The compositional diversity and temporal evolution of an active andesitic arc stratovolcano: Tongariro, Taupō Volcanic Zone, New Zealand

New geochemical data, including Sr–Nd–Pb isotopes for whole-rock and groundmass samples, are reported for edifice-forming eruptives at Tongariro volcano, New Zealand, which span its ~ 350 ka to late Holocene history. Tongariro eruptives are medium-K basaltic-andesites to dacites (53.0–66.2 wt% SiO2) that evolved via assimilation-fractional crystallisation (AFC) processes partly or mostly in the uppermost 15 km of the crust. When ordered chronologically using a high-resolution 40Ar/39Ar-dated eruptive stratigraphy, the compositional data show systematic 10–130 kyr cycles. Mafic replenishment events inferred from MgO values occurred at ~ 230, ~ 151, ~ 88 and ~ 56 ka and in the late Holocene, with high-MgO flank vents erupting at ~ 160, ~ 117, ~ 35 and ~ 17.5 ka. Cycles in Sm/Nd, 87Sr/86Sr, 143Nd/144Nd and Pb isotopic ratios, which are decoupled from MgO, K2O and Rb/Sr cycles, indicate periods of prolonged crustal residence of magmas from ~ 230 to ~ 100 ka and ~ 95 to ~ 30 ka. AFC modelling shows that intermediate and silicic melt compositions, with r-values between 0.1 and 1, are needed to reproduce Tongariro compositional arrays. AFC models also indicate that ~ 20% of the average Tongariro magma comprises assimilated (meta)sedimentary basement material. Locally, Tongariro and adjacent Ruapehu volcanoes attain their most crust-like 87Sr/86Sr and 143Nd/144Nd compositions at ~ 100 and ~ 30 ka, paralleling with zircon model-age crystallisation modes at the rhyolitic Taupō volcano ~ 50 km to the NNE. These coincidences suggest that the timing and tempo of magma assembly processes at all three volcanoes were contemporaneous and may have been tectonically influenced since at least 200 ka.


Introduction
Stratovolcanoes are iconic features of volcanism globally, mostly associated with arc volcanism at modern subduction zones (Davidson and de Silva 2000). Their eruptive products provide insights into magmatic processes associated with the creation and modification of the continental crust and their eruptions pose substantial hazards on and around the edifice. Stratovolcanoes are dominated by broadly intermediate (andesite/dacite) eruption products, generally inferred to be the end products of basalt magmas being injected into the crust. Basalt itself tends to erupt only in lesser amounts, either as a mixing end-member in hybrid intermediate lavas (e.g. Beier et al. 2017;Streck and Leeman 2018;Conway et al. 2020), or from peripheral vents that may be isolated (e.g. Izbekov et al. 2004) or form part of an adjacent basaltic field (e.g. Hildreth and Communicated by Daniela Rubatto. Lanphere 1994;Bacon et al. 1997;Hildreth 2007). Rhyolite is commonly found in the glassy mesostasis of intermediate composition lavas and as melt inclusions in the phenocryst assemblages of andesites (e.g. Donoghue et al. 1995), but only rarely is erupted as discrete products (e.g. Hildreth and Fierstein 2012;Calvert et al. 2018).
To place any individual stratovolcano into its proper context, there is a need to integrate information from mapping and geochronology with geochemical studies, isotopic studies and any available geophysical information (e.g. Dungan et al. 2001;Conway et al. 2018). In general, studies of individual composite volcanoes over the past three decades have addressed one or more of the following.
(a) The origins of the volumetrically dominant andesite magmas. A widely held view is that andesites in general represent hybrid magmas, composed of mixtures of basaltic and dacitic end-members (Eichelberger et al. 2006;Humphreys et al. 2006). This view arises from the presence of mafic enclaves in many andesites and the ubiquitous compositional diversity of crystals and interstitial melts, with the spectrum of melt compositions noted for the apparent scarcity of andesitic liquids (e.g. Reubi and Blundy 2009;Melekhova et al. 2013 and for New Zealand, Donoghue et al. 1995;Gamble et al. 1999;Price et al. 2005Price et al. , 2012. (b) The respective amounts of mantle-derived versus crust-derived material in andesitic magmas. Arc magmas often show direct petrographic evidence in their xenolithic and crystalline cargoes for both mantle and crustal inputs during magma assembly (e.g. Rudnick et al. 1986;Gruender et al. 2010;Tollan et al. 2017). Where the isotopic characteristics of mantle and crustal end-members are known, or can be inferred, mixing and assimilation-fractional crystallisation (AFC) models can be used to estimate their relative proportions (DePaolo 1981;Bohrson and Spera 2001). The amount of molten crust incorporated into arc magmas also has implications for the heat balances involved with magma assembly, such as energy-balanced crustal fusion and magma crystallisation. For example, the operation of multiple petrogenetic processes at a volcano, such as in situ differentiation paired with mixing, may be implied where both AFC and mixing models are needed to explain compositional arrays (e.g. Sisson et al. 2014).
(c) The diversity versus uniformity of erupted compositions over time, and likely implications for the degree of organisation of the magmatic system. At one extreme, substantial cones may be constructed from a restricted range of compositions over thousands to tens of thousands of years (Villarrica: Lara and Clavero 2004;North Sister: Schmidt and Grunder 2009;Klyuchevskoy: Bergal-Kuvikas et al. 2017), implying the presence of an organised and persistent magma chamber or chambers. At the other extreme, small eruptions like at Ruapehu in 1995-1996, may include juvenile ejecta that span the total compositional range known in the lifetime of the volcano Nakagawa et al. 1999Nakagawa et al. , 2002Kilgour et al. 2013). Comparably large compositional diversity is also seen in the Holocene eruptive record at Tongariro (e.g. in MgO concentrations), which cover much of the diversity seen in the longer-term Pleistocene eruptive record (cf. Hobden et al. 1999Hobden et al. , 2002Pure et al. 2020). However, these compositional trends are rapid over tens or hundreds of years in the Holocene, compared with gradual, systematic trends in the Pleistocene. Unusually large eruptions at stratovolcanoes, such as those from Katmai (Hildreth and Fierstein 2012) and Mazama (Bacon and Druitt 1988), reflect the periodic evacuation of organised systems capable of building magma bodies in the 10-100 km 3 range. However, the development of calderas at arc stratovolcanoes is uncommon (e.g. Hildreth 2007 for the Cascades) which implies that the development of large-scale organised magma systems necessary for such large eruptions is rare. The degree of organisation in an arc magma system, and its compositional evolution through time, has also been connected to thermal conditioning of the crust based on thermochemical models (e.g. Annen et al. 2006Annen et al. , 2015, but such models have not been tested with high-resolution time-composition datasets that span extended timescales (> 200 kyr).
(d) Temporal changes in eruption rates and erupted compositions between neighbouring volcanoes, with the implications these changes hold for the development of edifices and the long-term drivers of magmatic systems. In some examples, such changes have been attributed to the effects of ice loading and unloading on the edifice (e.g. Rawson et al. 2016). In other cases, however, temporal changes in compositions and apparent eruptive rates appear to be asynchronous with glacial/interglacial timings (e.g. Calvert et al. 2018;Pure et al. 2020) and instead reflect long-term fluctuations in the supply of mafic magmas into the system. Arising from the latter is the issue of whether the eruptive sequences recorded at a volcano reflect an overall maturation of the magmatic system(s), which may be expressed as unidirectional, acyclic trends in eruptive compositions (e.g. Annen et al. 2006). Such trends may arise when mafic magma intrusion rates into the crust are constant. Alternatively, the Fig. 1 Map of stratovolcanoes that dominate the southern Taupō Volcanic Zone (TVZ), which have ages from ~ 1 Ma to present. The main edifices and vents are labelled and the red box in the inset shows the southern TVZ's position in the central North Island. Red lines in the main image show Quaternary faults with one labelled in red text to support discussions (from GNS active fault database 1:250,000 version: Langridge et al. 2016) and water bodies are shown in blue. The digital surface model is from LINZ data (available at https:// data. linz. govt. nz/ layer/ 51768-nz-8m-digit al-eleva tion-model-2012/). Marginal ticks are in latitude and longitude (WGS 84 projection). The inset shows the boundary between the continental crust (CC) and oceanic crust (OC), which occurs offshore from New Zealand's North Island (after Gamble et al. 1993) ◂ 30 Page 4 of 36 occurrence of cyclic variations in eruptive compositions (or eruptive rates) may indicate variable mafic intrusion rates into the crust and/or variable tectonic stress states (e.g. Pure et al. 2020).
This paper considers these topic areas through a case study of Tongariro volcano in the North Island of New Zealand (Fig. 1). Tongariro, along with its neighbour Ruapehu, represent two iconic stratovolcanoes that have been among the most active centres since European arrival in New Zealand Leonard et al. 2021). We present a high-resolution temporal record of magma diversity over Tongariro's ~ 350 kyr lifespan with the aim of identifying and interpreting patterns in time-composition relationships. Geochemical data are arranged chronologically to examine relationships between major oxide, trace element and Sr-Nd-Pb isotopic data in individual samples with 40 Ar/ 39 Ar age determinations and also their assigned stratigraphic units ). These investigations focus on whether time-composition trends demonstrate cyclicity, whether composition-volume relationships can provide new insights into petrogenetic processes, and to provide a new dataset for assessing petrogenetic models for arc volcanoes (cf. Hildreth and Moorbath 1988;Annen et al. 2006;Eichelberger et al. 2006).

