Arc magma compositions controlled by linked thermal and chemical gradients above the subducting slab
Sebastian F. L. Watt,
Ocean and Earth Science, National Oceanography Centre Southampton, University of Southampton Waterfront Campus, Southampton, UK
Corresponding author: S. F. L. Watt, Ocean and Earth Science, National Oceanography Centre Southampton, University of Southampton Waterfront Campus, European Way, Southampton SO14 3ZH, UK. (Sebastian.Watt@noc.soton.ac.uk)
 Global arc magmatism is sustained by a continuous fluid flux that is returned to the mantle in subduction zones. Despite considerable advances in simulations of melting processes, models of arc magmatism remain incompletely tested against erupted products. Here, we show that a suite of primitive volcanic rocks from across the southern Chilean arc preserves the signature of a systematic down-slab gradient in fluid chemistry. The chemical gradient is consistent with predictions from modeling, geothermometry and experiments. We infer that increasing slab-surface temperatures cause the sub-arc slab flux to become less water-rich and increasingly dominated by hydrous melts over a distance of a few kilometers behind the arc front. This change exerts a first-order control on magma chemistry, and implies discrete melt-transport pathways through subduction zones. Our results replicate patterns in other arcs, implying common sub-arc slab-surface temperature ranges in thermally-diverse subduction zones.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Variations in arc magma compositions are a key tool for constraining element cycling and melt transport through subduction zones [Stolper and Newman, 1994; Schmidt and Poli, 1998]. Recent modeling of arc systems suggest that magmatic compositions reflect the nature of fluids released from the slab which, in turn, is determined by slab temperature [Plank et al., 2009; van Keken et al., 2011]. Aqueous fluids, derived from mineral breakdown, are released along a broad section of the slab interface. Above 700–800°C (at pressures of 3–4.5 GPa), these fluids promote wet melting of both sediments and mafic crust at the top of the subducting slab, producing a slab-derived component that is poorer in water and much richer in solutes [Hermann and Spandler, 2008; Skora and Blundy, 2010].
 Melt inclusions in primitive volcanic rocks, which have undergone limited modification in the crust, provide an opportunity to investigate primary magma compositions directly [Kelley et al., 2006, 2010; Portnyagin et al., 2007]. Here, we investigate variation in primitive rock compositions from a single arc, over short (10 km) across-arc length scales. Studying rocks from a single arc minimizes uncertainties associated with variable sediment and mantle-source compositions. Comparable studies have investigated the relationship between slab-derived material and volcanic rock compositions either from global data sets [Cooper et al., 2012; Ruscitto et al., 2012] or by investigating arc to back-arc or along-arc compositional changes [Kelley et al., 2010; Vigouroux et al., 2012]. There are few studies of across-arc chemical variation on within-arc (~10 km) length scales [Duggen et al., 2007; Portnyagin et al., 2007; Nichols et al., 2012]. Using the example of southern Chile, we demonstrate a sharp increase in slab-surface melting as temperatures increase behind the arc front, showing that changes in primary magmatic compositions are consistent with predictions from a range of modeling [Kelley et al., 2010; van Keken et al., 2011] and experimental data [Hermann and Spandler, 2008; Wood and Turner, 2009; Skora and Blundy, 2010].
2 Study Region
 This study is based on a suite of four picrites and high-Mg basalts erupted as bedded scoria deposits (eruptive units) from four volcanoes in southern Chile (Table 1). These volcanoes range from Apagado (Ap1 eruption) at the arc front (defined by a chain of stratovolcanoes; Figure 1) to distances progressively towards the rear arc (Hornopirén, Ho1 eruption; South Minchinmávida; and Palena) (Figure 1). Each unit comprises coarse-ash to fine-lapilli deposits, produced by the explosive eruption of magmas ascending rapidly from depth. The rapid quenching of such deposits is ideal for preserving olivine-hosted melt inclusions [Lloyd et al., 2013], which record equilibrium melt compositions at their depth of formation. The melt inclusions analysed here are from multiple free olivine crystals, picked from each unit (see Supporting Information for full description).
Table 1. Summary of Sampled Units and Derived Parameters (See Watt et al. [2011b] and Supporting Information)
 The Ap1 unit is notably primitive (MgO 12–15 wt%), containing unzoned, forsterite-rich (Fo87–90) olivines as the only phenocryst phase. The Palena and South Minchinmávida units are basalts (MgO 8 wt%), and also contain unzoned olivines with a narrow compositional range (Table 1). These units appear to be compositionally homogeneous. In contrast, Ho1 is divided into a more primitive basal basalt (Ho1a; with Fo84–87 olivines) and an overlying basaltic andesite (Ho1b; Fo81–83).
