Introduction

Kimberlites are complex, hybrid rocks comprising abundant xenocrystic/xenolithic material and are the primary source of diamonds worldwide (Kjarsgaard et al. 2019, 2022). Olivine is the most abundant mineral (~50 vol%; Skinner and Clement 1979) in kimberlites and displays complex compositional zoning characterised by xenocrystic cores surrounded by multiple magmatic growth zones (e.g., Moore 1988; Fedortchouk and Canil 2004; Pilbeam et al. 2013; Giuliani 2018; Howarth and Gross 2019; Casetta et al. 2023). The xenocrystic cores display a wide range in composition [Mg# = Mg/(Mg+Fe)*100 = 78-95] and represent various lithospheric mantle lithologies, including lherzolites, harzburgites, dunites, megacrysts, and sheared peridotites (e.g., Kamenetsky et al. 2008; Bussweiler et al. 2015; Sobolev et al. 2015; Giuliani 2018; Soltys et al. 2020; Howarth et al. 2022). It has now been well established that the composition of kimberlite melts is significantly affected by variable assimilation of depleted and metasomatised lithospheric mantle material, which acts to increase the contents of SiO2, MgO, FeO, and also some incompatible elements (e.g., Ti) in the melt (e.g., Russell et al. 2012; Stone and Luth 2016; Dalton et al. 2020; Giuliani et al. 2020; Tovey et al. 2021; Howarth et al. 2022). Notably, a strong correlation between the average compositions of olivine xenocrystic cores and magmatic rims in kimberlites worldwide suggests assimilation of metasomatic products related to early stages of pre-conditioning of the sub-continental lithospheric mantle (SCLM) by proto-kimberlite melts (e.g., Giuliani et al. 2020; Howarth et al. 2022). This metasomatism by early kimberlite-related (or proto-kimberlite) melts has also been linked to diamond destruction just prior to kimberlite emplacement (Giuliani et al. 2023) and, thus, plays a fundamental role in the diamond grades of kimberlites that reach the surface.

Diamond exploration and evaluation of kimberlite bodies relies heavily on kimberlite indicator minerals (KIMs), such as garnet, clinopyroxene, chromite, and ilmenite (e.g., Gurney and Zweistra 1995; Griffin and Ryan 1995; Nowicki et al. 2007; Kjarsgaard et al. 2019; Nimis 2022). These xenocrystic minerals provide direct insights into the chemical and physical properties of the SCLM traversed by kimberlites, including the geothermal gradient, lithospheric thickness and related depth of the diamond window (e.g., Grütter 2009), and the occurrence and relative abundance of potentially diamondiferous lithologies (e.g., Gurney 1986; Griffin and Ryan 1995). Although olivine is highly susceptible to deuteric (i.e., late-stage magmatic) and post-emplacement hydrothermal alteration near the Earth’s surface (e.g., Stripp et al. 2006), recently, there has been great interest in employing olivine as an indicator mineral (Fedortchouk et al. 2005, 2010; Harvey et al. 2013; Giuliani et al. 2023). Olivine has a similar hydrodynamic behaviour as diamond and can also inform on diamond sorting arising from volcanic processes (e.g., Scott-Smith and Smith 2009). Olivine composition can also serve to trace kimberlites in till samples (McClenaghan et al. 2007). It additionally provides constraints on the thermal and compositional state of the SCLM (de Hoog et al. 2010; Bussweiler et al. 2017; Giuliani et al. 2023). The benefit of using olivine to constrain the characteristics of the SCLM is that it is the most abundant mineral in the mantle and the most abundant of the KIM suite and, thus, should give a more representative indication of SCLM material sampled by kimberlites. The Al-in-olivine thermometer (Bussweiler et al. 2017) has proved to be particularly useful in constraining SCLM sampling depths for olivine and has now been effectively applied in numerous recent studies across different cratonic regions (e.g., Howarth and Taylor 2016; Bussweiler et al. 2017; Jaques and Foley 2018; Shaikh et al. 2019; Howarth and Giuliani 2020; Shaikh et al. 2021; Howarth et al. 2022; Veglio et al. 2022; Greene et al. 2023). In order to apply the Al-in-olivine thermometer, careful analysis and compositional screening are required (see Bussweiler et al. 2017). Iterative calculations need to be performed to reliably match the calculated Al-in-olivine temperatures to the local paleogeotherm, i.e., the relationship of temperature with depth at the time of kimberlite eruption. Therefore, a well-constrained local geotherm is required to effectively calculate both temperature and pressure of olivine xenocryst equilibration in the mantle. Thus, this approach is best used in combination with P-T constraints from other mantle xenoliths/xenocrysts such as those from clinopyroxene and/or garnet thermobarometry (e.g., Ryan et al. 1996; Nimis and Taylor 2000; Grütter et al. 2006; Grütter 2009; Mather et al. 2011; Nimis et al. 2024).

To further evaluate the utilisation of olivine as an indicator mineral to constrain sampling depth and composition of the SCLM, here we present a case study of olivine geochemistry in kimberlites from the Koidu diamond mine (Sierra Leone). We present electron-probe microanalysis (EPMA) and laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS) data for three coherent-textured hypabyssal kimberlites sampled from two dikes and one pipe in the Koidu cluster. We use these new data in combination with clinopyroxene and garnet indicator mineral data made available by Koidu Limited for this study as well as previous work on olivine inclusions in diamond (Lai 2022), diamondiferous eclogites (Hills and Haggerty 1989; Lai 2022), and non-diamondiferous eclogite xenoliths (Aulbach et al. 2019) from the Koidu mine. In addition, a robust local geotherm has been previously calculated at Koidu using clinopyroxene xenocrysts and FITPLOT (Mather et al. 2011) by Smit et al. (2016), hence providing a background to extract accurate depth constraints from Al-in-olivine thermometry. These results are employed to: 1) evaluate the use of olivine to understand whether the kimberlites predominantly sampled material within the diamond stability field or not; 2) compare P-T estimated for olivine xenocrysts and olivine included in diamond; 3) constrain the vertical distribution of lithologies, including diamond-rich regions, in the SCLM; and 4) evaluate the role of proto-kimberlite metasomatism to modify the SCLM and impact diamond preservation.

