Magmatic Ni–Cu–Co–platinum group element deposits are notoriously difficult exploration targets owing to a lack of alteration haloes or other extended distal footprints. Success requires prediction of prospective terranes, followed by identification of suitable host intrusions and deposition sites within those intrusions. At the regional scale, potential ore-hosting magmas tend to have lithophile trace element trends falling on mixing lines between primitive or slightly depleted source mantle and typical upper continental crust, with several significant exceptions. Most known deposits have parent magmas that are in the upper range of FeO content for given MgO compared with baseline data sets for continental large igneous province magmas. At the scale of individual intrusions, the presence of cumulate rocks, both mafic and ultramafic, is key. These can be recognized in regional datasets using combinations of magnesium number (molar MgO/(MgO + FeO), Al2O3, TiO2 and Zr contents. Combinations of alteration-resistant element ratios between Ni, Cr and Ti are also useful and can also be applied to moderately weathered samples. Concentrations and ratios of Cu and Zr are useful in discriminating chalcophile-enriched and depleted magma suites. In combination, these approaches can be combined to discriminate highly prospective cumulate-dominated magmatic suites and individual intrusions from non-cumulate suites with limited potential.

Supplementary material: Supplementary Appendix 1, procedure for determination of FeO–MgO contents of magmas from olivine-liquid mixing lines, and Supplementary Appendix 2, procedure and error analysis for correcting major element whole rock analyses volatile and sulfide free, are available at https://doi.org/10.6084/m9.figshare.c.6267664

Magmatic Ni–Cu–Co deposits hosted in mafic–ultramafic intrusions make up a large proportion of the global sulfide Ni resource endowment, and a large proportion of the global Ni–Co endowment overall (about 37%, based on figures from Mudd and Jowitt (2014). One single deposit, the Oktyabrsky system in the Norilsk–Talnakh region, includes well over half a trillion dollars’ worth of contained metals and a number of other large deposits exceed $100 billion in value (Mudd and Jowitt 2014; Barnes et al. 2020). With the expected increased demand for Ni, Cu and Co in the electric vehicle market, there is a greatly increased interest in exploring for this type of target.

Magmatic sulfide deposits are difficult exploration targets and deposits hosted in small conduit- or chonolith-type intrusions particularly so. The host intrusions in some cases are not much bigger than the deposits themselves (Lightfoot et al. 2012), there are no associated alteration haloes, structural controls are typically cryptic and the geological controls on deposition sites are not easily predictable. Electromagnetic methods are often confounded by the presence of barren conductors such as graphitic schists, and where deposits are steeply plunging or deeply buried their geophysical expression is minimal (Peters 2006). Successful exploration requires a comprehensive toolkit, of which lithogeochemistry can be a valuable part. Here we focus on applications of large modern routine multi-element assay datasets as used by Halley (2020) to address specific questions relevant to target selection and prospect-scale vectoring for Ni-Cu exploration (Fig. 1).

Lithogeochemical exploration takes two distinct but overlapping approaches: prediction, recognizing potentially ‘fertile’ host rocks; and detection, identifying the signatures of the ore-forming process. Obtaining the best value from the data requires addressing specific questions (Fig. 1).

  1. Are we in a prospective magmatic province where favourable ‘carrier magmas’ are present?

  2. Are suitable intrusive bodies present where magmas could carry and deposit sulfide liquids?

  3. Is there evidence that the magmas have been interacting with S-bearing country rocks along the flow pathway?

  4. Are the samples at or close to a deposition site, where silicate crystals are being deposited from the magma to form cumulate rocks?

  5. Is there evidence for the presence and/or deposition of sulfide liquid?

Clearly if the answer to the last question is yes, this is by far the best geochemical indicator of magmatic ore formation, but there are many examples of near-misses where lack of sulfide in an unmineralized part of an ore-bearing system could generate a false negative. Furthermore, signals of country rock interaction, in the form of geochemical indicators of wall-rock contamination of the magma, can be flushed out by uncontaminated fresh magma within the flow pathways, as is common in komatiite systems (Lesher and Arndt 1995). On the other hand, there are very few, if any, examples of ore-forming systems where there is not a positive answer to both questions 4 and 5. The geochemical proxies for these five questions are now considered, with the aim of producing some simple and reliable geochemical discriminants.

