Oceanic islands, situated amidst vast oceans and distant from continents, serve as unique observation points to understand the underlying geological processes. By studying these islands, it is possible to gain valuable insights into the formation of volcanoes, mechanisms of magma storage and ascent, and the evolution of magmatism in oceanic settings at depth. This is due to the fact that rising magma batches frequently contain mafic mineral phases that have been torn out from the deep cumulate layers, which form the basis of volcanic systems. Mafic minerals can originate from different parental melts, and can have different geochemical histories via different generations of silicate melt inclusions (SMIs) (Maclennan 2008). A basic study of bulk rocks provides only a partial insight into this information.

The Azores archipelago comprises the visible part of volcanic ridges, stretched for several tens of kilometres into the Atlantic Ocean, covering an area of about 52,000 km2. These volcanoes started to erupt on a plateau since the late Miocene (e.g., Luis et al. 1998; Cannat et al. 1999; Gente et al. 2003; Silveira et al. 2010). The formation of the volcanic ridges is linked to the movement of oblique transtensional systems, which are characterised by limited spreading over time (Marques et al. 2013). In this tectonic regime, magmas accumulate rather than erupt, resulting in the formation of cumulate layers and crystal mushes. The depth of these bodies, inferred from fluid inclusion studies, has been determined to range between 18 ± 0.31 and 20 ± 0.73 km beneath the islands east of the Mid-Atlantic Ridge (MAR) that are younger than 500 ka (Pico, Faial, Terceira, Graciosa). They have also been confirmed to extend up to 26 ± 0.66 km beneath the 1.9 Ma old island of São Jorge. On the eastern island of São Miguel, located about 400 km east of the MAR, this depth reaches up to 26 ± 0.51 km (Zanon et al. 2023).

In the Azores, olivine- and clinopyroxene-rich basalts with rare plagioclase dominate over evolved magmas such as trachyte, pantellerite, and comendite (Zanon 2015a). These basalts range from sub-aphyric to highly porphyritic and can contain up to 70% olivine. Olivine phenocrysts in these basalts are abundant and contain silicate melt inclusions (SMIs), whose study is crucial for defining both the petrogenetic processes that occur in the primary ponding zone and the conditions of crustal accretion.

The present study, carried out at the archipelago scale, aims to identify the petrogenetic processes that have influenced the geochemical signature of mafic magmas during stalling at the crust-mantle interface. It is based on the study of the geochemistry of SMIs, with particular emphasis on the role of volatiles. The SMIs, preserved as glass, are hosted in rapidly quenched, polyhedral olivine crystals. They belong to products of basaltic cinder cones formed over the last 40,000 years at fissure zones and lateral vents of central volcanoes on six islands east of the MAR (Table 1). They therefore cover a wide time span in the magmatic evolution of the Azores.

Table 1 Sample database

Geological setting

The Azores region in the central North Atlantic comprises nine volcanic islands and seamounts distributed across three plate boundaries and straddling the MAR, which extends at ~ N20° (Fig. 1a). Flores and Corvo are located on the North American plate, while all others are located along sub-parallel slow spreading systems oriented along a general N120° direction (e.g., Vogt and Jung 2004; Borges et al. 2007; Hildenbrand et al. 2014; Miranda et al. 2014; Fernandes et al. 2018). These systems, spreading away from Eurasia at rates ranging from 2 ± 1 mm/year (at the Terceira Rift) to 3.5 ± 0.5 mm/yr at the Pico-Faial system to the south, define the diffuse boundary between the Euroasia and Nubia plates (Marques et al. 2013).

Fig. 1
figure 1

Geographical location of the Azores, digital elevation model of the islands considered in this study and sampling sites. Panel (a) shows the location of the archipelago in the Atlantic Ocean. Panel (b) shows a bathymetric model of the Azores region and the main tectonic elements. The thick yellow line marks the Terceira Rift (TR), which is the northern boundary of the region, and the East–West Azores Fracture Zone transform fault (EAFZ). Thinner red lines mark the São Jorge and Faial-Pico spreading systems. The thick segmented black line is the Mid-Atlantic Ridge. Panel (c) shows sampling sites on the islands of São Miguel, Terceira, Pico-Faial, Graciosa and São Jorge. In these maps, blue triangles indicate the sampling sites at fissure zones, while yellow dots are at central volcanoes. Dashed lines represent fissure zones, and blue dashed lines are the main faults. Names indicate the location of central volcanoes referred in this paper

The intersection of local N45° to N75° strike-slip faults with the N120° fissure systems (Fig. 1b) served as a pathway for the ascent of poorly developed magmas, which rose rapidly from the deepest magma accumulation zones without further long-term ponding and differentiation in the crust (e.g., Zanon et al. 2020, 2023).

All volcanic systems are currently either inactive or extinct. The only geophysical activity of note is the recurrent seismic swarms along the tectonic systems, including the 2022 São Jorge crisis and the 2023 Terceira crisis.

