Discrete magma injections drive the 2021 La Palma eruption

Understanding the drivers of the onset, evolution, and end of eruptions and their impact on eruption style is critical in eruption forecasting and emergency management. The composition of erupted liquids is a key piece of the volcano puzzle, but untangling subtle melt variations remains an analytical challenge. Here, we apply rapid, high-resolution matrix geochemical analysis on samples of known eruption date spanning the entire 2021 La Palma eruption. Sr isotope signatures reveal distinct pulses of basanite melt driving the onset, restart, and evolution of the eruption. Elemental variations in matrix and microcrysts track progressive invasion, and draining, of a subcrustal crystal mush. Associated variations in lava flow rate, vent development, seismicity, and SO2 emission demonstrate that volcanic matrix resolves eruption patterns that could be expected in future basaltic eruptions globally.

pahoehoe-type lava flows. Eruptive activity stopped for ~1h on 12 th December, resuming with a 6,000-m high eruptive column and the fall of meter-size bombs. Finally, in the afternoon of 13 th December, a final explosive paroxysm produced the highest eruptive column of the eruption (8,500 m) and further bombs. After the paroxysm, the eruptive activity and seismic tremor stopped, marking the end of the eruption on 13 th Fig. 2), Sr (ppm) and MgO (wt%) with whole rock and glass compositions, including whole rock 187 Os/ 188 Os and Os (ppt; color scale) by (22) (excluding one outlier with >0.17 187 Os/ 188 Os and <50 ppt Os, as typical of samples related to crustal contamination). 87 Sr/ 86 Sr matrix data mirror 187 Os/ 188 Os whole rock variations, including decreasing signatures at the start of the eruption and a split in isotope signatures in the second half of the eruption, with unradiogenic values corresponding to samples from equivalent locations and eruption dates, including those from eccentric vents (note a one-day offset between the eruption dates of our Sr-unradiogenic samples and the Os-unradiogenic samples from (22), within sampling error; table S1, data S2). Overall, 187 Os/ 188 Os variations fall within the typical range of La Palma unaltered samples (>50 ppt Os; (41)) and the typical signature of the Canary plume, defined on non-contaminated samples with Os concentrations between 12 and 232 ppt (42) and reflecting HIMU-type metasomatized sources (e.g., (41,42)). This suggests minimal contamination of 2021 melts, with constant MgO compositions in matrix (note whole rocks are affected by accumulation of pyroxene and particularly olivine, common after the first eruption break at the end of September; table S1) yet variable isotope compositions suggesting subtle changes in the magma source. The Sr-Os-unradiogenic samples, including those from eccentric vents, have Os-isotope signatures trending towards the modern primitive upper mantle (43), suggesting a subtly less metasomatized source. The early, more radiogenic melts agree with typical Cumbre Vieja Sr-isotope signatures (Fig. 2), and we interpret the linear trend between radiogenic and unradiogenic signatures from late September to mid-October as mixing between end-member signatures through the first half of the eruption. Despite their lower elemental Os concentrations, radiogenic samples are unlikely related to assimilation of oceanic crust because they have relatively high Sr concentrations, whereas white xenopumice of sedimentary origin erupted in the first part of the eruption (bulk rhyolite composition; (25)) have Sr concentrations one order of magnitude lower (22). In contrast, grey xeno-pumice fragments of phonolitic composition, also noted in earlier La Palma eruptions (25) have high Sr concentrations (22) and may suggest early 2021 melts were mixed with phonolite compositions fractionated in shallow portions of the plumbing system (21) before the start of the eruption. Regardless, elemental and isotope relationships suggest mixing of melts with slightly different sources throughout the 2021 eruption.

