Oceanic intraplate explosive eruptions fed directly from the mantle

After performing postentrapment crystallization (PEC) corrections (SI Appendix, section 7, no more than ±1.2% in the corrected volatile concentrations depending on chosen model and fO₂), all the MI compositions are basanitic in composition, consistent with the whole rock data from samples FG04, FG18, and FG19 (Fig. 2A). These MIs are hosted in primitive olivine phenocrysts with Mg# [molar % Mg/(Fe²⁺+Mg)*100] of 86 to 88% for FG19 and FG18 with the most primitive close to equilibrium with the mantle (melt Mg# 70 to 71, SI Appendix, section 7). Sample FG04 shows more variability in olivine composition with some more primitive (Ol Mg# ~87), but the majority record a more evolved liquid (Ol Mg# 77 to 82). Trace element chemistry collected on these olivines is consistent with melts derived from melting of peridotite-dominated source for samples FG04 and FG18, and with a pyroxenitic component for FG19, both sources in the garnet stability field (ref. 44, SI Appendix, Figs. S13 and S14). Residual garnet is also recorded in the whole rock and MI rare-earth element concentrations (e.g., Gd/Yb > 3).

MIs at Fogo record highly oxidized mantle conditions even in the most primitive olivine (up to NiNiO +1; Fig. 2B) using the V-partitioning oximeter (D_{v (melt/ol)}; ref. 45) consistent with previous estimates of oxidation state at Fogo by Fe K-edge XANES (48). Silica undersaturation and extremely high CaO (up to 14 wt%) suggest derivation from a CO₂-rich source based on experimental petrology (Fig. 1A, ref. 27). Assuming that the most primitive MIs are representative of primary liquids, this experimental parametrization predicts an average 4.5 ± 2.7 wt% CO₂ in these melts. Considering the inherent uncertainty and the fact that other major volatiles like H₂O are not included in this model, the next logical step is to constrain these values in the MI. We identified that Fogo records extreme volatile contents (up to 2 wt% CO₂, 2.8 wt% H₂O, 6,000 ppm S, 1,900 ppm F, and 1,100 ppm Cl) in the most primitive samples (Fig. 3). These high S contents confirm the oxidizing conditions recorded by D_{v (melt/ol)}, and the overall increase in halogens and S with fractional crystallization (as Mg# in the host olivine decreases) suggests that system was closed to degassing until eruption (Fig. 3).

When compared to the available volatile data from other oceanic islands, these values represent the highest volatile contents measured thus far in primitive silica-undersaturated melts at any oceanic intraplate location on Earth (Figs. 3 and 4A). High volatile concentrations can often be explained due to melting style since low degrees of partial melting result in the concentration of highly incompatible elements in the melt, both volatile and involatile. However, such extreme volatile concentrations could also suggest that Fogo is sampling a volatile-rich source. Our data from Fogo highlight that Si-undersaturated melts have a higher solubility for volatile species as reported by experiments (27, 49). The fundamental question is whether these melts only represent low-degree partial melts with extremely high volatiles due to volatile incompatibility during melting or whether they also sampled a source higher in volatiles. To answer this question, we plotted ratios of elements with similar partitioning behavior during melting (e.g., H₂O/Ce, S/Dy, F/Nd, and Cl/K, ref. 3 and references therein) resulting in global trends that support a source higher in volatiles for Fogo compared to the depleted MORB mantle (DMM) and Hawaii (H₂O/Ce > 250, S/Dy > 400, F/Nd > 30; Fig. 4 B and C). While higher ratios can be a result of anomalous involatile trace element depletions (50), we stress that this is not the case for our suite of Fogo MIs, which exhibit high S/Dy, F/Nd, and H₂O/Ce despite having similar or slightly higher concentrations of these elements (Ce ~60 to 125 ppm, Dy ~4 to 8 ppm, and Nd ~35 to 65 ppm) as those found at other OIBs such as Hawaii, Azores, Antarctica, and Galápagos. In particular, the S/Dy ratio for Fogo MI is significantly higher than for all other comparable OIBs at the same range of Dy content. In contrast, CO₂/Ba and CO₂/Nb are very low on account of anomalously high concentrations of Ba (~400 to 825 ppm) and Nb (50 to 100 ppm), indicating that these may not be applicable to such enriched silica-undersaturated samples (SI Appendix, section 8), and thus, they require further experimental petrology exploration. Interestingly, we found variability in S/Dy (~400 to 1,200), F/Nd (~20 to 50), and Cl/K (~100 to 600) ratios in Fogo MI at the same olivine-host Mg#, indicating that these MIs likely also record heterogeneity in the sampled source consistent with trace-elements record in the olivine crystals (ref. 44, SI Appendix, Figs. S13 and S14). This could also explain some of the differences in volatile contents that are unexplained by typical degassing pathways (Figs. 2D and 3). Overall, we found clear differences in the global OIB trends for halogens: Hawaii (Kīlauea and Haleakalā) exhibits DMM-like Cl/K and F/Nd, another group (Antarctica, Galápagos, and the Canary Islands) extends to high Cl/K ratios and moderate F/Nd, and finally Fogo and some Haleakalā samples extend to high F/Nd but relatively low Cl/K. The CVA and the Canary Islands are two Atlantic oceanic intraplate locations where silica-undersaturated lavas are dominant (Fig. 1A), and where the high S and halogens (Figs. 3 and 4), as well as H₂O-CO₂ (Fig. 2D), are likely recycled by deep subduction and returned by the plume, or possibly even stored in the transition zone by previous subduction events and sampled by either the plume or small-scale instabilities derived from this boundary layer (40, 41).