Geological background
Tongariro volcano (Fig. 1) is situated in the andesite-dominated southern segment of the Taupō Volcanic Zone (TVZ) in the central North Island of New Zealand (Wilson et al. 1995;Leonard et al. 2021). The TVZ is part of an actively rifting arc that has been in its present locus for ~ 2 Myr, driven by westward subduction of the Pacific Plate beneath the Indo-Australian Plate. At the latitude of Tongariro, the plate convergence rate is ~ 45 mm/yr which decreases southwards to ~ 33 mm/yr in the South Island where the tectonic regime becomes transpressional (Nicol et al. 2007). On mainland North Island, the Taupō Rift is the expression of arc-related tectonic extension associated with the subduction zone, whereas the closely spatially overlapping TVZ denotes the area where related volcanism has occurred (Wilson et al. 1995). To the south of Ruapehu volcano, the rift terminates in a complex zone of variably oriented faulting (e.g. Villamor and Berryman 2006). Offshore to the north of New Zealand, the TVZ/rift arc continues into oceanic crust where it becomes the Kermadec arc and Havre Trough (Fig. 1).
Tongariro is a composite stratovolcano comprising many discrete vents (e.g. Pure et al. 2020). Previously, it has been defined as part of the Tongariro Volcanic Centre that included all Pleistocene-Holocene volcanoes south of Taupō volcano in the TVZ (i.e. including Ruapehu, Pihanga, Maungakau, Maungakatote, Pukeonake, Hauhungatahi, Ohakune Craters and Kakaramea: Cole 1978), which are labelled in Fig. 1. Some more recent studies have used the term 'Tongariro Volcanic Centre' to refer solely to the collection of vents on the Tongariro edifice and have treated each vent as a separate volcano (e.g. Shane et al. 2017). In contrast, we use the term 'Tongariro volcano' to describe all vents on the Tongariro edifice and the peripheral vent, Pukeonake, following the formal stratigraphic nomenclature proposed by Pure et al. (2020).
The total edifice volume of Tongariro is approximately 90 km 3 relative to a pre-volcanic datum at 750 m a.s.l. (the highest point on the edifice is 2291 m a.s.l.: Pure et al. 2020). Approximately 21% (19 km 3 ) of the edifice is represented by exposed eruptive material comprising predominantly lavas, with smaller volumes of pyroclastic flow deposits, agglutinates, breccias, hyaloclastites and intrusions including at least one prominent dike Cole et al. 2018;Pure et al. 2020). Pleistocene eruptive sequences at Tongariro have volumes ranging between 0.01 and 2 km 3 , with most formally defined stratigraphic members comprising 0.1-1 km 3 of exposed and buried material that can be inferred with confidence . Eruptive sequences contain non-depositional unconformities: some valley floor areas also contain minor incisional disconformities (i.e. paraconformities) but these are not well characterised because of burial from successive units. Individual Holocene eruptive sequences are ≤ 0.53 km 3 with many discrete units amounting to only 0.0003-0.0065 km 3 (Hobden et al. 2002;Pure et al. 2020). A key difference between Pleistocene and Holocene eruptive records, in edifice materials, is the scarcity of pyroclastic flow deposits in the Pleistocene . However, there is a pumice-rich ignimbrite within the 214-207 ka Upper Tama Member, interposed between more voluminous lava sequences, and ~ 17.5 ka scoriaceous eruption deposits atop the Pukekaikiore landform. These pyroclastic deposits demonstrate that 'Holocene'-style, small-volume eruptions occurred in the Pleistocene, even if they are poorly represented in the edifice record Pure et al. 2020).
In its tectonic setting, Tongariro sits in an active graben between uplifted eastern (Kaimanawa Mountains) and western shoulders, underlain by Mesozoic greywacke and which are locally covered by up to 200 m of Neogene sediments . The Tongariro section of the TVZ arc, also known as the Tongariro Graben, is rifting at ~ 7 mm/yr with several recently active normal faults crossing the volcano (Gómez-Vasconcelos et al. 2017;Villamor et al. 2017;Pure 2020). In contrast, there is less evidence of faulting activity on Ruapehu and rates of extension diminish to the south, towards the rift termination ( Fig. 1; Villamor and Berryman 2006;Villamor et al. 2007). This rifting is inferred to have lowered the central edifice areas, particularly of Tongariro, relative to the flanks (Gómez- Vasconcelos et al. 2017). Despite this subsidence, Tongariro and Ruapehu are the only stratovolcanoes in New Zealand that have repeatedly supported substantial glaciers (e.g. McArthur and Shepherd 1990) and deposits from three periods of major glacial advance are preserved in their 200-350 kyr histories (Conway et al. 2015;Eaves et al. 2016;Townsend et al. 2017).
The composite edifice of Tongariro is constructed from volcanic products erupted from multiple vents distributed over a ~ 12 × 5 km area (Fig. 2). Younger eruption Fig. 2 Eruptive units of Tongariro volcano (following Pure et al. 2020), which are also labelled for members and formations where appropriate for scale. Red and white dots show the locations of samples that were analysed compositionally in this study and by Hobden (1997) (Hobden et al. 2002;Shane et al. 2017), are obvious from their prominence in the landscape and the presence of lava flows that can be traced directly back to their vent source. The oldest products that are directly (physically) linked with their vent sources are those from North Crater (unit 'ari' in Fig. 2) that erupted between 45 and 36 ka . Older (> 50 ka) vent locations are less clear because of burial by younger eruptives and/or sediments, or can only be inferred from material preserved elsewhere such as on the ring plain.
Although the provenance of totally buried edifice material is largely unknown, the visible edifice materials at Tongariro are more likely to form a relatively complete inventory of erupted compositions than at some other stratovolcanoes, at least since ~ 230 ka . This is because eruptions at Tongariro were distributed across multiple vent areas which minimised the amount of stockpiling and edifice growth within confined areas, and therefore large eruptions at Tongariro were less likely to entirely bury smaller, less dispersed eruptive units. The sampled edifice record (Fig. 2) also captures a relatively continuous eruptive history, with few gaps between the analytical errors of 40 Ar/ 39 Ar and K-Ar age determinations since ~ 230 ka (Hobden et al. 1996;Pure et al. 2020). Andesite boulders dated at 304 ± 11 ka (all age errors henceforth are 2 sd) that were ejected during an ~ 11 ka eruption from southern Tongariro, and a cryptic inlier of 512 ± 59 ka basaltic-andesite on northwestern Tongariro, suggest that buried materials at Tongariro pre-date 230 ka; therefore, the samples from the surficially exposed edifice record should be representative of compositional variation since ~ 230 ka (Heinrich et al. 2020;Pure et al. 2020).
Previous petrological studies of Tongariro eruptives have considered general aspects of the petrology of much of the edifice (Cole 1978;Hobden 1997), but specific links to the long-term evolution of the volcano have not been made (Hobden et al. 1999;Pure et al. 2020). Locally detailed studies of geochemical and isotopic characteristics have been undertaken on the products of the Ngāuruhoe cone (Rogan 1996;Davidson et al. 2007;Price et al. 2010;Cocker et al. 2021) but otherwise published isotopic data are lacking for the volcano. In addition, numerous studies have focused on the young (< 20 ka) history of eruptive activity, particularly on the growth of the Ngāuruhoe cone and regionally distributed pyroclastic products (Topping 1973;Nairn et al. 1998;Nakagawa et al. 1998;Hobden et al. 1999Hobden et al. , 2002Moebis et al. 2011;Arpa et al. 2017;Heinrich et al. 2020). In contrast, we focus on presenting a representative suite of data from the entire edifice record of the volcano to address the issues considered in the introduction. The chronology and names of units described here are summarised in supplementary file 4.

Analytical methods
Whole-rock analyses were made on fresh lava and pyroclastic samples (100-500 g) from the interiors of flows, and otherwise always on unaltered material, which were crushed with hardened or high-Cr steel crushers. Xenoliths encountered during the crushing process were removed and reserved for separate analysis . Microcrystalline groundmass material in the 250-350 micron size fraction was separated by standard magnetic and hand-picking separation methods from suitable wholerock samples for 40 Ar/ 39 Ar age determinations . Strategically, aliquots of these same groundmass separates (that represent melt compositions) were reserved and also analysed for trace elements and isotopes for comparison with whole-rock samples (that comprise crystallised and quenched melts, crystal cargo and small country-rock xenoliths). The objective here was to ascertain the link between the whole-rock and the phenocrystic cargo. Where identifiable, crushed xenolith fragments were removed from devitrified groundmass material in the 250-350 micron size fraction. The xenolith-free groundmass separates (100-200 mg) were crushed with a clean agate mortar and pestle.
Major oxide concentrations of whole-rock samples were measured by X-ray fluorescence spectroscopy (XRF) using a Bruker AXS (WDS detector) S8 Tiger at the University of Waikato, Hamilton, New Zealand. Loss-onignition (LOI) mass changes were determined on 2 g of powdered rock after firing in alumina crucibles for 1 h at 1050 °C using a muffle furnace. Powdered rock (0.8 g) was fused with 8.0 g of BLiO 12:22 flux at 1050 °C for 20 min and moulded into glass discs. Analytical precision and accuracy were monitored with standards AGV-1, BCR-2, BHVO-2, OREAS, WS-E and an internal standard (andesite lava sample LP009B: Pure 2020). Analytical accuracy is within 5% (1 rsd) for each oxide concentration but generally ≤ 2.5% (1 rsd), as determined with rock standards compared with the recommended values of Jochum et al. (2016). Duplicate analyses of LP009B indicate an external precision (2 sd) better than 5% for all major oxides.
Trace element concentrations were determined via solution inductively coupled plasma mass spectrometry (ICP-MS) using a Thermo Scientific Element II at Victoria University of Wellington (VUW). Analyses were performed on 50-100 mg of powdered sample digested with concentrated HF and HNO 3 at a 20 HF:1 HNO 3 ratio, with the solution diluted appropriately in 1-3% HNO 3 for analysis. At the start of each analytical session, calibration curves (typical r 2 of 0.9994-1.0000) were determined with at least five multi-element standard solutions (0.2-20 ppb) prepared from certified 1000 ppm single-element solutions (Inorganic Ventures™). Trace element concentrations were calculated from drift-corrected and back-subtracted counting intensities. The results for AGV-2, BHVO-2 and WS-E were compared with, then, if accurate, normalised to the recommended values of these standards reported by Jochum et al. (2016). This procedure was necessary to quantify Lu concentrations in the first three (of ten) analytical sessions. Eu/Eu* (europium anomaly) values were calculated as Eu N /(sqrt(Sm N *Gd N )) after normalisation with the CI chondrite values of McDonough and Sun (1995). Overall uncertainties were calculated from the contributions from the weighing reproducibility (0.04-0.20%), internal analytical precision (0.5-2.0%, 1 rsd) and external reproducibility (0.6-13.3%) that was monitored with internationally recognised standards AGV-2, BHVO-2 and WS-E against the recommended values reported by Jochum et al. (2016). When combined quadratically, the overall uncertainty in trace element concentrations for most elements is 0.5-9% (2 sd). External reproducibilities are given in supplementary file 1.
Sr-Nd-Pb isotopic compositions were measured by thermal ionisation mass spectrometry (TIMS) on a Thermo Finnegan Triton TIMS at VUW. Strontium, Nd and Pb separates were obtained following conventional techniques (e.g. Pin et al. 2014; supplementary file 2). All measurements were performed on outgassed, zone-refined Re filaments. Strontium samples were dissolved in 1 µL of concentrated HNO 3 and were loaded with a Ta 2 F 5 activator onto single filaments, following Charlier et al. (2006). Nd samples were pre-dissolved in 0.01-1 M H 3 PO 4 and loaded onto one side of a double filament assembly. Pb samples were pre-dissolved in silica gel and loaded onto single filaments. Pb isotopic compositions were determined with an in-house 207 Pb-204 Pb double-spike 'Leeds DS' with the following isotopic composition: 29.15% 204 Pb, 6.26% 206 Pb, 57.53% 207 Pb, 7.06% 208 Pb. Measurements of spiked and non-spiked samples were performed on entirely separate TIMS componentry (i.e. filament holders and slit plates), and the spiked and non-spiked data were reduced together using a modified Newton-Raphson inversion to determine the natural Pb isotopic composition and a 4000-iteration Monte-Carlo simulation to determine the uncertainty. Further details are in supplementary file 2.

