What lies beneath? Reconstructing the primitive magmas fueling voluminous silicic volcanism using olivine-hosted melt inclusions

The magmatic systems that underpin large-scale silicic volcanism encompass large portions of the crust, with partially molten mushy reservoirs that can be thousands of cubic kilometers in volume (Bachmann and Huber, 2016). Although dominated by evolved compositions at upper-crustal levels, these systems are fundamentally driven from below by mantle-derived basaltic magmas. Therefore, the question arises: Are the basalts parental to the generation of large silicic volcanic eruptions derived from a different source compared to surrounding regional volcanism, or do they just represent locally enhanced (spatially and temporally) magma fluxes? This question is challenging to address because ascending primitive magmas are generally intercepted by large silicic reservoirs and are rarely erupted in unmodified form (Wiebe, 1994). Most evidence for mafic-silicic magma interactions therefore comes from mingled magmas, foreign crystal populations or zoned crystals, or co-erupted mafic enclaves (Bacon, 1986; Pritchard et al., 2013; Barker et al., 2016).

The central Taupō Volcanic Zone (TVZ; Fig. 1), New Zealand, is a frequently active and exceptionally productive region of Quaternary silicic volcanism, ultimately fueled by a basalt flux from the mantle that is unusually high for its continental arc setting (Wilson et al., 2009). The mantle processes driving this extreme flux are challenging to study because unmodified mantle-derived basalts are rarely erupted through the crustal silicic reservoirs. Over the past ∼60 k.y., a volume of >780 km³ magma (>99% silicic) has erupted from the central TVZ, almost entirely from two caldera volcanoes: Okataina and Taupō (Fig. 1; Wilson et al., 2009). Between Taupō and Okataina, volcanic activity since ca. 200 ka also includes scattered intercaldera mafic (basaltic to basaltic andesite), small-volume (collectively ∼1 km³) eruptive centers that are typically aligned along northeast-southwest–trending faults (Gamble et al., 1993; Table DR1 in the GSA Data Repository¹).

Here, we investigated the compositions of primitive melts feeding young volcanism in the central TVZ to see if there were any differences between the caldera centers and the less active areas in between. We used the novel approach of analyzing olivine-hosted melt inclusions (MIs) contained within juvenile mafic materials that were erupted during rhyolitic events at the caldera volcanoes and compared these with their counterparts from the interspersed intercaldera mafic centers (Fig. 2A; Fig. DR2).

We studied olivine crystals from mafic enclaves in deposits of the 25.5 ka Oruanui and 3.5 ka Waimihia eruptions from Taupō, and the 1314 Kaharoa eruption from Okataina (Fig. 2A; Fig. DR2). The enclaves are interpreted as juvenile because they have crenulated chilled margins and adhering rhyolitic glass, and they host rhyolite-derived crystals ingested during syneruptive interactions (Leonard et al., 2002; Rooyakkers et al., 2018). Olivines in these enclaves are Mg-rich (Fo_80–86) [Fo = molar Mg/(Mg + Fe²⁺) × 100] and commonly contain MIs that are mostly <20 μm, but sometimes up to 100 μm, across (Fig. 2B; Table DR3). The MIs are variably crystalline due to residence in the rhyolitic magmas, which promoted crystal growth from the inclusion wall (e.g., Danyushevsky et al., 2000). Two basaltic scoria units from Okataina were sampled for comparison with the 1886 Tarawera and 21.9 ka Okareka eruption products containing rare olivines (Fo_79–82; Table DR3). For contrast, we sampled olivines from six small-volume mafic centers between Taupō and Okataina (Fig. 1), the products of which represent the most primitive compositions to reach the surface in this area over the past ∼200 k.y. (Gamble et al., 1993; Table DR1). Olivines in the sampled units overlap in composition (Fo_78–90) with the caldera-related olivines (Table DR3), but their MIs are less common and tend to be smaller, and all have experienced some postentrapment crystallization.

MIs were homogenized through standard 1 atm heating experiments to remove postentrapment crystallization (Danyushevsky et al., 2002; Rowe et al., 2015). Rehomogenized MIs and olivine hosts were analyzed for major elements by electron microprobe (Table DR3), and then MIs >35 μm across were analyzed for trace-element concentrations by laser-ablation inductively coupled plasma–mass spectrometry (Table DR4). Following analysis, measured glass compositions were corrected for over/underheating (Rowe et al., 2015) and olivine-melt postentrapment reequilibration (Fe loss) using Petrolog3 (Danyushevsky and Plechov, 2011). A fundamental assumption required for Fe-loss corrections is that the whole-rock Fe content is representative of the MIs prior to entrapment. While a good approximation for basaltic lava, all enclaves showed clear macroscopic evidence for mixing with the silicic magma (Fig. 2A). In these instances, Fe contents of MIs closest to equilibrium with their host olivine with Fe-Mg distribution coefficients (K_D) of ∼0.3 were used as the corrected Fe values (Rowe et al., 2011; full analytical details in Table DR3).

