We present trace element and Sr-Nd-Pb isotope data on volcanic rocks recovered from the submarine Monowai volcanic center, which marks the midpoint of the ∼2500 km long Tonga-Kermadec arc. The center consists of a large (12 × 9 km) partly hydrothermally active caldera and a 12 km diameter ∼1500 m high volcanically and hydrothermally active stratovolcano. The stratovolcano lavas are tholeiitic basalts, which show variable evidence for plagioclase (±pyroxene) accumulation. The caldera lavas range from basalt to andesite, with the compositional variation being consistent with fractional crystallization as the dominant process. The mafic Monowai magmas were generated by relatively high degrees (12%–20%) of partial melting of a previously depleted MORB-type spinel-peridotitic mantle, metasomatized by slab-derived fluids. Strongly fluid mobile 87Sr/86Sr and 207Pb/204Pb ratios of the Monowai basaltic lavas are similar to those from the Putoto, Raoul, and Macauley volcanic centers 200–400 km to the south, suggesting derivation largely from subducted sediment. In contrast, variably fluid immobile 143Nd/144Nd ratios suggest an isotopically heterogeneous mantle along this segment of the arc. Higher 206Pb/204Pb in Monowai lavas imply some influence from the nearby subducting Louisville seamounts in melt generation. The formation of one of the Earth's largest submarine mafic calderas can best be explained through drainage of a single magma reservoir and subsequent collapse caused by trench-perpendicular extension, probably via southward progressive rifting of the northern Havre Trough.
Although there is broad consensus on the major processes occurring at intraoceanic subduction zones, where oceanic crust (including upper mantle and sediment cover) is recycled into the mantle, controversies still exist regarding some fundamental aspects of convergent margins. These controversies include the extent to which sediment on the subducted plate is involved in the refertilization of the mantle wedge, the relative contributions of fluids derived from dehydration of oceanic crust versus from overlying sediments, the depth of dehydration, extent of migration of hydrous fluids, element fractionation during generation of subduction zone fluids and, potentially, sediment melts [e.g., Pearce and Peate, 1995; Elliott et al., 1997; Plank and Langmuir, 1998; Stern, 2002]. Interest in submarine volcanic centers in oceanic arcs has increased in recent years with the discovery that oceanic arc and back-arc systems are commonly hydrothermally active and, in some cases host to massive sulphide mineralization [e.g., Fouquet et al., 1991; Kamenetsky et al., 2001; Moss et al., 2001; de Ronde et al., 2001, 2005, 2007, 2011]. Submarine arc volcanoes span the range from small cones, through large stratovolcanoes to large caldera systems [Worthington et al., 1999; Wright et al., 2003, 2006; Haase et al., 2006; Graham et al., 2008; Smith et al., 2009, 2010] and, as on land, felsic volcanic rocks dominate. There are few known examples of large mafic dominated calderas; thus the submarine Monowai volcanic center, located at the northern end of the Kermadec arc, is of particular scientific interest. In addition, focusing on single volcanic centers along the arc allows more detailed studies of magma chamber and mantle source processes beneath, which may give new insights undetected in large-scale studies.
The Tonga-Kermadec intraoceanic arc extends from northern Tonga ∼2500 km southwestward to the continental margin of New Zealand (Figure 1). Along the arc, Jurassic-Cretaceous Pacific lithosphere, including a ∼200 m thick sediment cover, has been subducting westward beneath the Australian Plate since the Oligocene [Clift et al., 2001; Turner and Hawkesworth, 1997; Turner et al., 1997]. The arc is located 100–120 km west of the trench, and is accompanied by crustal extension in the subparallel Lau Basin and Havre Trough back arc systems. The thickness of the mostly Eocene crust beneath the arc ranges from 10 to 18 km [Duncan and Clague, 1985]. The convergence rate increases from ∼50 mm/yr in the south to >240 mm/yr at the northernmost Tonga fore arc [DeMets et al., 1994; Bevis et al., 1995], the highest rate globally. The dip of the Pacific Plate slab below the arc is 28–30° down to around 100 km depth. Below 100 km, the slab steepens to 43–45° beneath the Tonga arc and 55–60° beneath the Kermadec arc [Isacks and Barazangi, 1977], corresponding to the change in convergence rates.
Seismic tomography has shown that the subducting slab beneath the Kermadec arc is still present below the transition zone (>660 km depth) and possibly extends to the core-mantle boundary [Billen and Gurnis, 2003; van der Hilst, 1995]. The composition and thickness of the sediment cover on the Pacific Plate has been documented at a drill site east of the trench (DSDP site 204 adjacent to the Tonga trench). DSDP site 204 sediments consist predominantly of clay, volcanic ash, sandstone and conglomerate [Shipboard Scientific Party, 1973; Ewart et al., 1998; Turner et al., 1997], whereas sediments collected by Gamble et al.  along the Kermadec trench between 28°S and 34°S are described as pale yellow-gray mud with variable carbonate content.
Near 25.36° S, the mantle plume-related Louisville seamount chain is undergoing subduction. Together with the ∼900 km E-W extending Osbourn Trough, an extinct Cretaceous spreading center [Worthington et al., 2006], the present position of Louisville seamount chain subduction marks the boundary between the Tonga (Tofua) arc to the north and the Kermadec arc to the south, and their respective associated back-arc basins, the Lau Basin and Havre Trough. Since ∼4 Ma, subduction of the Louisville seamount chain has progressively moved southward along the Tonga arc [e.g., Ruellan et al., 2003]. At present the ∼66 Ma old Osbourn seamount is also undergoing subduction, causing an indentation of the Australian Plate [Lonsdale, 1986]. It has been suggested that the subducting Louisville seamount chain marks the transition of back-arc spreading in the southern Lau Basin to back-arc rifting at the northern Havre Trough [Ruellan et al., 2003], causing local seafloor uplift.