Samples collection

We examined tephra samples and a submarine lava less than 40,000 years old, collected from the central volcanoes and fissure zones of six islands of the Azores archipelago, located east of the MAR. The samples were collected from the Pico-Faial and São Jorge fissure zones and the Terceira Rift which runs from the southwest to the northeast of the islands (as depicted in Fig. 1b). A sample over 800 ka old from the eastern segment of the São Jorge fissure zone (Marques et al. 2018) was included for comparison. As a whole, geochemical analyses were performed on 142 SMIs from 18 samples (Table 1).

In summary, we studied basaltic tephra from: (1) well-preserved Holocene cinder cones formed along fissure systems (11 samples; blue triangles in Fig. 1b). These samples include submarine lava, erupted during the 1998–2000 eruption off the coasts of Terceira; (2) lateral vents located along transtensive faults (7 samples; yellow dots in Fig. 1b) at the central volcanoes of Água de Pau and Sete Cidades on São Miguel Island, and at the central volcano of Pico. The remaining central volcanoes in the archipelago erupted lavas that were either too evolved or aphyric to be considered.

Analytical procedures

SMIs were found in polyhedral olivine ranging in size from 2 to 0.5 mm in diameter. These crystals were handpicked from the 0.25–1.00 cm grain size, after sieving samples of tephra fallouts, which included sandy (0.63 mm) to glassy lapilli (3.5 cm) deposits. These small-sized samples were specifically collected from explosive products to reduce the likelihood of post-entrapment chemical exchange between the trapped melt and the host crystal. SMIs were analysed for major, minor and trace elements, as well as sulfur and halogens (141 and 99 analyses) and water content (121 measurements), using a combination of vibrational microspectroscopy, electron microprobe and laser ablation inductively coupled mass spectrometry.

Fourier Transform Infrared (FT-IR) spectroscopy was used to measure the water concentration on double polished SMIs. A Nicolet Magna-IR 550 spectrometer, equipped with a Globar source, a liquid nitrogen cooled MCT/A detector and an X-KBr beam splitter was used at the Institut de Physique du Globe de Paris (France). Concentrations were calculated according to the Beer-Lambert’s law: H2O (wt.%) = [(100·A·M)/(ε ·ρ·e)], where A is the absorbance at the peak (measured after subtraction of the baseline extrapolated from the base of the peak), M the molar mass (g·mol−1), ε the molar absorptivity (L·mol−1·cm−1), e the thickness measured under an optical microscope (cm), and ρ the glass density (g·cm−3), calculated following Lange and Carmichael (1990). The total dissolved water (H2Omol + OH) was derived from the broadband at 3535 cm−1, using an absorption coefficient of 62.8 L·mol−1·cm−1 (Mercier et al. 2010). The thickness of the inclusions was measured using an electronic crystal thickness gauge (Supplementary Table 1), with relative errors estimated to be < 10% for H2O. All FT-IR spectra showed the characteristic bands of carbonate between 1610 and 1345 cm−1. However, the amount of CO2 (both that diffused from the melt into the bubble, and that still dissolved in the glass) was not determined.

Electron microprobe analysis was used to determine the major element compositions in SMIs and hosting olivines using a Cameca SXFive electron microprobe (Camparis, Paris, France). Glasses were analysed using a defocused beam (Ø = 10 μm) at 10 nA and counting times varying from 10 to 25 s. Three to four analytical spots were analysed, depending on the inclusion size. Concentrations of S, F, Cl and P were determined during dedicated analytical sessions using a 30 nA defocused beam (Ø = 15 μm) and counting times of 240 s at the peak. Fluorine, in particular, was analysed by coupling LTAP and TAP crystals. The relative errors are < 5% for F at the level of 0.04 wt.%, < 4% for Cl at the level of 0.02 wt.%, < 3% for S and < 6% for P, < 5% for K2O and < 3% for the other major elements. The accuracy of F (2495 ± 272 ppm, N = 12) was checked against the glass standard CF-A47 (2270 ± 10 ppm); and Cl (286 ± 17 ppm) and S (1419 ± 33 ppm, N = 14) against the international standard VG2 (303 ± 10 ppm Cl; 1450 ± 30 ppm S at 1σ; N > 100; Métrich et al. 2014). To calculate the melt-olivine equilibrium, the host olivine was analysed (3 analytical spots) around each SMI, at 15 keV and 20 nA, using a focused beam and counting times from 10 to 20 s on the peak. Accuracy was checked against the San Carlos Olivine international standard. The relative errors are < 7% for Ni and Mn, < 5% for Fe and Ca, and < 1% for Si and Mg. The detection limits for trace elements are as low as 18 ppm in olivine and 16 ppm in spinel. To calculate the redox state, olivine-spinel pairs were analysed for major element composition at 20 keV and 10 nA for 10 s (Wan et al. 2008). Trace elements in olivine (Cr, Al, Mn, Ni, Ca, Co) and spinel (Si, Mn, Co, Ni) were measured at 20 keV using a beam current of 300 nA for 30–40 s and 120 s (Supplementary Table 2). The error is better than 2% for olivine and spinel.

