Edinburgh Research Explorer Geodynamic controls on the contamination of Cenozoic arc magmas in the southern Central Andes: insights from the O and Hf isotopic composition of zircon

Subduction zones, such as the Andean convergent margin of South America, are sites of active continental growth and crustal recycling. The composition of arc magmas, and therefore new continental crust, reﬂects variable contributions from mantle, crustal and subducted reservoirs. Temporal (Ma) and spatial (km) variations in these contributions to southern Central Andean arc magmas are investigated in relation to the changing plate geometry and geodynamic setting of the southern Central Andes (28–32 (cid:2) S) during the Cenozoic. The in-situ analysis of O and Hf isotopes in zircon, from both intrusive (granitoids) and extrusive (basaltic andesites to rhyolites) Late Cretaceous – Late Miocene arc magmatic rocks, combined with high resolution U–Pb dating, demonstrates distinct across-arc variations. Mantle-like d 18 O (zircon) values (+5.4 & to +5.7 & (±0.4 (2 r ))) and juvenile initial e Hf (zircon) values (+8.3 (±0.8 (2 r )) to +10.0 (±0.9 (2 r ))), combined with a lack of zircon inheritance suggests that the Late Cretaceous ( (cid:2) 73 Ma) to Eocene ( (cid:2) 39 Ma) granitoids emplaced in the Principal Cordillera of Chile formed from mantle-derived melts with very limited interaction with continental crustal material, therefore representing a sustained period of upper crustal growth. Late Eocene ( (cid:2) 36 Ma) to Early Miocene ( (cid:2) 17 Ma) volcanic arc rocks present in the Frontal Cordillera have ‘mantle-like’ d 18 O (zircon) values (+4.8 & (±0.2 (2 r ) to +5.8 & (±0.5 (2 r ))), but less radiogenic initial e Hf (zircon) values (+1.0 (±1.1 (2 r )) to +4.0 (±0.6 (2 r ))) providing evidence for mixing of mantle-derived melts with the Late Paleozoic – Early Mesozoic basement (up to (cid:2) 20%). The assimilation of both Late Paleozoic – Early Mesozoic Andean crust and a Grenville-aged basement is required to produce the higher than ‘mantle-like’ d 18 O (zircon) values (+5.5 & (±0.6 (2 r ) to +7.2 & (±0.4 (2 r ))) and unradiogenic, initial e Hf (zircon) values ( (cid:3) 3.9 (±1.0 (2 r )) to +1.6 (±4.4 (2 r ))), obtained for the Late Oligocene ( (cid:2) 23 Ma) to Late Miocene ( (cid:2) 9 Ma) magmatic rocks located in the Argentinean Precordillera, and the Late Miocene ( (cid:2) 6 Ma) volcanic rocks present in the Frontal Cordillera. The observed isotopic variability demonstrates that the assimilation of pre-existing continental crust, which varies in both age

Geodynamic controls on the contamination of Cenozoic arc magmas in the southern Central Andes: Insights from the O and Hf isotopic composition of zircon

INTRODUCTION
Arc magmas, and hence new crust, potentially contain components derived from a number of different reservoirs (e.g., the mantle, subducted sediments, the subducting oceanic lithosphere, continental crust from subduction erosion, and the overlying crust). The proportions of these various contributions must be investigated and determined in order to quantify net crustal growth and the long-term recycling of crustal material. Resolving mantle and subduction components is more straight-forward in oceanic arcs than continental arcs because there is limited interaction with existing continental crust (e.g., Gill, 1981). The task is more challenging for arcs built on continental crust as the addition of subducting components is notoriously difficult to resolve from the contamination of magmas with the overlying crust (e.g., Davidson et al., 1991). However, as continental margins are considered the principal site for the growth of the continental crust, determining these contributions is key to understanding the generation, evolution and reworking of continental crust over time.
The Central Andes represents a type example of an ocean-continent convergent margin and is an active site of modern continental growth. Previous studies have identified an increased influence of crustal components on arc magmas from the southern Central Andes over the course of the Cenozoic on the basis of whole rock major and trace element geochemistry, and isotopic compositions (e.g., Kay et al., 1991;Litvak et al., 2007;Goss et al., 2013). However, discriminating between the processes involved (i.e., contamination of the mantle wedge with subducted components and/or crustal material from subduction erosion, versus arc magmas assimilating existing continental crust en route to the surface) remains unresolved.
