1 INTRODUCTION

The arc system is an active plate boundary where subduction-related magmatism occurs. However, the factors that control arc magmatism are still highly debated due in part to the complexity of convergent plates (Hawkesworth et al., 1993; Pearce and Peate, 1995; Stern, 2002; Elliott, 2003; Evans, 2012; Spandler and Pirard, 2013). Geochemical studies demonstrated distinct LILEs enrichment and HFSEs depletion in island arc lavas from the mid-ocean ridge basalts (MORBs). Addition of aqueous fluids/melt derived from subducting plates is considered to be responsible for these geochemical signatures (Johnson and Plank, 1999; Elliott, 2003). Other contributions, e.g. variably depleted mantle wedge melt and locally subducted seamounts or hotspots nearby, are also suggested to affect arc lavas (Turner et al., 2009; Timm et al., 2013; Price et al., 2016). Meanwhile, approaches on geophysical observations and computational models help to constrain the geometry and physical conditions of subduction process. Of particular essential is the thermal structure of subduction zones, which dominates the conditions for the generation of slab-derived fluids and arc volcanism (van Keken et al., 2002; Syracuse et al., 2010; van Keken et al., 2011). A wide range of physical parameters have been identified to influence the temperature distributions, including the age of the incoming lithosphere, shear heating, dig angle, and subduction rate (Peacock, 1990a, 1996; Zellmer et al., 2015). England and Katz, 2010 established a calculation formula to discriminate the scaling relations between these parameters and the temperature field within subduction zone.

Building a linkage between the various approaches is needed, because the degree and depth of dehydration of the subduction plates strongly dependent on the thermal structure of subduction zones (van Keken et al., 2011), thereby influence the melt productivity (Kelley et al., 2006) and arc magmatism. In this paper, we focus in particular on the role of the most obviously varying parameter, subduction rate, which increase from the south (~46 mm/a) to north (~83 mm/a) (Bevis et al., 1995), and explore the geochemical reflection of subduction rate in Tonga-Kermadec arc lavas.

2 GEOLOGICAL SETTING AND DATA SOURCE

The Tonga-Kermadec trench extends nearly linearly NNE from the north of New Zealand to the northern end of the Tongan Islands for some 2 800 km, delineates the collision zone between the Pacific and Indo-Australia plates (Fig. 1). A series of ridges and basins, including the Tonga-Kermadec ridge (frontal arc), the Lau-Havre trough (interarc basin), and the Lau-Colville ridge (third arc), are located to the west of the trench. The Louisville seamount chain intersects with the arc at 25°S, which constitutes the boundary between the Tonga (north) and Kermadec (south) arcs. A dozen of islands are distributed along the Tonga-Kermadec ridge, of which major islands from north to south include, e.g., Fonualei, Tofua, Hunga Ha'apai, Ata, Raoul Island Group, Macauley, and Healy. On the Kermadec ridge there are also a number of underwater volcanos, e.g., Monowai, Healy, Rumble Group.

The subduction of the Pacific plate downward to the Australian plate started after the early Eocene. The convergence rates vary linearly with latitude from 83 mm/a to the north of Tonga, which is a global maximum, to 46 mm/a at the south end of the Kermadec trench (Demets et al., 1990; Bevis et al., 1995). The dip of subduction zone changes from approximately 43°-45° beneath Tonga arc to 55°-60° beneath Kermadec arc (Pelletier and Louat, 1989). The age of the Pacific plate respectively increases from ~86 to 125 Ma north and south of the Osbourn trough, a paleo-spreading center (Sutherland and Hollis, 2001; Zhang and Li, 2016). The geochemical analyses of basalt samples drilled at IODP Site U1365 (Fig. 1) indicate that the subducting oceanic crust mainly consists of low-K tholeiitic N-MORB with significant low temperature alteration features (Zhang et al., 2012; Zhang and Smith-Duque, 2014). The overlying sediment portion decreases in thickness from south (≥500 m at Hikurangi Plateau) to north (~70 m at Site 596), and varies in composition from terrigenous source dominated to mainly pelagic materials (Carter et al., 1996; Plank and Langmuir, 1998). Tuffs and volcaniclastics from Louisville Seamount Chain also contributed to the sediment cover regionally (Regelous et al., 2010).