Petrographic characteristics
The petrographic characteristics of Tongariro edificeforming materials are reported in Pure et al. (2020) and representative features are shown in Fig. 3. In general, most whole-rock samples comprise plagioclase, orthopyroxene, clinopyroxene, olivine and Fe-Ti oxide phenocrysts within a glassy to holocrystalline matrix, which is sometimes flow-banded (compare points 4 and 5 in Fig. 3a). Silicate minerals, especially plagioclase, display a variety of internal textures and compositional zoning features, often with overgrowth rims, that indicate multiple generations or populations of the same mineral in individual samples (compare points 1 and 2 in Fig. 3a; see also Fig. 3d). Pyroxenes commonly occur as individual crystals but also as glomerocrysts, sometimes with plagioclase, Fe-Ti oxides and/or apatite (Fig. 3a, e). Crystals in glomerocrysts possess similar zoning and internal textures to co-occurring phenocrysts, which suggest a cogenetic origin. Crystals in glomerocrysts differ from the highly recrystallised, anhedral grains in metasedimentary xenoliths (Graham 1987;Pure et al. 2021). Olivine phenocrysts are uncommon but often contain Cr-spinels, and the host olivine is usually jacketed by an orthopyroxene reaction rim (Fig. 3b, c), as has been previously reported (e.g. Cole 1978). Fe-Ti oxides are present in ~ 90% of lavas (Hobden Page 9 of 36 30 1997) although highly vesicular, glassy and pyroclastic materials typically have lower abundances of Fe-Ti oxides owing to secondary crystallisation of these oxides in samples with completely devitrified groundmasses . Trace amounts of apatite (≤ 0.5 vol%) occur in dacites and high-silica andesites, but zircon, biotite and alkali feldspar are absent throughout. Hornblende is only significant in 349-293 ka Tupuna Formation andesites where it is the most abundant ferromagnesian mineral . Groundmass materials analysed are generally > 90% (micro)crystalline and comprise predominantly plagioclase (euhedral, subhedral and lath-like) with lesser amounts of pyroxene generally as orthopyroxene (subhedral and stubby) and equant, anhedral Fe-Ti oxides generally as magnetite (Hobden 1997). Groundmass grains are generally 1-20 µm but sometimes up to 50 µm in completely devitrified samples.

Geochemistry
Representative geochemical analyses for major oxides and trace elements are contained in Table 1 (whole-rock samples) and Table 2 (groundmass separates). Strontium, Nd and Pb isotopic compositions of whole-rock samples and groundmass separates are presented in Table 3. The full geochemical and isotopic dataset is presented in Figs. 4,5,6,7,8,9,10,11,12 and 13, and is provided in supplementary files 4-7. Radiogenic isotope data on samples from the Te Pupu Formation (Ngāuruhoe cone), Red Crater Formation and the Lower Tama Member (not represented in samples of this study) are available in Price et al. (2010), Shane et al. (2017) and Cocker et al. (2021).

Major oxides
Edifice-forming materials at Tongariro are medium-K andesites and less commonly basaltic-andesites and dacites ( Fig. 4; terminology following Gill 1981). SiO 2 is strongly correlated with K 2 O, CaO and Fe 2 O 3 * (representing all Fe as ferric hereafter), but SiO 2 vs. Na 2 O and vs. MgO show greater scatter. The compositional fields for Tongariro and Ruapehu closely overlap for these major oxides, except for SiO 2 vs. MgO data wherein Tongariro samples have generally lower MgO at a given SiO 2 concentration. Like other andesites globally (e.g. Gill 1981), Tongariro Fe 2 O 3 *, MgO and CaO concentrations decrease with increasing SiO 2 whereas K 2 O and Na 2 O increase.
Tongariro SiO 2 vs. MgO data define two curvilinear arrays for high-MgO and normal-MgO andesites (Fig. 4), which compare with a more prominent high-MgO array in Ruapehu samples ). The Tongariro high-MgO array begins at ~ 9 wt% MgO and ~ 55 wt% SiO 2 and converges towards the normal-MgO array at ~ 60 wt% SiO 2 and ~ 4 wt% MgO. Most of the Tongariro samples that follow the high-MgO andesite array were erupted from flank vents (e.g. the 40-30 ka Pukeonake Formation), but also include upper edifice material (e.g. the 152-150 ka Waiaruhairiki and 130-96 ka Mangahouhouiti members) that contain minor proportions (≤ 5 vol%) of olivine ). There is no significant relationship between whole-rock MgO concentrations and the relative abundance of pyroxenes because orthopyroxene occurs both more and less abundantly than clinopyroxene in samples across all MgO concentrations .

Trace elements
Trace element concentrations shown in Fig. 5 have Fe 2 O 3 * on the X-axis because SiO 2 concentrations could not be determined on HF-digested groundmass separates (analysed by solution ICP-MS), and because SiO 2 and Fe 2 O 3 * are demonstrably strongly correlated in the whole-rock data (Fig. 4c). Rb, Ba and Zr increase (Fig. 5) as whole-rock and groundmass compositions become more evolved (i.e. decreasing Fe 2 O 3 * which equates to increasing SiO 2 : Fig. 4c). Concentrations of P 2 O 5 and Y Fig. 3 Petrographic features in Tongariro edifice-forming lavas and pyroclastic deposits. a Image of sample LP018 (96-92 ka Otamangakau Member). Features of interest are 1) sieve-textured plagioclase; 2) 'pristine' plagioclase with no sieve textures; 3) metasedimentary xenolith containing plagioclase, alkali feldspar, pyroxene and Fe-Ti oxides (the latter being sometimes concentrated in areas of mineral replacement); 4) glassy area of flow-banding oriented left to right across image; 5) microcrystalline domain of the lava showing flowbanding to contrast with glassy areas, also oriented left to right across the image; 6) orthopyroxene (brown); 7) clinopyroxene (green); 8) glomerocryst containing plagioclase, Fe-Ti oxides, orthopyroxene and clinopyroxene. b Olivine crystals surrounded by pyroxene reaction rims and a glassy groundmass containing other crystals of clinopyroxene (cpx), orthopyroxene (opx) and plagioclase (plg) in sample LP011 (92-84 ka Te Rurunga Member). c Inset area from panel b showing Cr-spinels inside the olivine crystal. d A glomerocryst of plagioclase crystals containing Fe-Ti oxide and melt inclusions in sample LP002 (129-119 ka Rahuituki Member). Brown glass at the edge of the glomerocryst-matrix interface is indicated. e A clinopyroxene glomerocryst in sample MSR15034 (130-96 ka Mangahouhouiti Member). Note the abrupt compositional zoning in the topmost clinopyroxene grain ◂ (1) numbers in italics that were measured by instrumental neutron activation analysis (Hobden 1997) and (2) samples 24,266 and 24,269 which were measured by XRF (Cole 1979). All data for samples with the prefixes 'LP', 'MSR' and 'GL' and trace elements for TG084 and TG088 are from this study. All other data are from Hobden (1997) and Cole (1979). Stratigraphic classifications and codes are from Pure et al. (2020). GPS coordinates of all samples are in supplementary file 3  Sample  LP151  LP147  LP001  TG152  TG351  24,269  24,266  TG088  TG112  TG163  TG161  TG010   Formation  MM  MM  PN  MK  TM  TM  TM  RC  RC  TP  TP  TP   Member  amt  aww  --api  aht  mgt  ahi  mtr  mpa  mtk  Whole-rock samples display light rare earth element (REE) enrichment relative to heavy REEs, with (La/Yb) N ratios of 4.5-10.2, averaging 6.5 (Fig. 6). Negative Eu  (Fig. 7). Negative Ce anomalies are also more pronounced in REE-rich groundmass separates ( Fig. 7) than in corresponding whole-rock samples (Fig. 6a).  (Fig. 4c).
Fields for the Kaweka (KW) and Waipapa (WP) terrane sediments were defined from data in Price et al. (2015). The Ruapehu wholerock field (R) is after Conway et al. (2018).