The geodynamic setting of the Central Andean margin, specifically plate convergence rates and the angle of the subducting oceanic plate, as well as the thickness of the Central Andean continental crust, has changed significantly over the course of the Cenozoic (e.g., Pilger, 1981Pilger, , 1984Pardo Casas and Molnar, 1987;Kay et al., 1991Kay et al., , 2005Somoza, 1998;Yañ ez et al., 2001Yañ ez et al., , 2002Somoza and Ghidella, 2012;Goss et al., 2013). The Andean margin has also experienced high levels (50 km 3 /km/Myr (Stern, 2011)) of subduction erosion (Rutland, 1971;Stern, 1991) and variable sediment flux from the oceans and continents (Von Huene and Scholl, 1991a). This study utilises the isotopic composition of the robust accessory mineral zircon to investigate how the changing geodynamic setting, crustal thickness, sediment flux and levels of subduction erosion along the southern Central Andean margin has affected the contamination of southern Central Andean arc magmas, and thus the extent of crustal recycling over time.

GEODYNAMIC AND GEOLOGICAL SETTING
Subduction has been active along the western margin of South America since the earliest Jurassic, with subduction-related magmatism in the Central Andes initiated at $185 Ma (e.g., Pichowiak et al., 1990;Stern, 2004). The study area is located within the Pampean (Chilean) flat-slab segment ($28°-33°S) of the southern Central Andes (Fig. 1). This segment of the Andean margin has been volcanically inactive since the Late Miocene ($6 Ma) (e.g., Kay et al., 1999;Bissig et al., 2003;Litvak et al., 2007) due to a decrease in the angle (<10°a t 100 km depth) at which the oceanic Nazca plate subducts beneath the South American continent (e.g., Pilger, 1981Pilger, , 1984Yañ ez et al., 2001). The present day, low angle of subduction has been attributed to the subduction of the Juan Fernandez Ridge (JFR) which began intersecting the Andean margin during the early Miocene ($18 Ma) (e.g., Yañ ez et al., 2001;Jones et al., 2014). The shallowing of the Nazca plate caused the position of the volcanic arc to expand and migrate to the east (e.g., Kay et al., 1987;Kay and Mpodozis, 2002). Convergence rates and the relative plate motions between the oceanic (Farallon and Nazca) and South American plates have also changed during the course of the Cenozoic; in particular a major plate reconfiguration occurred in the Late Oligocene ($25 Ma) due to the break-up of the Farallon plate into the Nazca and Cocos plates (Lonsdale, 2005). This resulted in an increase in convergence rates (from $8 cm/yr to $15 cm/yr) between the Nazca and South American plates and a change from oblique (NE-SW) to orthogonal (ENE-WSW) convergence (Pardo Casas and Molnar, 1987;Somoza, 1998;Somoza and Ghidella, 2012). This increased convergence, combined with increased compression related to the subduction of the JFR, led to an increase in crustal thickness in the southern Central Andes from $30 km to >45 km (Kay et al., 1991;Haschke et al., 2002).
The Andean tectonic cycle (Jurassic to recent) represents only the latest in a series of orogenic cycles to affect the western edge of the South American continent. Earlier tectonic cycles (e.g., the Gondwanan tectonic cycle, latest Devonian to early Permian) along with episodes of extension, subduction and accretion formed the distinctive basement on which the modern Andes are built (Ramos et al., 1986;Charrier et al., 2007). In the Pampean flat-slab segment, the Andean margin is divided from west to east into the Coastal Cordillera, the Principal Cordillera, the Frontal Cordillera, the Precordillera and the Sierras Pampeanas (Fig. 1). The age and origin of the underlying basement varies across the Andean Cordillera depending on the extent of these north-south trending morphostructural units. The basement of the Principal and Frontal Cordillera is composed of highly deformed Paleozoic sediments which have been intruded and covered by extensive Late Palaeozoic -Early Mesozoic plutons and volcanic rocks (Kay et al., 1989). In the Argentinean Precordillera ( Fig. 1) a Grenville-aged basement (1200-1000 Ma), which is not exposed at the surface and has been identified through various geochemical studies, is overlain by Cambrian -Ordovician strata Thomas et al., 2004;Ramos, 2010). As a result of both the regional geology and the migration of the volcanic arc to the east there is a unique opportunity for a distinctive spatial relationship in the contamination of mantle-derived magmas, via assimilation and crustal reworking en route to the surface, to be revealed.