Lavas erupted from the Tonga arc are dominated by a series of basaltic andesites, with minor andesites and dacites, and sporadically developed felsic dacites. The Kermadec Islands are characterized by low-K tholeiitic basalts, basaltic andesites, dacites, and low-K rhyolites. Felsic pumiceous clasts occur in pyroclastic flow and deposits on Kermadec Islands (Falloon et al., 1987; Ewart et al., 1994, 1998; Smith and Price, 2006). Moreover, boninites appear at a volcano ~30 km near the Tonga arc island group of Hunga Ha'apai and the northernmost of Tongan ridge (Falloon and Crawford, 1991; Falloon et al., 2008; Cooper et al., 2010).

Tonga-Kermadec arc lavas have been extensively studied in the past few decades. Multiple components have been identified by previous geochemical investigations. The Tonga-Kermadec arc magma source is believed to be generated from the partial melting of the underlying mantle wedge, which is suggested to be increasingly depleted northward due to prior melt extraction in the back-arc (Woodhead et al., 1993; Ewart et al., 1998; Caulfield et al., 2008). Superimposed on the mantle wedge source is the aqueous fluid derived from the subducting Pacific plate by dehydration, which has been identified by high LILE/HFSE ratios (Turner and Hawkesworth, 1997; Ewart et al., 1998). These fluid signatures are noted to decrease southward along the arc, although they do not occur systematically from island to island. The geochemical tends were interpreted by the variations of component and amount of subducted materials (Regelous et al., 1997; Turner et al., 1997; Ewart et al., 1998; Castillo et al., 2009). Minor contributions of subducted local geochemical anomalies, e.g. seamounts, plateaus, and regional dominated terrigenous sources, were also added to the source of the arc lavas. Lavas from Tafahi and Niuatoputapu Islands on the northern edge of Tonga arc show evidences of either the influence of OIBmaterial from the Louisville Seamount Chain, or the plume-type mantle from Samoa plateau for their distinct high concentrations of Nb relative to other HFSEs (Wendt et al., 1997; Ewart et al., 1998; Tian et al., 2011; Haase et al., 2016). The more radiogenic Pb isotopes in lavas from volcanoes of southern Kermadec arc, including Rumble series, and Ngatoro ridge, reflect the addition of terrigenous sediments supplied by New Zealand (Ewart et al., 1998), and perhaps the subduction of Hikurangi Plateau (Timm et al., 2014, 2016). The influences of Louisville seamounts also occur in Monowai lavas evidenced by higher ²⁰⁶Pb/²⁰⁴Pb (Timm et al., 2011).

The major and trace element data used in this paper are collected from previously published studies (Ewart and Hawkesworth, 1987; Gamble et al., 1996; Turner et al., 1997, 2000, 2012; Ewart et al., 1998; Clift et al., 2001; Barker et al., 2013), and the samples used were collected from young volcanic eruptions (< 10 Ma). To exclude the influences from the adjacent plateaus and the subduction of seamounts, we eliminated data from Islands of Niua fo'ou, Tafahi, Niuatoputapu at the northernmost of Tonga Arc, and Rumble Ⅲ, Rumble Ⅳ and Ngatoro Ridge at the southernmost of Kermadec arc. We treat Tonga and Kermadec arcs as two separate populations and analyze the data as two groups.

A histogram of the collected lava data from TongaKermadec Islands shows that, the SiO₂ content forms a bimodal distribution consisting of two subpopulations and very few samples in the range of 58%-62%, which is previously observed by Smith et al., 2006 (Fig. 2). The lavas with SiO₂ content >60% have been suggested to reflect the mixing of a silicic melt component potentially generated from subducted crust melting (Smith et al., 2010). Therefore, we selected samples with SiO₂ content < 60% to better represent mantle-derived arc lavas source.

On primitive mantle normalized diagrams (Fig. 3), lavas from both Tongan and Kermadec Islands have typical features of LILEs enrichment and HFSEs depletion. Although Tongan and Kermadec samples have similar REEs patterns, integral differences do exist between the two segments. Kermadec samples show larger ranges of rare earth element (REEs) concentrations (Fig. 3a), while Tongan samples have lower averaged REEs concentrations (Fig. 3b). Moreover, Tongan lavas have obviously higher averaged concentrations of fluid mobile elements (e.g. Cs, Rb, Ba, U, K, Pb and Sr) and stronger depletion of incompatible fluid immobile elements (e.g. Nb, Ta, Zr, Hf and heavy REEs).