Although its extent is not fully known, Monowai is the second largest by area (∼530 km2), and the tallest (∼1500 m) submarine volcanic center in the northern Kermadec arc (Figure 2). It is located ∼49 km to the NNE of Hinepuia volcanic center at 177.5°W and 25.5°S, and ∼48 km south of ‘U’ volcanic center [Graham et al., 2008]. The center covers an area of ∼24 × 22 km and comprises a large caldera (Monowai caldera 1; MoC1), in which a smaller caldera is located (MoC2), a large stratovolcano (MoV) and several small parasitic cones (P1–P8), with a total volume of ∼42 km3 [Graham et al., 2008].
The northern part of Monowai volcanic center is dominated by the large NW-SE elongated caldera MoC1, covering an area of 11.8 × 9.1 km. The outer caldera rim rises to ∼800 mbsl and is strongly dissected by ring faults and associated wall collapse. The caldera floor, ∼220 m below the rim (1250 mbsl) is sediment covered. MoC2 is nested within MoC1 and covers an area of ∼7.9 × 5.7 km and is similarly elongated to that of MoC1 with a floor at ∼1590 mbsl, some 560 m lower than the caldera rim. Numerous small cones mark the rim, and a ∼350 m resurgent cone ∼2 km in diameter (MoR) occurs near the center. The two calderas are the largest observed along the northern Kermadec arc.
About 12 km to the SSW of the Monowai calderas is a large ∼12 km diameter stratovolcano (MoV). With a volume of ∼11 km3, MoV is the largest single volcanic edifice along the northern Kermadec arc [Graham et al., 2008]. MoV is characterized by relatively smooth slopes and has been active for at least several decades making it the youngest volcanic edifice at Monowai [Wright et al., 2008]. Although a summit crater is absent, intense hydrothermal venting occurs in the summit area, and there has been eruption of fresh scoria and pyroclastics over the past seven years. Strong T wave signals from MoV detected in May 2002 have been linked to a significant sector collapse, lowering the summit by ∼90 m to 130 mbsl [Wright et al., 2008]. The collapse probably initiated a strong eruption phase due to magma-water interaction within the shallow magma conduit. Between 2004 and 2007 the summit rose again to ∼69 mbsl [Chadwick et al., 2008], although two further collapse events have since occurred. These construction-destruction cycles have been linked to the high eruption rates of unconsolidated volcaniclastic material, which are prone to destabilization.
Extensive faulting dissects the Monowai volcanic center in two main directions: (1) fault traces following a NE-SW trend, similar to those seen in the nearby back arc and (2) prominent, circular caldera ring faults within MoC1 and MoC2 (Figure 2).
3. Petrological and Geochemical Characteristics
Since the first recorded evidence of volcanical activity in 1977, Monowai has been repeatedly sampled, including during the R/V Sonne expedition in 1998 [Haase et al., 2002], NZAPLUME III in 2004 [Graham et al., 2008] and NZAROF in 2005 [Embley et al., 2005; Embley and the Scientific Shipboard Party, 2006]. The latter two cruises carried out 14 dredge runs and several manned submersible dives on various parts of the center, including the newly discovered calderas (Figure 2). Recovered samples consist of blocky, pillowed lavas and cobbles and scoriaceous cinders, including pristine samples with glassy rims [Graham et al., 2008]. Due to consolidation under pelagic sediment cover, dredge sampling proved much more difficult at the MoC than at the recently volcanically active MoV, so older volcanic successions are only sparsely sampled.
As documented by Graham et al. , Monowai lavas are vesicular (12%–57%; average 31%), variably glassy (0%–65%; average 22%), and moderately phenocryst rich (1%–34%; average 13%). Identified mineral phases are plagioclase (12%–39%; average 23%), pyroxene (19%–74%; average 49%, with clinopyroxene > orthopyroxene), Fe-Ti oxides (0.1%–13%; average 4.5%), olivine (0%–2%; average 0.8%) and rare apatite. Disequilibrium textures are common, if not ubiquitous, with some plagioclase and pyroxene phenocrysts being sieved, having reaction rims and resorbed margins, or showing reverse zoning. Quench textures are ubiquitous. Groundmass is typically intersertal to intergranular, containing microlites of plagioclase (commonly swallow-tailed), pyroxene (typically sheave-like or plumose, but more rarely swallow tailed), and disseminated Fe-Ti oxides. Secondary alteration is less than 1 vol%, based on the occurrence of secondary minerals and devitrified glass.
Chemical analyses and analytical methods are included in Text S1 and Table S1 in the auxiliary material. One hundred percent volatile-free normalized rock compositions range from basalt (n = 19) through basaltic andesite (n = 6) to andesite (n = 4) (49.2 to 62.9 wt% SiO2), and belong to the tholeiitic low-K series [Le Maitre et al., 2002] (Figure 3). MoV lavas have a different compositional range from MoC lavas, with lower SiO2, [TiO2], [FeO], [MnO], Na2O, [K2O], [P2O5] and higher Al2O3, MgO, CaO contents (species in square brackets not shown; Figure 4). [TiO2], [FeO], [MnO], MgO, and Na2O contents increase with increasing SiO2, whereas Al2O3 and CaO decrease. MoC lavas instead generally show decreasing Al2O3, FeO, MgO (7.16–1.26 wt%), CaO and increasing [TiO2], Na2O, K2O, P2O5 with increasing SiO2.