The Koidu kimberlites and underlying lithospheric mantle

As described in detail by Moss et al. (2012), Harder et al. (2013), and Nowicki et al. (2018), the Koidu mine comprises multiple kimberlite dikes, two large pipes and several blows (i.e., dike enlargements resulting from locally explosive activity) (Fig. 1). The dikes have a general northeast strike and are known to extend up to 10 km along strike. Three prominent dikes are present and termed DZA (also termed DZB-east dike), DZB, and DZC. The largest pipe, Pipe 2/K2 cuts across the DZA dike whereas Pipe 1/K1 is associated with the DZB dike (Fig. 1). The dikes comprise multiple phases although the mineralogical variation across all dikes is minimal with some variation in the abundance and texture of groundmass carbonate, phlogopite and spinel (e.g., Viljoen et al. 2022). Pipe 1 comprises four dominant phases: a coherent-textured phase termed KIMB1 and two volcaniclastic phases KIMB2 and KIMB3, and the mixed coherent to volcaniclastic phase KIMB4 (see Nowicki et al. 2018). An indicated resource of 1.38 Mt of ore has been declared by Koidu Ltd. with an average diamond grade across the pipe of 0.67 ct/t. Pipe 2 comprises two dominant phases: the main eruptive phase, volcaniclastic unit KIMB1, and the texturally mixed coherent to volcaniclastic phase KIMB2 as well as several minor phases (see Harder et al. 2013). An indicated resource of 2.79 million tonnes of ore has been declared by Koidu Ltd. with an average diamond grade across the pipe of 0.34 ct/t. An inferred resource for the dike segments of the DZA and DZB dikes has been reported at 4.27 million tonnes at an average grade of 0.54 ct/t.

Fig. 1
figure 1

Map of the Koidu kimberlite cluster modified after Moss et al. (2012) illustrating the various dikes, pipes, and blows. The ages for Pipe 2 and the DZA dike are derived from isotope-dilution mica Rb/Sr dating (Fitzpayne et al. 2023) whereas the ages of the DZB-E dike and Pipe 1 are from in-situ Rb-Sr dating of mica by laser ablation ICP-MS/MS from Giuliani et al. (2024). The map inset in the top left corner is modified after Skinner et al. (2004). The GPS coordinates for the two mined bodies are: Pipe 1 - 8°37'50.8"N 10°58'19.7"W and Pipe 2 - 8°37'58.5"N 10°57'57.0"W

Full size image

The coherent hypabyssal kimberlites in dikes and pipes of the Koidu cluster are highly micaceous with geochemical characteristics transitional with cratonic lamproites (Tompkins and Haggerty 1984; Taylor et al. 1994; Skinner et al. 2004; Viljoen et al. 2022; Fitzpayne et al. 2023). The exceptionally high abundance of groundmass phlogopite relative to most archetypal kimberlites worldwide has been attributed to the assimilation of highly metasomatised K-rich material during kimberlite transit through the SCLM (Viljoen et al. 2022; Fitzpayne et al. 2023). While the groundmass mineralogy of the Koidu kimberlites is similar throughout the cluster, the olivine populations vary between different dikes and pipes. Viljoen et al. (2022) showed that olivine macrocrysts of the DZA dike are characterised by a higher proportion of Fe-rich (‘megacrystic’) cores and more Fe-rich rims relative to the DZB dike and coherent kimberlites in Pipe 1. They interpreted the more Fe-rich rim compositions of the DZA dike to represent a greater proportion of assimilation of Fe-rich, metasomatised SCLM material relative to the DZB dike. Furthermore, Fitzpayne et al. (2023) showed that, while the major and trace element geochemistry of hypabyssal kimberlites of the dikes and pipes are generally similar, the bulk-rock Nd-Hf isotopes vary suggesting kimberlite crystallisation from distinct batches of magma. These magmas likely tapped marginally different sources and/or interacted to different extents with metasomatised SCLM wall-rocks. Despite these isotopic variations, Giuliani et al. (2024) did not identify any statistically significant temporal variability in the emplacement of the Koidu kimberlites based on in-situ mica Rb/Sr dating of samples from different dikes and pipes, with a pooled age of 146.3 ± 1.1 Ma. This age is indistinguishable from previous age constraints from isotope-dilution Rb-Sr analyses of micas and bulk-rocks (Fitzpayne et al. 2023).

Mantle xenoliths described from the Koidu cluster are predominantly eclogitic (e.g., Hills and Haggerty 1989; Fung and Haggerty 1995; Rollinson 1997; Barth et al. 2001; Aulbach et al. 2019) whereas peridotitic xenoliths are extremely rare (Hills and Haggerty 1989). Diamondiferous eclogites reported by Hills and Haggerty (1989) and Lai (2022) were shown to be sourced from 150-190 km depth in the SCLM. The KIM population from Koidu contains both eclogitic and peridotitic garnets, including Cr-rich, subcalcic G10 garnets (Harder et al. 2013), indicating that depleted peridotitic SCLM lithologies were also sampled by the host kimberlite. In addition, abundant Cr-diopside is present in the KIM suite, again suggesting sampling of peridotitic material from the SCLM. These clinopyroxenes have been used to calculate a mantle geotherm of 38 mW/m2, indicating diamond stability at depths >118 km, and a lithosphere-asthenosphere boundary (LAB) at ~220 km beneath Koidu (Smit et al. 2016).

The diamond population at Koidu reflects the dominance of eclogitic xenoliths. Lai et al. (2022) analysed 105 inclusion-bearing diamonds from Koidu and found that 78% of these diamonds were of eclogitic affinity whereas 17% were peridotitic and the remaining 5% contained mixed parageneses. Forty-five olivine inclusions in diamond have been analysed from Koidu with a Mg# range of 92.2-94.7 and with 93% of these having Mg# >93, which was interpreted to reflect sampling of highly depleted SCLM lithologies (Lai 2022). Furthermore, Al concentrations were analysed in a subset of 19 olivine inclusions with a calculated sampling depth range of 119-188 km, but predominantly (68%) from a relatively shallow interval of 119-150 km (Lai 2022).

Samples and analytical techniques

Samples for this study were kindly made available by Koidu Limited and represent multiple intrusive phases in the Koidu cluster as well as from the private collection of Mike Skinner (ESM Table 1). Samples with exceptionally fresh olivine were available from two of the main dikes, DZA (n=3) and DZB (n=2). In addition, coherent kimberlite from Pipe 1 phase KIMB1 (n=4) was also obtained that contained partially fresh olivine. Samples of coherent kimberlite from Pipe 2 had no fresh olivine. Koidu Limited also provided garnet and clinopyroxene indicator mineral EPMA data from the main units of Pipe 1 (KIMB1 and KIMB2) and Pipe 2 (KIMB1 and KIMB2) (ESM Table 1), which were previously employed by Smit et al. (2016) to construct the Koidu mantle palaeogeotherm; however, no KIM data are available for the dikes. Thus, diamond potential and mantle sampling depths for the pipes were primarily constrained using garnet and clinopyroxene with fewer olivine xenocryst data whereas the dikes were evaluated using olivine only. In addition, we have compiled literature data of P-T estimates from the Koidu kimberlites for eclogite xenoliths (Aulbach et al. 2019), diamondiferous eclogites (Hills and Haggerty 1989; Lai 2022), and olivine diamond inclusions (Lai 2022) although no information is available revealing to which magmatic phase these xenoliths and diamonds belong.