Fig. S3
. Backscattered electron (BSE) images of clinopyroxene throughout the 2021 La Palma eruption, obtained via electron microprobe. Clinopyroxene phenocrysts show reverse zoning, defined by partly resorbed, patchy-zoned cores overgrown by Mg-rich rims with dark BSE contrast, which are often sector-zoned. Matrix microcrysts have similar composition and zoning to phenocryst rims. Our electron microprobe analyses and clinopyroxene-matrix thermobarometry focus on clinopyroxene phenocryst rims (circles; excepting the outermost rims with light-BSE contrast) and microcrysts (diamonds; <100 μm width), on both prism and hourglass sectors where observable. Symbols mark the location of electron microprobe spot analyses, with fill colors following those in Fig. S4. The middle-right image zooms into matrix with olivine (Ol), clinopyroxene (Cpx) and plagioclase (Pl) microcrysts. where concentrations are expressed on a molar basis, and Fet is total iron as Fe 2+ ). The left panels include all data from clinopyroxene microcrysts and phenocryst rims (excluding outermost zones where the rims are zoned), defining a single evolutionary trend consistent with final crystallization from the carrier (matrix) melt. Clinopyroxene microcrysts and phenocryst rims are often sector zoned, and the right panels show only analyses where the sector (hourglass or prism) could be identified following the morphological model for titanoaugite (30) (cf. example images in fig. S3). Sector zoning explains much of the variability in Al2O3, TiO2, SiO2 and Mg#, indicating dynamic crystallization conditions under a mild undercooling regime, such as during magma ascent (30). Cr2O3 concentrations are less dependent on sector zoning and reach maxima in inner phenocryst rims recording mafic recharge (16).  (20,21). Plagioclase antecrysts are evolved andesine, low in anorthite (An) relative to matrix microcrysts (which typically classify as labradorite, and have compositions up to bytownite in samples erupted at the end of November 2021; see also Fig. 2). Plagioclase antecrysts in this sample have more radiogenic 87 Sr/ 86 Sr signatures than rims and host matrix, indicating recycling of large, low-An plagioclase crystallized from evolved, radiogenic melts (high 87 Sr/ 86 Sr signature; Fig. 2) into later, more primitive and less radiogenic melts (mixing 87 Sr/ 86 Sr signature; Fig. 2). Symbols represent individual matrix rasters, color-coded with eruption date and with error bars representing analytical uncertainty (2x standard error). Matrix erupted before the 27 th September eruption break is distinctly more evolved (purple symbols have higher contents in high field strength elements, light ion lithophiles and rare earth elements, including Zr, Th, Rb and La; together with slightly lower concentrations in compatible metals like Ni) and represents remobilization and mixing with a hydrous (amphibole-bearing) tephrite mush. Toward the end of the eruption in mid-December, the mild decrease in compatible metals and increase in incompatible trace elements defined by green to yellow symbols agrees with ∼10% fractional crystallization of liquids erupted at the subtle maficity climax in late November (Fig. 2). Fractional crystallization modeling considers the phenocryst assemblage observed in erupted basanites (dominated by clinopyroxene and olivine; table S1) and mineral/melt partition coefficients in alkali basalt (54). The right panels show our matrix data broadly agrees with trends defined by whole rock literature data from Cumbre Vieja, with higher concentrations in incompatible elements for evolved samples (compiled and curated from GeoRoc: Geochemistry of Rocks of the Oceans and Continents, http://georoc.mpch-mainz.gwdg.de/georoc/; data S6). Some whole rock literature data are enriched in compatible metals (e.g., Ni) and depleted in incompatible elements (e.g., Zr, Th, Rb, La) due to accumulation of mafic minerals (14). In contrast, our matrix data represent crystal-free melts that are relatively processed and constant in composition.

Fig. S7.