The trace element composition was determined by laser ablation associated with an inductively coupled plasma mass spectrometer (ICP-MS) at the Istituto Nazionale di Geofisica e Vulcanologia (INGV), Palermo Section (Italy). The laser, a GeoLasPro 193 nm excimer system, equipped with an Agilent 7500ce quadrupole ICP-MS, operated at 10 Hz, with a spot size of 24–32 μm, associated with a fluence of 14 and 15 J·cm−2, and He fluxes of 800–820 ml·min−1 in the ablation cell. Thirty-five trace elements were measured in peak-hopping mode with a dwell time of 10 ms. The entire analysis took 2 min for each spot, including 1 min for background acquisition. Plasma conditions were adjusted to a ThO/Th ratio of less than 1% so that no further oxide corrections were required. NIST SRM 612 was measured as an external standard at the beginning and the end of each analysis session (20–30 spot analyses). 43Ca was used as the internal standard for the geochemical reference materials. For SMIs, we used CaO wt.% measured by electron microprobe. Raw data of major and trace element compositions are presented in Supplementary Table 1. Data were collected in time-resolved graphics mode to check for any compositional heterogeneity that may be present in the sample at the scale of the laser sampling and to check for inter-element fractionation. The data were processed using the GLITTER program (Van Achterbergh et al. 2001). Accuracy was checked by repeated measurements of BCR-2G basaltic reference glass from the USGS (Jenner and O’Neill 2012), during each analysis session (Supplementary Table 3).

Data description

Sample characteristics

The samples collected are proximal tephra fallouts containing juvenile porphyritic to hypocrystalline and holohyaline clasts as well as loose olivine and pyroxene crystals. The content of crystals is within the range of 20–40 vol.%. Fragments of the submarine lava are poorly porphyritic (~ 15–18 vol.%) with euhedral to skeletal clinopyroxene, olivine and plagioclase in the mineral assemblage. The groundmass ranges from vitrophyric to intersertal.

Studied olivines in these samples are typically either euhedral (polyhedral) or subhedral (Ø ≤  2 mm). Additionally, megacrysts measuring up to 4 mm, are a common feature in the products of Pico and Sete Cidades volcanoes. These have been discovered dispersed or accumulated in discontinuous layers or lenses, resulting from the disruption of consolidated crystal mushes, due to the ascent of magma along tectonic discontinuities (Zanon et al. 2020). Furthermore, glomeroporphyritic aggregates incorporating pyroxenes and oxides have been occasionally found in lavas (Fig. 2a).

Fig. 2
figure 2

Photomicrographs of representative studied lava samples and SMIs. a Highly porphyritic lava from the north coast of Pico volcano. All crystals present are olvines. A large euhedral twinned crystal is present in the upper right corner. Other crystals of different sizes form aggregates. b Large clinopyroxenes in the lava from the west coast of Sete Cidades volcano, São Miguel Island. These large crystals show patchy zoning. c A cluster of SMIs of variable size hosted in olivines from São Miguel Island. d Partially exposed melt inclusion from Faial Island, containing a fluid bubble and a sulfide globule

Silicate-melt inclusions and their host crystals

This study examines SMIs found in ~ 10% of olivine crystals with compositions ranging from Fo75.8 to Fo85.6, as reported in Supplementary Table 2. Chemical core-rim profiles of some crystals from Pico, São Jorge, and São Miguel demonstrate the absence of any zoning (Supplementary Fig. 1).

We followed the practises for selection, preparation and analysis guidelines for SMIs as recommended by Rose-Koga et al. (2021). SMIs are either isolated or form minor clusters (Fig. 2b). The inclusions are transparent with a diameter larger than 20 μm and there are no signs of fractures or decrepitation. The majority of them have a tiny shrinkage bubble that represents approximately 2–5% of the total inclusion volume and a small sulfide globule (Fig. 2c). Any potential post-entrapment modifications were corrected using the Petrolog3 program (Danyushevsky and Plechov 2011) with the following inputs:

  • The equilibrium condition between liquid and the host olivine has been computed using the model of Herzberg and O’Hara (2002). Temperatures have been corrected for the effect of olivine liquidus depression due to water content dissolved in SMIs, using the equation dT(°C) = 74.403H2O (wt.%)0.352 (Falloon and Danyushevsky 2000). This correction enables calculations under hydrous conditions using models developed for anhydrous conditions only;

  • The Fe2+/Fe3+ ratio of the melt that was calculated from the compositions of spinel and olivine pairs using the equation:

\(\frac{{{\varvec{F}}{\varvec{e}}}^{2+}}{{{\varvec{F}}{\varvec{e}}}^{3+}}={\varvec{a}}{\varvec{F}}{\varvec{o}}\left({\varvec{m}}{\varvec{o}}{\varvec{l}}\boldsymbol{\%}\right)+{\varvec{b}}\) (Danyushevsky and Sobolev 1996), where \({\varvec{a}}\) and \({\varvec{b}}\) are the coefficients of the interpolation line of Fe2+/Fe3+ from spinel composition versus Fo (mol%) in olivine (Supplementary Table 2);

The original FeOt of the trapped melt has been calculated by intercepting the FeOt-MgO linear data regression in bulk rocks obtained from literature data for every volcanic system (GEOROC database, with the trend created by the SMIs of each sample. On average, SMIs display 32% depletion in iron, regardless of their size. Two individual inclusions from São Jorge experienced a loss of 38%. In Terceira volcanic systems, SMIs exhibit a continuous trend of iron depletion, with a maximum of 41%. Fe gain was observed in 14% of SMIs from Faial (up to 23%) and in most SMIs from Graciosa (up to 27%).

