The nature of the Karoo mantle source region from the perspective of olivine in the Luenha picrites: an oxygen isotopic and chemical study

Abstract

The Karoo large igneous province has been divided into rift zone and basin-related groups, with picrites from the Luenha river, Mozambique, representing an end-member of the latter. New O isotope, major and trace element data for olivine have been combined with MELTS crystallisation modelling to deconvolve compositional diversity associated with magma differentiation from source-derived heterogeneity. Three olivines populations have been discerned as follows: (1) the ‘main trend’, which records crystallisation from a variety of magma compositions; (2) the ‘low Cu trend’, which is inferred to represent xenocrysts or antecrysts; and (3) the ‘high CaO’ olivines, which record polybaric crystallisation of a primitive, little fractionated magma. The trace element variability in olivine phenocrysts relates partially to sampling of different parts of the same overall magma transport and storage systems, and partly to heterogeneity of parental magmas and their mantle sources. When the measured δ¹⁸O_olivine values have been converted into δ¹⁸O_melt values, the mean δ¹⁸O_melt values for the ‘main trend’ and ‘low Cu’ groups are indistinguishable from each other (5.7 ± 0.1‰, 2σ); however, the mean δ¹⁸O_melt value of 6.1 ± 0.1‰ for the ‘high CaO’ group is distinctly enriched. These data record source heterogeneity and suggest contributions from two mantle sources, one with elevated δ¹⁸O, and another with more ‘typical’ mantle δ¹⁸O. Combining these data with previously reported trace element and Nd and Sr isotope data support derivation from a mantle source similar to non-chondritic bulk silicate earth, but with minor contributions (1–2% for the enriched magmas) from a recycled sedimentary component. This points to the importance of a primitive mantle source for the basin-related successions in Karoo.

Introduction

The Karoo large igneous province (LIP) is the result of extensive magmatism between 189 and 174 Ma (Ivanov et al. 2017; Jourdan et al. 2007b; Luttinen et al. 2022, 2015; Moulin et al. 2017; Muedi et al. 2022; Svensen et al. 2012) that produced compositionally diverse continental flood basalts (CFB), basaltic-picritic shallow intrusive complexes as well as subordinate silicic volcanics outcropping across southern Africa, the Falkland Islands and East Antarctica (Fig. 1; Eales et al. 1984; Galerne et al. 2008; Jourdan et al. 2007a; Luttinen and Furnes 2000; Manninen et al. 2008; Marsh et al. 1997; Mitchell et al. 1999; Riley et al. 2004; White 1997).

Fig. 1

Map of the Karoo Large Igneous Province (a) and the Luenha locality in greater detail, with sampling locations (present day coordinates), shown in (b); modified after Turunen et al. (2019). In a the map and inset show Gondwana reconstructed at 180 Ma. Note the marked picrite suites: Luenha, Mwenezi, Vestjella and Ahlmannryggen

Like many other LIPs (e.g., Bellieni et al. 1984; Callegaro et al. 2013; Haissen et al. 2021), the Karoo LIP is typified by lateral geochemical heterogeneity that probably reflects variable mantle source compositions. Several provincial scenarios have been applied to assess the geochemical structure of the Karoo LIP and its sources ever since the recognition of this systematic compositional variability by (Cox et al. 1967). Most scenarios are based on the distribution of highly enriched (high-Ti) and mildly enriched (low-Ti) CFB compositions (e.g. (Jourdan et al. 2007a; Sweeney et al. 1994). Luttinen (2018) recently pointed out, however, that a tectonic distinction between CFBs associated with the incipient Africa-Antarctica triple rift and CFBs found in the main Karoo, Kalahari, and Zambezi basins likely more accurately reflects the mantle source differences. Particularly, the igneous rocks related to the Karoo rift zone and the Karoo basins (main Karoo Basin, Kalahari Basin, and Zambezi Basin) are geochemically distinguished from each other by their abundances of Nb relative to Zr and Y. The notably uniform low-Ti CFBs in the basins typically have positive ΔNb (Nb-undepleted), where the ΔNb of a sample reflects its position relative to the linear Icelandic mantle plume array on a log–log plot of Nb/Y versus Zr/Y (Fig. 2; Fitton et al. 1997). In contrast, the high-Ti CFBs and related low-Ti CFBs of the fossil rift zone (conjugate African and Antarctic rifted margins and associated failed rifts; Fig. 1) have negative ΔNb values (Nb-depleted). This rift vs. basin division corresponds to the “South Karoo” and “North Karoo” sub-provinces of Luttinen (2018), and we favour it to avoid confusion with the historic geochemical North vs. South provincial scheme of Cox et al. (1967).

Fig. 2

Geochemical diversity of the Karoo LIP, using the compilation of Luttinen (2018) and data from Turunen et al. (2019) and Harris et al. (2015). a Distribution of the basin-related Karoo (including Luenha picrite samples) and rift-zone Karoo rocks in a log–log plot of Nb/Y vs Zr/Y. The plotted reference line represents the lower bound of the Icelandic mantle array, defining the boundary between positive and negative ΔNb (Fitton et al. 1997). Variation associated with partial melting is dominantly parallel to this array. Mantle endmembers and crustal compositions are shown for reference: DMM, depleted MORB mantle (Salters and Stracke 2004); PM, primitive mantle (McDonough and Sun 1995); NC-BSE, non-chondritic bulk silicate Earth, (Jackson and Jellinek 2013); CC, bulk continental crust (Taylor and McLennan 1985); and GLOSS, global subducted sediment (Plank and Langmuir 1998). Key samples for this study have been individually labelled. The D- and E-picrites represent parental magmas with strong influences from depleted mantle and enriched, subcontinental lithospheric mantle (Luttinen 2018). b Karoo compilation Nd and Sr isotope data. The elevated ⁸⁷Sr/⁸⁶Sr of most Luenha samples reflect the influence of assimilation in the crust. The arrow marked ‘assimilation’ shows the overall direction of two alternative AFC curves modelled by Turunen et al. (2019). The EMII composition shown here is based on the theoretical ‘pure’ EMII lava isotopic composition of Workman et al. (2004), age corrected to 180 Ma using their estimates of the EMII source Rb/Sr and Sm/Nd. c Plot of ΔNb vs εNd, showing that the Karoo LIP spans a quadrilateral with the Luenha samples as one potential endmember. The plotted average ΔNb value (0.14 ± 0.16, 1σ, n = 99) for EMII basalts (Fitton 2007) is within uncertainty of the ΔNb = 0.19 for the ‘pure’ EMII lava. d In a plot of Th/Nb vs La/Nb, the Luenha samples are positioned at the edge of the Karoo array. Sample T042 corresponds closely with NC-BSE and the other samples are spread roughly linearly towards CC and GLOSS

A variety of source compositions have been associated with the formation of CFBs and related rocks in LIPs such as Karoo (Ellam 2006; Harris et al. 2015; Heinonen et al. 2018; Jourdan et al. 2007a; Yu et al. 2017), the discussion of which is linked to mantle heterogeneity. Many geochemical ‘flavours’ of the mantle have been proposed based on the global diversity of basalts and related rocks (Jackson and Jellinek 2013; Salters and Stracke 2004; Stracke et al. 2005; Willbold and Stracke 2006). These, along with crustal components, constitute a broad set of potential source compositions tapped by the processes thought to be responsible for LIP magmatism. The ΔNb of mantle-derived rocks has been shown to be a source characteristic that varies among the different ‘flavours’ of the mantle (Baksi 2001; Fitton 2007; Fitton et al. 1997). This underscores the importance of the distinction between low-ΔNb rift zone and high-ΔNb basin-related CFBs in Karoo (Luttinen 2018).

Contamination with lithospheric material during magma ascent hampers the characterisation of the purported Nb-depleted and Nb-undepleted mantle sources. High-Mg rocks (such as picrites) can be useful probes of the mantle sources of continental LIPs, because the primitive magmas from which they are derived would typically be less overprinted by crustal contamination than other lithologies. Furthermore, phenocrysts in picrites have the potential to provide additional insights into parental magma compositions and source regions because early crystallising phases in such magmas can capture a primitive magmatic signature whereas assimilation by crustal material may have overprinted the bulk-rock compositions. These insights include the relationship between source composition and source lithology.