INSIGHTS FROM THE ISOTOPIC COMPOSITION OF ZIRCON
In continental arcs, the relative contributions to arc magmas, from mantle, crustal and subducted components can be difficult to constrain from whole rock geochemistry alone as it often reflects a complex history of fractional crystallisation and late stage alteration (e.g., McCarthy and Hasty, 1976;Marfil and Maiza, 2012). This is a particular problem in the southern Central Andes where some of the arc magmatic rocks (e.g., the Tilito Formation) have been affected by hydrothermal alteration (Bissig et al., 2001(Bissig et al., , 2003Kay and Mpodozis, 2001;Charchaflié et al., 2007;Litvak et al., 2007). A solution is to conduct high resolution, in-situ oxygen and hafnium isotopic analysis of the robust accessory mineral zircon and combine this with high resolution U-Pb dating and detailed cathodoluminescence imaging. Zircon grains also potentially preserve information about the nature of the contaminant (e.g., through the presence of xenocrystic cores), which are not apparent from whole rock geochemistry. The analysis of individual zircon grains, which unlike whole rocks can be readily dated and reliably analysed for d 18 O (robust to recent alteration), by in-situ techniques, enables a much more detailed picture of the relative contributions to arc magmas to be achieved.
Specifically, the combination of oxygen and hafnium isotopic variations in zircon can be used to investigate the interactions between mantle-derived melts and older crustal material (e.g., Hawkesworth and Kemp, 2006;Kemp et al., 2007). Assuming zircon crystallisation is cogenetic (i.e., the zircon has not been inherited from elsewhere), zircon will retain the oxygen isotope composition of the melt from the time of crystallisation (Peck et al., 2003). Zircons which crystallise in equilibrium with melts derived from an uncontaminated mantle source have a narrow range of d 18 O values (5.3 ± 0.6& (Valley et al., 1998;Valley, 2003)). As oxygen isotopes are sensitive to water-rock interactions and the fractionation of oxygen isotopes increases with decreasing temperature (Clayton et al., 1972), the interaction of magmas with sediments and/or low temperature altered crust leads to elevated d 18 O (zircon) values. Thus, higher than 'mantle like' d 18 O (zircon) values suggests the incorporation of a 18 O enriched supracrustal component, such as sediment, into the magma from which the zircon crystallised (e.g., Eiler, 2001).
Hafnium (Hf) is more incompatible than lutetium (Lu) during the melting of spinel and garnet peridotite. Thus crust is formed with low Lu/Hf and with time the radioactive decay of 176 Lu to 176 Hf results in the crust evolving with a less radiogenic 176 Hf/ 177 Hf composition relative to the mantle from which it separated (Patchett et al., 1982;Kinny and Maas, 2003). Therefore, the mixing of new mantle-derived melts with older, and hence less radiogenic crust (e.g., a Grenville-aged basement), will lead to lower epsilon Hf values (i.e., the 176 Hf/ 177 Hf expressed in parts per ten thousand difference relative to a chondritic reservoir) in the zircon crystallising from the magma. In addition to this, where there is no evidence for mixing with older crustal material (i.e., high initial eHf (zircon) values, 'mantle-like' d 18 O (zircon) values and no zircon inheritance) Hf model ages can be used to identify when a primary melt differentiated from the primitive mantle (e.g., Hawkesworth and Kemp, 2006). Combining zircon U-Pb crystallisation ages with Hf model ages therefore has the potential to reveal periods of new crustal growth (e.g., Belousova et al., 2010;Dhuime et al., 2011).
We have applied these techniques to zircons separated from representative samples of both intrusive (diorites to granites) and extrusive (basaltic andesites to rhyolites) Late Cretaceous -Late Miocene arc rocks. These samples were collected from an W-E transect across the Andean Cordillera from Chile into Argentina, between 29 and 31°S ( Fig. 1 and Table 1). All the major plutonic and volcanic formations, which had previously been mapped as between Late Cretaceous -Late Miocene in age (Maksaev et al., 1984;Mpodozis and Cornejo, 1988;Nasi et al., 1990;Cardó and Díaz, 1999;Emparan and Pineda, 1999;José Frutos et al., 2004;Pineda and Emparan, 2006;Cardó et al., 2007;Pineda and Calderó n, 2008), were sampled from this region. The arc rocks sampled from the Principal Cordillera of Chile, which are primarily plutonic and belong to the Cogotí Supergroup, range in age from the Late Cretaceous (72.6 ± 0.8 Ma) to the Eocene (38.9 ± 1.0 Ma), as determined by the U-Pb dating conducted as part of this study (Table 1). The samples collected from the Frontal Cordillera range in age from the Early Paleocene (61.9 ± 9.1 Ma) to the Late Miocene (6.2 ± 0.3 Ma) and are primarily extrusive, and the plutonic and volcanic rock samples collected from the Precordillera of Argentina (i.e., furthest east) range in age from the Late Oligocene (22.6 ± 0.3 Ma) to the Late Miocene (9.4 ± 0.2 Ma) ( Table 1). Zircons were also separated from a number of plutonic and volcanic rocks from the Late Palaeozoic -Early Mesozoic basement, which were collected in order to characterise potential contaminants of the Late Cretaceous -Late Miocene arc magmas. These samples range in age from 280.2 ± 3.5 Ma to 221.0 ± 4.4 Ma (Table 1).