MIs in olivines derived from the five caldera-related units (intracaldera samples) were among the most primitive melt compositions identified from the TVZ to date, forming a distinct compositional group with low TiO₂, P₂O₅, and Na₂O contents when compared to MIs from the intercaldera centers (Figs. 3A and 3B). Compositional differences between MIs from the two Taupō samples were minor, with Oruanui MIs having slightly lower TiO₂ and Na₂O contents than those from Waimihia. Okareka MIs from Okataina showed slightly higher TiO₂ than the other intracaldera samples, but otherwise Okataina and Taupō MIs were very similar in major-element compositions. Major-element compositions of MIs from the six intercaldera centers overlapped with published whole-rock data for monogenetic centers throughout the TVZ (red field in Fig. 3).

Trace-element data further highlighted contrasts between the intra- and intercaldera samples. MIs from Taupō and Okataina showed low abundances of incompatible trace elements (e.g., Zr = 30–60 ppm, Nb = 1–3 ppm, and Y = 10–19 ppm) and rare earth elements (Figs. 3C and 3D; Table DR4). Incompatible element concentrations were lowest in the Oruanui-derived MIs, which also had higher Sr/Nb and Th/Nb ratios than all other examples. In comparison, MIs from one of the intercaldera samples (Ongaroto) showed the highest concentrations of Zr (112–163 ppm), Nb (4–7 ppm), and Y (27–33 ppm). Incompatible trace-element concentrations in MIs from another intercaldera sample (Kinloch) fell between those of Ongaroto and the intracaldera values, overlapping with published whole-rock trace-element data from other intercaldera basalts from the TVZ (red field in Figs. 3C and 3D).

In general, the geochemical characteristics of mafic to intermediate arc magmas inevitably reflect variable degrees of crustal hybridization and differentiation, masking the primary magma compositions (Waight et al., 2017). Although we targeted MIs in the most primitive olivines, these crystals still recorded variable forsterite and NiO contents (Table DR3), consistent with varying degrees of crystallization at the time of melt entrapment. We therefore back-calculated olivine-hosted MI compositions to a primary mantle–derived magma composition using the PRIMACALC2 software (Kimura and Ariskin, 2014). Most of the MIs experienced only olivine fractionation (typically <20%), but some from Waimihia and Kinloch also experienced some clinopyroxene fractionation (Table DR5). Trace-element abundances, back-calculated using stepwise addition of the fractionated minerals, demonstrated, however, that compositional differences between the intra- and intercaldera MIs reflect actual differences in primary melt compositions and not just differentiation processes (Fig. DR6).

Despite the relatively small geographic distances involved (Fig. 1), there are major differences in the melt compositions entering the crust beneath Taupō and Okataina volcanoes versus those represented in the intercaldera centers. Intracaldera olivine-hosted MIs associated with silicic eruptions have lower high field strength element (HFSE) concentrations, which reflect fundamental differences in the mantle melting regimes beneath versus between the caldera volcanoes in the central TVZ. Compositions from the intracaldera MIs can be modeled by ∼10%–30% melting of a depleted mid-ocean-ridge mantle (DMM) source across a range of different pressures and mineralogies (Figs. 3C and 3D; Table DR4). Such compositions could reflect mantle source depletion through previous melt extraction, although MIs from the largest eruption considered (Oruanui) reflect the highest percentage of source melting (∼20%–30%) with a stronger subduction signature, suggesting that melting may be directly linked to enhanced fluid fluxes from the slab (e.g., flux melting; Fig. 3C; Rowe et al., 2009). In contrast, the intercaldera MI (and their host rock) compositions can be modeled by 3%–10% partial melting of DMM. Alternatively, the subarc mantle beneath the TVZ may be heterogeneous (e.g., Waight et al., 2017) and enriched for the intercaldera basalts. Intercaldera TVZ basalts have higher HFSE concentrations, with trace-element compositions that extend to those observed in primitive lavas erupted in the Havre Trough back-arc basin, offshore to the north of New Zealand (Figs. 1 and 3). There, magmatism is primarily driven by low-degree decompression melting with variable, but lesser, inputs from slab fluids, reflected in low Sr/Nb, Ba/Nb, and Th/Nb values (Wysoczanski et al., 2006). The central TVZ is a segment of rifted continental arc with extension rates of ∼8–15 mm/yr^–1, similar to those in the Havre Trough, which averages ∼20 mm/yr^–1 (Hamling et al., 2015; Caratori Tontini et al., 2019). Rift-associated decompression melting is therefore inferred to play a major role in central TVZ magmatism.