The prime mechanism for formation of Ni–Cu–platinum group element (PGE)-dominated magmatic sulfide ores (Fig. 1) has been generally agreed upon for several decades (Keays 1995; Barnes and Lightfoot 2005; Naldrett 2011; Barnes et al. 2016a). The essential elements are (1) a magma passing through some kind of trans-crustal conduit system, assimilating S, usually in the form of sulfide, from the country rocks; (2) the sulfide melt so formed reacting with this ‘carrier’ magma to sequester chalcophile elements; (3) a physical mechanism of segregation and accumulation of the sulfide liquid; and (4) a variety of physical processes including re-entrainment, gravity flow, country rock infiltration and in some cases tectonic mobilization, giving rise to the final disposition of the ores (Barnes et al. 2017, 2018). Of these processes, summarized in Figure 1, the first three all potentially leave geochemical imprints in the host rocks and magmas that are detectable and mappable as exploration proxies at a range of scales. However, determination of the extent to which these proxies can generate false positives and false negatives depends on proper understanding of the length- and time-scales at which the various component processes interact (Barnes and Robertson 2019). We consider here a variety of possible geochemical proxies, from regional to local scale, and assess their applicability and reliability.

‘Fertile magmas’

At the largest scale of exploration targeting and terrane selection, a widespread concept in economic geology is that of ‘fertile’ magmas: the capacity of a particular magmatic suite or episode to form ore deposits. This concept has been successfully applied in porphyry Cu exploration (e.g. Hao et al. 2017; Xu et al. 2021) and has proved to have some, albeit limited, application in komatiite-associated Ni–Cu-dominant magmatic sulfide mineral systems.

In the case of the komatiite association, the strongly endowed East Yilgarn province of Western Australia has some distinctive geochemical attributes, notably a strong signal of crustal contamination and a preponderance of highly magnesian olivine-rich cumulates compared with less-endowed provinces such as the Abitibi belt (Barnes et al. 2007; Barnes and Fiorentini 2012). However, these signals are regarded as a consequence of a craton–margin tectonic setting (Begg et al. 2010; Mole et al. 2014) and favourable volcanic facies (Lesher and Keays 2002; Barnes and Fiorentini 2012; Gole and Barnes 2020) rather than an input of particularly ‘fertile’ magma; very similar source magmas exist in the Abitibi and other less-endowed greenstone belts.

A corresponding global data analysis for deposits associated with mafic magmas is harder; unlike in komatiite terranes, unambiguous parent magmas to the host bodies are only rarely available for sampling. There are exceptions, where prospective intrusive bodies and overlying volcanic equivalents are both preserved within continental large igneous provinces, e.g. the Siberian Traps in the Norilsk–Talnakh region (Lightfoot et al. 1990; Naldrett et al. 1996c) and the Mid-Continent Rift (MCR) of central North America (Keays and Lightfoot 2015). Elsewhere, the geochemical character of the parent magmas must be inferred indirectly from the intrusions themselves, with complications due to local-scale wall-rock contamination and cumulus processes. In many exploration campaigns, particularly at the early greenfields stage, it is likely that most of the available geochemical information will come from sparsely sampled intrusive rocks, so it is useful to consider how much ‘fertility’ information can be extracted from such sampling. We consider two aspects, using a suite of lithophile and chalcophile trace elements from the databases of Barnes et al. (2015, 2021): firstly, the various lithophile and chalcophile trace element signatures that have been considered indicative of particular tectono-magmatic settings, and secondly the range of MgO–FeO contents (and hence temperature) of parent magmas inferred from whole-rock data and mineral chemistry.

Lithophile trace elements

The elements Nb, Yb, Th and the rare-earth elements (REEs), along with Zr and Y, are commonly used as indicators of tectonic setting. The latter application has been criticized by a number of recent publications (e.g. Condie 2015; Li et al. 2015a; Barnes et al. 2021), but the relative proportions of these elements are nonetheless important indicators of petrogenetic processes. Barnes et al. (2021) show that most of the useful variance in the entire suite can be captured in the commonly used plot of Th/Yb v. Nb/Yb devised by Pearce (2008, 2014) and we apply that approach here. Given that many of the samples used in this exercise are cumulates (see below), it is necessary for the purpose of this particular plot to apply some filters to eliminate the effect of crystal accumulation on the ratios. Hence the dataset in Figures 2, 3 is restricted to exclude adcumulate and mesocumulate rocks by filtering out samples with <0.5 ppm Th or <0.5 ppm Nb. (See Supplementary Appendix 2 for further discussion.)