The olivine major and minor element data were acquired using a CAMECA SX-100 electron probe micro-analyser (EPMA) at the University of Johannesburg, South Africa. Analyses were conducted using an acceleration potential of 15 kV, beam current of 20 nA, and beam diameter of 1 μm. Peak counting times varied between 10 s and 60 s depending on measured chemical element. All elements were measured on the Kα line. Reference materials used to calibrate the instrument included jadeite (Na), olivine (Mg), almandine (Al), diopside (Si), orthoclase (K), wollastonite (Ca), rhodonite (Mn), hematite (Fe), and synthetic Cr, Ni, and Ti oxides. Detection limits are ~ 0.01 wt% for all elements. Data reduction and matrix correction was performed using the ‘X-PHI’ method.

LA-ICP-MS analyses were performed at the Institute of Geology and Mineralogy at the University of Cologne, Germany, using an ESL GeoImage 193 nm laser ablation system coupled to a Thermo Scientific iCAP quadrupole ICP-MS. The laser parameters were: fluence of 3.0 J/cm2, spot size of 50 μm, and repetition rate of 5 Hz. Helium was used as the carrier gas and argon as makeup (i.e., mixed with the helium prior to entering the torch) and plasma gas. The background (gas blank) was recorded for 25 seconds before 30 seconds of ablation time, followed by 15 seconds of washout. The following isotopes were scanned in peak hopping mode: 7Li, 23Na, 27Al, 29Si, 31P, 47Ti, 51V, 52Cr, 53Cr, 55Mn, 59Co, 60Ni, 66Zn, 71Ga, 88Sr, 89Y, 90Zr, 93Nb, 140Ce, and 208Pb. Elements strongly incompatible in olivine such as Sr, Ce and Pb were monitored to assess potential contamination by inclusions or material in fractures and/or grain boundaries. The calibration material was NIST SRM 612 with 29Si as internal standard. The olivine standard 355OL was interspersed as a secondary standard (Bussweiler et al. 2019) and the results are broadly consistent with those from solution analyses (ESM Table 1). The data were processed using Iolite 4.0 (Paton et al. 2011). When setting the integration intervals, the first 2-3 seconds of ablation were omitted to avoid surface contamination, and signal spikes due to mineral/fluid inclusions were avoided as well. Data accuracy for the secondary standard was generally better than 10% in all sessions, except for ultra-low trace elements (e.g., Sr, Y, Ce).

Results

Petrography

The Koidu kimberlites are all highly micaceous and the same samples analysed in this study were previously described in detail by Viljoen et al. (2022). Here we summarise the petrographic features of these samples for context. The olivine abundance is variable (37-56 vol%) and comprises megacrysts (>10 mm), macrocrysts (>0.5 mm) and microcrysts (<0.5 mm) (Fig. 2). All olivines are generally fresh in the dike samples whereas microcrystic olivines are generally altered in the Pipe 1 coherent kimberlite (Fig. 2). Notably, the DZA dike samples generally contain a greater proportion of megacrystic olivine relative to the DZB dike and Pipe 1 samples (Fig. 2). In all samples, the olivine population is dominated by monocrystalline grains with a smaller proportion of polycrystalline grains occasionally observed (Fig. 2a). In BSE images, all olivines, regardless of size, contain distinct core and rim zones (ESM Figure 2). In rare cases, microcrystic olivines are observed without distinct core regions and BSE greyscale overlapping the rims on coarser grained olivine, i.e., true phenocrysts. Macrocrystic phlogopites are present in variable proportions whereas ilmenite macrocrysts are rare (Fig. 2a). The groundmass of all kimberlites is comprised of abundant phlogopite (22-39 vol%; Viljoen et al. 2022) along with spinel, perovskite, and apatite within an interstitial material dominated by carbonate and lesser serpentine (ESM Figure 1).

Fig. 2
figure 2

Photomicrographs of representative samples from: a) DZA dike – cross-polarised light, b) DZB dike – cross-polarised light, and c) Pipe 1 – plane polarised light. ol – olivine. olp – polycrystalline olivine. ilm – ilmenite

Full size image

Olivine composition

Here, we document only the compositions of the core regions of olivine as the primary goal is to constrain the SCLM sampling depth of the kimberlite using Al-in-olivine thermometry of olivine xenocrysts. See Viljoen et al. (2022) for a more comprehensive investigation into olivine core and rim compositions in the Koidu kimberlites.

DZB dike and Pipe 1

Olivine cores were analysed for major elements using EPMA in a sample from the DZB dike (DZB-2-122.95) and the coherent phase KIMB1 of Pipe 1 (KP14-6-190P). Sixty cores were analysed in the DZB dike and 25 in the Pipe 1 sample. The core compositions show strong overlap and thus are henceforth described together. The core populations display a broad range in Mg# from 81.7 to 95.5 (Fig. 3). Notably, 28% of the cores analysed exhibit Mg# >94. A positive correlation is observed between Mg# and NiO at Mg# <89, which is lacking for cores with Mg# >90. The average core Mg# for the DZB dike is 91.4 ± 1.3 (1 standard deviation, throughout the text) and 92.5 ± 2.7 for Pipe 1.

Fig. 3
figure 3

Mg# - NiO (wt.%) covariation diagrams of olivine EPMA data for a) DZB dike and Pipe 1 and b) DZA dike. All the rim data and some core data (black dots) are from Viljoen et al. (2022). Koidu olivine diamond inclusion data from Lai (2022)

Full size image

Trace element concentrations, measured via LA-ICP-MS, were obtained in 46 cores from the DZB dike sample, however, only 9 cores were analysed in the Pipe 1 sample (Fig. 4). Aluminium concentrations are generally low in the higher Mg# (>90) cores (<40 ppm) with some grains extending to higher values. The lower Mg# cores have high Al concentrations (up to ~180 ppm) at Mg# ~89 and decrease with decreasing Mg# to concentrations of 18 ppm Al for Mg# ~82, defining a strong positive correlation between Mg# and Al (Fig. 4). In these low-Mg# cores, a strong negative correlation between Mg# and Zn and Mn is observed (Fig. 4). Other elements such as Ti, V, and Co display initial negative correlations with decreasing Mg# to ~Mg# 85 followed by a sharp drop in concentrations at lower Mg# (Fig. 4). In contrast, there are limited relationships between Mg# and trace elements for the higher Mg# (>90) core population, exceptions being Zn and Co which display strong negative correlations (r2 = 0.90 and 0.88, respectively) with Mg#.