Kernel density estimates of clinopyroxene-liquid thermobarometry results (pressure left; temperature right) following thermobarometric calibrations appropriate for alkaline magmas (top-preferred: (24); middle: (56) water-dependent thermometer with (24) barometer) and recent machine learning approaches (bottom: (63)). Bin width represents model uncertainty. For the water-dependent thermobarometry approach, we consider melt water contents appropriate for undegassed Canary magmas: 1, 2 and 3 wt% H2O (59,60,61). Within this water range, results from traditional calibrations based on thermodynamic principles are in good agreement with each other (top and middle). Machine-learning (bottom) returns similar temperatures but lower pressures, which nevertheless agree with crystallization upon ascent yet with larger uncertainties. Pressure results from the top panel are transformed to crystallization depths for plotting in Fig. 3 and fig. S8 (see crust-mantle density model in Methods).  (32), together with crystallization pressures of clinopyroxene phenocryst rims and microcrysts converted to depths following the crust-mantle velocity model from each study. Crust-mantle boundary (dotted line) located at 10 km (26) or 11 km (32) in the region of magma intrusion feeding 2021 volcanism. Following the assessment of both seismic models by (28), we favor the dataset by (26), who also provide a local crust-mantle velocity model, in contrast with the regional model by (32). Fig. S9. 87 Sr/ 86 Sr ratios in 2021 La Palma matrix do not correlate with elemental MgO, SiO2 or Sr, suggesting Sr-radiogenic signatures are not related to crustal contamination. This is consistent with upper mantle storage, with fast magma ascent informed by shallow pre-eruptive seismicity, and with the lack of xenoliths in our samples. Matrix data broadly agree with trends defined by La Palma whole rock literature data. Interestingly, 87 Sr/ 86 Sr ratios in the early 2021 matrix (until the first eruptive break on 27 th September; top panel) follow a trend of increasing 87 Sr/ 86 Sr defined by historical eruptions at Cumbre Vieja. In contrast, some 87 Sr/ 86 Sr signatures from November and December, including those from eccentric vents, are less radiogenic, trending toward pre-Cumbre Vieja values. Colored symbols represent individual matrix rasters (this study), color-coded with eruption date. Whole rock Cumbre Vieja data compiled and curated from GeoRoc (data S6), and pre-Cumbre Vieja values from (39).  (70). REE patterns are similar across eruption dates, vents, and for matrix and whole rocks, suggesting a similar mantle source across the eruption. Subtle differences in isotope compositions (Fig. 2, fig. S2) unveil isotopically distinct melt batches variably mixed and tapped through time. Matrix analyses were undertaken via laser ablation quadrupole and multi-collector mass spectrometry. Mineral chemistry includes clinopyroxene phenocryst rims and microcrysts, plagioclase microcrysts and antecrysts, and glass. Table S2. Instrument parameters for matrix analysis by laser ablation quadrupole mass spectrometry (LA-Q-ICPMS) for major and trace elements, and laser ablation multi-collector mass spectrometry (LA-MC-ICPMS) for Sr-isotopes. Plagioclase Sr-isotope LA-MC-ICPMS analysis followed the same procedures, with distinct interference corrections as detailed.
Data S1. Geochemistry of volcanic matrix for the 2021 eruption at Cumbre Vieja rift, La Palma, from this study. Major and trace element data analyzed via LA-Q-ICPMS rasters (elemental concentrations in ppm; major element oxides in wt.%). Sr-isotope data analyzed via LA-MC-ICPMS rasters.
Data S2. Whole rock geochemistry for the 2021 eruption at Cumbre Vieja rift, La Palma, from this study, (7), and (22). Major elements in wt% oxides; trace elements in ppm unless otherwise specified (low abundance highly siderophile elements Os, Ir, Ru, Pt, Pd, Re; (22)); LOI loss on ignition. Data from this study were analyzed by solution ICP-OES for major elements and ICP-MS for trace elements. Data S5. Electron microprobe major element data on clinopyroxene phenocryst rims and microcrysts, plagioclase microcrysts and antecrysts, and volcanic glass (wt% oxides). Clinopyroxene data are paired with putative equilibrium liquids (matrix compositions) and tested for equilibrium on Fe-Mg exchange (56) and DiHd components (57). Only clinopyroxene-matrix pairs that pass both equilibrium tests are used to obtain thermobarometric estimates, following calibrations appropriate for alkaline magmas (24). Crystallization depths follow the crust-mantle density profile obtained from measurements on the speed of P-wave anomalies from the surface down to 30 km depth at La Palma, using data collected by (26). Data S6. Literature compilation of geochemical data from eruptions prior to 2021 at Cumbre Vieja, downloaded and curated from GeoRoc: Geochemistry of Rocks of the Oceans and Continents, http://georoc.mpch-mainz.gwdg.de/georoc/ (major oxides wt%; trace elements ppm).