All basalts and basaltic andesites are enriched compared with normal mid ocean ridge basalts (NMORB, after Sun and McDonough ) in incompatible fluid mobile elements (i.e., large ion lithophile elements; LILE, e.g., Rb, Ba, U, K, Pb; Figure 5). Thorium shows variable contents ranging from depleted (0.6× NMORB in the basalts) to enriched (6× NMORB in the andesites). By contrast, incompatible fluid immobile elements (e.g., Nb, Hf, Zr, Ti, Y, REE) are depleted in the basalts and basaltic andesites (<0.9× NMORB) and show a characteristic trough at Nb (Figure 5); the andesites are enriched with respect to the most primitive lava, although all have similar Sr contents. The REE patterns are almost flat with slight LREE versus HREE depletion (e.g., (La/Sm)N = 0.52–0.97; N = normalized to C1 chondrite after Sun and McDonough .
Radiogenic isotope compositions exhibit narrow ranges (87Sr/86Sr = 0.7034–0.7035; 143Nd/144Nd = 0.51303–0.51307; 206Pb/204Pb = 18.71–18.75; 207Pb/204Pb = 15.57–15.58; 208Pb/204Pb = 38.34–38.39; Table 1). Pb and Sr isotope ratios are generally more radiogenic than in East Pacific Rise (EPR) and South East Indian Ridge (SEIR) lavas [Meyzen et al., 2007], although Nd isotopes have similar values (Figure 6).
Table 1. Sr-Nd-Pb Isotope Ratios of the Monowai Volcanic Center Lavas
MoC, N wall
MoC, SW edge
MoC, SW edge
MoC, SW edge
4.1. Shallow-Level Processes
Plagioclase and pyroxene (clino- ortho-) are the main phenocryst phases in Monowai lavas, with olivine, titanomagnetite and apatite minor constituents in the more mafic lavas. Fractional crystallization of these mineral phases can explain the major and minor element variations from the most primitive MgO-rich MoV basalt (P72257; 7.1 wt% MgO; 50.6 wt% SiO2; Figure 4) to the most MgO-poor MoC andesite (1.3 wt% MgO; 62.9 wt% SiO2). Dominant plagioclase fractionation is supported by an increasingly negative Eu anomaly ((Eu/Eu*)N = 0.78–0.93), low Sr/Y and Sr/Zr, and enrichment of HFSE in the more evolved lavas (Figure 5).
The stratovolcano basalts show decreasing MgO (and Mg#) content with decreasing SiO2, TiO2, FeO and MnO, and to a lesser extent Na2O and P2O5 (Figure 4), which is inconsistent with fractional crystallization of the observed phenocryst phases. Decreasing MgO contents in the basalts are accompanied by increasing Sr/Y, (Eu/Eu*)N (1.0–1.3) and decreasing Y and total REE, which is best explained by accumulation of Ca-rich plagioclase together with minor pyroxene (Figure 4). Crystal accumulation is supported by a strong negative correlation between phenocryst content and silica content (Figure 4h). A low-Si anorthitic plagioclase, however, requires a parental lava of lower silica content than the most primitive Monowai sample (P72257), which itself has already undergone some fractional crystallization. The fractionated nature of the Monowai basalts is supported by their low Mg# (54–62), and low Ni (<34 ppm) and Cr contents (<46 ppm).
Notwithstanding the evidence for accumulation, the basaltic composition of MoV lavas and the rarity of crystal disequilibrium textures (e.g., resorption, reaction rims and sieve textures) imply relatively short magma storage times and/or high ascent rates [e.g., Reid, 2003]. The Sr-Nd-Pb isotopic compositions of the MoV lavas show no significant change with decreasing MgO, suggesting incorporation of phenocrysts with similar Sr-Nd-Pb isotopic signature. Possible scenarios to explain the accumulation process includes (1) incorporation of plagioclase and some pyroxene fractionated from the upper parts of the magma chamber; the density of Ca-rich plagioclase (<2.75 g/cm3) is less than that of tholeiitic magma (∼2.9 g/cm3), and (2) incorporation of plagioclase-rich gabbroic crustal fragments. As no gabbroic fragments have been observed as xenoliths in the Monowai lavas, scenario (1) seems the more likely.
4.2. Temporal Evolution of Monowai Volcanic Center
The northwestern and western flanks of MoC1 and the western parasitic cones (P1, P7 and P8; Figure 2) are strongly dissected by trench-parallel NE-SW trending faults; this sector clearly formed before the main caldera-forming faulting events. A high density of caldera ring faults is associated with the MoC1 and MoC2 calderas. The existence of two relatively large calderas either requires continuous evolution or two successive stages of subsidence. The volcanically active MoV is the youngest part of the center, implying a relative younging from NE to SW, except for the resurgent cone (MoR), which postdates caldera formation (Figure 2).