Fig. 4
figure 4

Covariation diagrams of Mg# vs. trace element concentrations (panels a-f) quantified by LA-ICP-MS in olivine from the Koidu kimberlites. The olivine compositions are divided between ‘peridotitic xenocrysts’ (xeno) and ‘megacrystic xenocrysts’ (mega) based on Mg# and Zn contents (see text). Diamond inclusion data from Stachel and Harris (1997), Jean et al. (2016), Smith et al. (2018), and De Hoog et al. (2019). Garnet peridotite data from De Hoog et al. (2010) and Bussweiler et al. (2017). Sheared peridotite data from Howarth et al. (2014). Dunite and wehrlite/websterite from Rehfeldt et al. (2008)

Full size image

DZA dike

Two samples (DZA-2-391.85 and Y5770) were analysed from the DZA dike. Forty-six cores were analysed using EPMA in sample DZA-391.85 and 69 cores in Y5770. The core compositions are nearly identical in both samples and are described together. The cores display a broad range in Mg# from 80.8 to 93.9 and notably do not contain very high Mg# cores (i.e., Mg#>94) such as those observed in the DZB dike and Pipe 1 samples (Fig. 3). A characteristic feature of the core compositions for the DZA dike is the notable low density of analyses between Mg# 89.0 and 90.8, suggesting effective discrimination of high- (Mg#>90.8) and low-Mg# (Mg#<89) core populations (Fig. 3). Both the higher Mg# and lower Mg# cores display positive correlations between Mg# and NiO (Fig. 3). The average core Mg# for samples DZA-391.85 and Y5770 are 90.3 ± 2.8 and 90.2 ± 2.8, respectively.

Trace element concentrations were analysed in 34 cores in DZA-391.85 and 18 cores in Y5770. Overall, they show similar trends to those observed in the DZB dike and Pipe 1 samples (Fig. 4). Aluminium concentrations in the higher Mg# cores are relatively low (<40 ppm) with only one core analysed above 40 ppm. The lower Mg# cores, however, display a strong positive correlation between Mg# and Al with high Al concentrations (up to ~190 ppm) at higher Mg# and decreasing to ~30 ppm at the lowest Mg# values (Fig. 4). The lower Mg# core population also displays negative correlation between Mg# and Zn. Vanadium and Ti show changing trends at ~Mg# 85 with a marked decrease in V and Ti with decreasing Mg# <85 (Fig. 4b).

Discussion

Discriminating megacrystic from peridotitic olivine cores

Olivine cores in kimberlites represent mantle-derived xenocrysts and are commonly divided into low-Mg (Mg# <89) and high-Mg (Mg#>89) populations based on contrasting trends in Mg# vs minor and trace element concentrations (e.g., Giuliani 2018; Lim et al. 2018; Howarth et al. 2022). However, in some cases there is no clear divide in the core data leading others to suggest such division is not warranted (e.g., Moore and Costin 2016; Arndt et al. 2022). Regardless, it is widely agreed that the lower Mg# cores (generally Mg# <89) are related to the Cr-poor megacryst suite or to sheared peridotites, as they are outside of the typical Mg# range of coarse-grained granular peridotite xenoliths (e.g., Kamenetsky et al. 2008; Moore and Costin 2016; Giuliani 2018; Soltys et al. 2020; Arndt et al. 2022; Abersteiner et al. 2022; Howarth et al. 2022; Casetta et al. 2023). The Koidu olivine cores from the DZA dike can be clearly divided into two populations with a divide at Mg# of approximately 90 (Fig. 3). Most trace element concentrations overlap completely in the two core populations with the notable exception of Zn: divide at ~90 ppm Zn (Figs. 4 and 5). Combinations of Zn, Al and Mn best discriminate these two populations (Fig. 5). The higher Zn population across both dikes and Pipe 1 (Mg# = 81.8-90.1) is characterised by strong correlations (positive or negative) with other trace elements (e.g., r2 = 0.97 for Al vs Zn in low-Mg# olivine from the DZA dike), which are typical of olivine of the Cr-poor megacryst suite (Howarth 2018; Howarth et al. 2022). If megacrysts represent the products of proto-kimberlite melt crystallisation (e.g., Moore and Belousova 2005; Giuliani et al. 2013; Kargin et al. 2017; Woodhead et al. 2017; Nkere et al. 2021), these trends reflect melt evolution during megacryst formation. For example, these Zn-rich cores display evidence for co-crystallization with ilmenite based on a marked decrease in elements compatible in ilmenite such as Ti, V, and Co with Mg# decreasing below 85 (Fig. 4), a feature typical of megacrystic olivine (Howarth 2018). Furthermore, the Zn concentrations are higher than those reported for olivine in sheared peridotites as well as metasomatised lithologies such as wehrlite/websterite (Figs. 4 and 5). Thus, these high-Zn (>90 ppm), low-Mg# cores are interpreted to reflect the disaggregation of megacrystic olivine into the transporting kimberlite magma, rather than olivines from sheared peridotites or other metasomatised lithologies. However, this is based on a limited trace-element dataset for olivine in sheared peridotites and metasomatic mantle lithologies; a more comprehensive dataset for olivine in mantle xenoliths will be crucial for further understanding the trace element variation in SCLM olivine. Furthermore, the range of Mg# up to 90.1 for the higher Zn cores suggests that dividing core populations based on Mg# (i.e. <89 and high >89, respectively) should be done with care and that trace elements may be more effective at discriminating megacryst-related olivine cores, consistent with the suggestions of Howarth et al. (2022). Ultimately, the DZA dike contains a significantly higher proportion of megacrystic cores (30%) relative to the DZB and Pipe 1 samples (14%) based on Mg# and trace element data (i.e., concentrations of Zn, Al and Mn; Fig. 5).

Fig. 5
figure 5

Al-Zn-Mn systematics (panels (a-b) of olivine for the Koidu kimberlites. The olivine compositions are divided between ‘peridotitic xenocrysts’ (xeno) and ‘megacrystic xenocrysts’ (mega) based on the results of this study. Diamond inclusion (Di inc) data from Stachel and Harris (1997), Jean et al. (2016), Smith et al. (2018), and De Hoog et al. (2019). Xenolith data from De Hoog et al. (2010) and Bussweiler et al. (2017). Sheared peridotite data from Howarth et al. (2014)

Full size image

The low-Zn (<90 ppm) cores with higher Mg# (89.7-94.8) have trace element concentrations similar to those of olivine in SCLM-derived peridotite xenoliths (Fig. 4). They predominantly fall within the lower temperature range of garnet peridotites with few from spinel peridotites based on their V-Al systematics (Fig. 6). Furthermore, of the 99 low-Zn cores analysed from across the Koidu cluster, only five fall within the field of high temperature (>1200oC) peridotites based on Al-V contents, which include sheared peridotites (Fig. 6). Thus, the low-Zn cores from the Koidu kimberlites are most similar to those of shallow and cold (<1000°C) garnet peridotites (Fig. 6). Within this low-Zn core population strong correlations are observed between trace elements associated with metasomatism (e.g., Zn, Mn, Co; e.g., Zn-Mn: r2 = 0.81; Zn-Co: r2 = 0.77) (ESM Figure 3). High-Mg# olivine (>93 – i.e., harzburgitic/dunitic olivine) is characterised by low concentrations of these elements whereas lower Mg# olivine (<93 – predominantly lherzolitic olivine) has elevated concentrations of these elements. This strong correlation is interpreted to be the result of metasomatic refertilization of harzburgite and dunite to form lherzolite.