Differences between Tongan and Kermadec lavas are best illustrated by the ratios Ba/Th, U/Th, and Th/Nb (Fig. 4). The Tonga segment shows significant greater values of Ba/Th and U/Th ratios than that of Kermadec, although the averaged values do not increase from south to north continually. The values of Ce/Pb ratio in Tonga lavas are slightly lower than that of Kermadec. Tofua on the Tonga arc show the greatest range and highest values of Ba/Th and U/Th, while L'Esperance seamount at the southernmost of Kermadec segment appear to have the highest values of Th/Nb. The only exception is the Ata Island, which has lavas with lower Ba/Th, U/Th, and higher Ce/Pb, Th/Nb values than most other lavas from both segments.

Similar differentiations occur in TiO₂ concentrations. For Fig. 5a, the lavas of Kermadec segment show relatively greater range of TiO₂ and MgO, and higher TiO₂ concentrations for a given MgO content than lavas of the Tonga arc. When potted versus Ba/Th ratio, TiO₂ concentrations appear to be negative correlated. Samples of Tofua at Tonga arc with higher Ba/Th ratios tend to show lower TiO₂ values.

3 DISCUSSION 3.1 Form of subduction addition

Before we discuss the influences of thermal structures on arc lavas, it is crucial to figure out the forms of contribution of subduction materials, because the dehydration or melting of mineral phases are significantly constrained by temperature.

This issue has been previously investigated by abundant studies. Two chemically distinct components are defined to be derived from the subducting slab in arc lavas records. The first is a sediment delivered partial melt, which contributes a range of incompatible elements to the mantle wedge and is responsible for the enrichment of most REEs (Elliott et al., 1997; van Keken et al., 2002). The second is an aqueous fluid, which selectively transport the highly fluid-mobile elements (e.g. Rb, Ba, Pb), and is suggested to be delivered mainly from the altered oceanic crust rather than sediment (Morris et al., 1990; Clift et al., 2001), or from all available components (Hoogewerff et al., 1997; Haase et al., 2002). ¹⁰Be is a cosmogenic radioisotope that is highly enriched in marine sediments. High ¹⁰Be/⁹Be in arc lavas is treated as conclusive evidence for the addition of subducted sediments into arc magma sources (e.g., Morris et al., 1990). For Tonga-Kermadec Arc, pelagic sediment contribution has been recognized by more recent ¹⁰Be isotope data (George et al., 2005), and fluid addition were inferred to be delivered from slab dehydration by the record of U-Th isotope disequilibrium (Turner et al., 1997).

As explained by Pearce et al., 2005, Ratios of elements of similar incompatibilities, one mobile and one immobile, are useful to constrain subduction inputs. Experiments show that Ba and U are significantly partitioned into aqueous fluids derived from the subduction slab, while both Ba, U and Th are significantly partitioned into siliceous melts (Plank and Langmuir, 1993; Keppler, 1996; Johnson and Plank, 1999; Plank, 2005). In contrast, Nb is relatively immobile unless a rutile-bearing slab melts at the highest temperature (Brenan et al., 1994). Consequently, high Ba/Th and U/Th indicate the lower-temperature fluid addition, as both elements are mobilized in melts but only Ba and U are mobilized in fluids. In contrast, Th enrichment, and hence high Th/ Nb, indicates the higher temperature melt component (probably sediment melt, as suggested by Elliott et al., 1997).

For Fig. 4d-e, samples with higher Ba/Th or U/Th values tend to have lower Th/Nb, implying physically separate sources for Ba and U relative to Th, which support the ideas reviewed above. Except for Ata lavas, which shall be discussed in a later section, Tongan lavas show stronger slab-derived fluid signature (higher Ba/Th, U/Th). Some of the Kermadec lavas show the highest values of Th/Nb. It is thus indicated that the proportion of slab dehydration is greater under Tongan arc during subduction. The highest values of Th/Nb appear among Kermadec samples represent greater probability of high temperature melting of subduction component.

Pb is considered to be ~10 times more mobile than U during slab dehydration (Kelley et al., 2005), and is preferential enriched in hydrothermal fluids over Ce, therefore Ce/Pb is also a proxy of slab-derived fluids. Most of the subducted Pb is hosted by sulfides in oceanic crust during hydrothermal alteration, and metalliferous deposits in sediments (PeuckerEhrenbrink et al., 1994; Mühe et al., 1997). In Fig. 4f, the slightly lower Ce/Pb of Tonga lavas indicate more Pb addition from subduction materials, also imply stronger slab-derived fluids contribution.

