Eruptive dynamics reflect crustal structure and mantle productivity beneath volcanoes

Volcanic island arcs, such as the Lesser Antilles arc (LAA), may show large along-strike variability in their eruptive frequency-magnitude relationships, crustal structure, geochemistry, seismicity, and isotopic signatures (Wadge, 1984; Hildreth and Moorbath, 1988; Schlaphorst et al., 2016; Melekhova et al., 2019; Cooper et al., 2020). The proposed mechanisms for such stark changes include different melt flux from the mantle (“melt flux”; Schlaphorst et al., 2018; Till et al., 2019; Cooper et al., 2020), chemically distinct mantellic melts (Pichavant et al., 2002), contrasting crustal differentiation mechanisms (Tamura et al., 2016), and shifts in wedge thermal structure (Turner et al., 2016). These processes can all modulate the thermal-chemical architecture of subvolcanic magma plumbing systems (Castruccio et al., 2017; Giordano and Caricchi, 2022), and subsequently the chemical and physical properties of magma via phase equilibria (Müntener and Ulmer, 2018). Presently, however, the connection between variegated crustal or mantle processes is less frequently extended to include the diverse chemistry, frequency, and magnitude of volcanic eruptions.

To explore the nested interplay among mantle, crustal, and volcanic processes, we turned to the well-studied LAA. We first recovered the generalized thermal-chemical structure of magma plumbing systems in the LAA and used this to contextualize variability in the regional geochemical, geophysical, isotopic, and volcanic record (see Supplemental Material¹ for compiled data sets, sources, and extended methods). In doing so, we highlighted a fundamental role for melt flux in controlling magma plumbing system architecture and, by extension, eruptive dynamics. Finally, we found that global data indicate that the first-order control of melt flux operates at volcanoes worldwide.

The LAA, resulting from the westward subduction of the Atlantic plate beneath the Caribbean plate, hosts a well-studied geological and geochemical record. The 11 active volcanoes, spaced ~50 km apart (4–122 km), express a range of magma compositions, eruptive styles, and magnitudes (Metcalfe et al., 2023). The LAA also fortuitously hosts a plutonic record via intrusive fragments (“xenoliths”; Melekhova et al., 2019). Their texture and geochemistry reflect separation of refractory cumulates during differentiation (cumulate intrusive fragments [CIFs]), mobilization of magmatic mush or solidified magma (plutonic intrusive fragments [PIFs]), and metamorphic overprinting (Stamper et al., 2014; Melekhova et al., 2017, 2022; Camejo-Harry et al., 2018; Brown et al., 2021). Their bulk chemistry (Table S1), mineralogy, and mineral chemistry (Table S2) dictate their origins under varied pressure–temperature–melt composition (P-T-X) conditions within the crust.

The limitations of using intrusive fragments from the LAA as a crustal archive are twofold. First, inversion of their P-T-X values via mineral chemistry cannot resolve subtle interisland differences in magma storage depths due to thermobarometer and chemometer uncertainty (Wieser et al., 2023). Second, mineral chemistry is absent for several islands and is weighted heavily toward CIFs. However, the data are sufficient to make general observations of crustal P-T-X structure in the LAA when combined with independent data sets. One such data set is the entrapment pressure from volatiles in melt inclusions (MIs) for the LAA (Balcone-Boissard et al., 2023). MI pressures have a smaller uncertainty compared to thermobarometry, yet they have biases associated with MI selection, as well as issues of postentrapment reset and modification. Geophysics is another valuable along-arc data source (Melekhova et al., 2019), although resolution may tail off at depth. However, by integrating records from MIs, intrusive fragments, and geophysics, we can limit the resolution and uncertainty shortcomings of individual methods. Together, these data can contextualize the geochemical and eruptive variability between adjacent LAA volcanoes.

Intrusive fragments from the LAA combine up to eight minerals, with most containing amphibole and/or clinopyroxene (Melekhova et al., 2019). Therefore, we applied the single-mineral amphibole and clinopyroxene thermobarometer and chemometer of Higgins et al. (2021a) to recover the pressure (P; kbar), temperature (T; °C), and major-element melt chemistry from which these minerals crystallized (X; SiO₂_liq; wt%; Table S2).