Fig. 6
figure 6

Covariation diagram of V and Al in peridotitic olivine from the Koidu kimberlites. Diamond inclusion data from Stachel and Harris (1997), Jean et al. (2016), Smith et al. (2018), and De Hoog et al. (2019). Olivine in sheared peridotites from Howarth et al. (2014). Olivine in peridotite fields and Al-based isotherms after Bussweiler et al. (2017)

Full size image

Titanium in olivine has been previously shown to be a good tracer of depleted vs. refertilised (i.e., metasomatised) SCLM lithologies (e.g., Rehfeldt et al. 2008; Foley et al. 2013). Titanium in olivine inclusions in diamonds, for example, is generally very low (<10 ppm), and coupled with high Mg# (>92) (Fig. 4) is consistent with derivation of olivine inclusions in diamonds from depleted peridotites (e.g., de Hoog et al. 2019). The DZA dike olivine cores with low Zn generally have higher Ti concentrations (>40 ppm) than olivine inclusions in diamonds, whereas 30% (14 of 47) of those from the DZB dike/pipe 1 samples have similar, low Ti to that of olivine inclusions in diamonds (Fig. 5). The deficiency of highly depleted olivine in the DZA dike is highlighted by the near absence of high Mg# (>93) cores (2% of peridotitic olivines) compared to the DZB dike/Pipe 1 where 48% of the peridotitic core population has Mg# >93 (Fig. 4).

In summary, while high-Zn, low-Mg# olivine cores typical of the low-Cr megacrysts suite occur in all the examined samples, the low-Zn, high-Mg# cores derived from peridotite disaggregation show differences: very high-Mg# (>93) cores with low Ti contents, similar to olivine included in diamonds, are limited to the DZB dike and Pipe 1 while they are near absent from the DZA dike (Fig. 7). Below, we explore variations in the vertical distribution of these peridotitic olivine cores taking advantage of the Al-in-olivine thermometer.

Fig. 7
figure 7

Koidu olivine xenocryst and diamond inclusion Mg# histograms: a) DZA dike, b) DZB dike and Pipe 1, and c) Koidu olivine inclusions within diamonds from Lai (2022)

Full size image

Al-in-olivine thermometry and lithospheric mantle sampling by kimberlite magmas

In order to constrain the depth at which the Koidu kimberlite magmas sampled olivine from the SCLM, we have applied the Al-in-olivine thermometer following the approach of Bussweiler et al. (2017). The high-Zn, low-Mg# megacrystic cores were screened out as the thermometer is not calibrated for these olivine compositions and does not give reliable temperatures (see Howarth and Giuliani 2020; Howarth et al. 2022, 2023). In addition, only olivine cores sourced from garnet peridotites rather than spinel peridotites are suitable for Al-in-olivine thermometry and were screened using Al-V systematics (Fig. 6), as suggested by Bussweiler et al. (2017). The calculated temperatures using the Al-in-olivine thermometer were matched to the local paleogeotherm of 38 mW/m2 (Fig. 8), which was calculated using clinopyroxene xenocrysts by Smit et al. (2016). The iterative Al-in-olivine thermometer calculations were done using the PTEXL spreadsheet (https://cms.eas.ualberta.ca/team-diamond/downloads/). It is important to note that olivine interpreted to be derived from spinel peridotites (i.e., <~80 km depth) have been filtered out. However, as only two olivine grains plotted in the spinel peridotite field of Fig. 6, the absence of these olivines in the plotted histograms would not affect the histogram data distribution discussed below.

Fig. 8
figure 8

Pressure-temperature conditions of peridotitic olivine in equilibrium with garnet peridotites based on the Al-in-olivine thermometer of Bussweiler et al. (2017) and the local paleo-geotherm of 38 mW/m2 from Smit et al. (2016). a) Olivine (this study) and clinopyroxene (Cpx; Smit et al. 2016) P-T estimates for DZB dike and Pipe 1 KIMB1 and KIMB2, respectively. b-c) Histograms showing the depth distribution of (b) peridotitic olivine and (c) clinopyroxene xenocrysts from DZB dike and Pipe 1. d) Same as (a) with olivine data (this study) for DZA dike and clinopyroxene data (Smit et al. 2016) for Pipe 2 KIMB1 and KIMB2. e-f) Same as (b) and (c) with olivine data for the DZA dike and clinopyroxene data for Pipe 2

Full size image

Histograms of the calculated equilibration depths for peridotitic olivine cores display a main mode of SCLM sampling at 110-140 km and tapering off at 160-170 km for both dike DZA and dike DZB/Pipe 1 samples (Fig. 8). Only 2% of the cores analysed for the DZA dike appear to have been sampled below 170 km, in contrast to 15% of the DZB dike/Pipe 1 cores, predominantly from the 180-190 km range (Fig. 8). This suggests that olivine from the Koidu cluster (DZA/DZB dikes and Pipe 1) predominantly represents sampling just within diamond stability (i.e., diamond stability at >118 ± 10 km; Smit et al. 2016). Peridotitic olivine is virtually unsampled in the lowermost 40 km of the lithosphere (lithosphere-asthenosphere boundary: LAB = 220 ± 10 km; Smit et al. 2016) for the DZA dike (Fig. 8). Variations in Mg# and trace element composition of peridotitic olivine at Koidu (Fig. 4) suggest the presence of both relatively Fe-rich (probably lherzolitic) lithologies and highly depleted (probably harzburgitic) lithologies in the 110-140 km interval sampled by the DZB and Pipe 1 magmas, whereas the DZA dike kimberlite sampled predominantly more Fe-rich refertilised peridotites.

There is no systematic correlation between olivine composition (e.g., Mg#) and depth (Fig. 9). However, all olivines sampled from below 170 km have Mg# ≤92, suggesting derivation from refertilized (probably lherzolitic) lithologies and suggesting that the highly depleted lithologies with olivine Mg# >93 are restricted to depths <170 km (Fig. 9). This is consistent with the “chemical stratigraphy” of the SCLM traversed by kimberlites elsewhere, which is largely based on garnet analyses, where the base of the SCLM commonly displays higher degrees of refertilization resulting in more Fe-rich garnet and olivine at greater depth and more depleted compositions at shallower depths (e.g., Griffin et al. 1999, 2003, 2004; Kobussen et al. 2008; Veglio et al. 2022).