3.2 Volume of subduction input fluxes

The most significant characteristic of TongaKermadec arc system is the gradually increasing rate of plate convergence and back-arc spreading from south to north. Faster subduction rates tend to supply more subducting materials per unit of time. The rate at which bound and free water enters the subduction zone is suggested to be approximately proportional to the convergence rate (Hyndman and Peacock, 2003). Consequently, the relationship of amounts of slab additions and geochemistry of arc lavas needs to be discussed.

Previous studies on mantle melting beneath backarc basins found that, positive correlations occur between the water concentration of the mantle and the extent of melting (Kelley et al., 2006). Titanium, like other HFSEs, tend to be retained in residual rutile in the slab during dehydration (Ryerson and Watson, 1987). It is thus approximately adopted that, the TiO₂ content of arc magma sources from a given mantle wedge are independent of the slab-derived component. Furthermore, TiO₂ decreases monotonically in the melt with increasing extent of melting of a single source, due to the incompatibility of titanium during mantle melting. Therefore, the variation of TiO₂ concentration in mafic arc lavas is useful to indicate the extent of melting, and shall be correlated with fluid signatures which represent water content of the mantle source.

The TiO₂ values of selected sample data are plotted versus MgO to eliminate the interference of fractional crystallization (Fig. 5a). Tongan lavas show lower TiO₂ at given MgO contents. Several studies argued that the low Ti and HFSEs might be generated by depleted mantle wedge reservoir which has been metamorphosed by strong back-arc melt extraction (Ewart et al., 1998). However, the Nb/Ta and La/Sm ratios, which represent mantle wedge depletion (Caulfield et al., 2008), show no obvious distinction between Tonga and Kermadec segments (Fig. 6). This contradiction may be interpreted by the southward migration of mantle wedge flow beneath Tonga-Kermadec Arc, as suggested by Turner and Hawkesworth, 1998. The diagram of Ba/Th versus TiO₂ shows negative correlation (Fig. 5b), which means greater fluid contributions tend to stimulate higher extent of partial melting in mantle wedge. It is indicated that (1) larger amount of subducted slab driven by faster subduction rates entered Tonga trench and contributed to the arc lavas, leading to stronger fluid signatures; (2) the lower HFSE concentrations in Tongan lavas are resulted from greater extents of melting rather than prior depletion of the source.

3.3 Thermal structure

The thermal structure of subduction zones has been of intriguing interest for a long time. Plenty of reasonable predictions have been achieved particularly from geochemical observations and numerical simulations. Syracuse et al. (2010) calculated the thermal model of Tonga and Kermadec arc segments under different partial coupling causes, obtained lower slab and Moho temperature, and higher maximum mantle temperature at deeper depth in Tonga segment beneath arc. This consequence is an important reference for the following discussion.

The highest values of Th/Nb ratios of Kermadec lavas represent greatest high temperature sediment melt addition. Turner et al. (1997) and Ewart et al. (1998) recognized the variation tendency of more radiogenic Pb isotope compositions and higher Th/U ratios in Kermadec lavas, and attributed it to the thicker sediment to the south. Indeed, the thickness of sediment entering the Tonga-Kermadec trench show increasing trend southward, which changes from ~150-300 m (Crawford et al., 2003) to < 200 m around Louisville Seamounts (Contreras-Reyes et al., 2011) and to ≥500 m at Hikurangi Plateau (Carter et al., 1996). However, there is no accurate continual measurement of sediment thickness from south to north. We thus make an assumption that sediment thickness varies continually and linearly. From Tofua to L'Esperance Island, the thickness of the subducted sediment may be 200 m to 250 m. As concluded in the last section, faster subduction rate shall introduce lager amount of subducting materials into mantle wedge sources. The calculated sediment fluxes per unit time for the two segments are almost the same. Therefore, the greater sediment signature in Kermadec lavas is better interpreted by warmer thermal structure, which lead to more sediment melt added to arc magma reservoir. This is consistent with the simulated results mentioned above.