Lavas from the upper north wall of MoC1 are basalts (6.3 wt% MgO), whereas those from the northeastern MoC2 (P3) rim are basaltic andesites. The latter are likely to have formed after the older Monowai volcano, but before the MoC1 caldera (P3 is cut by MoC1; Figure 2). The P3 basaltic andesites are slightly lower in Al2O3 than most of the other caldera samples (Figure 4) and may therefore represent a different eruption event. The presence of volcaniclastic material suggests phreatomagmatic eruptions along the MoC1 NE rim [Schwarz-Schampera et al., 2007], consistent with a higher vesicularity of the recovered samples. None of these areas show any signs of volcanic and/or hydrothermal activity [e.g., Embley et al., 2005]. They are therefore considered to be the next oldest formations within the Monowai volcanic center, as confirmed by thick sediment cover on the caldera floor and resurgent cone, and rough topography (including erosive structures) on the outer northern flanks of the caldera.
Lava from two ridge-like structures at the SW inner wall of MoC2 are basaltic andesites and andesites (SiO2 = 57–63 wt%). This region of the caldera is the most active in terms of heat flow and hydrothermal activity. Faults dissecting the western part of the caldera likely provide pathways for hydrothermal circulation. The existence of hydrothermal activity also indicates the presence of a heat source beneath the SW part of the caldera. Fractional crystallization processes to produce andesite require minimum stagnation times of several thousand years [e.g., Reagan et al., 2005], which supports the presence of a magma reservoir beneath this part of the caldera.
Within submarine intraoceanic arcs, including the Kermadec arc, large calderas are almost exclusively silicic [e.g., Worthington et al., 1999; Wright et al., 2003; Smith et al., 2010], and only few large mafic calderas (but all much smaller than MoC) are known (e.g., Haungaroa, Rumble II West; Wright et al. ). The formation of the ∼12 km wide, almost circular basaltic, Ambrym caldera in the New Hebrides arc has been attributed to the injection of hot, primitive, basaltic magma into more evolved magma, causing fracturing through uplift and subsequent inflow of water [Picard et al., 1994]. At Ambryn, the water-melt interaction resulted in a giant pheatomagmatic eruption that formed the caldera. Conversely, low loss of ignition (<1 wt%) and low glass contents (<7 vol%) of the oldest MoC basalts suggest predominantly effusive, rather nonexplosive eruptions of an early Monowai volcano. The presence of volcaniclastic material associated with basaltic andesites in the NE of MoC1 indicates progressively more explosive eruptions with time. Andesitic volcanism is confined to the SW part of the caldera and most likely represents the youngest volcanic activity of the older Monowai volcano. This early activity was followed by the eruption of a basaltic resurgent dome, which presumably formed via a late pulse of basaltic lava. The predominantly basaltic volcanic island of Fernandina, Galapagos, although about half the size of the Monowai caldera, also has a large outer caldera with a smaller caldera nested inside [e.g., Simkin and Howard, 1970; Howard, 2010]. The Fernandina caldera is characterized by outward dipping caldera rim faults, which have been interpreted to be a result of a deflating magma chamber [e.g., Lipman, 1997], similar to those shown by sandbox experiments [e.g., Roche et al., 2000]. The main collapse at Fernandina volcano has been attributed to drainage of magma from the magmatic system beneath the volcano through intrusion into other parts of the volcano [Simkin and Howard, 1970] or submarine flank eruptions [e.g., Geist et al., 2006]. At Monowai the existence of numerous outwards dipping ring faults suggests, similar to Fernandina, that the calderas formed as classic collapse structures most likely through a deflating magma chamber, possibly concurrent with the eruption of the Monowai stratovolcano. In addition, the Monowai calderas are elongated roughly perpendicular to the trench (NE-SW), which indicates that extension contemporaneously with trench-parallel faulting (Figure 2) also played a role in caldera formation. Extension may have weakened the volcanic structure and caused drainage of magma from the reservoir beneath the old Monowai stratovolcano and subsequent collapse of the gravitationally unstable roof rocks and the eruption of the young Monowai stratovolcano. A similar process has been proposed for the caldera formation at Miyakejima volcanic island, Japan [Nakada et al., 2005], where input of a pulse of basaltic magma into the basaltic andesite reservoir may have created overpressure, which caused lateral drainage of magma. What caused the shift of Monowai's volcanic activity toward the SW is uncertain, although it might have coincided with a NW-SE directed extensional event (resulting in the formation NE-SW fault traces) shutting off the old magma pathway and forming a new pathway to the surface. Such trench-perpendicular extension may be related to southward progressive rifting and opening of the significantly uplifted Havre Trough west of Monowai [Ruellan et al., 2003].
Predominately tectonic-controlled caldera formation is supported by the relatively minor occurrence of volcaniclastic material. This contrasts with the violent caldera-forming eruption proposed for formation of Ambryn, Raoul Island and Macauley calderas [Picard et al., 1994; Smith and Price, 2006]. As the most volcanically and hydrothermally active volcanic edifice within the northern Kermadec arc, the Monowai stratovolcano (MoV) shows smooth immature flank topography resulting from repeated recent coverage with pyroclastic and volcaniclastic material [Chadwick et al., 2008; Wright et al., 2008].