Fig. 9
figure 9

Mg# variations vs depth for olivine xenocryst (xeno.) and diamond inclusions (di. inc.) in the Koidu mine (panels a-b). Depth values calculated using the Al-in-olivine thermometer and matched to a 38 mW/m2 geotherm. Geotherm, graphite-diamond transition (118 km), and lithosphere asthenosphere boundary (LAB; 220 ±10 km) from Smit et al. (2016). Data for Koidu olivine inclusions in diamond from Lai et al. (2022)

Full size image

Vertical distribution of SCLM lithologies and location of diamond-rich regions

Olivine inclusions in diamond (n = 45) from Koidu have high Mg# with 93% of the grains showing Mg# >93 (Lai 2022) (Fig. 9). These data indicate that peridotitic diamonds are predominantly sourced from highly depleted SCLM lithologies. Olivine xenocrysts in the Koidu kimberlites with Mg# >93 are near absent in the DZA dike (2% of peridotitic olivine compositions) but are relatively abundant (48% of peridotitic olivine compositions) in the DZB dike/Pipe 1 kimberlites. This suggests that the DZB dike/Pipe 1 kimberlite magmas have a significantly higher potential to contain peridotitic diamonds compared to the DZA dike parent magma. Sampling depth estimates for olivine inclusions in diamonds, calculated using the same Al-in-olivine thermometer approach used in this study, have a main mode at 140-150 km and a secondary mode at 160-180 km but ~70% of diamonds (n = 19) represent sampling at between 119 and 150 km of depth (Fig. 10). This relatively shallow SCLM sampling is close to diamond stability at ~118 km and similar to the main mode of olivine xenocryst sampling at 120-140 km (Fig. 10). Both olivine inclusions in diamond and olivine xenocrysts with Mg# >93 are absent from the lower SCLM (170-220 km) at Koidu (Fig. 9), indicating the absence of highly depleted lithologies at these depths and consistent with metasomatic refertilization of the base of the SCLM.

Fig. 10
figure 10

Sampling depth histograms for olivine cores, olivine inclusions in diamonds, clinopyroxene indicator minerals, and eclogite xenoliths from the Koidu mine. Olivine data from this study. Koidu olivine diamond inclusion data from Lai et al. (2022) and diamondiferous eclogite data from Hills and Haggerty (1989) and Lai et al. (2022). Clinopyroxene data from Smit et al. (2016) and from Koidu ltd. Eclogite xenolith data from Aulbach et al. (2019). The depths of the graphite-diamond stability and LAB (lithosphere-asthenosphere boundary) are from Smit et al. (2016)

Full size image

Clinopyroxene indicator mineral data (n = 406) were made available by Koidu Limited from the largest volcanic units within Pipe 1 (KIMB1 and KIMB2) and Pipe 2 (KIMB 1 and KIMB2); no KIM data were available from the dikes. The clinopyroxene data were previously processed and filtered by Smit et al. (2016) using the screening criteria of Grütter (2009) to obtain reliable pressure and temperature estimates. P-T values for grains in equilibrium with garnet peridotites were calculated using the geothermobarometer of Nimis and Taylor (2000). The screening process filters out compositions corresponding to Cr-poor megacrysts (i.e., <0.5 wt% Cr2O3); however, in general it is not possible to discriminate the more Cr-rich megacrystic varieties from typical peridotitic clinopyroxene (e.g., Nkere et al. 2021). Thus, it is not possible to fully constrain the proportion of megacrystic clinopyroxene relative to peridotitic clinopyroxene in an indicator mineral suite. Clinopyroxenes showing Cr-poor megacryst compositions (i.e., <0.5 wt% Cr2O3) represent only 3% (12 of 406) of the total clinopyroxene dataset from the Koidu pipes. The P-T estimates for the filtered clinopyroxene dataset from the different phases in each pipe overlap completely and are discussed together. Histograms of the depth estimates of the clinopyroxenes display a bimodal distribution for both pipes (Fig. 8). The shallower peak corresponds to a sampling depth of ~120-150 km, which is similar to the range of peridotitic olivine and olivine inclusions in diamond. The second peak represents a deeper sampling depth of ~180-210 km, which extends to the LAB. The clinopyroxene data, therefore, indicate that the Koidu kimberlites did indeed sample material from close to the LAB despite the almost complete lack of peridotitic olivine from the lowermost lithosphere (Fig. 8). The clinopyroxene data show some compositional variation with depth (ESM Figure 4). In the deeper sampling window (180-210 km), clinopyroxene has relatively low Cr2O3 (0.9-1.5 wt%) and a narrow Ca# range (39.0-40.5) compared to 1.5-4.5 wt% and 43-51 for the shallower clinopyroxene. Nkere et al. (2021) showed that a similar thermobarometry approach used for peridotitic clinopyroxenes can be applied to Cr-poor megacrysts. The clinopyroxenes of the Koidu KIM suite with <0.5 wt% Cr2O3 (i.e., Cr-poor megacrysts) yield a calculated depth range of 190-230 km, indicating formation in the lowermost lithosphere. These results are consistent with the P-T estimates for Cr-poor megacrysts from southern African kimberlites (Nkere et al. 2021).

Mantle peridotite xenoliths are rarely preserved at Koidu (Tompkins and Haggerty 1984); however, eclogite xenoliths are abundant (e.g., Hills and Haggerty 1989; Fung and Haggerty 1995; Rollinson 1997; Barth et al. 2001, 2002; Aulbach et al. 2019). Aulbach et al. (2019) presented P-T estimates for 69 eclogite xenoliths derived by iterative solution of the garnet-clinopyroxene Krogh Ravna (2000) geothermometer using a 38 mW/m2 geotherm (Smit et al. 2016). The calculated depth estimates display a broad bimodal distribution with peaks at ~130-140 km and ~160-180 km (Fig. 10). The shallower depth is consistent with olivine and clinopyroxene sampling just below the graphite-diamond transition and indicates a complex heterogeneous SCLM at this depth consisting of highly depleted harzburgites/dunites (i.e., olivine Mg# >93), refertilised lherzolites (i.e., olivine Mg# = 89-93 and peridotitic clinopyroxene), and eclogites. The deeper eclogite sampling depth falls between the two major peaks of clinopyroxene sampling in a region where little peridotitic olivine was entrained (Fig. 10), suggesting a dominance of eclogites in this depth interval (~160-180 km). Diamondiferous eclogites (n = 10) have been analysed by Hills and Haggerty (1989) and Lai (2022) and sampling depth was estimated using the same approach as Aulbach et al. (2019). The calculated depth range for diamondiferous eclogites is 150-190 km with 8 samples out of 10 coming from a narrower range of 150-170 km (Fig. 10). This depth range of 150-170 km overlaps with the deeper interval of the eclogites analysed by Aulbach et al. (2019) (Fig. 10). Thus, it appears likely that eclogitic diamonds are predominantly sourced from this region of the SCLM (150-170 km) whereas the bulk (~70%) of the peridotitic diamonds are derived from shallower depths (119-150 km; Lai 2022). Both P-type and E-type diamonds are scarce in the lowermost reaches of the SCLM within 40 km of the LAB (i.e., 180-220 km) (Fig. 10).

