrations in peridotites compared to metasomatizing melts or fluids [Chesley et al., 2004; Bizimis et al., 2007]. Moreover, the relatively short amount of time (∼15 Ma) since metasomatism took place would have prohibited significant radiogenic 187Os ingrowth, even if Re were added by the metasomatic fluids. Melts that interact with the high Os content mantle are likely to inherit the low Os isotope ratios [Shirey and Walker, 1998]. Thus, the low 187Os/188Os of the lithospheric mantle may be preserved and it would likely be one of the best ways to establish a lithospheric origin for the rejuvenated lavas.
 Indeed, low Os isotope ratios are observed in the Samoan lithospheric mantle samples studied by Hauri and Hart  and Jackson et al. [2010b], even though noble gases and Sr-Nd-Pb isotopes indicate that they have been highly metasomatized [Hauri et al., 1993; Burnard et al., 1998]. We suggest that the Sr-Nd-Pb-He-Ne isotopes reflect a metasomatized Rarotonga plume signature (Figure 9) that is decoupled from Os isotopes that may reflect a lithospheric signature, owing to the incompatibility of Sr, Nd and Pb isotopes compared to Os. In fact, Samoan rejuvenated lavas host among the lowest 187Os/188Os ratios among ocean island basalts (OIBs) globally, with ratios that extend down to 0.123 [Hauri and Hart, 1993]. When compared to the Os isotope data on Samoan shield lavas, none of the rejuvenated lavas from Savai‘i (>30 ppt Os) have 187Os/188Os that overlap with Samoan shield lavas [Hauri and Hart, 1993; Jackson and Shirey, 2011] (Figure S3 in the auxiliary material). Critically, however, the Savai‘i rejuvenated lavas form a trend that extends to lower 187Os/188Os than shield lavas (Figure S3 in the auxiliary material), and we explore whether this can be explained with our metasomatized lithosphere-driven melting model for the rejuvenated stage.
 The lithospheric mantle in the Samoan region has remarkably unradiogenic Os signatures that overlap with the very low 187Os/188Os ratios measured in Samoan rejuvenated lavas. Jackson et al. [2010b] reported 187Os/188Os on 12 peridotite mantle xenoliths from Samoa, and half of the xenoliths yielded 187Os/188Os ratios equal to or lower than 0.123, the lowest 187Os/188Os ratio reported in Os-rich (>30 ppt Os) Savai‘i rejuvenated lavas (all < 0.127). Additionally, the same set of peridotite xenoliths examined for187Os/188Os were previously shown by Hauri and Hart to have pressures of <2.5 GPa, based on the Ca olivine-clinopyroxene exchange barometer ofKöhler and Brey , which places the xenoliths within the mantle lithosphere beneath Savai‘i. The shallow origin suggests that the xenoliths do not sample a sub-lithospheric (or “plume”) source, but instead represent the oceanic mantle lithosphere beneath Samoa. Therefore, the lithospheric mantle is a suitable geochemical reservoir for the low Os isotope ratios observed in the Samoan rejuvenated lavas.
 In Hawai‘i, Os-isotopic systematics were also used to argue that rejuvenated lavas are melts derived from the oceanic mantle lithosphere [Lassiter et al., 2000]. In Hawai‘i the depleted Sr-Nd-Pb isotopes, combined with a lack of correlation with Os isotopes, were used to argue against a plume source in rejuvenated lavas.Lassiter et al. argue that a plume origin for the Os isotope signature would imply significant contributions to the Sr-Nd-Pb isotopes as well, which would likely not generate the depleted signature observed in Hawaiian rejuvenated lavas.Lassiter et al.  also argue that metasomatized lithosphere is unlikely to be a source, since the expected relation of silica content with Os isotope ratios is opposite of what would be expected for this scenario. Finally, Lassiter et al. suggest that the Hawaiian rejuvenated stage Os isotope compositions may be the result of melting of a lithospheric pyroxenite component combined with a peridotitic component. In their model, large degree pyroxenite melts contribute to much of the Os-isotopic signature, while the other radiogenic isotope signatures are controlled by small degree peridotite melts. This argument hinges on pyroxenites being formed at the mid ocean ridge at ∼100 Ma as trapped melts or their cumulates. However, the Pacific Plate's previous passage over Rarotonga before it reached Samoa is vital here; in contrast to the Pacific plate “upstream” of the Samoan hot spot, no hot spot is known to have occupied the region of the Pacific plate now occupied by Hawai‘i. In the case of Samoa, old mid-ocean ridge-related pyroxenites are unlikely to be present due to the low pyroxenite solidus and high melt productivity [e.g.,Hirschmann and Stolper, 1996], which would have caused melting and removal of most of this component during passage over the Rarotonga hot spot. As a result, even though a lithospheric component may be present in Samoan rejuvenated lavas, its composition would likely not be the same as that suggested by Lassiter et al. for Hawai‘i. It is, therefore, not surprising that rejuvenated melts sampling the lithosphere beneath Hawai‘i exhibit a more depleted Sr-Nd-Pb isotope signature than Samoan rejuvenated lavas.