In the Karoo LIP, studies of high-Mg rocks have demonstrated that enriched and depleted sources in the lithospheric and convective upper mantle have contributed to the production of the Nb-depleted magmas of the Karoo rift (Ellam 2006; Harris et al. 2015; Heinonen et al. 2010). Heinonen et al. (2018) used olivine major and trace element and O-isotopic compositions along with bulk-rock Sr–Nd-Pb-Os isotope compositions of an Antarctic picrite suite (D-picrites in Fig. 2) to decipher the mantle sources of rift-related magmas. The wide range of δ¹⁸O_vsmow (4.4–7.5‰,) was interpreted to reflect variable amounts of subduction-related materials in a depleted mantle peridotite source. In comparison, Harris et al. (2015) and Howarth and Harris (2017) associated the olivine major and trace element compositions, bulk-rock Sr–Nd-Pb isotope compositions, and the high δ¹⁸O_vsmow (6.0–6.7‰) in the olivine and orthopyroxene phenocrysts of the rift-related Tuli and Mwenezi picrites (E-picrites in Fig. 2) with an enriched, pyroxenite-rich (eclogite-bearing) mantle source.

In contrast to an upper mantle origin for the rift-associated Nb-depleted rocks, the Nb-undepleted low-Ti lavas and intrusions of the main Karoo, Kalahari and Zambezi basins probably had different sources (Luttinen 2018). A recent study of the low-Ti Luenha picrites from Mozambique has demonstrated that a mantle source with a signature broadly similar to primitive mantle (PM) or non-chondritic bulk silicate Earth (NC-BSE) could also be involved and may be related to the origin of the Nb-undepleted Karoo basin lavas (Luttinen 2018; Turunen et al. 2019). In particular, the Sr–Nd isotope composition of the least contaminated sample of these picrites is very similar to PM (Fig. 2b), while the Nb-Zr-Y systematics show a closer affinity for NC-BSE (Fig. 2a and c). The characteristics of the Luenha picrites suggest possible derivation from a mantle plume source. Firstly, primitive mantle-like compositions have been associated with deep mantle plume sources (Jackson and Carlson 2011; Jackson and Jellinek 2013; Willhite et al. 2019). Secondly, the high-Mg whole rock and olivine compositions and trace element modelling indicate of high mantle potential temperatures (> 1500 °C) and very high degrees of melting (≥ 20%) for the Luenha source, suggestive of a mantle plume impinging the base of the lithosphere (Heinonen et al. 2022; Turunen et al. 2019). Moreover, if the Luenha picrites represent the source of the basin-related CFBs, a plume origin is compatible with the high volume and compositional uniformity (Galerne et al. 2008; Luttinen 2018; Marsh et al. 1997; Neumann et al. 2011) and rapid formation (possibly < 1 Ma; Ivanov et al. 2017; Luttinen et al. 2022; Svensen et al. 2012) of the basin-related CFBs as well as the position of the main Karoo Basin above present South Atlantic hotspots in Mid-Jurassic plate tectonic reconstructions (Svensen et al. 2018).

Our study focuses on the nature of the mantle source of the PM-affinity low-Ti Luenha picrites. The Luenha picrites represent a unique Nb-undepleted endmember composition in the Karoo LIP and they may well sample a significant mantle source of the voluminous low-Ti magmas of the Karoo basin sub-province. The existing bulk-rock and olivine data are suggestive of a peridotite source with PM-like isotopic and trace element composition, but several critical uncertainties render the nature of the Luenha source ambiguous. We use new data on the chemical and O isotopic compositions of their olivine as windows into magmatic processes and mantle sources. Interpretation of the major element data is aided by MELTS modelling, which allows magmatic processes such as fractionation and mixing to be identified. This informs the interpretation of the trace element data, which illuminate further details of the magmatic system and probe signatures of source lithology. We combine interpretation of the olivine compositions with published bulk-rock chemical and Sr–Nd isotope compositions to further explore the possible roles of crustal contamination and magma mixing in the magma transport and storage system, and mantle source heterogeneity. Furthermore, we discuss the significance of our results within the context of the Karoo LIP and its potential upper mantle and plume sources.

Samples and methods

Samples

This study focuses on olivines from a subset of the picrites documented by Turunen et al. (2019). The Luenha picrites are olivine-porphyritic rocks from localities around the Luenha River, Mozambique (Fig. 1b). Samples T026B, T026C and T026D represent vesicle-poor cores of three upper lava flow units, T026E represents the vesicular upper portion of an underlaying unit, while T026F and T026G represent the vesicle-poor cores of the lowermost lava flow unit of the exposed succession (full sample names and coordinates are provided by Turunen et al. 2019). Although not exposed as one continuous outcrop, field observations suggest that the units represented by these samples are part of a ~ 60 m thick lava succession. Geographically and geochemically associated with this succession are two fields of picritic boulders, interpreted by Turunen et al. (2019) to represent in-situ weathering of lava flows, from which samples T039 and T042 were collected ca. 25 km NNW from the Luenha outcrops. Euhedral to subhedral olivine is the main phenocryst in these samples, occupying 10–50% of the assemblage.

While the olivine in the upper lava flows is substantially altered, fresh olivine is common in the samples that are the focus of this study: T026E, T026F and T042. Olivine phenocrysts dominate sample T042 (Fig. 3c) and, to a lesser extent, sample T026F (Fig. 3b and d). Less olivine occurs in sample T026E, consistent with a general decrease in abundance of olivine phenocrysts up section. Nevertheless, many large fresh olivine grains are present in T026E (Fig. 3a). Additionally, euhedral plagioclase phenocrysts are observed in T026E, but rare in T026F and absent from T042. Spinel phenocrysts are present in all three of these samples. Based on major element analyses of olivine, Turunen et al. (2019) identified that T026E and T042 have the most primitive olivine (and thus focussed on these for estimation of parental and primary melt compositions), while T026F has a more diverse olivine population. In addition, sample T042 is the most primitive in terms of Sr–Nd isotope data. Thus, the subset chosen are well-suited to characterise the olivine systematics of the suite and elucidate mantle source signatures.

Fig. 3

Images of the olivines from the key samples. a Grains from T026E prior to mounting in epoxy. b Example of the euhedral olivine grains from a petrographic thick section of T026F. c A composite image showing the thick section of T042. Olivine occurs as phenocrysts of various sizes, but alteration along cracks and grain surfaces obscures the outlines. d A composite image of the thick section of T026F shows the outlines of olivine grains more clearly, but grains also have cracks

Sample preparation

Petrographic thick sections for samples T026F and T042 were prepared at the Environmental and Mineralogical Laboratories (HelLabs), Department of Geosciences and Geography, University of Helsinki. These were examined with an optical microscope at the Finnish Museum of Natural History (LUOMUS) to select target locations for analysis and photomicrographs were taken to document the samples (Supplementary Material). Following rock-crushing and gravity and magnetic separation of samples T026E and T042 at HelLabs, the freshest olivine grains were hand-picked under binocular microscope at LUOMUS. Grains were mounted in epoxy (mounts 1896 & 1972) along with the San Carlos olivine reference material, polished and coated in gold at the NordSIMS laboratory of the Swedish Museum of Natural History. The mounts were mapped with reflected light microscopy to select targets prior to analysis. A small selection of grains for samples T026E and T026F had been included in mount 1235 for a previous study, and the procedures used are described in Heinonen and Fusswinkel (2017) and Heinonen et al. (2018), which reported data for other samples included in the mount.

Electron microprobe analysis

Mineral compositions from petrographic thick sections and grain mounts 1896 & 1972 were obtained via wavelength-dispersive electron microprobe (EMP) analysis on the CAMECA SX100 at the Geological Survey of Finland (GTK). Samples were carbon coated (following removal of the gold coating from grain mounts) prior to analysis. EMP analyses were used to verify the identity of mineral domains in mounts analysed by SIMS. Analyses were performed with accelerating voltages, beam currents and beam diameters of 20 kV, 60 nA and 5 μm, respectively, for olivine and 15 kV, 20 nA and 1 μm for plagioclase and pyroxene. Standards were measured as unknowns and relative error for majors (concentrations > 10 wt%) was better than 3% and about 5% for minors (concentration range 1–10 wt%). A small subset of analyses of olivines from samples T026E and T026F (mount 1235) were derived from unpublished data from the same analytical session (using a JEOL JXA-8600 at HelLabs) as data presented by Heinonen and Fusswinkel (2017; analytical methods provided therein).

LA-ICPMS trace element analysis

Trace element analyses of olivines were performed via laser ablation inductively-coupled plasma mass spectrometry (LA-ICPMS) using an Agilent 7900 s quadrupole system coupled to a Coherent GeoLas Pro MV 193 nm laser ablation system at HelLabs, University of Helsinki. The analyses were performed in two sessions: thick sections in November 2019 and grain mounts 1235, 1896 & 1972 in June 2020. Instrumental settings and data collection and processing procedures followed the methods of Heinonen and Fusswinkel (2017).