4.3. Mantle Wedge Processes
Monowai lavas show typical arc-like trace and minor element distributions with negative Nb (less pronounced Zr and Hf) anomalies, flat normalized REE contents and significant LILE enrichments (Figure 5), the latter indicating a significant fluid component derived from the subducting slab added to the mantle wedge. The similar flat REE distribution pattern of the MoV and MoC basalts (Figure 5) suggests a common source, as do similar incompatible trace element ratios (e.g., (Nb, Hf, REE)/Zr; La/(Nb,Sm); (Nb,Th)/Yb; (Th,Ba)/La), and the relatively restricted range of the Sr-Nd-Pb isotope compositions (Figure 6).
To quantify slab input into the mantle source beneath Monowai we focus on the lavas erupted behind the volcanic front in the Kermadec back arc. The slab beneath the southern Lau Basin–northern Havre Trough is ∼250 km deep (i.e., ∼200 km behind the volcanic front [e.g., Wiens et al., 2008], so the slab contribution to magma production there is small [e.g., Haase et al., 2002; Todd et al., 2010; Gamble et al., 1996]. It has been proposed that mantle corner flow drags the back-arc mantle beneath the volcanic front via convection [e.g., Kelemen et al., 1993; Tatsumi and Kogiso, 2003]. The back-arc mantle can therefore approximate an end-member composition prior to subduction zone associated metasomatism. The southern Lau Basin–northern Havre Trough back-arc lavas show slightly higher more to less fluid immobile incompatible element ratios (e.g., MREE/HREE, Nb/Zr, Zr/Yb, Hf/Y) than Monowai lavas, either indicating lower degrees of partial melting and/or a less depleted mantle source beneath the back arc than beneath the arc front. The back-arc lavas show comparable isotopic compositions to the SE Indian Ridge (SEIR; see Meyzen et al.  for details), and generally higher 206Pb/204Pb and lower 143Nd/144Nd than lavas of the East Pacific Rise (EPR; data from the Easter microplate are excluded because of possible influence of the Easter plume [e.g., Hanan, 1989; Fontignie, 1991]). A broad negative trend of the southern Lau Basin–Havre Trough Sr and Nd isotopic compositions (although all plot within MORB) and positive 206Pb/204Pb versus 207Pb/204Pb and 208Pb/204Pb trends may either indicate a heterogeneous composition of the back-arc mantle [Haase et al., 2002] and/or little (∼3%) influence of sedimentary hydrous melt from deep parts of the subducting slab, as suggested by Todd et al. . This is consistent with a broad positive trend of Th/(Yb, Nd) versus 207Pb/204Pb and 208Pb/204Pb (not shown) and (Ce/Yb)N. The mantle transported beneath the volcanic front may therefore already carry a small sedimentary component.
The strong negative Nb anomaly together with Zr, Hf and Ti depletions relative to REE in many subduction-related arc lavas have led several authors to invoke the presence of residual minerals, such as rutile (TiO2) and zircon (ZrSiO4), retained either in the subducting eclogitic slab [e.g., Foley et al., 2000; Rudnick et al., 2000; Rubatto and Hermann, 2003] or in the mantle wedge after melt-wall rock interaction [e.g., Kelemen et al., 1993]. To form rutile in the mantle source, high TiO2 content in the basaltic liquid is required (to reach rutile saturation under solidus conditions [Ryerson and Watson, 1987; Gaetani et al., 2008]), which is in excess of that observed in the respective arc lavas. Audétat and Keppler  and Antignano and Manning  obtained relatively low solubilities of TiO2 in aqueous fluids at pressures up to 0.7–2.3 GPa and temperatures between 700 and 1000°C (expected P range but higher T for a subducting slab, e.g., Schmidt and Poli , consistent with rutile and/or zircon remaining in the subducting eclogitic slab rather than forming a residual phase in the mantle melting region.
Based on more depleted HFSE in the New Britain arc front lavas than in the related Manus back-arc lavas it has been shown that strong HFSE depletion (including a negative Nb anomaly) can be generated through relatively high degrees of partial melting of a mantle depleted through previous extraction of the back-arc lavas [Woodhead et al., 1998]. This is irrespective of the presence of rutile and/or zircon in the subducting slab and/or in the mantle wedge above. In similar way, Monowai basalts are more depleted in HFSE than the MORB-type southern Lau Basin–northern Havre Trough back-arc lavas, suggesting HFSE depletion through (repeated) previous melt extraction. Assuming a spinel-peridotitic mantle source depleted by ∼5% previous melt extraction (to form the back-arc lavas), the HFSE distribution in the most primitive Monowai basalts can be modeled through 10%–20% partial melting (Figure 7; batch melting [Shaw, 1970]), which is in good agreement with the results of Woodhead et al.  and Haase et al. . We therefore favor the formation of the Monowai primary magmas through partial melting of a previously depleted spinel peridotitic mantle, rather than invoking accessory minerals in the mantle melting region to explain the trace element distribution patterns.
4.4. Along-Arc Geochemical Variation
Compositional variations in arc lavas are commonly attributed to (1) mantle source heterogeneity, (2) metasomatism of the mantle through hydrous fluids derived from the subducting slab (via fluids and/or melts, including altered oceanic crust and sediment cover), and in the case of the northern Kermadec–southern Tonga arc (3) subduction of the Louisville seamounts and their related volcaniclastics. To constrain the influence of these components to the mantle wedge beneath Monowai we focus on spatial geochemical variations in volcanic rocks from volcanic centers 250–400 km south of Monowai and ∼150 km north (i.e., north of 35°S to minimize possible geochemical variations in the arc volcanoes due to the subducting Hikurangi Plateau; Figure 1).