Sr-Nd-Pb isotopes
Whole-rock and groundmass Sr, Nd and Pb isotopic compositions are given in Table 3 and are presented in Fig. 8. A comparison of isotopic ratios in the whole-rock vs. groundmass components from the same samples is shown in Fig. 9. For each isotopic system, Tongariro whole-rock and groundmass compositions form curvilinear arrays that scatter between regional basalts and metasedimentary basement rocks that likely comprise the Kaweka and Waipapa terranes ( Fig. 8; Pure et al. 2021). Notably, Tongariro whole-rock and groundmass 143 Nd/ 144 Nd ratios are within analytical error of each other and of the 1:1 equal ratios line, except for the sample with the highest 143 Nd/ 144 Nd ratios (Fig. 9b). The tie lines in Fig. 8a show that 143 Nd/ 144 Nd ratios in groundmass separates are sometimes more crust-like than whole-rock samples, but other pairs show the reverse relationship. For 87 Sr/ 86 Sr and Pb isotope ratios, whole rock-groundmass pairs commonly differ beyond analytical error, and there is no systematic affinity for more crustal or mantle-like isotopic compositions in the groundmass versus whole-rock samples (Figs. 8 and 9).

Magma assembly processes at Tongariro
The origins of volumetrically dominant andesitic magmas at arc stratovolcanoes have been debated using global datasets of rock and melt inclusion compositions (e.g. Reubi and Blundy 2009;Lee and Bachmann 2014  . The Ruapehu whole-rock field is after Price et al. (2012). Tongariro data from Hobden (1997) are only shown in panel a. Note that Pb isotopic ratios from Ruap-ehu (Price et al. 2012) and Tongariro (Hobden 1997) were corrected for fractionation externally and are not comparable to Tongariro data shown here, hence only Sr-Nd isotopes are shown in panel a.

Crystallisation differentiation
The mafic enclaves, crystal clots (containing plagioclase, pyroxene, Fe-Ti oxides and apatite) and country-rock xenoliths in Tongariro andesites indicate that recharge, crystallisation-differentiation and assimilation processes contributed to magma assembly ( Fig. 3; Pure et al. 2021). Of these processes, crystallisation differentiation is considered first using major oxide and trace element data in Figs. 5 and 10, which are interpreted noting the preference of certain elements to partition into certain minerals (i.e. following the Kd values compiled by Gill 1981 andRollinson 1993, and references therein). Ni vs. Fe 2 O 3 * (Fig. 5h) and Ni vs. MgO (Fig. 10a) arrays provide evidence for olivine fractionation in basalts and basaltic-andesite magmas above 8 wt% MgO. This is supported by the occurrence of olivine in primitive Tongariro eruptives (~ 8 wt% MgO) relative to its scarcity in eruptives with < 8 wt% MgO . Invariably, however, olivine crystals are jacketed by orthopyroxene overgrowths, indicating that they are not in equilibrium with the host melt. Below 8 wt% MgO, pyroxene fractionation likely drives further MgO decreases that occur concurrently with decreasing Sc, from ~ 35 to 15 ppm. Other compositional data (e.g. Y/ Co vs. MgO, not shown) provide no indication that clinopyroxene and orthopyroxene fractionation occur at different stages of magma evolution. This is also supported by Pure et al. (2020) who report that MgO concentrations are not correlated with orthopyroxene/clinopyroxene ratios in Tongariro whole-rock samples.
Rb/Sr vs. CaO/Al 2 O 3 data (Fig. 10c) suggest that plagioclase fractionation commenced at ~ 0.55 CaO/Al 2 O 3 (which corresponds to ~ 7 wt% MgO: see Fig. 10d), and scattered Rb/Sr ratios at andesitic SiO 2 concentrations suggest that plagioclase may have been recycled during magma assembly (Fig. 10e). However, Rb/Sr vs. Eu/Eu* relationships indicate that groundmasses generally have lower Eu/Eu* (i.e. a stronger negative Eu anomaly) and higher Rb/Sr than whole rocks (Fig. 10e). Outside of analytical uncertainty, groundmass-whole-rock tie lines show that Eu/Eu* values in the groundmass are generally lower (indicating plagioclase fractionation relative to the whole rock) in the same samples. Plagioclase fractionation is also indicated by groundmass Sr concentrations that are on average ~ 40 ppm lower than in whole rocks (Fig. 5c). In whole rocks, stronger negative Eu anomalies generally occur in samples with elevated Rb/ Sr ratios (Fig. 10f) that equate to higher SiO 2 concentrations (Fig. 10e), which suggests that plagioclase fractionation (as opposed to accumulation) is the more dominant process in Tongariro magmas.
Y vs. Fe 2 O 3 * and P 2 O 5 vs. Fe 2 O 3 * groundmass-wholerock tie lines in Fig. 5 support the idea that apatite fractionation and accumulation occurred in Tongariro magmas, with little overall net change in P 2 O 5 concentrations as magmas evolved. Increased Rb, Ba and Zr in the groundmass, relative to whole-rock concentrations (Fig. 5a, b, f), supports the observation that alkali feldspar, biotite and zircon did not crystallise as phenocryst phases in Tongariro magmas, which is shown by their complete absence in all lava and pyroclast samples from the combined collections of Hobden (1997) and this study. Although geochemical data do not clearly demonstrate Fe-Ti oxide fractionation, it probably occurred because Fe-Ti oxides are abundant in crystal clots with pyroxene, plagioclase and apatite that were likely removed from Tongariro magmas via crystallisation differentiation (Fig. 3). Based on the interpretations above, we adopt the following proportions for the fractionating mineral assemblages used to model the chemical evolution from basaltic-andesites to dacites: olivine (4%), plagioclase (32 wt%), orthopyroxene (32 wt%) and clinopyroxene (32 wt%).

Assimilation and recharge
Sr-Nd-Pb isotopic data indicate that both crustal and mantle-derived materials contributed to the generation of Tongariro magmas. This is supported by whole-rock and groundmass 87 Sr/ 86 Sr and 143 Nd/ 144 Nd ratios that form arrays between regional basalts and basement metasediments (Fig. 8), with the latter inferred to comprise the Jurassic Waipapa and Kaweka terranes . Bulk magmas also appear to have co-evolved with their coexisting melts as shown by similar (or indistinguishable) isotopic compositions in the whole-rock and groundmass components of each sample, which is observed across the full compositional ranges examined (Fig. 9a-d). Regression lines between whole-rock and groundmass concentrations are within error of 1:1 equal composition lines for Sr, Nd and Pb isotopic ratios (95% confidence). This result requires that either Sr-Nd-Pb isotopic compositions were initially set in magmas before they crystallised phenocrysts (with negligible crystal fractionation or subsequent crustal assimilation), or that evolving magmas progressively fractionated crystals whilst concurrently assimilating country-rock melts and restites that continued to modify Sr-Nd-Pb isotopic compositions.
Relevant to the latter interpretation is that metasedimentary basement xenoliths in Tongariro eruptives have been interpreted as restites comprising ~ 50% of their former masses, with the remaining ~ 50% assumed to be assimilated as melts (Graham 1987;Pure et al. 2021). The metasedimentary restites form up to 6%, but generally only 1-2%, of total whole-rock volumes of eruptives (Hobden 1997). Additionally, xenocrysts derived from disaggregated country-rock xenoliths also likely occur cryptically in Tongariro eruptives, which would add to the proportions of incorporated country-rock material just mentioned. Consequently, the latter interpretation (i.e. concurrent crystal fractionation and crustal assimilation) is favoured on petrographic evidence (restitic metasedimentary xenoliths) and on trace element data that imply the fractionation of olivine, clinopyroxene, orthopyroxene (Ni, MgO, Sc, CaO/Al 2 O 3 relationships: Fig. 10a-d) and plagioclase (Rb/Sr, SiO 2 and Eu/Eu* relationships: Fig. 10e, f).
Genetic relationships between melts and co-existing crystal cargoes are also revealed by comparing wholerock and groundmass trace element concentrations from the same samples (n = 34). On average, Sr concentrations in whole-rock samples are greater by 13 ± 20% (± 2 sd; -34 to + 113 ppm total range) whereas Nd concentrations are on average greater by 13 ± 20% in groundmass separates (± 2 sd; -0.8 to + 6.5 ppm total range). The opposing preferences for Sr in the whole-rock component (i.e. in crystals, especially plagioclase) vs. Nd in melts appears to have little effect on the general covariance of Sr and Nd isotope ratios in groundmass samples and their whole-rock hosts (Fig. 9a, b). Th and Zr relationships in paired whole-rock and groundmass samples also show statistically significant enrichments of these elements in the groundmass (Fig. 9e,  f). These results support the inference that the isotopic compositions of Tongariro magmas were continually modified as they evolved, which parallels the general correlations between Ca concentrations and 87 Sr/ 86 Sr ratios in plagioclase at Tongariro (Shane et al. 2019). Collectively, these findings challenge the idea that compositional diversity and mineralogical diversity are developed separately in arc magmas via decoupled processes, as has been suggested in popular petrogenetic models (e.g. Annen et al. 2006;Blundy 2022).