Laser ablation spot sizes of 60 and 90 μm in diameter were used with a fluence of 7 and 10 J/cm² (for the thick section and grain mount analytical sessions, respectively) at 10 Hz. A set of 19 isotopes (²⁴ Mg, ²⁷Al, ²⁹Si, ⁴³Ca, ⁴⁴Ca, ⁴⁵Sc, ⁴⁹Ti, ⁵¹ V, ⁵²Cr, ⁵⁵Mn, ⁵⁶Fe, ⁵⁷Fe, ⁵⁹Co, ⁶⁰Ni, ⁶²Ni, ⁶³Cu, ⁶⁶Zn, ⁸⁸Sr, ¹⁸⁹Os) was analysed with the dwell time for each mass set at 10 ms, and a 40 s washout prior to and after each analytical period (40 s) was used to ensure sufficient background levels. Analyses of olivines and secondary reference materials measured as unknowns (SRM NIST612 and BHVO2G) were bracketed by measurements of the SRM NIST610 standard reference material every 10–15 analysis. The instrument was tuned to ThO/Th ratios of < 0.3% and U/Th ratios of ~ 100%, doubly charged cations were tuned to < 0.3% using SRM NIST610, and SRM NIST612 as unknown. Trace element concentrations were internally standardised using Si contents determined by EMP for the same grain domains.

Repeated analyses of BHVO-2G during all analytical sessions used here (n = 15) corrected using SRM NIST610 deviate less than 5% relative to accepted values published in GEOREM except for Cu, Sr***** and Ti (< 7%), and Sc, Ni and Mn (< 12%). The relative reproducibility standard deviations for all elements is < 5% for BHVO2G over the analytical period. The long-term accuracy using SRM NIST612 (n = 339) deviates less than < 2% from the published GEOREM values for all isotopes used here. SRM NIST 612 has a relative reproducibility standard deviation of less than 3% for all elements except for V (5%) from the published GEOREM values.

SIMS O isotope analysis

Oxygen isotope analyses were performed using the Cameca IMS 1280 multicollector large geometry ion microprobe at the NordSIMS laboratory during two analytical sessions. The instrumental setup and analytical protocols follow those of Heinonen et al. (2018) for olivine, with more details provided in Heinonen et al. (2015) for the similar analysis of zircon. All data were normalised relative to δ¹⁸O_vsmow = 5.30‰ for the San Carlos olivine, which was measured throughout each analytical session, bracketing every 6 sample analyses. Although batch-related O-isotopic heterogeneity of San Carlos olivine is a known issue (Starkey et al. 2016), the composition of the batch used at NordSIMS has been confirmed by laser fluorination analyses (Heinonen et al. 2018). The value of 5.3 ± 0.4‰ (2σ, n = 4) reported by Heinonen et al. (2018) is consistent with δ¹⁸O_vsmow values determined by various conventional and laser fluorination techniques in published literature: e.g., 5.14 ± 0.22‰ (2σ, n = 5), 5.27 ± 0.04‰ (n = 21), 5.28 ± 0.06‰ (n = 4), 5.28 ± 0.18‰ (n = 15), 5.33 ± 0.32‰ (n = 9) and 5.4 ± 0.2‰ (n = 20) (Ahn et al. 2012; Gurenko et al. 2011; Harris et al. 2015; Widom and Farquhar 2003; Yu et al. 2017). Drift corrections were applied to the datasets for each session based upon minimising the session external uncertainty of the standards. For the 1st session, analyses of San Carlos olivine have a standard deviation of 0.11 and for the 2nd session the San Carlo Olivine analyses have a standard deviation of 0.13. These values are propagated onto the within run uncertainty for each analysis.

Results

Major and trace element data

The newly acquired major element data are provided in the Supplementary Material. Domains targeted for SIMS analysis that proved to be plagioclase or clinopyroxene have been excluded. Combining the new olivine data with those of Turunen et al. (2019) shows that the olivines across this suite of picrites exhibit coherent geochemical variations, with most samples recording overlapping ranges of Fo contents (Fig. 4). Mostly the centres of grains were targeted. The variation within individual grains is typically restricted (Fo varies by 0.1–3.5 within grains in which > 3 locations were analysed), and variations generally do not strongly relate to position relative to grain centre (i.e., those analyses potentially representing olivine “rim” domains usually do not record markedly lower Fo than grain “cores”).

Fig. 4

Olivine compositions by sample using the combined data of this study and those from Turunen et al. (2019). Element oxides plotted against Fo% show an overall coherency of the olivine populations from sample to sample, particularly in c where trends are well defined and co-linear. a With the exception of scattered outliers, the CaO contents of olivines are relatively uniform across a wide range of Fo contents, but decreases with decreasing Fo in sample T039, and CaO extents to elevated values with little change in Fo in sample T042. b Al₂O₃ contents are similarly uniform. In d, the olivines show an overall positive correlation between NiO and Fo contents, but with an increase in NiO diversity with decreasing Fo content

The LA-ICPMS data for Al, Ca, Co, Cr, Mn and Ni are consistent with the EMP data. The CaO, MnO and Al₂O₃ data from EMP analysis are utilised here in favour of the equivalent LA-ICPMS data. The full dataset is available in the Supplementary Material. In plots of trace element concentration versus Fo content the variations within samples generally pool to form coherent trends for the whole dataset (Fig. 5). Enrichment with decreasing Fo content is recorded in the olivine data for Sc, V, Co and Cu (except in a subset of grains discussed further below). An overall depletion occurs for Cr, although the pattern is irregular, and Ni shows both depletion and increased diversity with decreasing Fo content. The concentrations of Ti are low but variable and lack a correlation with Fo content.

Fig. 5

Olivine compositions by sample using the combined data from thick sections and grain mounts. In a–c The ranges of Sc, Ti and V contents in samples T042 and T026F are very similar at any particular Fo content, but T026F shows greater diversity because of its greater Fo range. In these plots, the diversity of sample T026E overlaps significantly with that of the combined other two. d These samples also show similar overlap with respect to Cr contents, although T042 is somewhat distinct in its large diversity with very little Fo variability. E Sample T026E encompasses essentially all of the Ni-Fo diversity of the other samples. F The distribution of olivine Cu concentrations relative to Fo is distinctive in that the data are split between a main trend that extends to high Cu contents and is evident in all three samples and a group of low Cu compositions only distinguishible in T026E

O isotope data

The complete O isotope dataset is provided in the Supplementary Material. The grain averages for all reliable analyses are listed in Table 1 along with summary statistics for these data. The individual analyses are shown in Fig. 6. For all samples the standard deviations of analyses are larger than the standard deviation of the reference material. For the 1st session, analyses of San Carlos olivine have a standard deviation of 0.11 and MSWD of 0.63, whereas these statistics are 0.36 and 6.8 for T026E. For the 2nd session the San Carlos olivine analyses have a standard deviation of 0.13 and an MSWD of 0.67, whereas these statistics are 0.26 and 2.5 for the analyses of T042 from that session. Thus, the olivine populations in both samples exhibit O isotope heterogeneity. The samples display approximately symmetric dispersion of O isotope data, as indicated by the similarity between the means and medians (Table 1). If the average δ¹⁸O of individual grains are used to estimate the central tendency and dispersion in the samples there are only minor differences in the statistics (Table 1).

Table 1 Oxygen isotope data, including single analyses or grain averages, followed by summary statistics

Fig. 6

Comparison of the Luenha olivine O isotope compositions. a Individual analyses shown with their 2σ external uncertainties, and the sample mean values shown as bands bracketed by their 2 s.e. uncertainties. b Individual analyses shown in comparison to the O isotope composition range for African Karoo olivine (E-picrites) reported by Harris et al. (2015), the range for the D-picrites reported by Heinonen et al. (2018) and the range and average for mantle peridotite olivine reported by Mattey et al. (1994). Note the change in scale

The number of analyses per grain range from 1 to 7. Within grain variability is discussed in greater detail in the Supplementary Material, but the salient observations are noted here. Examination of olivine grains for which at least four analyses were made shows that the within-grain O-isotopic heterogeneity is ~ 50% of the total sample heterogeneity. Sample T042 exhibits generally lower within-grain variability than T026E. Typically, in olivines that preserve intra-grain O isotope heterogeneity, there is no clear core to rim variation.

The analyses from mount 1235, acquired as part of the study of Heinonen et al. (2018), included three from TO26E, with δ¹⁸O of 5.05 ± 0.21‰, 5.08 ± 0.21‰ and 4.93 ± 0.21‰ (1σ ext.). These are within uncertainty of the mean for T026E from mount 1896, and result minor lowering of the sample mean and median (Table 1). That session also included five analyses of olivine from T026F, with δ¹⁸O ranging from 4.67 ± 0.22‰ to 5.51 ± 0.21‰, and having a mean of 5.08 ± 0.27‰ (2σ). These analyses are included in the ‘Main trend’ and ‘Low Cu’ groups according to their trace element compositions in the Discussion and for the calculation of the melt δ¹⁸O averages (Table 1).