Basaltic lavas from Putoto and Raoul Island (Figure 1) have low incompatible element contents (e.g., Ti, K, P, Ba, Rb, Th, U, Pb, Zr, Hf, Y, Nb, REE) and more to less incompatible fluid immobile trace element ratios (e.g., (Sm, Nd)/Yb, Zr/Y, Zr/Hf, Hf/Yb) similar to Monowai, consistent with a depleted mantle source. Basaltic lavas from Ata Island and Macauley (Figure 1), however, have slightly higher incompatible element contents (e.g., Th, Pb, La, Ce, Sm) than Monowai, resulting in e.g., higher (Th, Pb, La, Ce, Sm)/Yb (Figure 8).
Compared with Pacific and Indian MORB and the southern Lau Basin–northern Havre Trough lavas, northern Kermadec arc basalts have similar Nd isotope ratios, but significantly more radiogenic Sr and Pb isotopic compositions. The latter are consistent with 1%–4% addition of a fluid-transported Sr and Pb sediment isotope signature (pelagic sediments [Turner et al., 1997; Gamble et al., 1993, 1996]) to the back-arc magma source.
The Nd and Pb isotopic differences in Monowai lavas may be due to variable composition of the either subducting Pacific lithosphere (including crust and sediment cover), resulting in variable influence on the composition of the overlying mantle wedge. Dredged lavas from the Cretaceous Pacific crust adjacent to the Tonga-Kermadec trench at 24°36′S and 27°57′S have more radiogenic Nd and less radiogenic Pb isotope compositions [Castillo et al., 2009] which, if subducted, cannot explain the Nd and Pb isotope signature of the Monowai lavas. The sedimentary cover has been drilled at DSDP Site 204 and consists of ∼103 m pelagic sediments, underlain by ≥47 m of plume-related Louisville volcaniclastic material (an apron of at least 420 km in diameter [Turner et al., 1997]). The pelagic sediment has significantly higher 87Sr/86Sr (0.706–0.709), 207Pb/204Pb (15.6–15.7), 208Pb/204Pb (38.5–39.0), similar 206Pb/204Pb (18.6–19.2) and lower 143Nd/144Nd (0.5124–0.5130) than the arc lavas, whereas the Louisville volcaniclastics have higher 206Pb/204Pb (19.0–19.2), 208Pb/204Pb (38.7–38.9) and 87Sr/86Sr (0.7065–0.7073), similar 207Pb/204Pb (15.5–15.6) and lower 143Nd/144Nd (0.51235–0.51299) (Figure 6). To date a Pb and Sr isotope Louisville signature has only been identified in lavas from Tafahi and Niuatoputapu at the northern end of the Tonga arc, but not in volcanic centers in the south [Regelous et al., 1997, 2010; Turner et al., 1997; Turner and Hawkesworth, 1997; Wendt et al., 1997; Ewart et al., 1998; Hergt and Woodhead, 2007]. Given the minimum radius of ∼210 km (distance between DSDP Site 204 and the closest Louisville seamount) the Louisville volcaniclastic apron may have reached the mantle source beneath Monowai. Although some input of fluid-transported volcaniclastic-derived Pb could explain the slightly elevated 206Pb/204Pb, it is inconsistent with the similar 87Sr/86Sr values of the Monowai lavas to those of the Kermadec volcanoes to the south, located outside the influence of the Louisville volcaniclastics, unless the Pb and Sr isotopic compositions are decoupled [e.g., Woodhead and Fraser, 1985; Ishizuka et al., 2003]. To assume a decoupling of Pb and Sr isotopic composition is reasonable, because Pb is much more fluid mobile (85%) than Sr (41%) [Kogiso et al., 1997]. During dehydration of the subducting plate it is therefore likely that more Pb than Sr is released and transported to the overlying mantle wedge via aqueous fluids and thus Pb may have a stronger effect on the mantle wedge composition.
The Sr-Nd-Pb isotopic composition of Monowai lavas could also be explained through minor influence of the fresh to variably altered Louisville seamount lavas, which have much less radiogenic 87Sr/86Sr (87Sr/86Sr = 0.7032–0.7043; 143Nd/144Nd = 0.51283–0.51295; 206Pb/204Pb = 19.14–19.61; 207Pb/204Pb = 15.57–15.63; 208Pb/204Pb = 38.68–39.29 [Cheng et al., 1987; Vanderkluysen, 2008]. Input of partial melts from subducted Louisville material into the mantle source region (and bulk mixing between Louisville material and the overlying mantle wedge) beneath Monowai would result in high LREE/(MREE, HREE), Zr/Y, Th/Yb [see Hawkins et al., 1987], which is not observed in the Monowai lavas.
However, ∼3% fluid transported Pb and Sr derived from subducted Louisville material to the already metasomatised mantle could explain the slightly more radiogenic 206Pb/204Pb and similar 87Sr/86Sr in the Monowai lavas compared with those from the volcanic centers to the south. The location of the Louisville seamounts, if projected beneath the arc front, is north of Monowai, so input of decoupled Pb and (±Sr) from the Louisville volcaniclastics seems more likely. Alternatively arc-parallel mantle and/or fluid flow, as inferred by arc-parallel shear wave splitting beneath the northern Tonga arc and Lau back arc [Conder and Wiens, 2007] may transport a Louisville lava signature southwards to the mantle source beneath Monowai, consistent with the geochemical and geophysical evidence found in the Central American arc [Hoernle et al., 2008]. The similarity of 208Pb/204Pb and more radiogenic nature of 207Pb/204Pb in lavas from Monowai and the southern volcanoes are consistent with its derivation predominately from a sedimentary source, which equally affects all volcanic rocks along the arc (Figure 6 and 9). Ata Island basalts have similarly high 206Pb/204Pb and (Pb, Th, LREE-MREE)/Yb consistent with minor Louisville input into its mantle source (Figures 6 and 9). A two-step evolution of the mantle source is envisaged: (1) sediment contribution, which has been active since the initiation of subduction in the late Oligocene and (2) the influence of the Louisville seamount chain, which occurred much later.