Vapor bubbles in MI are generally formed by the growth of a denser phase (here olivine) from a less dense phase (silicate liquid) during PEC followed by shrinkage due to the differences in the coefficient of thermal expansion of the host olivine and liquid (9). That said, the complexity of a magmatic system can allow fluid exsolution at depth, if the melt is volatile-saturated resulting in heterogenous entrapment of a silicate liquid and exsolved fluid phase (10). Therefore, before the volatile data were interpreted in terms of magma storage depths, we accounted for this effect. A common test for heterogeneous entrapment of MI vapor bubbles consists of comparing bubble volume to MI volume on a log:log plot (ref. 8; SI Appendix, Fig. S17 in SI Appendix, section 9). This method assumes that vapor bubbles formed by the same process should approximately follow the same isovolumetric % line. After this, a threshold of 10% in the volume % (%vol) of bubbles in MI studies is typically used as an upper limit to exclude heterogeneously entrapped vapor bubbles (8), regardless of differences inherent to the samples (i.e., melt composition, glass transition temperature, and difference between entrapment and eruptive temperature). Under these premises, all but three of our MI would be considered shrinkage bubbles, and reconstructed CO₂ contents would be artificially high (up to ~3.7 wt%). Nevertheless, it is likely that smaller %vol bubbles could have also resulted from coentrapment of fluids (10). In such cases, it may be very difficult or even impossible to discern which MIs are likely to be heterogeneously entrapped solely based on the method and threshold described above.

To better assess for heterogeneous entrapment, we modeled the two-stage growth of vapor bubbles, representative of the physical processes that MIs undergo after their capture (e.g., refs. 9, 10, and 18; SI Appendix, section 9). 1) A stage of high-temperature PEC-driven growth, during which melt composition and density change and a significant amount of CO₂ can diffuse into the bubble (52) and 2) a stage of rapid syn-eruptive cooling growth, driven by the thermal contraction of the melt as it cools to the glass transition temperature and during which the bubble may grow to much larger extent. If PEC and syn-eruptive growth are the only two processes that MI undergo, bubbles with larger %vol should have lower CO₂ densities than bubbles with lower %vol in MI that experienced the same amount of PEC. In Fogo MI, bubbles with more PEC often show higher CO₂ densities (Fig. 2C); however, larger %vol bubbles often record higher densities (up to 0.7 g/mL) than lower bubble %vol bubbles with the same amount of PEC, inconsistent with solely postentrapment bubble growth. Additionally, the bubble growth model predicts a maximum volume of ~6 ± 1.5 vol% for Fogo MI (considering transition temperature and volume uncertainties). We note that in Fogo MI, the bubbles that occupy more than the maximum modeled bubble %vol for each MI also show higher CO₂ densities at similar amounts of PEC (Fig. 2C). We therefore conclude that ~10% of Fogo MI were heterogeneously entrapped as melt and exsolved fluid.