The calculated PEC varies from 2 to 5% in all islands, except in São Miguel, where it is 10%. Trace element compositions were calculated using the CaO content (internal standard) after the correction for PEC. Corrected major and trace element compositions are presented in Table 2 and all figures in this article. It is important to note that all subsequent data presentation and discussion are exclusively to the corrected compositions.

Table 2 Melt inclusions data

The \({D}_{Ca}^{ol-liq}\) was also computed using the CaO content of olivine and PEC-corrected SMIs. Of these values, 95% of them fall between 0.021 and 0.029 (Fig. 3d), in good agreement with literature data on olivine-liquid pairs found in basaltic liquids (e.g., Leeman and Scheidegger 1977; Beattie 1994). These values, which are characteristic of both MORB and oceanic basalts (Gavrilenko et al. 2016) indicate equilibrium conditions between the SMIs and their host. Roughly 9% of the values, represented by São Jorge data are < 0.021 while ~ 20% of the data (Pico and São Miguel) are > 0.029. Values from São Jorge, which show a reduced Fo content, are positively correlated with PEC. Therefore, the low partition coefficients might be an artefact of the calculation. The other data, especially those from Pico volcano, show a large variability in CaO (0.30–0.49 wt.%) associated with high Fo content (~ 85–88%).

Fig. 3
figure 3

Chemical variability of PEC-corrected compositions of SMIs and olivines from the Azores. a The TAS classification diagram shows the compositions of SMIs compared to bulk rock compositions from the GEOROC database ( Volcanic systems are distinguished into central volcanoes (c.v.) and fissure zones (f.z.), with the alkaline versus subalkaline field distinction following Middlemost (1975). b Evolution of SMIs revealed by FeOt/MgO and CaO/Al2O3 differentiation indices. Geochemical data of SMIs from the Pico fissure zone (Métrich et al. 2014) are shown for comparison. c Comparison between CaO/Al2O3 and the forsterite content in the host olivines. d Variability of the Ca partition coefficient between olivine and melt, calculated from SMIs, allowing the identification of antecrysts between the hosts

SMIs exhibit a dominant compositional range from basalt to trachybasalt, with a few tephrite exceptions (Fig. 3a). They are mostly alkaline, except for some inclusions from Terceira, Graciosa and Pico, which display a sub-alkaline affinity. Their alkalinity is highly pronounced, except for a few inclusions from Terceira, Graciosa and Pico, which show a sub-alkaline affinity. The major and trace element compositions of the SMIs are in the bulk rocks domain and are considered to be representative of the erupted magmas and the mafic lavas of the Azores archipelago, based on the GEOROC database (Supplementary Figs. 2 and 3).

The SMIs exhibit a wide spectrum of CaO/Al2O3 and FeOt/MgO major element compositions (Fig. 3b), while their MgO content varies from 11.6 to 4.4 wt.% (Table 2). The most evolved compositions are found in samples from São Jorge, Graciosa, Terceira and Faial islands (Fig. 3b). The decrease of CaO/Al2O3 ratios suggests clinopyroxene fractionation, alongside olivine (Fig. 3c). A reduction in Sc, Ni, and Co concentrations, with a simultaneous increase in Nb from 4 to 82 ppm, provides further evidence supporting this interpretation (Supplementary Fig. 3).

The least evolved compositions (CaO/Al2O3 = 0.77–0.99; FeOt/MgO = 1.04–1.31) are from the pyroclastic deposits from the central volcanoes in the islands of Pico and São Miguel, and are trapped in forsterite-rich olivine (Fo = 82–85 mol.%). These inclusions also show the highest contents of TiO2, and FeOt and the lowest concentration of Na2O and Al2O3 (Supplementary Fig. 2). They record a large range of Nb concentrations for a narrow range of transition elements as marked by the grey field in Supplementary Fig. 3.