 Despite the agreement with our predictions, the data support our model but do not exclude other interpretations. For example, we cannot exclude a model where rejuvenated lavas are generated by contamination of plume melts with lithosphere containing high Os concentration and low Os isotope ratios, a model that would imply mostly sub-lithospheric melting that is at odds with our lithosphere-driven model. We consider this model less likely given the eruption of the rejuvenated lavas starts several million years after the initial construction of a volcano, implying the volcano has been transported up to hundreds of km away from the original mantle (plume) source. Additionally, we consider such a model to be unlikely because it does not explain why shield stage lavas have more radiogenic187Os/188Os ratios even though they pass through the same lithosphere as rejuvenated melts.
7. Evidence for Lithospheric Involvement From Lower Temperatures
 Another approach for evaluating the lithospheric contribution of rejuvenated volcanism stems from P-T estimates using major element compositions for primitive lavas. Therefore, we assess whether thermobarometry for the melts can be used to test for the model of involvement of lithosphere previously affected by Rarotonga. We might expect melting depths and temperatures to shift from deeper plume-based melting to shallow flexurally driven melting, and such a shift does seem present in temperature estimates based on lava major element compositions.
 Melting is expected to occur at shallower levels for rejuvenated compared to shield lavas based on the 3D numerical model of Govers and Wortel  that shows upwelling velocities concentrated in the lithosphere and decreasing into the top 80 km of the asthenosphere while the estimates of Farnetani and Hofmann  for plume melting in Hawai‘i suggest depths down to ∼175 km. Therefore, a shift in temperatures and pressures might be expected between the shield and rejuvenated stages, if our model is correct. As a first step, we calculate conditions assuming a peridotite source [Herzberg and Asimow, 2008; Lee et al., 2009], assuming that the low Os isotope signatures in the Samoan lavas [Hauri and Hart, 1993; Jackson and Shirey, 2011] signifies that pyroxenite is not likely to make up a significant component of the mantle source for Samoan shield and rejuvenated lavas.
 To test this idea, we use CaO (and Al2O3) versus MgO to filter for primitive lavas that have not experienced clinopyroxene fractionation (see Text S1 and Figure S4 in the auxiliary material) and exclude any picritic samples in order to obtain P-T estimates for the melts using the technique ofLee et al. . Unfortunately, the majority of the estimated pressures are around 4 GPa, based on the silica activity in the melt, and assuming orthopyroxene and olivine are present [Lee et al., 2009]. However, at these pressures orthopyroxene is not stable, except in combination with olivine and melt at a degree of melting over 25% [Walter, 1998; Herzberg and O'Hara, 2002], which is likely not relevant to hot spots such as Samoa (where the degree of melting is thought to be relatively low). As a result, we only consider the temperature estimates reliable for the samples. Each temperature estimate is offset from its solidus along a melting adiabat, reflecting the degree of melting that took place, and implying that the solidus was actually crossed at a higher temperature than calculated. Therefore, it is not the average temperature, reflecting average degrees of melting, but the highest temperatures for each stage that provide insight into the difference in conditions between shield and rejuvenated lavas. This argument assumes that the highest temperatures reflect the lowest degree melts from a peridotite source with similar solidus, therefore being closest to the solidus intersection. This is probably a reasonable assumption, since rejuvenated lavas are generally expected to be low-degree melts [e.g.,Clague and Dalrymple, 1987].