To correct the CO₂ contents of heterogeneously entrapped MI, we ran simulations of CO₂ loss to basanitic MI vapor bubbles containing 0 to 10 vol% cotrapped fluid, undergoing 0 to 20% PEC, and starting at 500 and 700 MPa with the model of refs. 52 and 53. Using the amount of PEC for each MI, the estimated cotrapped bubble %vol from our bubble growth model, and average saturation pressure for each eruption, we recalculated the original dissolved CO₂ contents and saturation pressures for MI with cotrapped bubbles (SI Appendix, sections 9 and 10). Saturation pressures for MI at Fogo indicate that they were trapped at variable depths from relatively shallow in the crust (~10 km) to mantle pressures up to ~600 to 900 MPa or ~20 to 30 km using a pressure gradient of ~30 MPa/km (Fig. 2D). We note that CO₂ is the predominant exsolved volatile component in our MI (>90 mol%). Additionally, corrected pressures for MI that were heterogeneously entrapped indicate that nearly pure CO₂ exsolved fluids were present in the plumbing system at Fogo at depths of up to ~30 km, which is consistent with recent findings of near-pure CO₂ fluid inclusions captured at up to ~24 km at Fogo (34). Trapping of Fogo MI at mantle depths of up to 20 to 30 km is also consistent with the presence of wehrlite xenoliths that record pressures of 22 to 42 km carried by other Fogo eruptions (36) as well as the estimated depth of the 1995 magma reservoir at >16.5 km based on an InSAR deformation model (54).

Our MI entrapment pressures are broadly consistent with estimates from clinopyroxene-melt barometry (Fig. 1D) and CO₂-rich fluid inclusions (23, 24, 30, 34, 35) and indicate melt migration from mantle pressures around 20 to 30 km (where the most mafic record the highest pressures) to magma temporarily stalling at ~15 to 18 km), where the majority of MIs were entrapped. Our data are also consistent with a transient shallower magma reservoir at the Moho (~10 to 13 km) represented by MI hosted in the most fractionated olivine (sample FG04, Mg# 77 to 82) and that has been reported previously (24, 30, 34, 35, 55). While we recognize that the uncertainty in melt inclusion saturation pressures is fairly large, especially given bubble volume and solubility model uncertainties, these depths of entrapment are consistent with previously reported storage regions identified by the presence of near-pure CO₂-rich fluid inclusions for which relatively small uncertainties reliably permit the identification of several storage regions (24, 30, 34, 35, 56). Our findings also indicate the presence of exsolved fluids, rich in CO₂ (>90 mol%, calculated fluid composition in equilibrium with the MI), at pressures of up to ~900 MPa, indicating an already bubbly magma at ~30 km depth. At Fogo, earthquakes have been identified at mantle depths (~30 to 40 km, ref. 31, and at shallower depths <10 km, while our MIs seem to be ubiquitously trapped at depths in between, suggesting that most of the melt storage is between those hypocenter locations. Deep magma storage has been identified as a common feature of ocean islands (55), although most of this evidence comes from clinopyroxene-melt barometry, seismicity, and sparse fluid inclusion data (mainly in the Azores and Cabo Verde). In many locations, shallow magma storage also appears to be common. However, this is not the case at Fogo where evidence is lacking for anything other than transient syn-eruptive shallow storage and geophysical evidence rather points to melting well below 10 km (57) and eruptions being fed directly from sub-Moho storage as was the case of the 2014 to 2015 eruption (30, 55, 58). Our MI data appear to be direct geochemical evidence of this as well.