Discussion

In the following discussion, we firstly distinguish patterns within the major element data then use crystallisation modelling to decipher the different causes of these patterns. This delineates the key mechanisms for magmatic differentiation involved in the Luenha system and provides a framework for understanding the trace element data, in the first instance, and, ultimately, the O isotope data.

Major and trace element constraints on magmatic differentiation

Major element modelling

From the major element data three main trends occur with respect to the Fo number of the olivine domains: (1) the enrichment of FeO and MnO coupled to depletion of NiO with decreasing Fo while CaO and Al₂O₃ contents (for all samples except TO42) show scatter but no correlation with Fo; (2) the large variation in CaO contents in T042 olivine with relatively little change in Fo, creating a separate trend in Fig. 4a that intersects that formed by the other samples; and (3) the general broadening of the NiO concentrations with decreasing Fo such that the NiO-Fo plot (Fig. 4d, see also Fig. 5e) shows a spread of compositions between the upper- and lower-bound trends. While enrichments of Mn and Fe and a depletion of Ni in olivine are expected for magmas dominated by olivine crystallisation, the shape of the NiO-Fo trends can be affected by pressure and the appearance of clinopyroxene and other phases on the liquidus, and the CaO contents of olivine are sensitive to pressure and the water content of the magma (Hole 2018; Matzen et al. 2017; Putirka et al. 2018).

To investigate the causes of these variations, crystallisation modelling was undertaken using the rhyolite-MELTS phase equilibria software (Ghiorso and Gualda 2015; Gualda et al. 2012). The basis of the models was the “preferred primary melt” composition of Turunen et al. (2019), calculated on the basis of bulk-rock and olivine major element compositions presented therein for sample T042. Turunen et al. (2019) suggested that their calculated primary melt composition is consistent with melt extraction pressure of 1.5–4.0 GPa and would likely have a H₂O concentration of ≤ 1 wt%. Accordingly, modelling tested compositions with 0 wt%, 1 wt% and 1.5 wt% H₂O at pressures between 1 bar and 2GPa (higher pressures were deemed inappropriate as the 1.5 and 2 GPa models proved inconsistent with the olivine data). A NiO content of ~ 0.09 wt% for the primary melt was used (see Supplementary Material). Models without Ni, with slightly lower NiO contents, and with different MnO and CaO contents were also run. A summary of the starting compositions and conditions is provided in Table 2.

Table 2 Parameters for the 7 starting parental melt compositions for MELTS modelling

The modelling produces a variety of trends of NiO and CaO versus Fo (Fig. 7) the shapes of which vary with H₂O content and pressure according to the effects of these parameters on the appearance of clinopyroxene and orthopyroxene on the liquidus. Orthopyroxene crystallises earlier at high pressure compared to low pressure, resulting in marked variations in the olivine crystal lines of descent (CLDs) with changing pressure. The appearance of orthopyroxene on the liquidus results in a sharp turning point in the NiO-Fo CLD (Fig. 7b). Importantly, the lack of orthopyroxene phenocrysts in the samples argues against the viability of high-pressure models with a prevalence of orthopyroxene. Thus, only models for crystallisation below ~ 0.4 GPa are consistent with our olivine data. Pressure has a similar effect on the point at which clinopyroxene starts to crystallise. Onset of clinopyroxene crystallisation causes a shallowing of the NiO-Fo trends and a turning point of the CaO-Fo trends. As illustrated by Fig. 7, the upper bound of the NiO-Fo variation in the samples is well matched by low-pressure equilibrium crystallisation models (particularly 0.2 GPa). The more steeply positively-correlated lower bound of the data is only reproduced by fractional crystallisation models (Fig. 7d, particularly at 0.05 GPa and 1 bar), in which olivine composition evolves to lower Fo and NiO before clinopyroxene appears on the liquidus compared to equilibrium models. In fact, a similar variation is produced in the modelling of Hole (2018) for olivine crystallisation at 1 atm pressure.

Fig. 7

Comparison of the Luenha olivine compositions with olivine crystal lines of descent derived through MELTS modelling. a-c MnO, NiO and CaO versus Fo for equilubrium crystallisation models in which the same Ni-bearing initial magma compositions vary only in H₂O content (compositions 2, 4 & 5 in Table 2) and model pressures. For each initial compositions (C2, 5 & 7), stars mark the earliest primitive olivine for one pressure case. In a MnO enrichment is suppressed by decreases in pressure, except for very low pressure (e.g., 1 bar), and is also sensitive (if less systematically) to changes in H₂O content. Pressure changes have small effects on the MnO content of the earliest olivine (as shown with arrows). b The evolution of the CLDs for NiO are strongly influenced by the effect of pressure on the crystallisation of orthopyroxene (opx). The effect of pressure on the composition of the earliest olivine is slightly more pronouced. The 0% H2O, 1 GPa model is off-scale for this plot (opx-in early and earliest olivine NiO > 0.7 wt%). c Note that descreasing pressure increases olivine CaO contents in the models and shift the appearance of clinopyroxene (cpx). d, e Luenha olivine compositions compared to CLDs for equilbrium and fractional crystallisation models and lines representing the various compositions of the first olivine that would crystallise from melts formed by mixing the initial magma composition with the liquid from a more evolved stage (stars mark olivines from the endmembers). The lines ‘0.2 GPa mixing 1’ and ‘0.2 GPa mixing 2’ represent the olivines associated with mixing two different evolved liquids from the 1% H2O (composition 4), 0.2 GPa equilibrium model with the initial magma. The lines ‘0.05 GPa mixing 1’ and ‘0.05 GPa mixing 2’ similarly represent mixing of the composition 7 (see Table 2) initial magma with evolved liquids (both with 5 wt% MgO) from equilibrium and fractional crystallisation, respectively, at 0.05 GPa

Although the modelling supports low pressure crystallisation on the basis of the Fo-NiO-MnO relationships, the CaO contents of the olivines are notably lower than those predicted by low-pressure models. Two important observations can be made from the modelling with respect to the CaO of the olivine CLDs: (1) the CaO content of the earliest-crystallising olivine increases with decreasing pressure (along with a subtle increase in Fo) for any particular starting magma composition (consistent with the findings of Stormer 1973); and (2) the turning point of the CLD caused by clinopyroxene crystallisation is delayed at lower pressures, shifting the peak olivine CaO concentration upwards and to lower Fo values. These observations suggest that the elevated CaO contents exhibited by olivines from TO42 and their variability within a limited Fo range are due to polybaric crystallisation of olivine after the magma had evolved slightly to a composition in equilibrium with Fo₈₈ olivine.

The remaining olivine compositions have CaO concentrations more in line with equilibrium crystallisation at higher pressure (e.g., the 1 GPa model); however, the general shape of even the high-pressure model trends is not reflected in the Luenha olivine data. Rather, the olivine compositions suggest equilibrium with melts that change very little in CaO (and in fact Al₂O₃) content with decreasing MgO.

The CaO data can be reconciled with the low-pressure models if the olivine compositions represent olivines in equilibrium with melts along mixing lines between primitive and evolved melt compositions. These have been tested with MELTS modelling of a variety of mixed melt compositions. As illustrated in Fig. 7e, the olivine compositions associated with mixing lines between the parental magma and ~ 5 wt% MgO melts from the 0.05 GPa equilibrium and fractional crystallisation LLDs bracket many of the measured CaO contents. A similar mixing line based on the 0.2 GPa equilibrium crystallisation model expands the ranges of CaO contents captured. Additionally, the equivalent MnO and NiO variations differ little from the CLDs, thus they, also, are plausible matches to the data. That the NiO-Fo variations are spread between equilibrium and fractional crystallisation trends suggests variable degrees of equilibration between olivine and melt. This could represent a situation in which segregation of crystals from melt is not efficient, thus creating an imperfect equilibrium crystallisation regime (a hybrid between fractional and equilibrium crystallisation).

The two mixing models used do not extend to Fo contents as low as those exhibited by olivines from sample T039; however, the slight decrease in CaO with decreasing Fo shown by those olivines is paralleled by the CLDs for the 0.2 and 0.05 GPa models, suggesting that these olivines record the most evolved stages of equilibrium crystallisation in this system. Only the most forsteritic olivines (Fo_90–91) are not explained by the mixing model. Their MnO-FeO-MgO-NiO compositions are matched by models at pressures 1–0.05 GPa, but only the highest pressure models match their CaO contents, suggesting that these are remnants of an early stage of crystallisation at depth.