Detailed petrographic analysis and cathodoluminescence imaging ( Fig. 2) of individual zircon grains was conducted in order to identify the magmatic origin of the zircon grains, and suitable grains and specific locations for isotopic analysis (refer to Section 2 of the Appendix). This included the identification of magmatic zircon grains with simple histories and those with distinct xenocrystic cores which may reveal the nature and age of the contaminating crustal material (Fig. 2). Oxygen isotope compositions and U-Pb crystallisation ages were determined from the same zircon growth zones using a Cameca ims 1270 secondary ion mass spectrometer (SIMS) at the NERC Ion Microprobe Facility, University of Edinburgh ( Fig. 2 and Appendix). Subsequently, hafnium isotope analysis was conducted on a multi-collector inductively-coupled plasma mass spectrometer combined with a laser ablation system (LA-MC-ICPMS), at the University of Bristol. Full details of analytical methods and data correction are presented in the Appendix. Table 1. Full results are presented in Figs. 3 and 4, and Table A1 and A2 of the Appendix.

Oxygen isotopes
No correlation is apparent between the d 18 O (zircon) values and the whole rock SiO 2 content as shown in Fig. 3a. The d 18 O values obtained for individual zircon grains from the Late Cretaceous (72.6 Ma) to Eocene (38.9 Ma) plutonic rocks located in the Principal Cordillera of Chile, range between 4.96& (±0.33 (2r)) and 6.13& (±0.33 (2r)) ( Fig. 4). These values are all within analytical uncertainty of those expected for zircon crystallising in isotopic equilibrium with mantle-derived melts (5.3& ± 0.6 (Valley et al., 1998)). With the exception of the most easterly sample of the Tilito Formation (PC14), the majority of d 18 O (zircon) values obtained for the Early Paleocene (61.9 Ma) to Late Miocene (6.2 Ma) arc rocks located in Table 1 Summary sample information. The reported U-Pb ages (Ma) are sample ages as given by Concordia diagrams and Terra-Wasserberg plots (Jones, 2014). The whole rock major element compositions were determined by X-ray florescence spectrometry (XRF). Whole rock data for sample DI095 is from Litvak et al. (2007).

Hafnium isotopes
Initial eHf (zircon) values obtained for individual zircon grains from the plutonic rocks located in the Principal Cordillera, are distinctly positive and range between +7.7 (±1.0 (2r)) and +10.8 (±0.8 (2r)) (Fig. 4). These values lie close to the values projected for mantle-derived melts (i.e., new crust (NC) (Dhuime et al., 2011)), and yield hafnium model ages (T(NC)) ranging between 190 and 416 Ma ( Fig. 5 and Table A1, Appendix). With the exception of the Paleocene (61.9 Ma) sample of the Los Elquinos Formation, which has eHf T(zircon) values within the range of values obtained for samples of similar age located in the Principal Cordillera (Fig. 4), the initial eHf (zircon) values of the Late Eocene (35.6 Ma) -Late Miocene (6.2 Ma) volcanic rocks present in the Frontal Cordillera range between À3.4 (±0.9 (2r)) and +4.8 (±1.0 (2r)), with the lowest values obtained for the youngest sample (DI095, Vacas Heladas Ignimbrites) (Fig. 4). The initial eHf (zircon) values produced for individual zircon grains from the Precordillera are generally lower than those obtained for the Principal and Fontal Cordillera and range between À3.9 (±0.6 (2r)) and +4.1 (±0.8 (2r)) (Figs. 4 and 5). The initial eHf (zircon) values obtained from sample PC14, the most easterly sample of the Tilito Formation (23.6 Ma), lie within this range and are lower than the eHf T(zircon) values obtained for other samples of the Tilito Formation located in the Frontal Cordillera (Fig. 4). Analyses of zircons obtained from plutonic and volcanic rocks comprising the Late Palaeozoic to Early Mesozoic basement produced initial eHf (zircon) values ranging between À2.2 (±1.0 (2r)) and +3.0 (±1.0 (2r)) ( Fig. 5 and Table A1 of the Appendix). No correlation is observed between the initial eHf (zircon) values and the whole rock SiO 2 content (Fig. 3b).