Assimilation with fractional crystallisation
Chemical and thermodynamic AFC models are often applied to compositional datasets to estimate the proportions of mantle-derived and crust-derived materials in magmas under scenarios of country-rock fusion and concurrent fractional crystallisation (e.g. DePaolo 1981; Bohrson and Spera 2001). In their common application, the ratio of the mass assimilation rate (M a ) against the mass crystallisation rate (M c ), known as the r-value (= M a /M c ), is usually predetermined and limited to less than 1. Whether such assumptions are appropriate in continental arc settings is questionable because mixing and in situ crystallisation differentiation (without crystal fractionation) are common magma assembly processes at some arc volcanoes. An example is at Mount Rainier in the Cascades where evolved in situ melts were mixed into less evolved magmas that ascended through the system, which have compositions that are predicted by AFC evolution at r-values greater than 1 (Sisson et al. 2014). In such cases, crystallisation and country-rock fusion may be thermodynamically balanced across the broader open system, but this may not be reflected in chemical and isotopic arrays. In the context of Tongariro (and Ruapehu), the presence of restitic metasedimentary basement xenoliths in eruptives indicates that r-values determined from compositional arrays will not necessarily reflect thermodynamically balanced crystallisation and country-rock fusion. This is because country-rock fragments were physically incorporated into magmas in full but did not melt completely . Accordingly, our estimates of the proportion of assimilated crust in Tongariro samples is pursued from a chemical and isotopic perspective, and not from a thermodynamic perspective.
Additionally, although the percentage of assimilated crust in an arc magma can be estimated from δ 18 O data, there are few studies that report oxygen isotopic compositions for Tongariro eruptives, appropriate mafic end-member magmas and assimilated country rocks. Hobden (1997) reports δ 18 O for pyroxene grains, which may include up to ~ 15% of xenocrystic pyroxene from disaggregated xenolithic restites based on modal proportions of xenoliths in Tongariro eruptives (< 2%: Hobden 1997) and pyroxene in metasedimentary xenoliths (< 30%: Pure et al. 2021). Beier et al. (2017) report δ 18 O for mantle-like olivines in the Pukeonake Formation basaltic-andesites of Tongariro, but how these values transpose to basaltic arc magma compositions is unclear because the groundmass glass has mixed with more evolved melts. Other crystals in Pukeonake Formation basaltic-andesites were likely derived or crystallised from magmas that were more evolved than basalts. Regarding crustal assimilants, Graham et al. (1990) report δ 18 O for xenoliths in Ruapehu eruptives, which Fig. 10 Major oxide and trace element arrays, and interpretations of fractional crystallisation. The directions of residual magma evolution, after crystals are removed, are shown by arrows. Minus (-) signs indicate mineral removal; plus ( +) signs indicate accumulation, with d following Gamble et al. (1990). Abbreviations are plagioclase (plag), clinopyroxene (cpx), orthopyroxene (opx) and olivine (ol). Tie-lines are shown for paired WR (blue diamond) and GM (white diamond) analyses from the same samples. The arrays indicate that decreases in MgO, Ni and Sc are predominantly driven by pyroxene fractionation, with some contribution from olivine removal. Rb/Sr, CaO/Al 2 O 3 and SiO 2 arrays, and Eu/Eu* vs. Rb/Sr WR-GM tie lines in panel f, suggest that plagioclase is both fractionated and re-entrained (i.e. plagioclase is both phenocrystic and antecrystic). The key in panel f applies to all panels. Fields for the Kaweka (KW) and Waipapa (WP) terrane sediments were defined from data in Price et al. (2015) and the Ruapehu whole-rock field (R) is after Price et al. (2012). Regional basalt samples are the Kakuki (K), Waimarino (W) and Havre Trough (H) basalts and the Te Rongo basaltic-andesite (a.k.a. 'Red Crater basalt') after Gamble et al. (1993), supplemented with trace element data from Barker et al. (2015) and Waight et al. (2017) ◂ are similar to values reported by McCulloch et al. (1994) for the Waipapa and Kaweka terranes that comprise the upper crust (Price et al. 2015;Pure et al. 2021). Estimating the percentage of assimilated crust in mafic magmas that replenish the Tongariro system would therefore require arbitrary choices of mineral-melt fractionation correction values. Notwithstanding these limitations, mass balance with δ 18 O values reported by Graham et al. (1990), Hobden (1997) and Beier et al. (2017) would likely yield around ~ 10-25% of assimilated upper crustal metasedimentary rocks in Tongariro magmas, provided that there is no contribution from lower crustal contaminants that have largely unknown lithologies and compositions.

Geochemical modelling of magma evolution
Tongariro whole-rock and groundmass compositions are compared with AFC and mixing model results in Fig. 11. The models also use a fractionating mineral assemblage of olivine (4%), plagioclase (32 wt%), orthopyroxene (32 wt%) and clinopyroxene (32 wt%), following the discussions above. A mix of 50% average Kaweka terrane and 50% average Waipapa terrane is adopted for the bulk composition of the crustal contaminant, using the values of Price et al. (2015) and following Pure et al. (2021). The most primitive (53.1 wt% SiO 2 ) eruptive sampled from the Tongariro edifice is adopted as the starting composition for the models, which is the Te Rongo (TR) basalticandesite ). This end-member composition is denoted by the yellow squares in Figs. 4,5,6,7,8 and 10.
Although the Tongariro magma system was likely replenished by magmas more primitive than the TR basalticandesite, such as the Kakuki and Waimarino arc basalts, neither of these basalts individually represent a plausible basaltic end-member across key trace elements (e.g. Sr, Rb, Nd) and isotope ratios (Figs. 5,6,7,8,9 and 10). For example, the Kakuki basalt may be an appropriate end-member for Sr and Rb (Fig. 10e, g), however, its Nd concentration is higher than the TR basaltic-andesite (Fig. 10h). Additionally, the Waimarino basalt could be considered an appropriate end-member from Nd concentrations (Fig. 10h), but its 143 Nd/ 144 Nd ratio is more crust-like than many higher-SiO 2 Tongariro andesites (Fig. 8a). The Waimarino basalt also has an elevated 87 Sr/ 86 Sr ratio (0.7044) that is more crust-like than other regional basalts (Fig. 8a: Gamble et al. 1993).
Whole-rock and groundmass samples generally plot between mixing curves and low r-value (= 0.1) AFC curves (Fig. 11); however, both intermediate and silicic melt Kd values are needed to model the entire Tongariro compositional array. Mixing and fractional crystallisation pathways for SiO 2 concentrations and 87 Sr/ 86 Sr ratios show that varied amounts of crystal fractionation and assimilation are needed to explain the compositional diversity of Tongariro samples because Tongariro data do not follow the 'crystal fractionation only' trajectory (Fig. 11a). Trial and error plotting shows that AFC curves adopt shapes and positions similar to mixing curves when r-values approach 1 (not shown).
Mixing curves and AFC curves follow similar trajectories in 87 Sr/ 86 Sr vs. 143 Nd/ 144 Nd space (Fig. 11b). At r-values of 0.3, AFC models predict that magmas will attain evolved 87 Sr/ 86 Sr (= 0.7062) and 143 Nd/ 144 Nd (= 0.51263) ratios with 10-50% remaining melt, or by mixing subequal proportions of TR basaltic-andesite and metasedimentary basement rocks. Variations in whole-rock Nd concentrations and 143 Nd/ 144 Nd ratios cannot be reproduced by a single AFC or mixing model, and require models that use both silicic and intermediate melt Kd values and r-values between 0.1 and 1 (Fig. 11c). Other modelling results for 87 Sr/ 86 Sr, La/Yb, Rb/ Sr, Sc and Zr further support the inference that Tongariro magmas evolved along AFC pathways with r-values of 0.1 to 1 in the presence of silicic and intermediate composition melts (Fig. 11d-f).
Using an r-value of 0.5 for a typical AFC pathway of a Tongariro magma that evolved from a TR basaltic-andesite composition onwards, the percentage of remaining melt may be between ~ 20 and 100%, depending on the sample (Fig. 11). In a simple hypothetical case where only mantle-derived material is fractionated (and all assimilated materials remain in the evolving magma), the percentage of assimilated crust acquired in this stage of magma evolution is approximated by ~ 0.5 (average r-value) multiplied  Gamble et al. (1993). The choice of fractionating mineral proportions, based on information in Fig. 5 and the interpretations of Fig. 10, are: olivine (4 wt%), plagioclase (32 wt%), orthopyroxene (32 wt%) and clinopyroxene (32 wt%). The modelled assimilant is a 50:50 average of the average Kaweka and Waipapa compositions defined by Price et al. (2015), with the proportion based on the subequal representation of the two terranes in metasedimentary xenoliths at Tongariro . The mixing line (heavy line) between the Te Rongo basaltic-andesite and the 50:50 Kaweka-Waipapa composition shows the percentage of mixed assimilant at 10% intervals. All AFC curves are for r = 0.1, except in panel b (r = 0.3), and are shown for intermediate and silicic composition partitioning coefficients compiled by Gill (1981), Rollinson (1993) and Ersoy and Helvacı (2010). Labelled increments on the AFC curves represent the fraction of remaining melt. The AFC curves for r-values between 0.1 and 1 (not shown) plot between the mixing curve and the r = 0.1 curves. The key in panel d applies to all panels. Errors are shown for isotopic data which are all 2 sd. Other 2 sd errors are approximately as follows: SiO 2 (0.75 wt%), Nd (1 ppm), La/Yb (0.7), Rb/Sr (0.01), Sc (1.5 ppm) and Zr (8 ppm In some cases, r-values calculated from compositional arrays may not represent thermodynamically balanced processes because assimilation can occur without total melting of wall rocks and because conductive heat exchange can occur in magmas without mixing. On the former point, detailed comparisons of xenoliths and likely protolithic materials at Tongariro suggest that approximately 50% of each assimilated (meta)sedimentary xenolith was melted when incorporated in its host magma ). The other ~ 50% of chemically incorporated material comprises disaggregated, restitic, country-rock xenoliths and xenocrysts that are seen petrographically in erupted materials. With this constraint, the AFC modelling results in Fig. 11 show that magmas probably evolved at r-values of 0.1 to 1, or thermodynamically balanced r-values of ~ 0.05-0.5 because of the incomplete melting of incorporated country rock. Conversely, assimilation that occurs at peripheral areas of a magma may be fuelled by heat conducted from inner areas of the magma, which may not be followed by mixing and homogenisation. In such cases, incorporated wall rock melts and restites may be concentrated in peripheral areas of a magma body that could ascend and erupt independently, and be falsely interpreted as the products of closed-system AFC evolution at r-values that exceed 1.
Collectively, the AFC modelling results (and highly approximate δ 18 O mass balance estimates) above indicate that the average Tongariro magma (basaltic-andesite to dacite) contains ~ 20% of assimilated metasedimentary basement rocks. While this estimate provides some constraints on the heat budget of the Tongariro magmatic system, the unerupted proportion of each magma that remains as cumulates or mush is difficult to estimate because the r-values in the early stages of magma evolution, at basaltic compositions, are poorly constrained. Moreover, although geophysical surveys (in particular using magnetotelluric methods) indicate that magmatic bodies containing up to ~ 45% melt exist down to ~ 12 km depth beneath the Tongariro edifice (Hill et al. 2015;Miller et al. 2018), the volume of intruded but unerupted magma is also largely unconstrained.