The central TVZ is a complex rifting arc (Fig. 1), making it challenging to explain temporal and spatial changes in volcanism. Compositional contrasts between olivine-hosted MIs from intercaldera centers versus intracaldera eruptive products show that the basaltic feedstocks entering the central TVZ crust reflect both lower degrees of more decompression-driven melting (intercaldera examples) and higher degrees of more subduction-related flux melting (intracaldera examples). These findings highlight two important aspects of modern silicic volcanism in the central TVZ. First, there is a fundamental mantle control on the locations and productivities of Taupō and Okataina, driven by higher degrees of mantle melting and supply rates of primitive magmas into the crust (Fig. 4). This is supported by seismic imaging of the mantle wedge beneath the TVZ, which shows that there are large spatial variations in mantle melting, consistent with variable fluid flux from the subducting slab (Eberhart-Phillips et al., 2020). Mantle melting beneath the TVZ calderas may therefore reflect a combination of both rift-induced decompression melting and enhanced fluid-induced flux melting. Thermal calculations of magma input to the crust modeled using both volcanic and geothermal outputs indicate that the relative degrees of partial melting broadly match the relative rates of mafic magma supply, whereby 4–10 times more mafic melt per unit length of arc is focused into the crust beneath the caldera volcanoes, which ultimately provides the magma flux required to sustain and drive such large-scale silicic reservoirs (Fig. 4; Table DR7). In contrast, measurements of crustal seismic anisotropy suggest that geothermal activity in the area between Taupō and Okataina is driven by lower-crustal magmatism, and that shallow magma reservoirs of a comparable size do not exist in this region (Illsley-Kemp et al., 2019). Pervasive normal faulting in the area between Okataina and Taupō may also help the small-degree basaltic melts to erupt (Leonard et al., 2010). Second, the compositional contrasts between the two data suites imply that compositions of mafic magmas feeding caldera systems are distinctly different than those of peripheral mafic centers, even over distances of <10–20 km. Compositions of the monogenetic intercaldera basalts should thus not be used to infer the mantle melting conditions for caldera-related silicic volcanism in the central TVZ (cf. Rooney and Deering, 2014). A complicating factor is that the foci of silicic volcanism have shifted through time, with multiple caldera centers active at different times and locations in the central TVZ (Fig. 1; Gravley et al., 2016). This history would suggest that the degrees of mantle melting and/or delivery pathways of melt to the base of the crust have changed through time.

Despite occurring only in trace amounts, inherited olivines and their MI cargoes provide unique insights into the subcrustal melt compositions that ultimately have given rise to large-scale, caldera-related silicic volcanism in the central TVZ. We took advantage of two factors in this study: (1) the entrapment and preservation of the most primitive melts in early crystallized high-Fo olivines, and (2) the survival of these olivines into the crustally evolved and contaminated enclaves in silicic eruptions or in basaltic magmas that ascended rapidly. Targeting high-Mg olivines in the trace mafic “contaminants” of silicic deposits thus provides a new way to see through the crustal overprints in silicic magmatic systems in general, not just in the extreme example in the central TVZ. Our approach is applicable to any magmatic system where mafic enclaves have been documented, and it offers new possibilities to investigate the dynamics of silicic magmatic systems associated with caldera-forming events.

We thank Adam Kent, Tyrone Rooney, Roger Nielsen, and James Muirhead for helpful discussions, and Joel Baker for input during the early stages of this research. Frank Tepley, Melissa Drignon, Bruce Charlier, and Stuart Morrow are thanked for analytical assistance. We thank Esteban Gazel, Audrey Bouvier, Maxim Portnyagin, Florence Bégué, and two anonymous reviewers for their helpful comments and suggestions, and James Schmitt for editorial handling. We acknowledge the Marsden Fund of the Royal Society of New Zealand (grant VUW1627) awarded to Barker and ongoing support from the ECLIPSE project (contract RTVU1704) funded by the New Zealand Ministry of Business, Innovation and Employment.