 The subducting plate beneath southern Chile is relatively young and warm (thermal parameter φ = 430 km [Syracuse et al., 2010]), and models predict extensive dehydration, with pulses of aqueous fluid at depths of ~80 km and ~130 km [van Keken et al., 2011] (Figure 1). Beneath Apagado, the slab-surface depth is 105 km [Lange et al., 2008], increasing to 115 km beneath Hornopirén, and up to 140 km beneath Palena (Table 1). Thermal models [Syracuse et al., 2010] indicate (model-dependent) down-slab temperature increases of 20–80°C between Apagado, Hornopirén, and Palena, with a slab-surface temperature of 810 ± 50°C beneath Apagado.
3 Primitive Magma Compositions
 Olivine-hosted melt inclusion compositions were analysed by electron microprobe (major elements) and ion microprobe (secondary-ion mass spectrometry; CO2, H2O, and selected trace elements) (see Supporting Information for extended methods, details of sample corrections, and full results). Correction was made for post-entrapment crystallization against the inclusion walls, by adding increasingly forsteritic olivine incrementally until the glass was in equilibrium with the host (using PETROLOG software [Danyushevsky and Plechov, 2011]). The amount of olivine added was notably lower for CO2-rich samples; samples showing signs of degassing (<120 ppm CO2) were therefore rejected, along with four low-H2O outliers. This substantially reduced the data set, but remaining compositions were tightly clustered, in contrast to the scatter of degassed samples. Less than 9% olivine addition was required to correct (undegassed) Ap1 samples, and <5% for all other samples except one Ho1a analysis (11%).
 Melt-inclusion compositions lie at the high-MgO end of southern Chilean arc lava compositions (Figure 2a and Supporting Information), suggesting they are representative of parental magmas. Ap1 inclusions are richer in water (~3 wt%), but relatively poorer in CO2 (500 ppm maximum), than those from behind the arc front (~1.5 wt% H2O, 1000 ppm CO2; Figure 2b). Melt-inclusion major element compositions are nearly identical to the whole rock for each unit except Ap1 (Figure 2a), for which some samples appear to have been affected by olivine accumulation.
 We estimate parental magma compositions by adding olivine to the melt inclusion compositions, until they were in equilibrium with mantle olivine (Fo90) [Stolper and Newman, 1994], using PETROLOG with the Ford et al.  olivine model and a QFM + 1 buffer. Melt-inclusion major element compositions show no evidence for significant plagioclase or clinopyroxene fractionation in any unit (CaO increases with decreasing MgO), so we assume olivine was the only liquidus phase removed prior to entrapment (water suppresses plagioclase crystallization at these MgO contents). The small addition required for Ap1 (0–2 wt%), in contrast with the other units (10–16 wt%), confirms that this rock is close to a primary mantle melt. Average parental magma compositions are high-Mg basalt for the water-rich Ap1, and picrite for the other, drier units (Figure 2a), consistent with the expected effect of water on the Mg:Si ratio of mantle melts [Wood and Turner, 2009].
 The trace element content of parental magmas varies systematically between Ap1 and the other units. High water contents in Ap1 correlate positively with Sr and negatively with Ce, Ba, and La (Figure 2c). Ratios involving H2O and Ba as the numerators show an inverse correlation. Sr is highly mobile at relatively lower slab-surface temperatures, being enriched in the aqueous fluid derived from altered oceanic crust [Vigouroux et al., 2012] and without a significant host accessory-phase in subducting sediment [Hermann and Rubatto, 2009]. In contrast, sediment melts are an important source for Ce, Ba, and La [Hermann and Spandler, 2008; Skora and Blundy, 2010]. The compositional differences between Ap1 and the other units suggest that aqueous fluids are a relatively dominant component of the slab flux beneath the arc front, but that slab-surface melts dominate the flux behind the arc front. However, an exclusively aqueous slab flux may be too solute-poor to impart the trace element concentrations in Ap1 [cf. Hermann and Rubatto, 2009; Plank et al., 2009]. We therefore suggest that the sub-Apagado flux may be a mixture of aqueous fluid and a near-solidus sediment melt component (with relatively higher Sr:Ba ratios than hotter, supra-solidus melts; cf. Hermann and Rubatto ), contrasting with higher-temperature melts that dominate the flux beneath the other volcanoes.
 The increase in La/Y behind the arc front (Figure 2) may indicate reduced partial melting of the mantle source, but this cannot be unambiguously discriminated from variation in slab-derived La. Slab-derived melts, rather than increased mantle fertility, may also cause the elevated Nb and Zr for units other than Ap1. Differences in these elements between Ho1a and Ho1b suggest potentially complex source heterogeneities for magmas involved in the same eruption.