Trace element relationships

On the basis of the major element data and modelling, the olivine compositions can be divided into those with high CaO and Fo > 88 (mostly from sample TO42) and those with lower CaO that show no correlation with Fo (except perhaps at the low Fo end of the dataset, with the subtle decrease in T039). The latter group accounts for most of the geochemical variability, and it exhibits coherent trends for most elements versus Fo content (Figs. 4, 5 and 8); however, a subset of this group is distinct with respect to Cu vs Fo. While most compositions fall on a trend of increasing Cu with decreasing Fo, roughly half of the grains from sample T026E fall in a restricted range of Cu concentrations below 5.5 ppm and lack correlation with Fo content (Fig. 5f). Thus, we define three geochemical groups within the olivines:

• The main trend, which is represented in all samples and has CaO mostly between 0.3 and 0.35 wt% and a negative correlation between Cu and Fo contents (5–83 ppm Cu and Fo_76-91).

• The high CaO group, which constitutes most of the olivine in sample T042 and to which only a few olivines from other samples have been assigned, is characterised by steep increase in CaO content away from the main trend in concert with a subtle increase in Fo from ~ 88 to ~ 89

• The low Cu group, which is exclusive to sample TO26E and is distinguished from the main trend by lower Cu at a given Fo content (thus, given the Cu-Fo correlation of the main trend, the low Cu compositions can only be distinguished for Fo < 88)

Fig. 8

Olivine compositions divided into ‘main trend’, ‘high Ca’ and ‘low Cu’ groups. a, b The CaO and Al contents distinguish the high Ca group from other olivines in the samples. c The groups have essentially the same Ti diversity, without any trend. d-h The high Ca group is consitent with the trends in the rest of the dataset in Sc-Ti-V-Co-Ni-Mn-Fo space. i While the high Ca olivines are consitent with the main trend for Cu vs Fo contents, the low Cu group is clearly distinct

The geochemical relationships in the dataset are henceforth discussed according to these groups.

The main trend olivines display negative correlations between most analysed trace elements (Co, Co, Mn, Sc & V) and Fo content (Fig. 8). In contrast, Ca, Al and Ti show no correlation with Fo in this group. In the main trend, Ni is positively correlated with Fo content (Fig. 8f). As is Cr, although this is largely due to a small number of analyses domains at the extremes of the Fo range (note the lack of correlation among T026F in Fig. 5d). Among the high CaO group, many trace elements (Al, Co, Sc, Ti, V) show large variations within the group’s limited Fo content range (Fig. 8), similar to the variation in CaO concentrations; however, the concentrations of trace elements do not correlate with CaO. This implies a cause for the trace element variability other than the pressure variation shown to explain the CaO range. Since the high CaO group are at the primitive end of the wider dataset and the variability of Al, Co, Sc, Ti and V for this group are similar to the variability these elements exhibit in the main trend at a given Fo content (particularly at high Fo), it is likely that the scatter represents chemical heterogeneity in the parent magmas. In contrast, the variability in Cu, Mn and Ni in the highest Fo content olivines is small relative to the range of concentrations produced with decreasing Fo content; however, for these elements the concentrations in high CaO olivine are consistent with those of the main trend. Thus, trace element data support the assumption made in the major element modelling that most of the olivine compositions (and hence the different samples in which they are found) can be derived from the same (or at least very similar) parental magmas.

In general, the low Cu group plots inside the main trend on element vs Fo plots, suggesting consanguinity with the other groups, yet the group’s departure from the main trend indicates a key difference in their origin. For the Cu concentrations of this group to be buffered to low values suggests coprecipitation of olivine and a phase in which Cu is compatible, such as a sulphide, yet sulphides are not observed in these samples. We posit, therefore, that the low Cu group olivines did not form in the same magma as the main trend olivines. This could reflect a xenocrystic origin. Nevertheless, since their compositions are otherwise similar, it is possible that these xenocrysts originate from a similar parental magma to the other Luenha olivines, but represent a distinct magma batch that underwent sulphide precipitation or separation of an immiscible sulphide liquid. Alternatively, their origin may be antecrystic—that is, they may have formed within the same plumbing system as the main trend olivines. The lack of sulphides in the samples suggests that the main trend precipitated from S-undersaturated magmas, which would dissolve sulphide phases present in the material from which the low Cu olivines are derived. This would not lead to sulphide saturation if the consumed material was relatively S-poor due to effective separation of olivine from the sulphide phase (which might be more likely in the case of an immiscible sulphide). The low Cu group plots along the lower bound of the Ni vs Fo plot (Fig. 8f), which is consistent with fractional crystallisation at low pressure (< 0.2GPa) in the modelling presented above (Fig. 7d); however, sulphide fractionation would also deplete Ni, and the models do not account for how strongly this might influence the trends.

Olivine compositions as indicators of source lithology and heterogeneity

Chemical tracers of mantle characteristics

The chemical compositions of olivine phenocrysts (particularly MnO/FeO and Ni*FeO/MgO) have been widely used as indicators of the proportions of pyroxenite and peridotite in the source regions, based on the effect of mineral assemblage upon element partitioning during partial melting (Gurenko et al. 2009; Sobolev et al. 2007).

In the case of the Luenha picrites, all three olivine geochemical groups record similar values of the pyroxenite indicators, MnO/FeO and Ni*FeO/MgO Fig. 9). The MnO/FeO ratios in olivine are not correlated with the Fo contents, suggesting that all samples preserve the chemical signature without modification due to crystallisation. In contrast, the Ni*FeO/MgO of the olivine are correlated with Fo content, albeit weakly (Fig. 9b); therefore, only the most forsteritic olivine domains (Fo > 86%) are used as source indicators in the following discussion. If these two parameters are plotted against each other, no correlation is observed in any of the three groups or in the dataset as a whole (Fig. 9c).

Fig. 9

a, b Pyroxenite source indicator ratios, MnO/FeO and Ni*FeO/MgO, plotted against forsterite content, Fo, with olivine compositions divided into ‘main trend’, ‘high Ca’ and ‘low Cu’ groups. c Pyroxenite indicators plotted against eachother, showing the variation between peridotite- and pyroxenite derived olivine in the dataset of Sobolev et al. (2007) and the distribution of Luenha olivine compositions relative to the peridotite-derived field. The three Luenha olivine groups show a large degree of overlap for pyroxenite source indicators in Mg-Fe-Mn-Ni space

In the model of Sobolev et al. (2007), these parameters have a negative correlation caused by the effect of variations in pyroxenite vs peridotite contributions to the source upon melt composition (Fig. 9c); therefore, the lack of any correlation between them in our dataset implies that variability of MnO/FeO and Ni*FeO/MgO is not caused the presence of pyroxenite in the source region. Different equations for the mass fractions of pyroxenite (X_py) using one or both of the parameters have been developed (Gurenko et al. 2009; Sobolev et al. 2008, 2007). Using the MnO/FeO, Ni*FeO/MgO and combined parameterisations give X_py ranges of -0.60 to 0.92, 0.26 to 0.59 and -0.06 to 0.75, respectively, for the various analysed olivine domains (these include only olivine for which both EMP and ICPMS data are available). These parameterisations have been called into question by Heinonen and Fusswinkel (2017). Given that the variation in X_py estimates is large, strongly dependent on choice of parameterisation and includes several negative mass fractions (associated with olivine compositions outside the calibration range), it is inappropriate to use these parameters to quantify even an average mass fraction of pyroxenite in the mantle source. Instead, it is possible only to say that the data permit the presence of a minor pyroxenite component in the source. Since the data strongly overlap with the peridotite field of Sobolev et al. (2007) in the MnO/FeO vs Ni*FeO/MgO plot (Fig. 9c) it is likely that pyroxenite constitutes only a very minor lithology in the source.

Matzen et al. (2017) suggested that the global anticorrelation between the Ni and Mn of mantle-derived olivine could alternatively reflect the pressure and degree of melting of a peridotitic source, without the need for a pyroxenitic component as proposed by Sobolev et al. (2007). Matzen et al. (2017) showed that decreasing the pressure of melting would result in decreased NiO and increased MnO of Fo₈₉ olivine crystallised at near surface pressures. Furthermore, varying degrees of melting (~ 5–30%) at a particular pressure partially contributes to this trend and partially produces scatter in the trend. The dispersion in the Luenha olivine NiO contents at Fo₈₉ covers the 1–4.5 GPa melting pressure range of Matzen et al. (2017). Thus, mixing of melt batches formed by various degrees of melting of peridotite at different pressures to produce the Luenha parental could partially explain the heterogeneity observed in the olivine, and the pyroxenite fraction estimates. Two observations suggest, however that an additional source of heterogeneity (such as spatial diversity in source composition) is required: (1) the dispersion in the NiO and MnO contents in the Luenha olivine are not correlated; and (2) the dispersion in MnO in particular is wider than that which can be accounted for by the experiments and modelling of Matzen et al. (2017).