Garnet major element indicator mineral major element compositions have been provided by Koidu Limited for the main phases of Pipe 1 (KIMB1 and KIMB2) and Pipe 2 (KIMB1 and KIMB2), however, no P-T estimates could be calculated as NiO was not analysed. These data have been classified following the schemes of Schulze (2003) and Grütter et al. (2004) (ESM Figure 5). The distribution of garnet compositions overlapped completely from the different phases from each pipe so that in ESM Figure 5 the data have been plotted as Pipe 1 and Pipe 2. The garnet dataset is dominated by G1 garnets of the Cr-poor megacrystic suite (50% for Pipe 1 and 35% for Pipe 2), equivalent to the Zn-rich, low-Mg# olivine cores interpreted to derive from Cr-poor megacrystic suite (Fig. 5). Depleted harzburgitic G10 and refertilized lherzolitic G9 garnets represent the second and third most abundant garnet types at Pipe 1 (G10: 13.2 % and G9: 15.2 %) and Pipe 2 (G10: 22.4 % and G9: 18.8 %) (ESM Figure 6). Unexpectedly, eclogitic garnets (G3) only account for 2.9% and 2.5% of the Pipe 1 and Pipe 2 dataset, respectively. This suggests that the limited recovery of peridotite compared to eclogite xenoliths at Koidu may be due to preferential disaggregation of peridotite during transport and/or alteration during/after eruption. Comparison in this regard may be made to Roberts Victor of the Kaapvaal craton, likely the most famous location worldwide for containing abundant eclogite xenoliths. Early studies at Roberts Victor showed that eclogites dominated the overall xenolith population (~90%; MacGregor and Carter 1970; Hatton and Gurney 1979) whereas it was later shown to be an overestimate due to the high degree of alteration of peridotitic xenoliths (Viljoen et al. 1994).

Implications for lithospheric mantle modification and diamond survival

One of the major outstanding questions related to the origin of olivine cores at Koidu is the depth of sampling of megacrystic grains and the implications for modification of the deep lithosphere including its relationship to diamond grades (Agashev et al. 2018; Fedortchouk et al. 2022; Giuliani et al. 2023). Megacrysts are a common component of kimberlites worldwide, however, detailed investigations on megacrysts from West African kimberlites have not been conducted to date. In contrast, the P-T conditions of megacryst formation are well-constrained in other, although few cratonic regions (e.g., Gurney et al. 1979; Bussweiler et al. 2018; Tappe et al. 2021). For example, Nkere et al. (2021) presented a comprehensive study and review of megacrysts across the Kaapvaal craton. They showed that Cr-poor megacrysts equilibrated in the deep SCLM at temperatures of 1200-1450 oC and depths of 140-215 km (4.5-7.0 GPa). Cr-poor megacrysts at Koidu comprise 14-30% of the olivine population and 35-50% of the garnet population and, thus, represent a major component of the SCLM sampled by the Koidu kimberlites. While Cr-poor megacrysts comprise a lower proportion of the Koidu clinopyroxene indicator suite (~3%), they were shown to have equilibrated at 190-230 km in the previous section, consistent with Cr-poor megacrysts in the Kaapvaal craton (Nkere et al. 2021). Olivine, clinopyroxene, and garnet megacrysts of the Cr-poor suite generally form at the same time from the same evolving melt(s) (e.g., Gurney et al. 1979; Hops et al. 1992; Moore et al. 1992; Howarth 2018) and, thus, the Koidu Cr-poor olivine megacrysts probably also formed at similar P-T conditions as Cr-poor clinopyroxenes near the base of the SCLM. Such an interpretation is consistent with the low abundance of peridotitic olivine sampled from depths >170 km at Koidu, indicating a likely dominance of Cr-poor megacryst sampling near the base of the SCLM. If Cr-poor megacrysts represent the product of interaction between failed pulses of kimberlite melt (i.e., proto-kimberlite) and lithospheric mantle wall-rock (e.g., Moore and Belousova 2005; Giuliani et al. 2013; Kargin et al. 2017; Nkere et al. 2021), the lowermost 40 km of the SCLM beneath Koidu have been extensively modified by proto-kimberlite pre-conditioning. Thus, the absence of olivine in the depth histograms from the lower SCLM reflect megacrystic grains that are not calibrated for Al-in-olivine thermometry.

Here, we present a model for kimberlite melt evolution for the Koidu cluster, which is illustrated in Fig. 11. In the early stages of magmatism, (proto-)kimberlite melts likely formed stockwork-like networks that extended from the base of the SCLM upward over ~30-40 km. The lateral extent of this melt network cannot be estimated here, although we believe this likely to be a local feature and armouring of the kimberlitic conduit rather than an extensive megacrystic layer extending across the base of the SCLM. The overall tight evolutionary trends of olivine megacrysts (Fig. 4) are interpreted to reflect progressive melt evolution through fractional crystallisation of megacryst phases rather than metasomatic replacement of peridotitic lithologies (see detailed discussion in Nkere et al. 2021). Specifically, the progressive decrease of Ca and Al (typically incompatible in olivine) with decreasing Ni (compatible in olivine) reflects co-crystallisation with other silicate phases such as garnet and clinopyroxene whereas the notable inflection in Ti and V contents at Mg# ~85 tracks the onset of ilmenite crystallisation (Fig. 4). This early stage of proto-kimberlite evolution resulted in Fe enrichment in the lower SCLM based on the composition of megacrystic olivine and garnet compared to the same phases in typical cratonic peridotites. In previous studies, such a region of mantle refertilization has been interpreted as the asthenosphere-lithosphere transition zone and linked with the formation of sheared peridotites (e.g., Tappe et al. 2021). Furthermore, extensive proto-kimberlite pre-conditioning has been shown to be diamond-destructive (Giuliani et al. 2023) and, thus, we would expect diamonds from Koidu to be predominantly sourced from shallower depths rather than near the base of the SCLM. This is indeed the case at Koidu where both peridotitic and eclogitic diamonds are sourced from <180 km.