4 Melting Conditions and Fluid Input
 The mantle-source melt fraction, F, and water content, CH2O, can be estimated for each parental magma using batch-melting relationships [Kelley et al., 2006, 2010]. The major uncertainty in these calculations, which assume that TiO2 is wedge-derived, is the Ti concentration of the mantle source. To account for this, we make all calculations with both high end-member (estimated from melt Ti:Y; see Supporting Information) and low end-member (normal mid-ocean ridge basalt (NMORB) mantle Ti of 0.133 wt%; Salters and Stracke ) values (Figure 3), which produce the same relative results. Values of F (based on Ti:Y) are 17–20% for Ap1 and Palena, and mostly 14–17% for the other units, corresponding to maximum CH2O of 0.63 wt% for Ap1, 0.33 wt% for Palena, and <0.3 wt% for the others (Figure 3a). The slightly higher (relative to Ho1) melt fraction and water content for Palena, which lies well behind the arc front, may relate to a modeled second pulse of slab dehydration at ~130 km depth [van Keken et al., 2011].
 The reduction in F and CH2O between Ap1 and Ho1 occurs across a down-slab distance of only ~10 km (illustrated in Figure 1c). The equilibration conditions of each parental magma are constrained further by Si-Mg thermobarometry [Lee et al., 2009] (Figure 3b), using our estimated parental magma compositions. This shows that Ap1 parental magmas pooled toward the base of the crust, and lie below or on the dry-mantle solidus, confirming that the observed melt fractions beneath Apagado require wet melting. Equilibration conditions for the other magmas lie above the dry-mantle solidus (Figure 3b) and suggest a potential role for decompression melting, although a slab-derived component is still required to explain their enrichment in fluid-mobile elements. For all units, pressures estimated from the thermobarometer of Lee et al.  are broadly consistent with those derived using silica contents from lherzolite melting experiments (total range: 0.9–2.1 GPa) [Wood and Turner, 2009].
 The consistency between our thermobarometric and batch-melting calculations can be assessed via hydrous-melting parameterizations. For example, F-CH2O relationships calculated from our thermobarometric results using the Kelley et al.  model, and assuming an NMORB source (see Supporting Information), agree well with F and CH2O from batch-melting calculations, particularly for Ap1 (Figure 3a). The offsets for the other units may reflect a more fertile mantle source. Our thermobarometric results also agree with the Portnyagin et al.  parameterized melting model, which provides a relationship between mantle temperature, melt fraction (F), and parental-magma water content. This model indicates water-fluxed melting for Ap1, but a component of dry melting (for F derived from both sets of Ti values) for the other units (Figure 3c). Our results are consistent with lower melt productivity (for a given water content) for the Ap1 parental magma, relative to the other units (Figure 3a), but high melt fractions due to higher source water contents.
 Estimates of slab-surface temperatures (Table 1 and Supporting Information), from Ce and K geothermometry [Plank et al., 2009; Cooper et al., 2012], suggest a steep temperature increase behind the arc front, from 720°C beneath Apagado (from Ce; 800°C for the K-thermometer) to ~870°C for the other volcanoes. This temperature range agrees well with modeled slab temperatures [Syracuse et al., 2010]. The apparently rapid temperature increase behind the arc front may reflect heating of the slab following slab-mantle decoupling [Wada and Wang, 2009]. Within model uncertainties, our estimates cannot resolve whether the sub-Apagado slab lies above or below the wet-sediment solidus (700–800°C at 3–4.5 GPa) [Hermann and Spandler, 2008; Skora and Blundy, 2010], but are nevertheless consistent with our inference that the sub-Apagado mantle is fed by a relatively water-rich flux, potentially incorporating near-solidus slab-surface melts. Importantly, our estimates suggest that the slab beneath the other volcanoes exceeds the wet-sediment solidus, consistent with a melt-dominated flux behind the arc front driving the change in magma compositions between Ap1 and the other units.
 Thermal models suggest the arc front lies beyond the main pulse of slab dehydration [van Keken et al., 2011], but given uncertainties in the models, slab depth, and fluid transport dynamics, Ap1 may be affected by this initially high fluid flux. Figure 1c depicts vertical ascent of slab-derived material through the mantle wedge, but this is poorly constrained.