Oxygen isotope heterogeneity in the mantle source

In order to evaluate the oxygen isotope characteristics of the mantle source of the Luenha picrites it is necessary to estimate the oxygen isotope composition of the parental magmas. Thus, the melt O isotope compositions in equilibrium with those measured in olivine must be calculated. Fractionation between olivine and liquid (the difference between their δ¹⁸O) is expressed as follows:

$$1000\times \mathrm{ln}{\alpha }_{\mathrm{liquid}-\mathrm{olivine}} =\frac{A\times {10}^{6}}{{T}^{2}}$$

where temperature is in Kelvin and the A factor is dependent on the magma composition. The combination of literature constraints (Eiler 2001; Zhao and Zheng 2003) and MELTS modelling used to calculate appropriate fractionation factors are described in the Supplementary Material. The fractionation factors range from 0.41 to 0.89 and result in melt δ¹⁸O values from 4.95 to 6.78‰. As illustrated in Fig. 10, the main trend, high CaO and low Cu groups overlap with each other, with the most enriched compositions being recorded by two olivines from the high CaO group and the most depleted being recorded in 4 olivines from the low Cu group (note that one of these olivine domains is from a grain with δ¹⁸O_melt from 5.16 to 6.33‰).

Fig. 10

Oxygen isotope compositions measured in situ from olivine and recalculated to melt compositions. While a slight correlation between measured δ¹⁸O and Fo appears in the main trend olivines (a), there is no correlation in any groups for the calculated melt δ¹⁸O vs the Fo of olivines (b). In c, d, the data show no correlation between melt δ¹⁸O and Mg-Fe-Mn-Ni pyroxenite source indicators

Previously, Turunen et al. (2019) associated the wide range of initial Sr isotopic compositions of the Luenha picrites with crustal contamination. Several lines of evidence suggest the variable oxygen isotopic compositions of the analysed olivines represent source heterogeneity rather than the result of crustal contamination in the magma transport and storage systems. First, none of the three groups exhibit a correlation between the δ¹⁸O_melt values and the Fo content of the olivines that would be the expected as a result of contamination (Fig. 10b). Second, sample T026F exhibits high bulk-rock Th/Nb and ⁸⁷Sr/⁸⁶Sr indicative of crustal contamination (Fig. 2), but its olivine phenocrysts have generally lower δ¹⁸O values (i.e., less contaminated) than those in T042, which indicates that the contamination responsible for the bulk-rock signature took place after the generation of olivine phenocrysts. This is also supported by a study of plagioclase phenocrysts in the Luenha picrites that showed that the radiogenic initial Sr isotopic compositions typify phenocryst rims and the groundmass (Aaltonen et al. 2018). Third, there is not any other significant correlation between the O isotope compositions and olivine geochemical characteristics (Fig. 10c and d). Thus, the variability in δ¹⁸O_melt values reflects the original oxygen isotope heterogeneity of the magmas, as inherited from the mantle source. The means and medians of the δ¹⁸O_melt for the three groups are reported in Table 1, which shows that the high CaO group has a slightly but distinguishably higher central tendency than the nearly identical main trend and low Cu groups, despite the extensive overlap in oxygen isotope compositional ranges.

Previous research has linked the oxygen isotopic heterogeneity recorded in suites of olivines to variations in the contributions of pyroxenite and peridotite lithologies in the source of the magmas (Day et al. 2012; Gurenko et al. 2011, 2012). As previously discussed, on the basis of MnO/FeO and Ni*FeO/MgO the olivine data presented here could be consistent with a small pyroxenite component in the source of the Luenha picrites. Correlations between these parameters and the δ¹⁸O_melt values would be expected if variations in the amount of pyroxenite were the cause of the observed oxygen isotopic heterogeneity; however, for neither MnO/FeO nor Ni*FeO/MgO are such correlations present in any of the three groups or dataset at a whole (Fig. 10c and d). Thus, the observed oxygen isotopic heterogeneity does not show a direct link with a possible pyroxenite component in the source. Overall, we consider that the Luenha picrites were derived from a mantle source that was mainly peridotitic, which is also compatible with recent bulk-rock trace element modelling which constrained a possible pyroxenite component in the Luenha source to be less than 10 wt.% (Heinonen et al. 2022).

Petrogenesis of the Luenha picrites

Magma differentiation

The olivine geochemical record of the Luenha picrites shows that two mechanisms are principally responsible for the differentiation of magma in this system: crystallisation without complete equilibration of olivine with their host magma and mixing of primitive and evolved magmas. We envisage crystallisation starting at depth (at an approximate pressure of 1 GPa, and in equilibrium with Fo_90-91 olivine) but continuing either during ascent (as in the case of sample T042) or after ascent of an only slightly evolved magma (in equilibrium with Fo_88-89 olivine) to a shallower level (0.05–0.2 GPa). Crystallisation in this shallow magma chamber occurs with varying degrees of melt-crystal segregation (and hence varying degrees of equilibration) and is punctuated by recharge events, which bring fresh primitive magma batches that mix with the more evolved resident magma. The compositions of the olivine depend on the degree of differentiation of the resident magma at time of recharge, the extent of re-equilibration of olivine phenocrysts from the evolved magmas with the mixed melts and the degree of hybridisation between primitive and evolved melt, producing the diversity of compositions in the main trend. Further differentiation of the mixed magmas prior to eruption would add to the geochemical variation.

In this model each sample captures slightly different parts of the magmatic system’s history. Individual olivine grains generally only record relatively small portions of this history: the minimum, average and maximum Fo variation observed within grains (with > 3 locations analysed) are 0.1, 0.9 and 3.5 units, respectively, compared to the whole suite range of ~ 14 units (Fo_77-91). On the basis of the olivine major element data, samples T026F and T026G capture very similar aspects of the system. Each preserves a variety of compositions produced by mixing, with a slight skew towards more evolved compositions in T026G, and some record of the CLDs in T026F (rare high CaO associated with crystallisation before mixing, and low CaO associated with crystallisation from more evolved melt after mixing). Sample T026E captures the most diverse range of the magmatic history, preserving similar compositions to T026F as well as the earliest crystals formed at higher pressure. In addition, T026E appears to include a xenocrystic or antecrystic population (the low Cu compositions). Since this sample records the mixing-derived olivine, which formed at low pressure, it is inferred that assimilation of the low Cu olivine occurred after mixing, with little else contributing to the sample before eruption. This would explain why there is no indication of re-equilibration of these olivines. Sample T039 records only the late stages of crystallisation (Fo_76-79), while T042 records mostly crystallisation of near parental melt (albeit at multiple pressures).

Source heterogeneity

Some of the geochemical diversity observed is not attributable to these differentiation processes, rather representing heterogeneity original to the parental magmas and their mantle sources. We interpret olivine grains that belong to the low-Cu group as possible xenocrysts and exclude them from the discussion on mantle sources.

Correction of the oxygen isotope compositions measured in olivine for liquid-olivine isotope fractionation shows that the magmas were characterised by heterogeneity in δ¹⁸O above the typical variability of mantle rocks. Compared to the δ¹⁸O of 5.5 ± 0.4‰ (2 s.d.) for a variety of mantle peridotites (Mattey et al. 1994), the Luenha melt δ¹⁸O values range from 4.95 to 6.78‰, with means of 5.74 and 6.11‰ for the main trend and high CaO groups, respectively. This suggests a source region in which an O isotope enriched component is mixed with a composition having more typical mantle δ¹⁸O values. Clear correlations between the oxygen isotope compositions and element concentrations and pyroxenite source indicators might have been expected if the enriched component occurred as discrete lithological heterogeneities within the mantle source, but such correlations are not observed. Several processes can lead to decoherence of such relationships and potentially explain the scatter in these geochemical parameters.

Since the observed heterogeneities record the state of the parental magmas at the onset of crystallisation, the heterogeneity inferred in the source must be of a magnitude and scale to survive hybridisation processes during melting in the mantle, melt extraction and magma ascent.

Studies of mixed peridotite-pyroxenite melting scenarios have pointed out the roles of diffusion rates, solid–liquid partition coefficients, and the dimensions of heterogeneities and melting columns. Kogiso et al. (2004) demonstrated that decoupling between the major element and isotopic signatures of a pyroxenitic component in a mixed peridotite-pyroxenite source could arise due to the different rates associated with diffusion of trace elements such as Sr, Nd, Pb and Os and for Mg-Fe exchange. Furthermore, this decoupling is dependent on the size of lithological heterogeneity, such that as the size of pyroxenite domains decrease the geochemical signature imparted upon melts would tend towards peridotite while varying degrees of heterogeneity could be preserved for different elements and isotopes. Stracke and Bourdon (2009) explored the geochemical signatures associated with melting of a mixed pyroxenite-peridotite mantle column. The modelling suggested that mixing during melt extraction will result in different degrees of homogenisation of the inherent source diversity for various trace element and isotope ratios, depending on the incompatibility of the elements. While ratios of highly incompatible elements were effectively homogenised by the extraction and mixing of melts, with increasing compatibility melts preserved or magnified the source heterogeneity. A further consequence of the melting and extraction processes modelled was that the strength of correlations between various geochemical signatures is greater for melts extracted over a wider range of depths than for extraction over a smaller interval. When applied to the Luenha picrite data, these scenarios seem to favour the presence of relatively small pyroxenitic heterogeneities and extraction of melts generated over a relatively narrow pressure range.