Fig. 11
figure 11

Schematic illustration of the vertical distribution of SCLM lithologies at Koidu based on olivine xenocrysts (this study), olivine inclusions in diamond (Lai 2022), clinopyroxene xenocrysts (Smit et al. 2016), and eclogite xenoliths (Hills and Haggerty 1989; Aulbach et al. 2019; Lai 2022). Graphite-diamond stability and depth of lithosphere-asthenosphere boundary (LAB) from Smit et al. (2016)

Full size image

In three other localities where diamond P-T data along with clinopyroxene indicator data exist, the Voorspoed mine (South Africa) shows similar systematics as Koidu with inclusion-bearing diamonds predominantly sourced from mid-lithospheric depths and notably absent in the lowermost lithosphere whereas there is no clear pattern of diamond sampling depth in the Cullinan mine (Nimis et al. 2020). However, at Kimberley (South Africa) diamonds are predominantly sampled from the lower lithosphere (Phillips et al. 2004; Nimis et al. 2020). While trace elements in olivine xenocrysts are not available for Voorspoed and Cullinan, Howarth et al. (2022) reported trace elements in olivines from several coherent-textured kimberlites of the Kimberley cluster. The Kimberley kimberlites peridotitic olivine data, which represent 60% of the analysed olivines, can be used to estimate sampling depths to be compared with those obtained from garnet and clinopyroxene, including diamond inclusions. Nimis et al. (2020) showed that P-T estimates for garnet and clinopyroxene indicator minerals from the Kimberley kimberlites are clustered at low to moderate depths with main modes at ~145 km and ~135 km, respectively, and an abrupt decrease in the abundance of garnet and clinopyroxene sampled below ~160 km. This observation is consistent with previous studies on both xenocrysts and xenoliths from the Kimberley kimberlites (e.g., Nimis and Taylor 2000; Griffin et al. 2003). Depths estimated using Al-in-olivine thermometry for Kimberley olivine xenocrysts (data from Howarth et al. 2022) screened following the same procedure as here and a 38 mW/m2 geotherm (Nimis et al. 2020) exhibit a main mode at 160-170 km, and predominant sampling (~60% of peridotitic olivine grains) below 160 km (Fig. 12). Thus, olivine xenocrysts appear to represent deeper sampling than other xenocrysts. This cannot reflect an overestimate of temperature based on the use of the Al-in-olivine thermometer, which can overestimate T by ~50oC relative to clinopyroxene thermometry only at lower temperatures (i.e., <950°C; Bussweiler et al. 2017) equivalent to depths of <120 km.

Fig. 12
figure 12

Sampling depth distribution for olivine and diamond inclusions as well as clinopyroxene and garnet indicator minerals from the Kimberley kimberlites (South Africa). a) Depth calculated using olivine thermometry and a 38 mW/m2 geotherm as suggested by Nimis et al. (2020). Olivine data from Howarth et al. (2022). b) Diamond inclusion data from Phillips et al. (2004) and pressure-temperature estimates for diamond inclusions, clinopyroxene, and garnet from Kimberley Pool kimberlites from Nimis et al. (2020). Graphite-diamond stability and depth of lithosphere-asthenosphere boundary (LAB) from Nimis et al. (2020)

Full size image

Pressure estimates of diamond inclusions (garnet and pyroxenes) from the Kimberley kimberlites, based on data from Phillips et al. (2004) and recalculated by Nimis et al. (2020), predominantly exceed 160 km of depth (~75%) and show a main mode at 180-190 km (Fig. 12). Thus, both olivine and diamond are predominantly sampled from the lower SCLM within ~50 km of the LAB (~212 km at Kimberley; Nimis et al. 2020). Conversely, garnet and clinopyroxene are predominantly sampled at shallower depths in the mid-SCLM (<160 km). The olivine population for the Kimberley kimberlites is also characterised by a high abundance of Cr-poor megacrysts (~40%; data from Howarth et al. 2022), which is consistent with the low diamond grade of the Kimberley kimberlites (Giuliani et al. 2023). However, the impact of kimberlite-related metasomatism on the vertical distribution of peridotitic olivine and diamond appears to be rather different from that observed in the Koidu mine with mid-lithospheric depths potentially more affected than the deep lithosphere (compare Fig. 12 with Fig. 10). The lithospheric mantle beneath Kimberley has witnessed a protracted history of metasomatic enrichment (e.g., Konzett et al. 1998; Giuliani et al. 2014a) with addition of abundant mica and other metasomatic phases through multiple episodes of melt/fluid enrichment (e.g., Erlank et al. 1987; Fitzpayne et al. 2020 and references therein). This peculiar setting might have fostered ascent and preferential stalling of kimberlite-related melts at mid-lithospheric depths as recorded by the occurrence of sheared peridotites (Katayama et al. 2009; Heckel et al. 2022) and mantle polymict breccias (Giuliani et al. 2014b) at these depths rather than in proximity of the LAB.

Conclusions

The Koidu indicator mineral population is dominated by megacrystic phases accounting for 14-30% of the olivine xenocrysts and 35-50% of the garnet xenocrysts. However, only ~3% of the clinopyroxene xenocrysts belong to the megacryst suite (the caveat here being that discriminating Cr-rich megacrystic from peridotitic clinopyroxene is difficult). The sampling depth distribution of olivine based on Al-in-olivine thermometry and pressures estimated by iterative calculations to match the local paleogeotherm indicates predominant sampling of the SCLM at depths of 110-150 km and a near absence of peridotitic olivine sampled in the lowermost 40 km of lithosphere. Clinopyroxene xenocrysts and eclogite xenoliths show bimodal depth distributions and, in combination with olivine, allow estimation of the vertical distribution of lithologies in the lithospheric mantle beneath Koidu. We outline three lithologically distinct regions in the SCLM: 1) the lowermost SCLM (~180-220 km) dominated by megacrystic material, 2) a layer containing abundant diamondiferous eclogites at depths of 160-180 km, and 3) a heterogeneous region containing lherzolite, locally diamondiferous harzburgite/dunite, and eclogite, at 110-150 km of depth. Notably, highly depleted lithologies (i.e., olivine Mg# >93) are only observed at depth <160 km, consistent with Fe-enrichment and metasomatic refertilization of the deeper SCLM.

The abundance of megacrystic material at Koidu indicates a significant stage of proto-kimberlite metasomatism in the deepest reaches of the lithosphere. At these depths diamonds appear to be absent – diamonds are predominantly sourced from between ~120 km (graphite-diamond transition) and 170 km of depth. This suggests that diamonds were either not present or, more likely, were destroyed during kimberlite-related modification of the deepest (170-220 km) SCLM. This finding is consistent with recent suggestions that carbonate-rich metasomatism of the deep lithosphere, probably mediated by failed pulses of kimberlite melt, is detrimental towards diamond preservation (Giuliani et al. 2023). This hypothesis for a major role of proto-kimberlite metasomatism in diamond destruction needs to be tested further in other localities exhibiting variable megacrysts contents and correlated diamond grades.

Ultimately, it is clearly apparent that olivine and the use of Al-in-olivine thermometry, in combination with other traditional indicator minerals (e.g., garnet, clinopyroxene), shows great potential as an evaluation tool to constrain sampling of potentially diamondiferous material during transit of kimberlite magmas through the SCLM.