P-T-X data from intrusive fragments collectively highlighted noteworthy properties of LAA magma plumbing systems (Fig. 1A). P-T distributions overall were found to be subvertical between 3.5 and 7.1 kbar (~10.5 to 21 km, where 1 kbar ≈ 3 km from the average xenolith density in Melekhova et al., 2019), a region dominated by CIFs (Fig. 1B). The temperature range was limited (981 ± 52 °C; ranges are 2σ), and SiO₂_liq values were basaltic-andesitic (56 ± 8 wt%). Amphibole was stabilized with decreasing temperature at the expense of clinopyroxene, agreeing with experiments (Melekhova et al., 2015) and intrusive fragment reaction textures (Melekhova et al., 2017). Thermal modeling results have implied that the lower to middle crust beneath active volcanoes is the thermal engine of a magmatic system (Annen et al., 2006; Karakas et al., 2017; Glazner, 2021). CIF data showed that this thermal engine runs at a rather constant temperature (981 ± 52 °C) in the LAA, maintaining a limited range of SiO₂_liq values throughout the middle crust (3.5–7.1 kbar). These features are indicative of buffering by a low-variance mineral assemblage (Blundy, 2022). At pressures between 1.2 and 3.5 kbar (3.6–10.5 km), PIFs and some CIFs (Fig. 1B) recorded a wider range of temperature (955 ± 136 °C) and SiO₂_liq (63 ± 15 wt%; Fig. 1A). The dominance of evolved melts in the upper crust is in good agreement with MIs, which record entrapment conditions in the uppermost LAA crust (~2.5–0 kbar) with a tail to higher pressure (Fig. 1C). While rhyolitic melts are present throughout the crust, they preferentially reside in the upper crust (< 3.5 kbar). These general features concur with other compilations of LAA magma storage (Metcalfe et al., 2023).

Together, the P-T-X conditions of PIFs and CIFs from the LAA revealed a stratification, namely, a rather homogeneous thermochemical structure in the middle crust and a more thermochemically diverse upper crust (Figs. 1A and 1C). The middle crust (3.5–7.1 kbar) transfers heat and mass to the upper crust, where melts differentiate to SiO₂_liq values spanning to more evolved compositions (Metcalfe et al., 2023). Insights from intrusive fragments into the lowermost crust (>7.1 kbar) are lacking (Fig. 1A). However, they likely comprise olivine + clinopyroxene + spinel (Melekhova et al., 2017).

Intrusive fragments and MIs indicated that the upper crust (<3.5 kbar) is the locus of rhyolitic melt storage in the LAA (Fig. 1). The mean upper-crustal thickness is ~11 km (~3.7 kbar), where the geophysical upper crust is defined as the distance (km) between layers 2 and 3 in Melekhova et al. (2019) (see also Supplemental Material). A depth of ~11 km corresponds closely to the average depth at which the SiO₂_liq range increases (Fig. 1A), demonstrating good agreement between geophysical and geochemical data in the LAA. While unbalanced mineral chemistry data and thermobarometer uncertainty preclude a robust comparison of P-T-X values between islands, LAA receiver functions clearly show that the upper-crustal thickness varies along the strike of the arc (Melekhova et al., 2019).

The thickness of the upper crust could reasonably relate to the volume of evolved melt present, given that: (1) the upper crust is a factory for evolved melts (dacitic-rhyolitic; Fig. 1A), a common MI and matrix melt composition in LAA erupted products; and (2) evolved MIs overwhelmingly record entrapment pressures <2.5 kbar, and many indicate upper-crustal storage before eruption (Fig. 1C). To explore this hypothesis and assess if upper-crustal thickness correlates with frequency-magnitude-chemistry distributions of along-arc eruptions, we investigated the relationships between volcanic and crustal signatures in the LAA (Table S3). Indeed, we found a good correlation between the upper-crustal thickness beneath each island, the height of the volcanic edifice, and the erupted volume in the past 100 k.y. (Fig. 2A; Wadge, 1984; Melekhova et al., 2019). This assessment was limited to LAA stratovolcanoes to minimize the effect of fluctuations in dome height during the lifetime of a given volcano. While edifice shape data come with several caveats (see Supplemental Material), to a first order, the edifice height relates to the thickness of a magma column via the hydrostatic (“magmastatic”) pressure equation (P = Δρgh). This fundamental relationship dictates that the volcanic edifice height increases with the vertical extent of a column of magma, with lower density than the crust and with interconnected melt (Castruccio et al., 2017).

Significant transcrustal connectivity in a magmatic system requires a sufficiently high melt fraction, which can only be sustained by heat input (i.e., injected high-T melt; Castruccio et al., 2017; Karakas et al., 2017). Thus, for comparable crustal thickness, volcanic edifice height should correlate with melt flux (Karakas et al., 2017; Weber and Sheldrake, 2022). Thermal modeling has independently shown that melt flux relates strongly to geochemical variance and eruptive magnitude (Caricchi et al., 2014; Karakas et al., 2017; Weber et al., 2020). However, melt flux is a highly elusive parameter, and so linking it to the diversity of magmatic systems and volcanoes beyond a conceptual model is problematic (Caricchi et al., 2014). Therefore, to assert that melt flux plays a central role in controlling geochemical and volcanological behavior in the LAA would require rare evidence of changing along-arc melt flux. Incidentally, recent geophysical and geochemical studies have revealed this may indeed be the case (Schlaphorst et al., 2016; Cooper et al., 2020).