The content of High Field-Strength Elements (HFSE: Zr, Nb, Hf, Ta), U, Th, the Rare Earth Elements (REE: La → Lu) and most of the Large-Ion Lithophile Elements (LILE: Ba, K, Rb, Cs) is positively correlated with each other. There is no evidence of Eu anomaly in the studied compositions. Spider diagrams of SMIs show similar trace element patterns, with each other, as well as fractionated REE, enrichment in LILE, and HFSE relative to the Primordial Mantle (McDonough and Sun 1995). Specifically, the SMIs from São Miguel exhibit the highest degree of enrichment in incompatible elements and are also the least evolved compared to the other samples. The sub-alkaline samples from Terceira are the most depleted in incompatible elements, despite they are more evolved than the SMIs from São Miguel (Fig. 3). These trace element patterns are akin to the Enriched MORB end-member (Hémond et al. 2006—Fig. 4a), although they show different degrees of enrichment in most elements, except for the Heavy REE, due to their evolutionary degree. Four SMIs from Graciosa, Terceira and Pico show distinct patterns of trace elements. These SMIs show variable degrees of depletion in HFSE and most LILE as compared to the Enriched MORB. However, exceptions are observed for Ba, Sr and Light REE (Fig. 4b).

Fig. 4
figure 4

Primordial mantle normalised trace element patterns of SMIs. a Most patterns of SMIs recall that of E-MORB (Hémond et al. 2006). b A few SMIs from Graciosa, Terceira and Pico volcanoes show intermediate patterns between E-MORB and N-MORB, with specifically Ba, Sr, Ti enrichments and Zr and Hf depletions, evidencing possible episodes of melting of deep-seated syenites occurring at local scale

The volatile element concentrations are variable. H2O ranges from 0.2 to 2.8 wt.%, Cl from 330 to 1826 ppm, F from 405 to 1778 ppm, and S from 290 to 2568 ppm (Table 2). These values fall within the same range as those measured on SMIs at Pico and São Miguel (Métrich et al. 2014; Rose-Koga et al. 2017; Turner et al. 2017; van Gerve et al. 2024). Furthermore, H2O does not reveal any correlation with Ce (Fig. 5a), La, or any other incompatible element. This absence of correlation was previously noted in the literature as a sign of water loss (Hartley et al. 2015).

Fig. 5
figure 5

Behaviour of the volatile component in PEC-corrected SMIs compositions. a Effect of proton loss/gain shown by the variability of the H2O/Ce ratio. The blue asterisk refers to the averaged value calculated for the submarine basalts (and associated standard deviation is represented by blue bars) dredged at the MAR crossing the Azores region (Dixon et al. 2002). The black line shows the value of H2O/Ce = 259 (± 11, dashed lines) measured at the Pico fissure zone (Métrich et al. 2014). Chlorine (b) and sulfur (c) contents are not related to the melt differentiation (FeOt/MgO), except for a few SMIs

The poorly evolved SMIs from Pico, Faial, and Graciosa exhibit a H2O/Ce ratio reaching ~ 248 that is comparable to the value of 259 ± 11 determined for the SMIs originating from the Pico fissure zone, which remains largely unaffected (or minimally affected) by H+ gain/loss (Métrich et al. 2014). Using this value as a reference, we infer initial water contents ranging from 1.1 to 1.9 wt.% for the least evolved SMIs compositions and up to an 85% H2O loss for the SMIs studied.

There is no correlation between water, halogens, and sulfur with the Fo content of the host (Supplementary Fig. 4). This indicates diverse volatile contents throughout the crystallisation path of olivine, similarly to the findings of van Gerve et al. (2023). Moreover, these trends show no correlation with differentiation indexes, like FeOt/MgO, on the island scale. Therefore, any connection between degassing and possible magma evolution during ascent can be excluded. This is also supported by the lack of intracrustal magma ponding stages where ascending magmas had time to rest and significantly degas (Zanon and Frezzotti 2013; Zanon 2015b; Zanon and Pimentel 2015). A few inclusions from Pico, Graciosa, and Terceira show a chlorine decrease related to the evolutionary degree (Fig. 5b).

Sulfur content differs in each volcano and at a regional scale and shows no correlation with FeOt, as illustrated in Fig. 5c. At Pico-Faial and Terceira, a decrease in sulfur follows that of MgO, indicating the presence of immiscible sulfide globules in numerous inclusions. The high sulfur values found in a few inclusions from all the islands and unrelated to differentiation, suggest the occasional entrapment of a melt enriched in sulfur, probably derived from a sulfide-enriched mantle source below the archipelago, as discussed in Waters et al. (2020).


Entrapment conditions of silicate melt inclusions

Without CO2 data, it is unattainable to determine the entrapment pressure of inclusions; nonetheless, certain observations can still be inferred. Published data on CO2-rich fluid inclusion study, carried out on the same samples (Zanon et al. 2023), reveal an area of magma accumulation located in the deep crust, between ~ 18 and 20 km below the islands of Faial, Pico, Graciosa and Terceira and ~ 26 km below São Jorge and ~ 29 for São Miguel. These depths mark the boundary between the lithospheric mantle and the deep crust (Zanon et al. 2023). Furthermore, the complete re-equilibration of densities of fluid inclusions is indicative of prolonged magma ponding. This conclusion is also supported by the absence of zoning in olivine crystals, which is consistent with their chemical equilibrium with the hosting melt(s). This leads us to infer that these crystals are most likely antecrysts—leftover crystals from prior magma differentiation (Charlier et al. 2005; Davidson et al. 2007; Jerram and Martin 2008). Their origin has been already discussed at Pico and Faial (Zanon and Frezzotti 2013) and Corvo (Larrea et al. 2013).