Time-composition relationships and the organisation of the magmatic system
The sequential evolution of magma compositions at a volcano reflects the timing and tempo of petrogenetic processes and the degree of organisation of the magmatic system. It is generally accepted that geochemical data should be ordered sequentially when making petrogenetic interpretations because compositions rarely evolve unidirectionally over a volcano's entire lifespan (e.g. Eichelberger et al. 2006). To address this issue, Tongariro compositional data are integrated with a high-resolution edifice construction record, from ~ 350 ka to present .
Time-composition data for MgO and K 2 O are shown in Fig. 12 for individual samples (points) and stratigraphic unit ranges (crosses). The MgO data display irregular cycles that have been interpreted as mafic replenishment events that commenced at ~ 230, ~ 151, ~ 88 and ~ 56 ka and in the late Holocene, with high-MgO flank vents erupting at ~ 160, ~ 117, ~ 35 and ~ 17.5 ka ). These events are each followed by 10-70 kyr intervals where MgO concentrations gradually decrease from ~ 9-5 wt% to ~ 4-2 wt%. These intervals are inferred to reflect prolonged crystal fractionation because the opposite pattern occurs in K 2 O vs. time data; that is, K 2 O increases as MgO decreases. Time-composition data prior to 250 ka have a lower temporal resolution and are not interpreted.
In contrast, time-composition trends in Sm/Nd and Sr-Nd-Pb isotopic compositions, which are shown for samples (points) and stratigraphic unit ranges (crosses) in Fig. 12, occur on 30-130 kyr cycles and do not align with the MgO and K 2 O trends. This result does not appear to be an artefact of lower temporal resolution in the Sr-Nd-Pb isotopic dataset. For example, the Waiaruhairiki Member and its sample (LP245) define an MgO maximum (5.5 wt%) at ~ 150 ka (Fig. 12a) but also occur in the middle of the cycle to increasingly crust-like 87 Sr/ 86 Sr (0.7054), 143 Nd/ 144 Nd (0.51272) and 206 Pb/ 204 Pb (18.848) ratios at ~ 150 ka (Fig. 13d-f). In Fig. 12 Time-composition trends for Tongariro whole-rock and groundmass samples, plus ranges for overall stratigraphic units. Error bars show 2 sd ranges for individual groundmass and wholerock samples (that have individual 40 Ar/ 39 Ar age determinations: Pure et al. 2020). The crosses show the full range of compositions and possible ages for stratigraphic units, which take into account field relationships and radiometric age constraints. The average composition of the crust used for modelling in Fig. 11 is shown as a green line. The encircled blue stratigraphic unit (Waiaruhairiki Member) and sample from that unit (LP245) are shown for comparison across compositional parameters (e.g. compare trends for MgO in a with 143 Nd/ 144 Nd in e). Data for stratigraphic units are from this study, Hobden (1997) and Cole (1979) Fig. 13; Price et al. 2015;Pure et al. 2021), is interpreted to show periods of prolonged crustal residence and crustal assimilation from ~ 230 to ~ 100 ka and ~ 95 to ~ 30 ka. The timings of these cycles, relative to the MgO vs. time trends, indicate that the episodes of mafic replenishment and prolonged crustal residence (with associated crustal assimilation) are partially decoupled at Tongariro.
Such systematic time-composition trends indicate a moderate degree of organisation in Tongariro's magma system, at least beneath the central edifice area which is the inferred vent area for most eruptive units represented in the long-term record . The gradual and cyclic shifts in the concentrations of compatible elements (e.g. MgO in olivine and pyroxenes), incompatible elements (e.g. K 2 O) and radiogenic isotopes show that magma bodies of at least 1 km 3 (i.e. the erupted component) both evolved and intermittently erupted over ≥ 10 kyr periods. Notably, the relatively low rate of compositional variation in the long-term eruptive record since ~ 350 ka (Fig. 12) contrasts with the rapid compositional change observed in small-volume (< 1 km 3 ) Holocene eruptives (e.g. Hobden et al. 1999;2002). For example, the entire range of 87 Sr/ 86 Sr ratios over the long-term record occurs within eruptives from the Ngāuruhoe cone at Tongariro alone, which have ages of ≤ 7 ka and individual flow volumes of < 0.1 km 3 , but more often < 0.01 km 3 (Hobden et al. 1999(Hobden et al. , 2002Pure et al. 2020).
Similar composition vs. volume relationships occur at Ruapehu Conway et al. 2018), and it appears that the smaller the eruptive-volume window that is examined, the larger the compositional diversity. The modern situation at Tongariro and Ruapehu may in fact represent the smaller-scale 'background' activity that has been present through much of the lifetime of both volcanoes but remains generally unrecorded in the long-term eruptive record (e.g. Gamble et al. 1999). With the caveat that erupted volumes may not correlate with magma batch volume, because magma bodies feeding eruptions may only be partially evacuated, we conclude that there is a lower chance of observing rapid, compositional heterogeneity in the edifice record because smaller-volume eruptives are less likely to be preserved or sampled.
The observations above can be reconciled in a model of the Tongariro magma system wherein magma batches of diverse volumes and compositions coexist in the crust. The rapid compositional variation observed in smallervolume eruptives, such as from the Ngāuruhoe cone, may arise when ascending magmas avoid or only briefly transit through longer-lived (and potentially larger and more integrated) parts of the magma system. Concurrently, other magmas may instead stall and homogenise to produce the gradual, systematic compositional trends that are recorded periodically when the longer-lived system is tapped during eruptions (e.g. Figure 12). This process has been inferred elsewhere, such as at Mount St. Helens, where antecrystic > 100 ka zircons occur in < 100 ka eruptives, despite thermal perturbations to 820-940 °C having occurred at ~ 100 ka (above zircon saturation temperatures of 740-780 °C), which would have dissolved zircons in ~ 100 years (Claiborne et al. 2010). Reconciling these observations requires brief (< 100 year) interactions between the hot, mafic magmas that ascended through the cooler, zircon-saturated magma system. Similarly, ascending mafic magmas at Tongariro appear to either recharge the longer-lived system where they hybridise and evolve or, instead, they may erupt peripherally after brief, limited interactions with the longer-lived, predominantly andesitic system.
Although the repeated 10-130 kyr compositional cycles demonstrate a degree of organisation in Tongariro's magmatic system (Fig. 12), their cyclicity suggests that any long-term, thermal conditioning of the crust occurs at nonconstant rates. This observation contrasts with simpler cases of uniform heating that are adopted by thermochemical models of crustal evolution in arc settings Annen et al. 2006). The time-composition trends at Tongariro show that magmas were produced by a moderately organised magmatic system (i.e. showing gradual compositional changes), but that each time-composition cycle was of a different duration. Any compositional trends in magma compositions induced by thermal conditioning of the crust, that arise from repeated intrusions of mantle-derived magmas (Annen et al. 2006), appear to be undetectable or overprinted by stronger time-composition variations that were induced by cyclic petrogenetic processes. We propose that the cyclic compositional trends at Tongariro (Fig. 12) likely arose from open-system processes that include temporally variable mafic intrusion rates and crustal residence durations Fig. 13 Time-composition data for stratigraphic units (total ranges) shown by crosses for Tongariro (grey) and Ruapehu (blue) that are compared with zircon crystallisation model ages from Taupō volcano in panel e, which are grouped into 5-kyr bin widths from ages reported at 1-kyr precision by Wilson and Charlier (2009). Data fields for the Te Herenga (TH) and Wahianoa (WA) formations from Ruapehu are shown as coloured boxes. Interpreted trend lines are shown in grey for the Tongariro data in panels a, b and d. At Tongariro, systematic trends occur in MgO, Rb/Sr and 143 Nd/ 144 Nd. The Rb/ Sr trends largely mirror MgO trends whereas isotopic compositional trends appear decoupled from MgO and Rb/Sr trends. Tongariro and Ruapehu 87 Sr/ 86 Sr and 143 Nd/ 144 Nd values become more crust-like at ~ 95 and 30 ka, coinciding with peak modes in zircon model ages at Taupō volcano. Tongariro compositional data are from this study, Hobden (1997) and Cole (1979). Tongariro age data are from Pure et al. (2020). Ruapehu compositional and radiometric data are from Gamble et al. (2003); Price et al. (2012) and Conway et al. (2016Conway et al. ( , 2018 ◂ (when AFC evolution occurred) that can both be modulated by tectonic processes.
Other modelling and across-arc studies have indicated that most compositional diversity in arc magmas is acquired in the deep crust, with petrographic diversity being developed later (Hildreth and Moorbath 1988;Annen et al. 2015). The Tongariro dataset presented here is, however, difficult to fully reconcile with that view. Partial conflict arises because (i) several (if not tens of) percent of total whole-rock material comprises assimilated crust in the form of restitic metasedimentary xenoliths and melts (Fig. 3;Hobden 1997;Graham 1987;Pure et al. 2021); (ii) sample petrography (Fig. 3) and trace element and major oxide relationships indicate that olivine, clinopyroxene, orthopyroxene and plagioclase fractionated from magmas (Figs. 5 and 10); (iii) correlated Ca concentrations and 87 Sr /86 Sr ratios in plagioclase (Shane et al. 2019) and similar (but often indistinguishable) Sr-Nd-Pb isotopic compositions in crystal cargoes and co-existing melts ( Fig. 9) suggest that crystal growth and fractionation occurred concurrently with crustal assimilation; (iv) the compositions of restitic metasedimentary xenoliths at Tongariro are similar to the compositions of the Kaweka and Waipapa terranes that are the inferred sources of the xenoliths (Graham 1987;Hobden 1997;Price et al. 2015;Pure et al. 2021); (v) the Kaweka and Waipapa terranes are inferred to comprise the uppermost 15 km of the crust beneath the Tongariro-Ruapehu area which indicates that assimilation occurred in the upper crust (Harrison and White 2006;Behr et al. 2011;Stern and Benson 2011); and (vi) Tongariro Sr-Nd-Pb isotopic compositions in wholerock and groundmass samples plot between a mafic basalticandesite end-member and the Sr-Nd-Pb isotopic compositions of Kaweka and Waipapa terrane reference materials which shows that these terranes are plausible contaminants of evolving Tongariro magmas (Fig. 8). In contrast with the findings of Hildreth and Moorbath (1988) and Annen et al. (2015), we conclude that crystal fractionation and upper crustal assimilation at Tongariro occurred concurrently and in sufficient amounts to explain the full range of Sr-Nd-Pb isotopic compositions in the combined datasets of Hobden (1997) and this study.
We note that the assimilation of the lower crust into Tongariro magmas cannot be discounted because the composition and lithology of lower crustal materials are largely unknown, so their influence on Tongariro magma compositions cannot be accurately assessed. Geobarometry estimates from a limited number of Tongariro samples suggest that some crystal growth (and presumably crustal assimilation) occurred in lower crustal areas at depths of < 25 km or < 7 kb (Hobden 1997;Arpa et al. 2017). However, representative geobarometry estimates for the greater Tongariro eruptive history are lacking. Therefore, on the available evidence, it is most likely that the radiogenic isotopic compositions of Tongariro eruptives were acquired from assimilating subequal (if not predominant) amounts of upper crustal rocks relative to lower crustal rocks.