Temporal and spatial variations in isotopic compositions
The isotopic compositions of zircons, and the occurrence and age of inherited, xenocrystic zircon cores, vary as a function of sample age and across-arc position (i.e., location in the Chilean Principal Cordillera (most westerly), Frontal Cordillera or Argentinean Precordillera (most Fig. 2. Cathodoluminescence (CL) images of select zircon grains highlighting the presence of inherited cores (outlined), internal growth zoning, and the locations of in-situ isotopic analysis. The relative areas analysed by the different techniques (SIMS and LA-MC-ICPMS) are also indicated. The oxygen isotope ratios are expressed in d 18 O notation (&) relative to VSMOW and the hafnium isotope values are expressed as initial eHf values (calculated using the 206 Pb/ 238 U age obtained by SIMS). The U-Pb ages are presented as 206 Pb/ 238 U ages for the individual zircon grains and the errors are quoted at the 2r level. easterly)), as shown in Fig. 4. Due to the eastward migration of the volcanic arc in the Pampean flat-slab segment, during the Cenozoic, the across arc position and sample age are also related.

Late Cretaceous -Eocene mantle derived melts
The 'mantle-like' d 18 O (zircon) values obtained for the Late Cretaceous to Eocene granitoids located in the Principal Cordillera, combined with the distinctly positive, initial eHf (zircon) values, which lie close to the values projected for mantle-derived melts (Figs. 4 and 5), suggests the Late Cretaceous to Eocene arc magmas experienced very little contamination from significantly older, upper continental crust, either in the melt source region or via crustal assimilation. This interpretation is consistent with the low strontium isotope values (initial 87 Sr/ 86 Sr ratios of between 0.7035 and 0.7045) obtained for these granitoid belts (Parada et al., 1988;Parada, 1990). The absence of any significantly older, xenocrystic zircon cores (Fig. 4) provides further evidence to indicate limited interaction of these arc magmas with old continental crust. During this time interval the southern Central Andean margin is thought to have been more extensional due to the highly oblique angle of convergence between the Farallon and South American plates and low rates of convergence (e.g., Pardo Casas and Molnar, 1987;Charrier et al., 2007). Therefore, a more limited interaction between the ascending mantle-derived melts and the overlying Andean crust might be expected in comparison to later in the Cenozoic, when the continental crust became thicker due to increased compression (Kay et al., 1991).
Devonian to Triassic hafnium models ages were obtained for zircons from the Late Cretaceous -Eocene magmatic belts (Fig. 5). During this time interval the western margin of South America was located along the western margin of Gondwana and formed part of the Gondwanan magmatic arc (e.g., Kay et al., 1989;Charrier et al., 2007). On the basis of the Hf model ages, combined with the oxygen and hafnium isotope data, and whole rock geochemistry (Jones, 2014), it is suggested that the Late Cretaceous to Eocene arc magmas were derived from the asthenospheric mantle wedge with a very minor addition from Paleozoic -Early Mesozoic crust. This crustal addition could be derived from the Gondwanan magmatic arc, either due to subduction erosion or the subduction of terrigenous sediment from the arc, and/or a lower crustal reservoir which separated from the mantle during the Gondwanan tectonic cycle. An addition of older crustal material is required to account for initial eHf (zircon) values which are slightly lower (i.e., less radiogenic) than those expected for melts derived directly from the mantle (Fig. 5). Sediments have high d 18 O values, reflecting low temperature processes, and therefore the addition of sediment to the melt source region might be expected to raise the d 18 O (zircon) values. Bulk mixing models suggest that a maximum addition of 4% Late Paleozoic -Early Mesozoic sediments, derived from the Andean arc, to the source of the Late Cretaceous to Eocene arc magmas could account for their initial eHf (zircon) values, whilst retaining 'mantle-like' d 18 O (zircon) values (Fig. 6). A lower crustal reservoir which separated from the mantle during the Gondwanan tectonic cycle is also likely to have 'mantle-like' d 18 O values and therefore a much greater addition from such a reservoir could account for the isotopic composition of the Late Cretaceous to Eocene arc magmas. Whichever mechanism is responsible, the evidence presented by this study suggests that during the early stages of Andean-type subduction (Late Cretaceous -Eocene) there was a sustained period ($35 Ma) of upper crustal growth, with extensive magmatic belts formed from melts derived from the mantle with little recycling of pre-existing crustal material.