An adminiculary evidence for the dominant role of thermal structure is the geochemical exception of Ata lavas. The Ata Island, located at the junction of Tonga and Kermadec segment, was expected to represent intermediate geochemical characteristics between slab dehydration characterized composition in the north and sediment melting characterized composition in the south. However, Ata lavas show high values of Th/Nb, low values of Ba/Th and U/Th similar with those of the Kermadecs, implying abnormal high temperature beneath Ata Island. While at the east of Ata, the adjacent Osbourn Trough, which is interpreted as a fossil spreading center, is subducting westward into the trench. The subduction of spreading center was suggested to generate extraordinary hot subduction zones (Hibbard et al., 1993; Peacock, 1996). What's more, in the model of Kirby et al. (1991), the slab temperature also depends on slab age, whereas older lithosphere may cool the mantle wedge more efficiently. Therefore, the geochemical anomalies of Ata lavas may be inferred as the result of Osbourn Trough or young hotter slab subduction, which lead to a thermal structure hot enough to induce sediment partial melting, similar to the consequences of slower subduction rates at south Kermadec.

The thermal structure of subduction zone could be affected by many factors, i.e. convergence rate, slab dip, slab age, and mantle thermal diffusivity (Kirby et al., 1991; England and Katz, 2010). There are different opinions on the thermal impacts of convergence rate. Someone suggested faster subduction rates tend to produce cooler subduction zones through introducing cold slabs (Peacock, 1990b). Others hold the idea that faster subduction rates may warm the subduction zones by inducing stronger corner flow (Conder and Wiens, 2007), hence more heat from the wedge is drawn to the slab interface (van Keken et al., 2002). However, for Tonga-Kermadec arc system, faster subduction rate is correlated to cooler thermal structure and less sedimentderived melt, which indicate a dominant role of the cooling effect of cold subduction slabs.

3.4 Modified subduction models

Based on the discussions above, we propose a modified contrast model for the Tonga and Kermadec arc segments (Fig. 7).

In Tonga segment, faster subduction rate introduced more slab into the subduction zone, and descent the temperature of mantle interface. Larger amounts of fluids derived from the slab through dehydration, with high concentrations of fluid-mobile elements (Ba, U, and Pb) were added to the mantle wedge. The more hydrated mantle wedge melt in higher extension, but the temperature was not high enough for massive sediment partial melting. Thus, the Tongan lavas show stronger fluid signatures rather than hightemperature melts.

In contrast, the thermal structure beneath Kermadec segment was less altered by descending slab. The higher temperature allowed subducted sediments to melt more efficiently, and transfer significant amount of Th, which is dominantly supplied by sediments, into the reservoir of arc lavas. Limited amounts of subduction slab entered the subduction zone and contributed less dehydrated fluids into the mantle wedge, thus the fluid signatures (Ba/Th and U/Th)are lower than those of Tonga segment.

4 CONCLUSION

We analyze the major and trace element compositions from the Tonga-Kermadec subduction system, which has an increasing subduction rate from south to north, to investigate the potential effects of subduction rate on arc lava composition. Geochemical differences between the Tonga and Kermadec segments are revealed based on the comparisons of the collected major and trace element data of lavas from 9 islands and underwater volcanoes on the Tonga-Kermadec arc. The Tonga arc lavas are more depleted in HFSEs, with lower TiO₂ content, Ce/Pb ratio, higher Ba/Th and U/Th ratios than those of Kermadec arc. Some of the Kermadec lavas show the highest values of Th/Nb. It is indicated that:

(1) The high Ba/Th and U/Th ratios are mainly attributed to fluid addition derived from slab dehydration, of which the latter dominated by altered oceanic crust. The high Th/Nb ratio component is mainly contributed from subducted sediments by high temperature melts.

(2) The lower TiO₂ content with higher Ba/Th ratio in Tonga lavas indicate a higher extent of melting in the underlying mantle wedge, caused by larger amount of fluid derived from the subducted slab, which is controlled by the subduction rate.

(3) The high values of Th/Nb in Kermadec lavas indicate warmer thermal structures under Kermadec segment, which is less altered by the cold slab.

We suggest that the northward increasing subduction rate is the primary cause to the stronger fluid signatures in northern arc lavas. The contributions from subducted slab are mainly characterized by dehydrated fluids at Tonga arc segment, which have cooler thermal structure due to the faster convergence rate, and by sediment partial melts at Kermadec segment with slower convergence rate.

5 DATA AVAILABILITY STATEMENT

The data analyzed during the current study are cited from these published article (Ewart and Hawkesworth, 1987; Gamble et al., 1996; Turner et al., 1997, 2000, 2012; Ewart et al., 1998; Clift et al., 2001; Barker et al., 2013) and the following public repository: http://www.earthchem.org/petdb.