Pelagic sediments north of 35°S have less radiogenic Nd isotopic compositions (0.5124–0.5130 [Turner et al., 1997; Gamble et al., 1996]), which can account for the observed variation in 143Nd/144Nd values. On isotope correlation diagrams (Figure 6) all Monowai lavas plot between the back-arc mantle field and the pelagic sediment field. The transport of the less radiogenic Nd component to the mantle wedge can be either within hydrous sediment melts and/or hydrous fluids. High Sr/Y, together with high (LREE, MREE)/LREE and high Al2O3 contents, are commonly attributed to melting of subducted oceanic crust (eclogite) [Bindeman et al., 2005; Condie, 2005; Martin, 1999], or high pressure differentiation of ‘normal’ basalt [Macpherson et al., 2006]. Low Al2O3, Sr/Y (both <16, excluding samples that have experienced plagioclase accumulation) and flat REE patterns of the Monowai lavas, argue against the presence of garnet in the source (which would fractionate the HREE resulting in LREE enrichment), and also argue against an adakitic and/or tonalitic signature in these rocks [e.g., Hickey-Vargas, 2005]. The narrow range and unradiogenic nature of the Sr-Nd-Pb isotopes also argues against significant contamination by sediments, or remelted amphibolite (after reaction of crustal gabbro with seawater). In addition, partial melting of amphibolite would generate high Zr/Sm (>70; DSm > DZr [Foley et al., 2002]), which is not observed in the Monowai lavas (Zr/Sm <26). It has been suggested that Th can only be mobilised in hydrous melts and therefore represents a sensitive tracer for Th addition via sediment melt to the mantle source [e.g., Elliott et al., 1997; Turner et al., 1997]. Monowai lavas (and those to the south) show significantly lower Th/(HREE, Zr, Y) than pelagic sediments, which suggests insignificant input of Th into the underlying mantle wedge. Minimal sediment melt input is also consistent with low HFSE concentrations and low (La/Sm)N and (Sm/Yb)N in the Monowai lavas. High ratios of more to less fluid mobile element ratios, e.g., (U, Pb)/Yb, Ba/(Th, La), Sr/Nd, Rb/Zr in Monowai lavas, however, suggests a fluid dominated system beneath Monowai (Figure 7). Because MREE and LREE become increasingly fluid mobile with increasing pressure and temperature during the dehydration of the subducting slab [e.g., Pearce and Peate, 1995; Kogiso et al., 1997; Kessel et al., 2005], the variation in Nd isotope ratios may reflect a variation in fluid flux from the Pacific plate. The Monowai lavas are generally more depleted in REE than those to the south, which is inconsistent with mobilisation of LREE-MREE (including Nd) via an increased fluid flux. It is, though, consistent with a dilute nature of subsolidus aqueous fluids [e.g., Spandler et al., 2007; Plank et al., 2009]. The similarity in Nd isotopic composition (and generally similar fluid immobile incompatible element ratios, e.g., Th/Yb, (LREE, MREE)/HREE, (Zr, Hf)/Yb) in the Putoto and back-arc lavas suggests neither sediment nor Louisville-derived input into the mantle wedge beneath the northern Kermadec arc. Slightly higher Th/(Nd, Yb), (La/(Nd, Yb))N, ((Nd, Sm)/Yb)N, together with generally lower Nd isotope ratios (although similar to those from Monowai lavas) in mafic lavas from Raoul and Macauley suggests some sedimentary and/or crustal influence toward the south [e.g., Gamble et al., 1993; Smith and Price, 2006; Smith et al., 2009; Wysoczanski et al., 2006]. More sedimentary and/or crustal influence is consistent with location of the volcanic front on the Kermadec ridge between ∼32°S and ∼29°S (Figure 1), and eruption through thicker crust and subduction of a thicker sediment package on the Pacific plate and stronger sedimentary and/or crustal influence there, compared with the more rear-arc location and subduction of a relatively thinner sediment package beneath the northern Kermadec arc volcano lavas [Gamble et al., 1996; de Ronde et al., 2007; Wright et al., 2006; Graham et al., 2008].