Fluid inclusions evidenced rapid ascent without significant ponding at a shallow crustal level (Zanon and Frezzotti 2013; Métrich et al. 2014; Zanon 2015b; Zanon and Pimentel 2015; Zanon et al. 2020). Therefore, we conclude that the olivine-hosted SMIs were trapped at such depths. Other important consequences of this rapid ascent are that the effects of late degassing, magma mixing and slow cooling would be minimal. As a consequence, minimal post-entrapment modification of SMIs (Danyushevsky et al. 2002; Portnyagin et al. 2008; Gaetani et al. 2012; Bucholz et al. 2013; Lloyd et al. 2013) is expected.

Constraints on CO2 saturation condition

The amount of CO2 required to saturate magma batches at the crust-mantle transition, as recorded by olivine-hosted SMIs, was calculated using the saturation model of Shiskina et al. (2014), together with the published fluid pressures deduced from CO2-rich fluid inclusions defined in Zanon et al. (2023), resumed in Supplementary Table 4. This results in a range of CO2 values from 0.47 wt.% in the Terceira volcanic systems, to 0.78 wt.% in the oldest São Jorge volcanic system, and 0.99 wt.% in the São Miguel volcanic systems. In all the other volcanic centres, the maximum CO2 content ranges from 0.54 to 0.58 wt.%. These values are in agreement with the recent results from van Gerve et al. (2023) for Pico Island.

Using the ratios CO2/Ba = 81.3, CO2/Rb = 991 and CO2/Nb = 515, obtained from un-degassed Atlantic MORBs and melt inclusions (e.g. Cartigny et al. 2008; Le Voyer et al. 2019), the resulting original CO2 values range from 2.0 wt.% for the Terceira volcanic systems, to 3.2–3.6 wt.% for the other islands near the MAR, up to 3.9 wt.% for the São Miguel volcanoes. These values are unrealistic for magmas that should have undergone significant degassing during prolonged ponding in deep reservoirs, and cast doubt on the validity of these ratios for calculating the dissolved CO2 in melt inclusions.

Crystal fractionation versus mantle processes

In this study, we focus on SMIs hosted in olivine crystals with a composition of Fo < 86. These show inverse relationship between CaO/Al2O3 and FeOt/MgO ratios and positive correlations observed between incompatible elements which can be attributed to a process of fractional crystallisation of both olivine and clinopyroxene (Fig. 3b, c). Samples collected at Faial, Graciosa, and Terceira show the presence of plagioclase as a stable phase. However, the absence of any Eu anomaly in the REE patterns, and the positive correlation between Sr and other incompatible elements in our data set, suggest that the crystallisation of plagioclase occurred at a later stage in the history of these magmas and did not influence the geochemistry of the studied SMIs.

On the opposite, binary diagrams between highly incompatible elements such as Zr, Nb, Hf, Ta, U, Th, Rb reveal a variability which cannot be explained by fractional crystallization or melting conditions (supplementary Fig. 3) and suggest the presence of geochemical heterogeneities. Mantle melting conditions could be revealed by comparing the variability of ratios between REE (e.g. La/Yb vs Tb/Yb), but Light REE are not perfectly incompatible and therefore their concentrations are slightly affected by the process of crystal fractionation. In plots where incompatible elements are compared with the ratio of the same element to another incompatible element, the effect of mantle melting is evidenced by a positive correlation. In Fig. 6, the lack of a positive correlation of Rb and Ba with Rb/La and Ba/Sr ratios, respectively, is consistent with crystal fractionation of mafic phases. This process was modelled for the samples from Graciosa and Faial SMIs, which have the highest concentrations of both Rb (up to 53–56 ppm; Fig. 6a) and Ba (up to 555–612 ppm; Fig. 6b). Fractional crystallisation trends were modelled by using the partition coefficients of clinopyroxene for the elements considered, i.e., DLa = 0.14–0.47; DRb = 0.011 (Vannucci et al. 1998); DSr = 0.2–0.46 (Ching-Oh et al. 1974); DBa = 0.0058 (Hauri et al. 1994). Olivine-liquid partition coefficients are, in fact, two to three orders of magnitude lower and do not alter these elemental ratios. The compositional variations of the chemical elements considered for these two islands are justified by 40–50% crystal fractionation.

Fig. 6
figure 6

Discrimination between primary and secondary processes (i.e., partial melting versus fractional crystallisation). In these plots, using elements with different degrees of incompatibility, mantle melting produces trends that intercept the y-axis near zero (black dashed lines); fractional crystallisation of olivine produces a horizontal trend; and the addition of clinopyroxene to the fractionating assemblage produces gently sloping trends. The simulated models relate only to the most evolved composition, which are found on the islands of Graciosa and Faial. The models were generated by using a range of partition coefficients from the literature. a Model of the variation in Rb/La and b Ba/Sr ratios

However, as suggested by the narrow compositional range of the host olivines, and their antecrystic nature, the SMIs could represent small volumes of intracumulus melts, whose chemical evolution could also be successfully modelled by considering the in-situ crystallisation model (Langmuir 1989). In both cases of fractional crystallisation and in-situ crystallisation, the degree of fractionation is up to 50 vol.%, close to the upper boundary of the “loose mush” definition, where crystals are jammed at the maximum packing (Bretagne et al. 2023). This element would therefore justify the formation of voluminous crystal mush layers.