 When the results for the temperature calculation are examined, it is clear that the shield lavas range to higher temperatures than the rejuvenated lavas by ∼50°C (Figure 10). Such a shift to lower temperature is at least qualitatively consistent with our model, where plume-driven deep melting gives way to shallower, flexure-based melting. We tested our results againstHerzberg and Asimow's temperature estimates, which filter for pyroxenite source material, clinopyroxene fractionation or accumulation, volatile-rich sources, and allows for adjustment of Fe oxidation state. Only a subset of samples meet all the filter criteria of theHerzberg and Asimow  model (only 2 rejuvenated lavas). This model suggests a similar difference in temperature between shield and rejuvenated lavas (∼75°C) even when Fe2O3/TiO2 is varied, although absolute temperatures are slightly different than the method by Lee et al. . Therefore, these models seem to be in agreement with melting under lower temperature conditions during the rejuvenated stage.
 However, there is an important caveat. Since we argue that the rejuvenated lavas may have a metasomatized component in their source, it is possible that the volatile content of this source is significantly different from the shield stage source. As a result, the solidus for the rejuvenated stage may be different from that of the shield stage. In particular, the presence of metasomatic phases such as amphibole or phlogopite can affect the solidus while still producing the observed alkalic melts [e.g., Pilet et al., 2008]. However, hydrous phases were never observed in the metasomatized Samoan xenoliths [Hauri et al., 1993]. Nonetheless, this uncertainty implies that the temperature estimates do not provide strong constraints on our model, but they do seem consistent with it, to first order.
 The isotopic composition of new samples from 200 m down-section in a canyon on Savai‘i (Samoa) allow us, for the first time, to constrain the unusually large volumes of rejuvenated lavas that are erupted in Samoa. Such volumes and their unusual isotopic compositions lead us to suggest that the location of Samoa near the northern Tonga Trench terminus plays an important role in the origin of this stage of volcanism. We propose a model where subduction related upwelling of the shallow mantle due to flexure causes decompression melting involving a metasomatized lithospheric component. Evidence for involvement of such a lithospheric component comes from the composition of both rejuvenated lavas and lithospheric xenoliths. These compositions are in agreement with melts that contain a lithospheric component that includes a signature from the plate's prior passage over the Rarotonga hot spot. Therefore, the two features that may explain the difference in erupted rejuvenated lavas between Hawai‘i and Samoa are: (1.) the location and flexure near a trench, and (2.) the potential for metasomatism of the lithosphere, suggesting the lithosphere plays an active role during this phase of volcanism. Evidence from Sr-Nd-Pb isotope compositions of the rejuvenated lavas suggests a mixture involving a metasomatized MORB component that incorporated a Rarotonga-type composition and a subsequent Samoan component. Os isotope signatures of lavas and xenoliths also suggest involvement of a lithospheric mantle component. Furthermore, He-Ne isotope compostions argue for a metasomatic component corresponding to an EM1 mantle source in the Samoan lithosphere. Last, temperature estimates for Savai‘i Island suggest a shift to lower temperatures during rejuvenated stage volcanism, which might be expected if the origin for magmatism shifts from a plume source to a shallow source dominated by melting just below and within the lithosphere. Data for each of these areas by itself does not provide conclusive evidence for our model, however the preponderance of data that may be explained with our model argues that it is likely the simplest explanation that explains the most data, although additional data would allow us to evaluate our model for each of these areas while ruling out alternative models that cannot currently be excluded.
 We would like to thank Warren Jopling, the good people of the Safua Hotel, and the chief of Sili village and his sons for making the field component a success. We are thankful for the reviews by Loÿc Vanderkluysen, Christoph Beier, as well as anonymous reviews and comments on various versions of this work, comments from Claude Herzberg and Jim Natland, and the editorial efforts by Joel Baker. We acknowledge NSF support, through grant EAR-0946752.