Another large volcanic structure on the Pacific subducting plate 25–30 km NW of Monowai may also contribute to the observed isotopic variations in the Monowai lavas. The Osbourn Trough, a mid Cretaceous paleospreading center [Worthington et al., 2006] has a 10–20 km wide axial valley bounded by several inward dipping faults, typical of slow spreading ridges [e.g., Macdonald, 1998]. The floor of the axial valley is bounded by ridge mountains 200–600 m higher than the valley floor, forming steep sections exposed to marine weathering and serpentinization since ∼105 Ma [Worthington et al., 2006]. Subduction of the Osbourn Trough, together with Cretaceous altered oceanic crust, may therefore provide increased fluid flux into the mantle wedge beneath Monowai. Basaltic lavas dredged from the Osbourn Trough have higher 206Pb/204Pb (18.6–19.1) and 87Sr/86Sr (0.7045–0.7047), similar 208Pb/204Pb (37.9–38.5) and 207Pb/204Pb (15.55–15.59) and lower 143Nd/144Nd (0.51293–0.51302) ([Worthington et al., 2006] unleached data) than the Monowai lavas, which may reflect some minor influence of the Manihiki-Hikurangi Plateau separating plume-type source [e.g., Worthington et al., 2006; Hoernle et al., 2010; Timm et al., 2011]. A high fluid flux and/or variation in fluid composition derived from the highly serpentenized Osbourn Trough (but note their relatively unradiogenic Sr isotope composition), fails to explain the similar 87Sr/86Sr of the Monowai lavas and those to the south, which erupted well outside a possible influence from subducted Osbourn Trough. However, additional to the Louisville contribution a minor contribution of fluid-transported Pb from the subducted Osbourn Trough basalts to the overlying mantle wedge may also have contributed to the Pb isotopic composition of the Monowai lavas.
Because no geochemical evidence for contribution of sediments and/or altered oceanic crust to the mantle wedge can be identified in the Monowai lavas, the lower Nd isotopic compositions (when compared to the southern and northern volcanoes) most likely reflect preexisting isotopic mantle wedge source heterogeneities (including a sedimentary component in the back-arc lavas). We therefore speculate that a higher flux of very dilute (LILE-poor) aqueous fluids derived from highly serpentinized mantle of the altered oceanic crust and Osbourn Trough, which starts to dehydrate at slab depth of ∼120 Ma [e.g., Stern et al., 2006; Rüpke et al., 2004] may contribute to the high melt productivity beneath Monowai (Figure 8).
In summary, based on the available major and trace element and Sr-Nd-Pb isotope data the Monowai volcanic center was formed through relatively high degrees of partial melting of a previously depleted spinel-peridotite mantle source, with aqueous fluids being the predominant melting trigger. Sr and Pb isotopic variations can be explained through fluid transported input of a few percent of subducted sediments and/or Louisville seamount signature (±a Osbourn Trough signature). Variation in Nd isotopic composition relates to preexisting mantle heterogeneity. The subducting Osbourn Trough is presumably adding significant amounts of fluid, resulting in a different melting behavior and a continuous supply of basaltic magma. Extension-related closure of the existing magma pathway to the magma reservoir (of up to few thousand years) and drainage of magma has led to fractional crystallization and concomitant crystal accumulation beneath the early Monowai stratovolcano and ultimately to the collapse of the volcano and the subsequent formation of one of the largest mafic calderas on Earth. The occurrence of petrogenetically linked basaltic and basaltic to andesitic lava compositions within 10 km of each other, both associated with hydrothermal activity, suggest the existence of a single magma reservoir feeding two separate volcanic systems (Figure 10).
Trace element and isotope data combined with major element, petrographic and structural data [Graham et al., 2008] has enabled new insights into the formation of the predominately basaltic Monowai volcanic center, and the complex interplay of mantle components under the northern Kermadec arc. The Monowai volcanic center is composed of two basaltic-andesitic calderas (MoC1 and MoC2) and a basaltic stratovolcano (MoV), ∼12 km SW of the calderas. MoV is volcanically and hydrothermally active, and hydrothermal activity in also present along the SW inner caldera wall. The caldera andesites have formed through fractional crystallization of the most primitive (MgO-rich) MoV basalt, whereas the more silica- and MgO-poor MoV basalts are a result of predominantly plagioclase accumulation. In contrast to typical silicic caldera formation through phreatomagmatic eruption, the Monowai caldera most likely represents a collapse caldera, which was formed through trench-perpendicular extension and drainage of a magma reservoir, presumably during southward progressive opening of the Havre Trough. This extension may have cut the existing magma ascent, opening a new pathway beneath MoV.
Trace element modeling suggests that the most primitive Monowai magmas result from relatively high degrees of partial melting (12%–20%) of a previously depleted spinel-peridotite mantle [Haase et al., 2002] refertilized by input of fluid mobile elements. Sr-Nd-Pb isotope compositions range narrowly; Sr and Pb isotope ratios are more radiogenic and Nd isotope ratios similar to unmetasomatised southern Lau Basin–northern Havre Trough back-arc mantle, consistent with 1%–4% influx of pelagic sediment-derived aqueous fluid. Elevated 206Pb/204Pb ratios could be explained by a 3% fluid transported component of subducted Louisville basalt. Nd isotope ratios of Monowai lavas are similar to those from Putoto volcanic center to the south, suggesting mantle heterogeneity beneath the northern Kermadec arc. The additional influx of fluids from subduction of the Osbourn Trough could explain the maintenance of large-volume basaltic volcanism during the formation of the Monowai volcanic center.
We wish to acknowledge the contribution of major and trace element data from Ian Smith, University of Auckland, New Zealand. Roland Maas is thanked for laboratory assistance at Melbourne University, Australia. Discussions with C. J. N. Wilson, R. Sutherland, and R. Wysoczanski helped to develop ideas in this paper. We also would like to thank John Gamble and an anonymous reviewer for constructive and helpful reviews, which improved the paper. Joel Baker is thanked for editorial handling. This work is a part of the MWE program funded by FRST (Foundation of Research and Technology, New Zealand) contract C05X0406.