Long-term drivers and aspects of magmatic systems at Tongariro and neighbouring volcanoes
Tongariro and neighbouring Ruapehu are predominantly andesitic composite stratovolcanoes comprising several cones and vent systems (Cole 1978;Hobden et al. 1996;Gamble et al. 1999Gamble et al. , 2003Conway et al. 2016;Pure et al. 2020). The two volcanoes are separated by ~ 20 km along the NNE-SSW-trending axis of the TVZ arc and have independent magmatic roots ). Tongariro's closest neighbouring volcano to the north that was active during the Holocene is Taupō volcano, which has been a source of voluminous rhyolitic eruptions, distinguishing it from the andesitic southern TVZ volcanoes (Wilson et al. 1995).
Compared with Tongariro, Ruapehu has a potentially more organised magma system that is reflected in its more systematic time-composition trends in major oxide, trace element and radiogenic isotopic compositions in eruptives from ~ 230 ka to present (Price et al. 2012;Conway et al. 2018). Like Tongariro, the compositions of historic and small-volume Ruapehu eruptives also span the entire range seen in the long-term edifice record at Ruapehu since ~ 230 ka . Overall, the compositional records suggest that magmatic systems at Tongariro and Ruapehu comprise multiple magma bodies with a range of sizes, compositions and longevities that coexist at any one time (Nakagawa et al. 1999(Nakagawa et al. , 2002Gamble et al. 2003;Kilgour et al. 2013).

Time-composition patterns at Tongariro vs. Ruapehu
A brief comparison of time-composition trends at Tongariro and Ruapehu, as inferred from their erupted products, is presented in Leonard et al. (2021). Here, we expand on this comparison with a compilation of Rb/Sr, 87 Sr/ 86 Sr and 143 Nd/ 144 Nd vs. age data from both volcanoes (Fig. 13;Hobden 1997;Gamble et al. 2003;Price et al. 2012;Conway et al. 2018;Pure et al. 2020). In general, time-composition trends at Tongariro and Ruapehu are similar and occur contemporaneously for 87 Sr/ 86 Sr and 143 Nd/ 144 Nd ratios but are less coordinated for MgO and Rb/Sr. Temporal MgO and Rb/Sr trends in ~ 200-100 ka eruptives from Tongariro and Ruapehu do not overlap. MgO concentrations at Ruapehu decrease from ~ 8 to ~ 2 wt% between ~ 45 and 40 ka and are mirrored by increased Rb/Sr ratios from ~ 0.2 to ~ 0.6 Price et al. 2012;Conway et al. 2016Conway et al. , 2018. Covariance in MgO values and Rb/Sr ratios at Tongariro is also observed, as shown by the interpreted grey trend lines (Fig. 13a, b) Gamble et al. 2003;Price et al. 2012;Conway et al. 2016Conway et al. , 2018Townsend et al. 2017). At the two volcanoes, broadly contemporaneous satellite vent eruptions occurred at 40-30 ka (Pukeonake Formation on western Tongariro: Fig. 1) and at 35.4 ± 0.6 ka (Ohakune Formation southwest of Ruapehu: Fig. 1), both of which comprise eruptives with elevated MgO concentrations of ~ 6-11 wt% (Froggatt and Lowe 1990;Beier et al. 2017;Pure et al. 2020;Svoboda et al. 2022).
Time vs. 143 Nd/ 144 Nd trends at Tongariro and Ruapehu display simultaneous excursions to more crustal values for the ~ 230-80 ka and 50 ka to Holocene periods (Fig. 13d) Together, the extensive geochemical and geochronological datasets that cover most of Tongariro's and Ruapehu's edifices show broadly contemporaneous variation in MgO, Rb/Sr, 87 Sr/ 86 Sr and 143 Nd/ 144 Nd over their long-term eruptive records (Fig. 13). Given that the magmatic roots at Tongariro and Ruapehu are physically disconnected , the contemporaneous compositional cycles at the two volcanoes suggest that the timings of mafic replenishment events and intervals of protracted magma residence in the crust (which appear to be decoupled at Tongariro: see Fig. 12 and associated discussion) are both controlled by a shared external process, most likely to be tectonic, that can cause shifts in crustal stress states.
At one extreme, synchronised eruptive activity in the ~ 11 ka Pahoka-Mangamate eruptions involved several explosive eruptions from compositionally distinct magmas at Tongariro and Ruapehu and their satellite vent systems, where the clustered timing of the eruptions has been attributed to tectonic processes Nakagawa et al. 1998;Heinrich et al. 2020). This inference has been recently disputed, however, because tephras from these eruptions contain crystals with similar reverse zoning patterns, amongst other compositional similarities (Heinrich et al. 2022). These features were interpreted by Heinrich et al. (2022) as showing that the entire Tongariro magmatic system was interconnected and partially homogenised. However, our view is that these similarities in reverse zoning patterns could also be achieved by contemporaneous mafic replenishment without the requirement for lateral interconnectivity between shallow crustal magmas. Such contemporaneous mafic replenishment into compositionally distinctive but isolated magma reservoirs at Tongariro has been reported for the Te Maari, Red Crater and Te Pupu (Ngāuruhoe) formations, without lateral connectivity between their shallow crustal magma bodies (Shane et al. 2017;Pure et al. 2020).