 In the context of the flux, the sub-arc slab as discussed here is not necessarily vertically beneath the arc, and we simply argue that discrete across-arc pathways are maintained. A component of transport away from the trench [e.g., Marschall and Schumacher, 2012] would produce greater consistency between modeled depths of slab dehydration [van Keken et al., 2011] and the (greater) slab-surface depth at the arc front, although others have argued for lateral transport in the opposite direction [cf. Cooper et al., 2012]. In any case, we suggest that a relatively water-rich flux drives mantle melting beneath Apagado. Beneath Hornopirén, although slab dehydration may be reduced, temperatures are high enough for it to drive significant slab-surface melting [Hermann and Spandler, 2008; Skora and Blundy, 2010; Cooper et al., 2012]. The resultant melt-dominated fluid will be enriched in K, Ba, and light rare earth elements from the breakdown of phases such as phengite [Hermann and Spandler, 2008] and may also contain more CO2 from carbonate breakdown (Figure 2b) [Cooper et al., 2012; Kerrick and Connolly2001].
 Although we invoke slab-surface melting, with sediment melts dominating the trace element signature, to explain our across-arc chemical trends, a sediment-diapirism model [Behn et al., 2011] provides a possible alternative if Ap1 only marginally overlaps with the locus of diapirism (and subsequent sediment melting), in contrast to the other units. The depth of diapir initiation calculated by Behn et al.  is shallower than the arc front, but dependent on (uncertain) slab thermal conditions. Better constraints on thermal structure and wedge dynamics are required to fully understand how our observations fit with current models of material transport through arcs.
5 Implications of Across-arc Changes in Slab Flux
 The chemical imprint of a changing across-arc slab flux may be preserved in more evolved volcanic rocks. In southern Chile, arc-frontal volcanoes are less potassic than those 10–20 km behind the front (Figure 1, Figure 4). This is consistent with the expected steady increase in K2O/H2O above the wet-sediment solidus [Hermann and Spandler, 2008] (up to ~1000°C, beyond sub-arc depths). Preservation of this systematic pattern, in spite of later modifying processes, suggests a general lack of across-arc mixing during magma transport through the mantle wedge. The slab flux may also influence the phase mineralogy of volcanic rocks within the arc. Amphibole is present in lavas only at three volcanoes in this arc segment (stable at Huequi; more commonly with disequilibrium textures at Calbuco and Mentolat; Watt et al. [2011a]), which have the lowest K-contents and all lie immediately at the arc front (Figure 4). We predict that their parental magmas are also the most water-rich and relatively more likely to evolve during crustal modification into melts where amphibole is stable.
 Given the global diversity in the temperature of subducting slabs (described by the thermal parameter, φ; Syracuse et al. ), we may expect arc magmas from cooler subduction zones (with high φ) to carry a stronger aqueous-fluid slab-derived signature than hotter systems, where we may expect a melt-dominated flux. However, across-arc chemical changes in Kamchatka [Duggen et al., 2007], a relatively cool system (φ = 5400 km), are very similar to those in southern Chile (φ = 430 km), with a transition just behind the arc front from wetter, higher melt-fraction magmas to magmas with a strong hydrous-melt signature. A similar across-arc pattern also occurs in the Izu-Bonin arc (φ = 4000–4500 km) [Nichols et al., 2012]. These observations suggest that the sub-arc slab spans a very similar temperature range in thermally-diverse subduction zones. This is supported by Ce-thermometry, which predicts near-identical across-arc temperature ranges in Kamchatka [Cooper et al., 2012] and southern Chile.
 Slab temperatures dictate the water content and quantity of slab flux [van Keken et al., 2011], and thereby influence mantle melt fractions. The observation of similar systematic across-arc chemical variations in different subduction zones suggests a relationship between temperature-driven processes at the slab-surface and the position and width of volcanic arcs. How these processes act in combination with mantle-wedge dynamics, to control the position where extractable quantities of magma are formed [Schmidt and Poli, 1998; England and Katz, 2010], requires further investigation.
 Although we cannot unambiguously distinguish the effects of current versus past subduction on the sub-arc mantle, the across-arc pattern we observe in southern Chile is remarkably systematic. Our understanding of current subduction zone structure and temperature-dependent variations in slab input correlate well with primary magma compositions. Our results confirm that the slab flux becomes less water-rich and increasingly dominated by slab-surface melts as slab temperature increases. This change occurs sharply, over a down-slab distance of a few kilometers, marking a distinct shift in the suite of elements transported through the arc. Preservation of these patterns at the surface indicates limited across-arc mixing during melt transport through the mantle wedge. Replication of these patterns in other arcs suggests a similar sub-arc temperature range in thermally-diverse subduction zones, and a connection between the position of arcs and processes at the slab surface.
 Glass analyses were funded by NERC EIMF Grant IMF377/0509; we thank C. Guzman for field assistance, and R. Hinton and J. C. M. de Hoog for analytical advice. We are grateful to M. Portnyagin and two anonymous reviewers for insightful reviews. This work was supported by a NERC studentship and grant NE/I02044X/1 (SW).
 The Editor thanks James Gill and Maxim Portnyagin for their assistance in evaluating this paper.