Although the potential for generating large melt compositional diversity from a particular mantle source region is well documented (Gurenko and Chaussidon 1995; Oliveira et al. 2020; Stracke and Bourdon 2009), the extent to which this diversity is preserved in mantle-derived rocks is variable. In the olivines from some Etendeka picrites and the Karoo D-picrites studied by Jennings et al. (2017) it was demonstrated that the trace element variability is dramatically reduced compared from that expected from fractional melting of the mantle, as a result of significant mixing prior to the onset of olivine crystallisation. They highlighted that mixing could take place during transport through the mantle and within deep crustal sills, but also could be associated with trapping of instantaneous melts within the source region before channels developed to allow transport from the mantle to the crust. In contrast, other studies have emphasised concurrent mixing and crystallisation (CMC), which tends to reduce compositional variability with decreasing host olivine Fo content (Maclennan et al. 2003; Shorttle 2015).

The compositional heterogeneity of the Luenha olivine compositions may be compared with the investigation of Maclennan et al. (2003) into olivine and clinopyroxene from Theistareykir, Iceland, which document evidence for CMC. One feature in common is heterogeneity in olivine O isotope compositions and a lack of correlation between δ¹⁸O and Fo; however, the Luenha samples lack the reduction of heterogeneity during olivine crystallisation, evident in the Theistareykir samples. In the Luenha olivine many elements show a consistent degree of variability across most of the Fo range, particularly CaO, Al₂O₃, MnO, Sc, V and Cu (Figs. 4, 5 and 8), while Ni variability broadens (see Major Element Modelling section). It is important to note that the reduction in compositional variability associated with CMC is countered by the introduction of additional variability when the magmatic system experiences recharge (Maclennan et al. 2003). Thus, the mixing of evolved magmas with recharge of primitive magmas in the Luenha system may not only account for the shapes of major element trends, but also for the maintenance mantle-derived compositional diversity. A more thorough investigation of compositional heterogeneity within the Luenha magmas and their mantle source could be undertaken through a melt inclusion study, involving a wider range of trace elements than have been measured in the Luenha olivines. This would permit better comparison with the studies of Jennings et al. (2017) and Neave et al. (2018), for example.

What is the geochemical flavour of the Luenha mantle?

A complication to the origin of the Luenha picrites is the possibility that they represent hybrid magmas from relatively enriched mantle (such as EMI, EMII and HIMU) and relatively depleted mantle domains (DM, PREMA). We assess this by comparing isotopic and trace element characteristics of the Luenha picrites with these domains.

Applying a liquid-olivine isotope fractionation of 0.5‰ to the compositions reported by Eiler (2001; the same fractionation as used therein for comparing N-MORB glasses with OIB olivine) gives δ¹⁸O from 5.8 to 6.6‰ for EMII magmas. The range of δ¹⁸O for N-MORB reported by Eiler (2001) is 5.4 to 5.8‰, and most other OIBs examined in that study exhibit similar ranges (again assuming a liquid-olivine fractionation of 0.5‰). When the potentially xenocrystic low Cu olivine compositions are excluded from our dataset the melt δ¹⁸O range is 5.27–6.78‰, which is comparable to the N-MORB and EMII range. The similarity between the oxygen isotope compositions of the high CaO group (the dominant component of the T042 olivine cargo) and EMII suggests there may be some similarity between the origin of the Luenha and EMII sources. Various studies have highlighted that δ¹⁸O-⁸⁷Sr/⁸⁶Sr of EMII basalts require incorporation into the mantle source of a component originating in the crust (Eiler 2001; Eiler et al. 1997; Widom and Farquhar 2003; Workman et al. 2008). Eiler et al. (1997) showed that the δ¹⁸O-⁸⁷Sr/⁸⁶Sr of EMII basalts could be accounted for by the incorporation of 2–6% sediment with δ¹⁸O of 15–25‰ into a depleted mantle source with δ¹⁸O of 5.5‰ (see also Eiler 2001). Similarly, in a more detailed study of Samoan EMII lavas, Workman et al. (2008) estimated that addition of 2–3% of a component with δ¹⁸O of 20–25‰ (the upper range of clastic marine sediments) to the most depleted lavas (δ¹⁸O_olivine of 5.3‰, corresponding to 5.8‰ for the melt) could account for the composition of the most enriched EMII lava (δ¹⁸O_olivine of 5.7‰). For the oxygen isotope compositions of the Luenha high-Ca olivine group the calculation is essentially the same as that of Eiler et al. (1997); however, additional information is necessary to distinguish between the different possible sources of the typical mantle δ¹⁸O of 5.5‰ (i.e., for Luenha, the mantle endmember need not be DMM).

The similarity between the Luenha picrites and EMII with respect to O δ¹⁸O-⁸⁷Sr/⁸⁶Sr systematics does not extend to Nd isotope compositions. Turunen et al. (2019) illustrated that the Luenha picrites (and particularly the most primitive, least contaminated bulk-rock sample, T042) are trace element depleted relative to OIBs and have notably higher εNd and ΔNb than EMI and EMII rocks (Fig. 2c). Indeed, because crustal compositions have ΔNb similar to or lower than enriched mantle and DMM, few mantle domains have sufficiently high ΔNb to account for the Luenha picrites, with or without mixing of an additional component. The ΔNb values of PM, HIMU and NC-BSE are ~ 0.2, 0.3 and 0.4, respectively (Fitton 2007; Jackson and Jellinek 2013; McDonough and Sun 1995), while the values for Luenha are 0.4–0.6. Since the low U/Pb (< 0.2) of T042 argues against the involvement of a HIMU source (Turunen et al. 2022, 2019), a primitive mantle composition such as the NC-BSE composition of Jackson and Jellinek (2013) for the mantle source of the Luenha picrites is favoured. Such non-chondritic, primitive reservoirs in the mantle are associated with the high ³He/⁴He rocks of Baffin Island, which preserve olivine δ¹⁸O values of 5.0–5.2‰ (Willhite et al. 2019), consistent with typical mantle compositions (Eiler 2001). Thus, we posit derivation of the Luenha magmas from a mantle reservoir similar to NC-BSE, and incorporation of small amounts of an enriched oxygen component (such as sedimentary material) into this source region to generate the elevated δ¹⁸O values of the high-Ca group compared to the main trend olivines.

To assess this model we have calculated mixing lines between NC-BSE and three crustal endmembers: bulk continental crust (CC) of Taylor and McLennan (1985), global subducting sediment (GLOSS) of Plank and Langmuir (1998), and the terrigenous sediments (TS) of Ben Othman et al. (1989) (Fig. 11). For CC, the Sr and Nd isotope compositions follow the estimates of Carlson and Boyet (2008). The isotope compositions of the endmembers were age-corrected to 180 Ma, using either the reported ⁸⁷Rb/⁸⁶Sr and ¹⁴⁷Sm/¹⁴⁴Nd isotopic ratios or estimates from the Rb/Sr and Sm/Nd trace element ratios.

Fig. 11

Luenha samples compared to other Karoo samples and various mantle and crust endmember compositions and three alternative mantle-crust mixing lines. In addition to the compositions used in Fig. 2, we have plotted the lower and upper crust (LC & UC) compositions estimated by Heinonen et al. (2010) for Karoo and the terriginous sediments (TS) of Ben Othman et al. (1989). a Uncertainties in the ΔNb values have been plotted for some of the endmember compositions to illustrate the compatibility of the Luenha compositions with NC-BSE and mixtures therewith, compared with involving other mantle endmembers such as DMM. b Note that, although mixing curves 1 and 2 pass through the samples with relatively high ⁸⁷Sr/⁸⁶Sr (thought to reflect later assimilation; see Fig. 2), only curve 3 passes through the least contaminated Luenha samples. c The δ¹⁸O values represent the average melt values estimated for the samples, melt derived from the displayed mantle sources (as discussed in text) a consistent bulk δ¹⁸O of 25‰ for the three crustal endmembers

To check that these mixing lines can be reconciled with the unusually high ΔNb values of the samples, the ΔNb and uncertainties of the endmembers were constrained in a few different ways (see Supplementary Material). These estimates of uncertainties are not necessarily symmetric; therefore, where necessary, the 1 s.d. uncertainties are quoted using a non-parametric equivalent (the 64.2% interval about the median). With these constraints, the ΔNb value of 0.4 ± 0.4 for NC-BSE is within uncertainty of those of the Luenha samples (0.4–0.6), as are any mixtures of NC-BSE with modest amounts (< 10% given the O isotope considerations) of either GLOSS (ΔNb = 0 + 0.1/− 0.2) or TS (ΔNb = 0 + 0.4/− 0.3), as illustrated in Fig. 11a.