Contemporaneous magmatic events at Tongariro, Ruapehu and Taupō volcanoes
Evidence for tectonically synchronised volcanic activity between Taupō volcano and southern TVZ andesite volcanoes was first proposed by Kohn and Topping (1978). Based largely on interfingered tephra deposits, they reported elevated transition metal concentrations in titanomagnetite crystals that distinguished the voluminous ~ 12-11 ka tephras erupted at Tongariro and Taupō from subsequent eruptives from both volcanoes. These contemporaneous shifts in tephra compositions suggested that magmatic variations between separate volcanoes (and unconnected magma systems) could be synchronised by an external, common influence. Subsequent studies by Hodder (1983) and Nairn et al. (1998) have further supported this hypothesis. Here, we propose that contemporaneous fluctuations in magmatic activity at Tongariro, Ruapehu and Taupō are a long-term feature over their > 200 kyr histories.
In Fig. 13, the time-composition data from Tongariro and Ruapehu are compared with zircon crystallisation model ages (obtained by U-Th disequilibrium methods) from rhyolitic eruptives at Taupō (Charlier et al. 2005;Wilson and Charlier 2009). At Tongariro, minima in 143 Nd/ 144 Nd ratios and maxima in 87 Sr/ 86 Sr ratios show that magmas attained their most crustal isotopic compositions at ~ 100 ka and ~ 30 ka, broadly coinciding with the ~ 95 ka and ~ 35 ka peaks in zircon model ages at Taupō (Fig. 13c-e). Sr and Nd isotope ratios at Ruapehu also reach their most crust-like values in the ~ 40-20 ka interval, hence broadly coinciding with the ~ 35 ka zircon model-age peak at Taupō. This contemporaneity is interpreted to show that magmas erupted from the three volcanoes resided in the crust for prolonged periods where they progressively underwent country-rock assimilation (Tongariro and Ruapehu) and cooling (all three volcanoes) from ~ 230 to ~ 100 ka and from ~ 95 to ~ 30 ka.
The idea that magmas experienced prolonged residence in the crust within the ~ 95-30 ka period is further supported by Ruapehu U-Th isotope data reported by Hughes (1999). The sole Ruapehu sample erupted within the ~ 95-30 ka period that was analysed has a U-Th disequilibrium model age of 81 +13/-12 ka, which is within error of the eruption age of 74 ± 37 ka determined via 40 Ar/ 39 Ar techniques . These overlapping ages imply a relatively short residence time within the crust that is likely to be ~ 10-20 kyr or less, given the U-Th and 40 Ar/ 39 Ar age uncertainties. Younger Ruapehu eruptives with 40 Ar/ 39 Ar age determinations between ~ 22 and 0 ka have U-Th disequilibria model ages of 72 +23/-19 ka and 62 +18/-15 ka that imply longer crustal times of ~ 40-50 kyr (Hughes 1999;Gamble et al. 2003;Conway et al. 2016). These results indicate that the magmas feeding these eruptives were formed at a similar time but some erupted ~ 50 kyr apart, provided that the bulk of the crystal population was not inherited and of substantially older age. On this evidence, crustal residence was potentially short (i.e. < 10 kyr) for Ruapehu magmas erupted between ~ 95-30 ka relative to longer crustal residence times (i.e. ~ 40-50 kyr) for younger (< 30 ka) eruptives.
Evidence for influxes of hot, mafic magma into the Taupō system between ~ 85 and 45 ka and after ~ 35 ka, that coincide with inferred mafic replenishment events at Tongariro and Ruapehu (Fig. 13), is provided by zircon crystallisation model ages and zircon saturation modelling. Zircon U-Th model ages from late Pleistocene Taupō eruptives have two peak crystallisation modes at 95-86 ka and 41-34 ka (Charlier et al. 2005;Wilson and Charlier 2009). Additionally, zircon saturation modelling for these Taupō eruptives by Charlier et al. (2005, using the model of Watson and Harrison 1983) shows that temperature was a predominant factor controlling the crystallisation and/or survival of zircons in Taupō magmas. We therefore propose that zircon crystallisation was suppressed in the periods between ~ 85 and 45 ka, and after ~ 35 ka, by temporarily elevated magmatic temperatures (Fig. 13e). The reduction in the number of zircon model ages in these periods, together with the time vs. MgO data, suggests that mafic replenishment events at ~ 88 and ~ 56 ka (Tongariro) and ~ 45 ka (Ruapehu) may have also occurred at Taupō (Fig. 13a, e). Shifts to more crust-like compositions at ~ 30 ka for Tongariro and Ruapehu also broadly coincide with the 41-34 ka zircon age peak at Taupō, and subsequent shifts to less crust-like 87 Sr/ 86 Sr and 143 Nd/ 144 Nd ratios at Tongariro and Ruapehu after ~ 30 ka align with a reduction in zircon model ages at Taupō (Fig. 13c-e). In general, the temporal alignment of peak zircon age abundances at Taupō and more crust-like eruptive compositions at Tongariro and Ruapehu, and vice versa, can be explained by tectonically modulated, contemporaneous mafic recharge events at all three volcanoes, with interposed periods of reduced mafic input wherein magmas cool and evolve to more crustlike compositions.
However, the cooler, more crust-like Tongariro and Ruapehu magmas erupted at ~ 40-30 ka temporally coincide with the eruption of mafic magmas from flank vents, which indicates further complexity exists than given in the interpretation above. These mafic flank vent eruptives are the high-MgO 40-30 ka Pukeonake Formation and ~ 17.5 ka Makahikatoa Formation satellite cones on western Tongariro and the 35 ka Ohakune Formation southwest of Ruapehu (see Ohakune Craters in Fig. 1). The ages of these mafic vents bracket and coincide with the 41-34 ka zircon age peak at Taupō and the excursion to more crust-like compositions at Tongariro and Ruapehu (Fig. 13) (Froggatt and Lowe 1990;Conway et al. 2016;Townsend et al. 2017;Pure et al. 2020;Svoboda et al. 2022). It is unclear, however, if the timing of mafic flank vent eruptions should be rapidly followed by reductions in crust-like character of central edifice magma systems (i.e. within ~ 1 kyr). Potentially, mafic flank vent eruptions could be sourced from magmas in the deeper crust that bypassed the predominantly andesitic central edifice magmatic systems at Tongariro and Ruapehu when they ascended (e.g. Price et al. 2012), as is suggested by the volume-composition relationships discussed above. If ascending mafic magmas largely avoid the cooler, more evolved central edifice magmatic systems then there would be reduced heating and mafic replenishment of these central systems. This would effectively preserve the cooler, more crust-like character of magmas in the Tongariro and Ruapehu central edifice systems. Collectively, the zircon age dataset from Taupō (Charlier et al. 2005;Wilson and Charlier 2009) and the time-composition datasets of Tongariro and Ruapehu Price et al. 2012;Conway et al. 2016Conway et al. , 2018Pure et al. 2020) support a model wherein tectonic processes modulate the timing of certain magma assembly processes at the three volcanoes, subject to minor deviations from the general patterns described above, such as mafic magmas erupting from flank vents instead of perturbing centralised magmatic systems.
A role for the tectonic modulation of magmatic activity in the crust is supported by measurements of crustal stress before, during and after recent volcanic eruptions at Ruapehu (Gerst and Savage 2004). Elevated slip rates on the Rangipo Fault in the late Pleistocene have also been associated with volcanic-tectonic interactions at Ruapehu (Fig. 1: Villamor et al. 2007). Additionally, the action of tectonics on regulating eruption rates and timings have been inferred in many other studies elsewhere in the TVZ, the western United States, the Chilean Andes and in Japan (Lanphere and Sisson 2003;Allan et al. 2012;Geshi and Oikawa 2016;Myers et al. 2016;Rawson et al. 2016). Interrelated behaviour between rifting and magmatism is also known in the central TVZ where rifting is aided by thermally assisted crustal weakening, and vice versa, which can result in positive feedbacks (Rowland et al. 2010;Allan et al. 2012;Ellis et al. 2014). At Tongariro specifically, rifting and tectonic processes have been used to explain the distribution of vents at Tongariro and the timing of eruptions (Gómez-Vasconcelos et al. 2017;Heinrich et al. 2020;Pure et al. 2020). On these grounds, we propose that in the central and southern TVZ, the duration of magma residence in the crust and the timing of mafic magma replenishment events are likely to be regulated by tectonic processes, such as episodic variations in crustal stress regimes. On current data (Fig. 13), such processes appear to be capable of synchronising the most volumetrically significant magma-forming and magma evolution events at adjacent arc volcanoes.

Conclusions
Chemical and isotopic data for Tongariro whole-rock and groundmass samples are reported for ~ 220 new samples that add to an existing database of ~ 370 samples (Cole 1979;Hobden 1997). We considered these data both with and independently of their stratigraphic context, which is provided by the high-resolution eruptive stratigraphy of Pure et al. (2020), and draw the following conclusions.
(2) AFC modelling results, and highly approximate δ 18 O mass balance estimates, indicate that typical Tongariro magmas contain ~ 20% of assimilated basement metasediments, derived in subequal proportions from the Kaweka and Waipapa terranes. Assimilated materials were variably melted and occur as both restitic xenoliths (including xenocrysts from disaggregated xenoliths) and incorporated melts. AFC modelling indicates that Tongariro magmas contained melts ranging from intermediate to silicic compositions, which evolved with r-values of 0.1-1 (Fig. 11). (3) A large proportion of Tongariro's compositional diversity in major oxides, trace elements and Sr-Nd-Pb isotopes was acquired partly or mostly in the upper 15 km of the 30-40 km thick crust, via AFC processes. (4) When ordered sequentially, both individual samples with 40 Ar/ 39 Ar age determinations (whole-rock and groundmass samples) and entire stratigraphic units show systematic time-compositional variation on 10-130 kyr cycles. Following interpretations by Pure et al. (2020), increases in MgO at ~ 230, ~ 151, ~ 88, ~ 56 ka and in the late Holocene, and high-MgO flank vents that erupted at ~ 160, ~ 117, ~ 35 and ~ 17.5 ka, represent mafic replenishment events (Fig. 12). These shifts in MgO concentrations are mirrored by decreases in incompat-ible element markers (e.g. K 2 O, Rb/Sr). Other timecomposition cycles in Sm/Nd, 87 Sr/ 86 Sr, 143 Nd/ 144 Nd and 206 Pb/ 204 Pb ratios, which are decoupled from MgO cycles, indicate periods of prolonged crustal residence and contamination from ~ 230 to ~ 100 ka and ~ 95 to ~ 30 ka. (5) The range of compositions in eruptives from ~ 350 ka to the Holocene is comparable to the range of compositions from the late Holocene alone, such as for 87 Sr/ 86 Sr (but not for all major oxides or trace elements). However, the volumes represented by the late Holocene eruptive record are generally far smaller (typically < 0.01 km 3 per lava flow relative to ≥ 1 km 3 for units in the longer-term eruptive record). In contrast to the longer-term record, the late Holocene record is also typified by relatively unsystematic time-composition trends (Hobden et al. 1999(Hobden et al. , 2002. On this basis, we propose that the smaller-volume eruptives are less likely to be representative of the degree of organisation in their source volcano's magmatic system. (6) There is a repeated temporal alignment of compositional cycles and evidence for contemporaneous magmatic cyclicity and/or eruptive activity at Tongariro, Ruapehu and Taupō (Fig. 13). This alignment suggests that an external process (i.e. tectonics) has regulated the assembly and eruption of the most significant and voluminous magma bodies at each of these volcanoes since at least 200 ka. Cycles of prolonged crustal residence and assimilation in magmas at these volcanoes occurred from ~ 230 to ~ 100 ka and ~ 95 to 30 ka, which are partially decoupled from cyclic mafic replenishment events (see point 4, above).