Controls on Late Eocene -Late Miocene isotopic variability
The contamination of arc magmas with older crustal material can occur either during magma ascent through the crust via crustal assimilation or via the subduction of   (zircon) , eHf T(zircon) (of rims or zircons without inherited cores) and inherited core 206 Pb/ 238 U ages. Error bars represent 2r values. The isotopic data is displayed relative to a schematic cross section of the modern day Andean margin at 30°S showing the extent of the different basement terranes (Ramos et al., 2002;Gilbert et al., 2006;Heit et al., 2008). The locations of the Principal Cordillera, Frontal Cordillera, Precordillera and the Chile/Argentina border are also shown. Full results are presented in Table A1 of the Appendix. upper crustal material (including that derived from subduction erosion) to depth. If the melt source was being contaminated, little intra-and inter-grain variation in isotopic values for samples of similar ages/from the same formations might be expected, compared to the variation which might be generated by a crystallising magma progressively assimilating continental crust (e.g., as demonstrated by core to rim changes). Temporal variations in d 18 O (zircon) and eHf (zircon) values might also be anticipated, reflecting the subduction of different amounts of sediment and/or continental crust derived from subduction erosion, over time. Subduction erosion has been identified as an episodic process along the Central Andean margin (Kay et al., 2005;Stern, 2011), and the JFR is thought to have acted as a barrier to the transport of sediments into the trench from farther south (e.g., Bangs and Cande, 1997), after its intersection with the margin at $20 Ma (Yañ ez et al., 2001(Yañ ez et al., , 2002. These changing dynamics might be expected to have affected the composition of the melt source region over time, by either starving or adding sediment/crustal material to the mantle wedge. However, the variability in d 18 O (zircon) and eHf (zircon) values obtained for Late Eocene to Late Miocene arc magmatic rocks appears to be related to geographic position relative to the Chilean margin, rather than sample age (Fig. 4). For example, sample PC14 (23.6 Ma) is contemporaneous with the Tilito Formation (Late Oligocene) and is located close to the boundary with the Precordillera. It displays distinctly different d 18 O (zircon) and eHf T(zircon) values compared to samples of the Tilito Formation and of the same age located in the Frontal Cordillera (Fig. 4). Furthermore, the youngest samples (Vacas Heladas Ignimbrites), are located in the High Andes of the Frontal Cordillera and have d 18 O (zircon) values which lie within the range of values obtained for other, older samples present in the Frontal Cordillera Fig. 5. Initial eHf (zircon) values plotted against crystallisation ages ( 206 Pb/ 238 U) (Ma) for all zircon grains and inherited cores. The data highlighted in the red box is expanded to the bottom right. Depleted mantle (DM) and new crust (NC) evolution lines are shown along with model ages for the Late Cretaceous -Eocene samples and the Proterozoic inherited cores from the Tertiary Intrusives. The initial eHf (zircon) values are presented relative to present day chondritic values (CHUR) (Bouvier et al., 2008). Error bars represent 2r values. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) (Fig. 4). Intra-and inter-grain variability, particularly in eHf (zircon) values (Fig. 4), is also displayed in a number of Late Eocene -Late Miocene samples, which suggests the isotopic composition of the magma was evolving during zircon  (Dhuime et al., 2011)) and average Late Paleozoic -Early Mesozoic sediments is shown in a. Bulk mixing between mantle-derived melts and the Permian -Triassic basement (using a Hf NC / Hf ACC of 0.07) is shown to generate the observed isotopic variability in the Early Paleocene -Late Miocene samples from the Frontal Cordillera (b). Up to 15% bulk mixing between mantle-derived melts and average sediment (AS) (Savin and Epstein, 1970;Plank and Langmuir, 1998) (using a Hf NC /Hf AS of 0.05) explains the isotopic variability observed in the Late Oligocene -Late Miocene samples from the Argentinean Precordillera (c). Average sediment values are used to represent the basement present in the Precordillera which is suggested to be of Grenville-age and to have a sedimentary component (i.e., high d 18 O values) (e.g., Thomas and Astini, 2003). The shaded areas represent mixing between mantle-derived melts and the upper and lower values. Error bars represent 2r values. crystallisation and was not solely determined in the melt source region.
The basement underlying the Frontal Cordillera is composed of highly deformed Paleozoic sediments which have been intruded and covered by extensive Late Palaeozoic -Early Mesozoic plutons and volcanic rocks (e.g., Kay et al., 1989). The initial eHf (zircon) values obtained for the Late Eocene to Late Miocene volcanic rocks present in the Frontal Cordillera are significantly lower (i.e., less radiogenic), than those expected for a primary melt derived from the mantle wedge (i.e., new crust (NC)) (Fig. 5) (Table A2). Bulk mixing models using the average d 18 O (zircon) values obtained for the Permian -Triassic basement samples and the projected Permian -Triassic eHf (zircon) values (Fig. 5), suggests mixing of typically <20% of this crust with new mantle-derived melts (i.e., new crust (NC) (Dhuime et al., 2011)) can produce the modest changes d 18 O, but more marked Hf isotopic variability observed in the Late Eocene to Late Miocene samples from the Frontal Cordillera (Fig. 6). The presence of primarily Permian -Triassic inherited zircon cores in these samples (Fig. 4) supports the assimilation of continental crust of this age.