The isotopic data provide tighter constraints. While the εNd values of the Luenha samples are quite uniform (Fig. 11b), the spread of bulk-rock ⁸⁷Sr/⁸⁶Sr ratios is thought to reflect varying degrees of assimilation during magma differentiation (Aaltonen et al. 2018; Turunen et al. 2019); therefore, the composition of T042 (the least contaminated sample) is the focus of the following calculations. With a δ¹⁸O_bulk of 25‰ for TS, 1–2% addition of this sediment composition to NC-BSE produces ⁸⁷Sr/⁸⁶Sr 0.70391–0.70474, εNd 3.2–0.9 and δ¹⁸O_melt 5.9–6.1‰, which are consistent with the T042 ⁸⁷Sr/⁸⁶Sr 0.704096, εNd 1.4 and δ¹⁸O_melt 6.1‰ (Fig. 11). In contrast, with GLOSS as an endmember, achieving the T042 ⁸⁷Sr/⁸⁶Sr would require a mere 0.5% sediment addition (because GLOSS has a more radiogenic and Sr-rich composition than TS), which would not impart the enriched O isotope signature and would produce a εNd value too high (~ 5.2) for T042. The influence of CC as the crustal endmember is intermediate between that of GLOSS and TS: a 1% addition of CC would approximately match the Sr isotope composition of T042 but not the Nd and O isotope compositions.

In the context of the Karoo LIP, a lithospheric mantle domain with an enriched O isotope signature has been identified as the source of the E-picrites (Harris et al. 2015; Luttinen 2018). A similar domain might be an alternative source for the elevated δ¹⁸O of sample T042. This would require a significant component of the E-picrite lithospheric mantle source because it has only a mildly higher δ¹⁸O than the Luenha pictrites (Fig. 6). However, the E-picrites have such a distinctly low ΔNb that any mixing with the NC-BSE would generate ΔNb too low to be compatible with the Luenha picrites. While this precludes the E-picrite source, other enriched lithospheric domains cannot be discounted; however, a similar problem will arise for any source with a distinctly Nb-depleted character.

Since the isotopic composition of T042 is consistent with mixing of merely 1–2% terrigenous sediment into a NC-BSE mantle source, the magmas from which the Luenha samples with lower, ‘typical mantle’ O isotope compositions are derived are inferred to have a source representing essentially pure NC-BSE. This implies that the radiogenic isotope composition of the parental magmas for most of the Luenha samples is not the T042 bulk-rock composition, but rather the marginally less enriched NC-BSE composition.

Implications for upper mantle and plume sources in the Karoo LIP

The presently available data on the Karoo LIP are compatible with distinction between two principal subsystems. On one hand, an extensive continental triple rift system formed across the Karoo LIP. Magmatism in the Karoo rift probably commenced already at ca. 189 Ma, well before the two main phases of Karoo CFB magmatism that occurred at 182–183 Ma and 178–181 Ma, and continued until ca. 172 Ma (Luttinen et al. 2022). Geochemically, the rift-related magmatism was notably diverse, producing rocks ranging from highly alkaline nephelinites to low-Ti and high-Ti basalts and picrites, as well as late phase rhyolites (Eales et al. 1984; Galerne et al. 2008; Jourdan et al. 2007a; Luttinen and Furnes 2000; Manninen et al. 2008; Marsh et al. 1997; Mitchell et al. 1999; Riley et al. 2004; White 1997). The picrites and basalts of the rift system are typified by Nb-depleted compositions. Detailed studies of rift-related picrites, critical evaluation of the age data, and recent geochemical modelling suggest that (1) initial magmatism was derived from convective depleted upper mantle that contained recycled subduction-related silicate and fluid components and (2) melting of strongly enriched lithospheric mantle became increasingly voluminous in the triple junction region during the main phase of magmatism (Heinonen et al. 2022; Luttinen et al. 2022).

On the other hand, voluminous low-Ti magmatism with Nb-undepleted characteristics spread across the main Karoo, Kalahari, and Zambezi basins during the brief main phase of Karoo magmatism at 182–183 Ma. Subtle but systematic variations in trace element and isotopic ratios in the basin-related CFBs have been associated with crustal contamination (Neumann et al. 2011). Within the context of the diverse rock compositions of the Karoo LIP, the low-Ti Luenha picrites stand out as the high ΔNb endmember necessary to explain the relative Nb enrichment that characterises the basin-related Karoo low-Ti lavas and intrusions (Luttinen 2018; Turunen et al. 2019). By extension, our results on the mantle source region of the Luenha picrites presently provide the most reliable insight into the source that likely contributed at least partially to many basin-related Karoo magmas. On the basis of the geochemical and isotopic data for the picrites and their olivine cargo, and the calculations discussed above, this source region overall has a signature consistent with the NC-BSE composition of Jackson and Jellinek (2013) and is largely peridotitic. However, it contains heterogeneously dispersed small contributions from an enriched component (possibly terrigenous sediments).

Whether or not this source was a part of a mantle plume remains uncertain. Nevertheless, a plume origin for the Nb-undepleted Karoo basin CFBs has been argued on the basis of the high volumes and uniformity of the magmas, the short duration of magmatism, the palaeoposition of the main Karoo Basin above present plumes of South Atlantic Ocean, and the compositions of the least-contaminated CFBs (see Luttinen 2018). Plume sources for the Luenha picrites are supported by the association between NC-BSE mantle reservoirs and mantle plumes (Jackson and Carlson 2011; Jackson and Jellinek 2013; Willhite et al. 2019) and the high temperatures estimated for the primary melts (Turunen et al. 2018). Furthermore, geochemical modelling suggests the HREE patterns of the low-Ti Luenha picrites and Karoo basin CFBs require very high degrees of partial melting and high temperatures under ~ 100 km-thick lithosphere, which have been associated with a scenario involving impingement of a mantle plume progressive thinning of the lithosphere (Heinonen et al. 2022). Our new results support these arguments: (1) the most forsteritic olivines discovered in the samples (Fo91) correspond to those predicted to be in equilibrium with the high temperature primary melts; (2) olivine compositions are compatible with a NC-BSE mantle source. Thus, the geochemical portrait of the mantle source of the Luenha picrites provides a plausible insight into the nature of the hypothesised large mantle plume head underneath Mid-Jurassic Karoo LIP.

Conclusion

Studying the olivine populations of the Luenha samples provides new insights into the magmatic system in three principal respects:

• Major and trace element analyses reveal three groups of olivines (one of which is likely xenocrystic) within the Luenha picrites, representing contributions from various magmatic processes (crystallisation driving magma evolution, magma mixing, polybaric crystallisation, assimilation). The trace element variability in olivine relates partially to sampling of different parts of the same overall magma transport and storage systems, and partially to heterogeneity of parental magmas.

• The olivine compositions do not show evidence for contamination of magmas by material from the continental crust, in contrast to the whole rock, groundmass and plagioclase Sr isotope data. This has implications for the timing of contamination. The olivine data provide a framework for future investigations of magma differentiation and assimilation in the Luenha system.

• The compositions of olivines highlight parental magma heterogeneity, not apparent from whole rock data alone, and reveal information about the Luenha magma mantle source. The olivines record a signature of a predominantly peridotitic source with only a weak, ambiguous chemical signal for a pyroxenitic component, and this is unrelated to the variation in O isotope compositions. While the main trend olivines have δ¹⁸O_melt values (5.7 ± 0.1, 2σ) within range of typical mantle O (Eiler 2001), the high-CaO olivines have elevated δ¹⁸O_melt values (6.1 ± 0.1, 2σ).

The insights from olivine augment our understanding of the origin of Luenha samples and the Karoo LIP more broadly:

• The δ¹⁸O_melt values of the main trend are consistent with the non-chondritic bulk silicate Earth origin inferred from bulk-rock trace element and radiogenic isotope evidence by Turunen et al. (2019), but elevated values of the high-CaO group indicate involvement of an additional component. The bulk-rock εNd-ΔNb-⁸⁷Sr/⁸⁶Sr composition of T042 and the O isotope compositions of the high-CaO olivines hosted in T042 can be explained by 1–2% addition of terrigenous sediment to a NC-BSE mantle domain. Thus, the olivine data support the view that the parental magmas of the Luenha picrites likely tap an ancient mantle domain mostly unmodified by addition of recycled components.

• Given the Luenha source appears to be an endmember within the Karoo compositional diversity, these findings imply that the Karoo LIP may have tapped a relatively primitive mantle domain,thought to be an important component in mantle plumes (Boyet and Carlson 2006; Jackson and Carlson 2011).