We estimated decompression rates for these eruptions using the method of ref. 59 based on MI sizes, location, and H₂O loss. Our results indicate that these magmas ascended with decompression rates of ~0.005 to 0.05 MPa/s for FG18 and FG19 and 0.005 to 0.0005 MPa/s for FG04 (SI Appendix, section 11), consistent with explosive eruptions of relatively low intensity (Volcanic Explosivity Index – VEI 2-3) like the 1992 eruption of Cerro Negro (11), the 1977 eruption of Seguam or Kīlauea Iki (59). These decompression rates correspond to 2 to 22 h from Moho depth (~13 km from ref. 37; 9.05 ± 3.80 km at Fogo via receiver functions in this study; SI Appendix, section 12 and Fig. S26) for FG18 and FG19 assuming a constant decompression rate for the entire ascent and 1 to 9 d for FG04. This suggests that the prehistoric eruptions represented by samples FG18 and FG19 were more explosive than the eruption of 1951, which is consistent with their larger cinder cones. Although we recognize that this approach assumes that H₂O starts to degas significantly at these depths, we note that these estimates agree with the record of re-equilibration of fluid inclusions from Fogo that indicated stagnation at Moho depths in the order of hours to days prior to recent eruptions (24, 35) and with the zonation of olivine in the 2014 to 2015 eruption that recorded stagnation for no more than a day and ascent from the Moho in <12 h (30).

While H₂O exsolution is often invoked as the main driver of decompression and ascent in explosive eruptions due to the late degassing behavior compared to CO₂ (18), a role for CO₂ in the ascent and explosivity of some mafic eruptions such as the sub-Plinian eruption of Sunset Crater (continental intraplate), as well as Cerro Negro (subduction volcano), has previously been proposed (10, 11). Additionally, explosive eruptions at Kīlauea were statistically linked to higher volatile concentrations in primary melts (22) and suggested to bypass the shallow storage being fed directly from a deeper source. The high saturation pressure and CO₂ contents indicated by our MI data show that CO₂ clearly plays a major role at Fogo volcano. Specifically, at these high pressures, solubility models indicate that CO₂ is the dominant volatile component (90% mol) in the exsolved fluid at Fogo volcano. Geophysical evidence also indicates that eruptions are fed by vertical propagation of dikes from sub-Moho depths to the surface, with a lack of evidence for shallow magma reservoirs (where H₂O would begin to dominate in the fluid phase) (55, 57, 58). Finally, magma ascent assisted by exsolved CO₂ is expected to be a self-accelerating process by buoyancy decrease. A likely scenario starts with CO₂ exsolution initiating dike fracture by overpressure, and as the melt continues to ascend, the density of the exsolved CO₂ decreases, enhancing buoyancy and driving an accelerated ascent of the magma which in turn facilitates further degassing. Therefore, CO₂ exsolution must play a critical role in driving the ascent of mafic explosive eruptions that evolve at mantle depths in volcanoes that sample volatile-rich sources, making the eruption style in such cases an inherited characteristic from the mantle source.

C.L.D. and E.G. conceptualized the study, designed research and wrote the main paper; C.L.D., S.V., J.H., C.M.A., L.R.M., and B.M. performed research; C.L.D., R.S.R., M.S.-M., and J.C.C. contributed new reagents/analytic tools; C.L.D., C.M.A., L.R.M., and B.M. analyzed data; C.L.D. conceptualization, experimental design, sample preparation, data acquisition, data processing and analyses, original full draft manuscript writing (including supplementary materials), review, and figure preparation. Python bubble growth model writing and implementation; E.G. conceptualization, experimental design, main text manuscript writing, manuscript review, Fig. 1A preparation, and funding PI; R.S.R. field sample collection, Fig. 1C panel preparation, and manuscript review; S.V. sample preparation assistance and processing, data compilation assistance, and manuscript review; M.S.-M. thermodynamic modelling and manuscript review; J.H. receiver function analysis and manuscript review; C.M.A. tenerife melt inclusion data acquisition and manuscript review; L.R.M. tenerife melt inclusion sample preparation and analysis and manuscript review; J.C.C. tenerife sample field collection and manuscript review; B.M. SIMS data acquisition and manuscript review and edition.

The authors declare no competing interest.