A subset of the most primitive SMIs hosted in olivine Fo82-85 have variable contents of both Nb (and in other incompatible elements) and transition elements (such as Sc ≥ 27 ppm; Supplementary Fig. 3). Such trace element variations cannot be explained by simple fractional crystallisation alone, but they require processes, such as the partial melting of a heterogeneous mantle (e.g., Beier et al. 2008, 2012), or the variation in the degree of melting of a homogeneous source, or both.

As crystal fractionation may be superimposed to mantle melting processes, we used ratios between highly incompatible elements, which are not, or are only weakly, affected by olivine fractionation, to explore mantle processes (Fig. 7). Apart from a few outliers, the Nb/Zr ratio range from 0.19 to 0.29 (Fig. 7a), and the Nb/Ba and Rb/Nb ratios range from 0.10 to 0.17, and from 0.59 to 0.99 respectively (Fig. 7b, c). These geochemical characteristics are comparable with the data of SMIs published for São Miguel and Pico islands (Métrich et al. 2014; Rose-Koga et al. 2017, van Gerve et al. 2024). No correlations are observed with La/Yb (Supplementary Fig. 5), which is a monitor of the degree of melting besides the non-perfect incompatibility of La. It strongly suggests that these values depict a chemical variability of the mantle source that supplied magmas studied here and thus corroborate the heterogeneous character of the mantle beneath the Azores archipelago, according to previous studies (Madureira et al. 2011). Thus, the SMIs preserved the geochemical signature of the Azores mantle as reported for whole-rock analyses of samples from lava flows at the scale of the archipelago (e.g., Beier et al. 2007, 2008, 2012; Elliott et al. 2007; Millet et al. 2009; Hildenbrand et al. 2014; Madureira et al. 2014; Béguelin et al. 2017, among others). However, from the present dataset, we show that mantle heterogeneity can be preserved at the scale of a single island (Fig. 7), a feature that suggests a vertical (continuous or patchy) zoned mantle. This could imply that the spatial distribution of mantle heterogeneity is non-continuous and/or vertically extended.

Fig. 7
figure 7

Source heterogeneity from incompatible element variations in poorly evolved SMIs compositions, corrected after PEC. The whole geochemical dataset falls within a narrow range of incompatible element ratios. a Zr/Nb, b Nb/Ba and c Rb/Nb indicate the presence of geochemical heterogeneities at the source, with the mixing between different end members

Appropriate modelling requires the analysis of the primitive melt composition (i.e. SMIs trapped in olivine Fo ~ 88), which will be published elsewhere.

Magma mixing and source enrichment: inference from halogens composition

Chlorine behaves like K during the fractional crystallisation process as illustrated in Fig. 8a. A large part of SMIs from all volcanic systems show a constant Cl/K ratio at 0.064 ± 0.006, that fall in the domain of E-MORB (0.05–0.08; Michael and Cornell 1998). It also includes the data from the Pico fissure zone (Métrich et al. 2014). Only a few SMIs have Cl/K ratios very close to the average unaltered N-MORB on a global scale (Cl/K =  ~ 0.01; Michael and Cornell 1998). On the other hand, Cl is also positively correlated with other highly incompatible elements with Cl/Rb from 25.5 to 11.0) (Fig. 8b). The absence of a single correlation with K, and with all the incompatible elements, supports the mixing between melts generated from sources characterised by different degrees of enrichment in Cl and other incompatible elements.

Fig. 8
figure 8

Halogen behaviour in SMIs compositions corrected after PEC. a In binary plots against K the Cl/K = 0.064 characterises some SMIs from the Terceira, Graciosa and Pico (both the central volcano and the fissure zone) volcanic systems. The grey area represents the range of ratios found for E-MORBs, while the red dashed line indicates the averaged value for N-MORB. b In plot versus LILE, chlorine is dispersed and included between Cl/Rb ~ 25.5 and Cl/Rb = 11.0. b Plot of Cl versus K. c The values of Cl and F show different behaviour due to the different variability of these two chemical species. (d) The variability of F/Nd and K of poorly evolved SMIs from the Azores (MgO > 6 wt.%) is limited compared to that of other volcanic archipelagos (Canary, Hawaii and Galapagos) and Iceland. However, the F/Nd ratios of these SMIs are systematically higher than the average of the MORB basalts erupted by the segments of the MAR crossing the Azores archipelago