The Late Oligocene -Late Miocene granitoids, andesites and rhyolites, located in the Precordillera of Argentina have higher than 'mantle-like' d 18 O (zircon) values and un-radiogenic values of eHf T(zircon) (Figs. 4 and 5), which are quite distinct from the values obtained in the Principal and Frontal Cordillera. This suggests the interaction of these arc magmas with an ancient crustal reservoir (therefore with low eHf T values) which has experienced low temperature interaction with the near surface, accounting for the elevated d 18 O values. Although no crystalline basement is exposed in the Precordillera, evidence from high-grade metamorphic, xenoliths present in the Miocene volcanic rocks indicates a Grenville-aged basement (Abbruzzi et al., 1993;. It has been suggested that this basement forms part of a rifted fragment of Laurentian crust and passive margin cover which was accreted onto the margin of Gondwana during the Ordovician (e.g., Thomas and Astini, 2003). Convincing evidence for the contamination of the Late Oligocene -Late Miocene magmas with this Grenville-aged crust is provided by the presence of inherited Proterozoic cores in a number of the samples present in the Precordillera (Fig. 4). This contamination must have happened via crustal assimilation, rather than the recycling of sediments through the subduction zone, as Proterozoic crust is not exposed at the surface in the main Andean Cordillera and therefore is unlikely to have been eroded and transported into the Chilean trench.
A Grenville-aged basement is likely to have unradiogenic, negative eHf values, as shown by projecting the eHf (zircon) values obtained for the Proterozoic cores to the Cenozoic (Fig. 5), and high d 18 O values reflecting a sedimentary origin. Bulk mixing suggests that contamination of mantle-derived melts with up to 15% Proterozoic aged, sedimentary crust can account for the observed isotopic array displayed in zircons from the Late Oligocene -Late Miocene arc rocks of the Argentinian Precordillera (Fig. 6). This corroborates the work of , who suggest that mixing between melts derived from the mantle wedge and a Grenville-aged basement accounts for the distinct geochemistry of the Miocene arc magmatic rocks present in the Argentinean Precordillera. The presence of inherited cores of Permian to Cretaceous age in these Late Oligocene -Late Miocene arc rocks from the westernmost extent of the Precordillera (Fig. 4) suggests that the mantle-derived melts also assimilated Late Paleozoic -Mesozoic crust en route to the surface. Hence the O and Hf isotopic values are more likely to reflect three way mixing processes between mantle-derived melts, a Grenville-aged basement and Late Paleozoic -Mesozoic Andean crust.
The lower eHf T(zircon) values obtained for the youngest volcanic arc samples in the study area, the Vacas Heladas Ignimbrites ($6 Ma) located in the Frontal Cordillera, suggests that these arc magmas may also have interacted with an ancient portion of Andean continental crust in addition to the Late Paleozoic -Early Mesozoic crust. Structural models suggest that the Grenville-aged basement currently underthrusts the Frontal Cordillera (e.g., Ramos et al., 2004;Gilbert et al., 2006;Gans et al., 2011). Consequently the eHf T(zircon) values obtained for the Vacas Heladas Ignimbrites, which overlap those obtained for magmatic rocks present in the Precordillera, may reflect the arrival of the Grenville-aged basement under this region of the Frontal Cordillera due to increased compression and crustal shortening over the later part of the Miocene (e.g., Allmendinger et al., 1990;Kley and Monaldi, 1998;Ramos et al., 2002 and references therein). A similar argument has been made to explain evolving eNd values to the south of the study region in the southern volcanic zone (SVZ) (e.g., Muñ oz et al., 2013). The development of enriched eNd values is also observed in the Pampean flat slab segment; a eNd value of À2.0 has been obtained for the Late Miocene, Vacas Heladas Ignimbrites compared to eNd values of between +1.2 and À0.1 for the Late Oligocene, Tilito Formation (Kay et al., 1991). This provides supporting evidence for the contamination of the Late Miocene arc magmas erupted in the Frontal Cordillera with the Grenville-aged basement.