In the LAA, slow subduction (2–4 cm/yr) creates lithospheric-scale fractures of the subducting slab, which expose the mantle to seawater. The ensuing serpentinization of the mantle leads to a heterogeneous water content in the subducting lithosphere. Higher water contents increase the degree of partial melting of the mantle, leading to a higher melt flux to the crust (Grove et al., 2012). Boron isotopes in melt inclusions express the relative contribution of these serpentine-derived fluids (Table S4; Cooper et al., 2020), with high δ¹¹B reflecting a serpentine-rich mantle source. Figure 2B shows that mean δ¹¹B (‰) in LAA MIs correlates with the upper-crustal thickness and the long-term volcanic output. These data suggest that higher water flux to the mantle, and by extension melt flux to the crust, thickens the upper crust. As the upper crust hosts mostly evolved melts (Fig. 1), higher melt fluxes appear to favor the growth of thicker, more chemically evolved upper-crustal magma reservoirs, underlying taller volcanic edifices. A similarly good relationship is observed between island area, a reasonable proxy for long-term productivity of volcanic systems built on oceanic crust, and both the upper-crustal thickness and mean δ¹¹B value (Supplemental Material).

Based on our observations (Fig. 2), LAA volcanoes atop a thicker upper crust are prone to generate more voluminous quantities of chemically evolved melts. Heat loss via hydrothermal circulation and advection in the upper crust is sustained by high rates of melt input, maintaining a thicker, active, upper-crustal magma reservoir (Glazner, 2021). Thermal modeling additionally shows that high melt fluxes lead to greater volumes of potentially eruptible upper-crustal mush (fig. 3b in Karakas et al., 2017). Upper-crustal melts do not necessarily erupt directly in the LAA; instead, evolved melts entrain various minerals to form hybridized andesitic-dacitic magmas (e.g., Dominica). In contrast, lower melt fluxes are associated with a thinner upper crust, promoting only the sporadic accumulation of potentially eruptible rhyolitic melts (e.g., Saint Kitts; Higgins et al., 2021b). In situ plagioclase Sr isotopes from intrusive fragments from Martinique further implicate the upper crust as a zone where melts are stored and blended prior to eruption (Brown et al., 2021). Our findings ratify previous studies (Cooper et al., 2020; Metcalfe et al., 2023) showing that volcanoes in the central LAA (Guadeloupe to Dominica) present the most significant modern-day risk in the region (Fig. 2A).

Melt flux modulation is demonstrably a viable method for controlling geochemical and eruptive behavior in the LAA (Fig. 2). To explore if our observations extend globally, we used the LaMEVE database of large-magnitude (volcanic explosivity index [VEI] >4; Table S5) explosive volcanic eruptions (Crosweller et al., 2012). It reports volcanic eruption parameters including magnitude, VEI, deposit volume, eruption age, and magma type. Principally, we wanted to gauge if we could scale up the relationship in the LAA (predominantly VEI <4; Metcalfe et al., 2023) between melt flux, volcanic output, and geochemistry (Fig. 2) to more productive systems worldwide. Giordano and Caricchi (2022) demonstrated that the average volumetric eruption rate (aVER; total erupted volume/volcano longevity) is proportional to the long-term rate of magma input into the crust (i.e., melt flux) despite deposit preservation biases. This is independent of the extrusive/intrusive ratio, which stabilizes in volcanoes that have been active for longer than a few hundred thousand years. Further, Giordano and Caricchi (2022) showed that the volume of the largest eruption from a volcano scales with aVER (Fig. 3). Larger eruptions are fed by magmatic systems with higher melt flux, agreeing with our observations from the LAA (Fig. 2). However, not only is there a proportionality between aVER and erupted volume, but also between aVER and the average erupted magma composition (color shading in Fig. 3). In other words, larger eruptions are systematically associated with volcanoes that erupt more chemically evolved (dacitic-rhyolitic) magmas (Mason et al., 2004).

Collectively, global data concur with our observations from the LAA indicating that higher melt flux conceivably drives geochemistry of magma reservoirs toward more evolved compositions, culminating in larger-volume eruptions on average (Figs. 2 and 3). In the LAA, this productivity manifests itself as a thicker geophysically defined upper crust (Fig. 2), representing the factory for evolved melt production (Fig. 1). We propose that the characteristic behavior of a volcanic system is strongly influenced by the mantle melt productivity and can be linked to a convolution of directly or indirectly measurable parameters such as the thickness of the upper crust, the active volcano height, and the geochemistry of melts in upper-crustal magma reservoirs (Figs. 1 and 2). Such first-order relationships may permit the rapid assessment of the potential for magmatic systems to feed large explosive eruptions, both at the regional and global scale, which is vital for appraising more remote and poorly monitored active volcanoes.

Thanks go to T. Sheldrake, M. Edmonds, A. Germa, G. Cooper, and E. Melekhova for insightful discussion; C. Stamper, H. Balcone-Boissard, and E. Melekhova for sharing their full chemical data sets; and A. Metcalfe and an anonymous reviewer for thorough reviews. O. Higgins would like to thank the Swiss National Science Foundation for funding through an Early Postdoc Mobility Fellowship (Project Number: P500PN_210239). L. Caricchi received additional support from the Swiss National Science Foundation (grant no. 200021_184632). This project has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No [677493]).