Very few SMIs show extreme Cl enrichments (up to 0.39 wt.%; Table 2). Similar Cl enrichments have already been observed in the Loihi Seamount (Hawaii), some MORB glasses (Michael and Cornell 1998; Kent et al. 1999) and SMIs from Raivavae and Rapa, Austral Islands (Lassiter et al. 2002). At these sites, the Cl-enrichment was related to the assimilation of Cl-rich brines and/or brine-saturated melts (Stroncik and Haase 2004) released by subducted sediments and hydrothermally weathered oceanic crust. However, we discard this possibility for the Azores, because Cl-enrichment is limited to a few and MgO-poor SMIs of Terceira and Graciosa. Not all the coeval SMIs from these two islands show the same feature, and no Cl-rich FIs have been yet found, among the thousands analysed. This Cl-enrichment could be due to the occasional interaction with fluids released at the base of the hydrothermally weathered oceanic crust, where magmas possibly ponded, as also suggested by the values of δ7Li and δ11B in olivines from the Azores (Genske et al. 2014). It should be noted that the few SMIs with this extreme Cl enrichment are from volcanoes on the islands of Graciosa and Terceira, which are close to each other, located on the same spreading system and with similar crust thicknesses. Therefore, this brine-magma interaction should be limited to peculiar conditions existing only under these two islands.

The more primitive SMIs (MgO > 6 wt.%) show a range of fluorine content from 0.04 to 0.15 wt.%, which is not correlated with the variations in compatible elements monitoring fractional crystallisation of mafic phases, and with parameters monitoring melting conditions (i.e., La/Yb and Tb/Yb). The two halogens are not correlated with each other, suggesting that the process responsible for the Cl variability did not affect F (Fig. 8c).

The ranges of the F/Nd ratio and the K content discriminate well between volcanic systems from different geographical areas (Fig. 8d). The elements in this ratio behave similarly during the melting process due to similar melt-solid partition coefficients (Workman et al. 2006). Therefore, the overall high F content of the Azores SMIs is not due to late-stage enrichment, but is a characteristic derived from the mantle source. The Azores magmas are characterised by high K contents and F/Nd ratios ranging from 15.1 to 45.4. The lowermost values which are from São Jorge and a few SMIs from Terceira, Graciosa and Pico, straddle F/Nd = 18.84, characteristic of the lavas erupted by the MAR segments at the intersection with the Azores archipelago and resulting from the averaged F of 389 ± 131 (Schilling et al. 1980) and Nd = 20.6 ± 2.5 (Gale et al. 2013). Therefore, the fluorine variability at the Azores seems to be related to the mixing of two mantle endmembers: a MORB component and a F-enriched mantle source. A comparison with the composition of lavas from the Hawaiian and Galapagos archipelagos and Iceland (GEOROC database) shows that the very variable K contents of the Azores SMIs are higher anyway and the F/Nd range is similar. On the other hand, in a comparison with the nearby Canarian lavas, the two Atlantic archipelagos show the same F/Nd range, while the K content of the Canarian lavas is overall comparable to the highest content of the Azores SMIs, but the variability is reduced.

Concluding remarks

Magmas erupted in the Azores during the last 40 ka contain olivine crystals hosting fluid inclusions or melt inclusions (SMIs), or both. The geochemical dataset of olivine-hosted SMIs reported here from six islands east of the MAR shows that SMIs suffered minimal post-entrapment modifications, and allows us to discuss magmatic processes beneath the Azores archipelago.

The cross-check with published data on coexisting CO2-rich fluid inclusions, and on textures of the host minerals led us to propose that SMIs were trapped in crystal from mush layers in the lower crust (~ 18 to 29 km beneath different islands), at or nearby the crust-mantle boundary. At these depths CO2 values required to saturate magma batches range from 0.47 to 0.99 wt.%, varying across different volcanic systems. These values are significantly lower than those (4.5–4.7 wt.%) recently found in basanites erupted at Fogo volcano at Cape Verde and La Palma at Canary Islands (Burton et al. 2023; Lo Forte et al. 2024), but also in French continental basanites (4.3 wt.%; Buso et al. 2022). Pre-eruptive degassing significantly reduced the CO2 content in melts, prior to SMIs trapping and intracrustal ascent, as suggested by the CO2 values retrieved from ratios with Ba, Rb and Nb incompatible elements. This extensive degassing could be the responsible for the formation of carbonic fluid inclusions commonly found in olivine and clinopyroxene antecrysts (Zanon et al. 2023).

Fractional crystallisation process of olivine and further olivine + clinopyroxene well explain the chemical evolution recorded by SMIs, while ratios between highly incompatible trace elements reveal mantle compositional heterogeneity. Such a heterogeneity has been demonstrated by high resolution studies of bulk rocks (e.g. Beier et al. 2007, 2012; Beguelin et al. 2017). Here we found mantle heterogeneity at the scale of one single magmatic system.

The process of fractional crystallisation of up to 50% of mafic phases should be the responsible for the formation of voluminous ultramafic cumulates (dunites, wehrlites, and clinopyroxenites) and crystal mushes at the base of the crust.

Both trace elements and halogens demonstrate the interaction between melts produced by melting of a MORB-like source and a more enriched source, probably of a deeper origin.