During the Late Eocene to Late Miocene the products of subduction erosion and sediments present in the Chilean trench may well have entered and influenced the melt source region in the asthenospheric mantle wedge, as is suggested by high rates of subduction erosion (e.g., Stern, 2011) and exhumation along the increasingly compressional Central Andean margin (e.g., Kurtz et al., 1997;Maksaev et al., 2003;Spikings et al., 2008). However, any effect which this subducted continental material had on the isotopic composition of the arc magmas in the melt source region appears to have been overprinted by the assimilation of the overlying crust en route to the surface, as is evidenced by the across-arc O and Hf isotopic variations which reflect the distinct basement terranes. Subduction erosion could potentially move continental crust laterally, however, this is still unlikely to generate the observed across-arc isotopic array that correlates so well with the composition of the Andean basement.

Punctuated versus steady state crustal evolution -an unbiased record
Subduction zones are considered to be sites of continental growth (e.g., Rudnick, 1995). Net growth, however, may be punctuated by periods of crustal loss due to tectonic erosion (e.g., von Huene and Scholl, 1991b;von Huene, 2007, 2009), as is evident along the Andean margin where mass flux, and hence rates of crustal recycling have varied over time (e.g., Kay et al., 2005). It has been proposed that the Andean margin has lost $250 km of crust over the past 150 Ma, with 30 km of trench retreat at 33°S in the last 10 Ma (Laursen et al., 2002;Clift and Hartley, 2007). If we assume an average crustal thickness of $40 km over the study area, this equates to a loss of 2.4 Â 10 6 km 3 of crust in the past 150 Ma. If continuous addition of crustal material to the mantle wedge occurred at a rate of 3 km/Ma as estimated at 33°S, then implicitly variable but recognizable signatures of crustal contamination should be observable. This is not the case for the samples analysed from the southern Central Andes. Intrusive rocks dated from Late Cretaceous ($73 Ma) to Eocene ($39 Ma) record a sustained period of upper crustal growth, with very limited evidence of contamination by older continental crust. From $36 Ma to 17 Ma however there is good evidence for mixing of mantle-derived melts with Late Palaeozoic -Early Mesozoic basement. This could be the result of increased sediment flux to the trench and ultimately the mantle wedge, or as a result of increased subduction erosion of the margin. In the Late Oligocene ($23 Ma) to Late Miocene ($6 Ma) at least two types of basement that differ in age are required to generate the isotopic variability observed, i.e., a mixture of Grenville and Palaeozoic basement. As no crust of Grenville age occurs close to the margin it is indicative of crustal assimilation and recycling. This is consistent with the conclusions of Clift and Hartley (2007) that the forearc has eroded slowly since $20 Ma and that subduction erosion is punctuated over time. Interestingly the timing of episodes of high flux do not correspond with those further north between 15 and 25°S which peaked at $80, 40 and <10 Ma ago (DeCelles et al., 2009); suggesting considerable spatial variability in flux along the length of the margin.
Models for the growth of continental crust rely on knowing the balance between the generation of new crust and the reworking of old crust. A number of studies (e.g., Kemp et al., 2006;Belousova et al., 2010;Dhuime et al., 2012;Wang et al., 2013) have used similar techniques to those applied here to unravel the preserved sedimentary record, which is just a snap shot of the overall signal. In this study, where the spatial relationship between samples is well constrained, along and across margin variability in the zircon record is revealed. Such spatial variability could be averaged by sediment transport processes and thus be masked if only using sediments to reveal the growth of continental crust over geological time (Allègre and Rousseau, 1984).

CONCLUSIONS
It is evident that combining the in-situ analysis of the oxygen and hafnium isotopic composition of magmatic zircon, with high resolution U-Pb dating, can be used to identify the spatial contamination of arc magmas with specific portions of crust during one tectonic cycle. This has significant implications for models of continental growth, particularly in settings where the geodynamics and plate tectonics are poorly understood, for example in the early Earth, demonstrating that wide variations in d 18 O (zircon) and eHf (zircon) values can be generated over relatively short distances (10's kms). This approach demonstrates that as the geodynamic setting of the southern Central Andean margin changed (the subducting Nazca plate shallowed, the continental crust thickened, and the position of arc magmatism migrated to the east), the Late Eocene to Late Miocene mantle-derived magmas became contaminated (via assimilation) with distinct basement domains. The age and composition of the pre-existing continental crust, which is spatially variable over the Andean Cordillera, is shown to play the dominant role (over the contamination of mantle wedge with subducting components or the products of subduction erosion) in modifying the isotopic composition of the Late Eocene to Late Miocene mantle-derived melts. In comparison to the early Cenozoic, which is shown to represent a period of upper crustal growth with little recycling of crustal material, the Late Eocene to Late Miocene is identified as representing a sustained period of crustal reworking in the southern Central Andes.