The Th–Nb–Yb signature of continental large igneous provinces (LIPs) and their Archean and Early Proterozoic equivalents has changed gradually through Earth history (Barnes et al. 2021). The signature of Archean and Early Proterozoic mafic–ultramafic sequences, mineralized or not, is a consistent broadly linear trend ranging from primitive to slightly depleted mantle along a steep trend of variable Th/Nb ratio representing mixing of primary mantle melts with continental crust (Fig. 2a), and generally lacking the linear trend of constant Th/Nb from depleted MORB to OIB mantle, the ‘mantle array’ of Pearce (2008) (Fig. 2). The Early Proterozoic terranes of the northern Fennoscandian shield, specifically the mineralized Pechenga belt (Green and Melezhik 1999; Brugmann et al. 2000; Latypov et al. 2001; Hanski et al. 2011) (Fig. 2b) are the unique exception. In contrast, Phanerozoic LIPs such as the Emeishan and Siberian flood basalt provinces Figure 2c–e) have broad data clouds, interpreted here as mixtures of magmas derived along the length of the mantle array with superimposed continental contaminants. (This differs from the original interpretation of Pearce (2008), of derivation from metasomatized sub-continental mantle lithosphere.) The ‘mantle array’ component as defined by Pearce (2008) appears gradually in the geological record, first appearing clearly in the Early Proterozoic and becoming dominant by around 1100 Ma as represented by the MCR of North America (Fig. 2).

For the two most extensively mineralized LIPs, the MCR of North America, containing Eagle, Tamarack, Marathon and the very large but very low-grade deposits of the Duluth Complex (Fig. 2c), and the Siberian LIP (Fig. 2d, e), the mineralized intrusions tend to fall toward the low-Nb/Yb end of the array for the LIP as a whole, and have patterns broadly consistent with upper crustal contamination of basalt with a source similar to ideal ‘Primitive Mantle’ (McDonough and Sun 1995). This linear trend (‘Mixing PM–UCC’ in Fig. 2) is similar to that seen in Archean komatiite–basalt suites (Barnes et al. 2021). The same is true of the three units of the Siberian flood basalt sequence that are thought to be most closely associated with the Norilsk–Talnakh ores (Fig. 2e). The final plot in Figure 2 shows the data cloud from the sparsely mineralized Emeishan LIP, along with the data from hosts to several of the small Ni–Cu and PGE sulfide deposits known in the province. There is a clear difference from Siberia and the MCR, but a closer affinity with Pechenga, in that the province is deficient in the low-Nb/Yb component and dominated by the high-Nb/Yb end of the mantle array. It is noteworthy that the Siberian LIP also contains a high proportion of magmas plotting along the mantle array towards high Nb/Yb, but known mineralization is exclusively associated with magmas falling close to the PM–UCC mixing line.

Data on Proterozoic orogenic belts such as the Halls Creek Orogen (HCO; Le Vaillant et al. 2020), and the Fraser Zone of the Albany–Fraser Orogen (AFO; Maier et al. 2016; Taranovic et al. 2022b) show variable distributions (Fig. 3). The HCO data overlap the indistinguishable continental arc and continental LIP field, whereas the AFO data match more closely with typical Archean basalt–komatiite patterns with a population extending into the modern oceanic arc field, above the mantle array in the lower left of the diagram. In both, the mineralized intrusions define a narrower trend parallel to the PM–UCC mixing line, but somewhat to the low-Nb/Yb side of it in the Nova case. Broadly similar trends parallel to the PM–UCC mixing trend are shown by the host intrusions to Jinchuan, Xiarihamu and various deposits of the Central Asia Orogenic Belt (CAOB) in NW China. The main exception is Kalatongke, which also shows a parallel trend, but significantly displaced to higher Nb/Yb, implying a distinctly e-MORB-type mantle source (Fig. 3e). The displacement of the other CAOB deposits and Xiarihamu towards lower Nb/Yb may indicate a metasomatized arc mantle source contribution (Lu et al. 2019).

In conclusion, there is some evidence to suggest that magmas with trace element compositions falling close to mixing lines between primitive mantle and average continental crust have a greater tendency to occur in ore-bearing terranes and to be responsible for ore formation. The Morongovsky–Mokulaevsky suite of the Siberian Traps is a good example of this ‘sweet spot’. The main outlying exception is the Pechenga belt in far northwestern Russia (Green and Melezhik 1999; Brugmann et al. 2000; Latypov et al. 2001; Hanski et al. 2011), where the unusual ferropicrite magmas have trace element characteristics very similar to the ocean island end of the mantle array (Hanski et al. 2011). It is likely that they were derived from anomalous source mantle (Hanski and Smolkin 1995). Overall, the data are consistent with most deposits forming from contaminated primitive asthenospheric melts. However, this approach, while indicative, does not lead to a single reliable discriminant of ‘fertile’ magmas.

Chalcophile element depletion as a regional footprint

The chalcophile element contents, particularly the PGE contents, of mantle derived and ore-associated magmas have been extensively reviewed, for komatiites by Fiorentini et al. (2010, 2011) and for basaltic magmas by Barnes et al. (2015). The potential association of PGE-depleted basalts with the super-giant Ni–Cu–PGE deposits of the Norilsk–Talnakh camp has been identified in a number of studies (Brugmann et al. 1993; Naldrett et al. 1996b; Lightfoot and Keays 2005) and was widely assimilated into exploration strategies in the 1990s, but there is ongoing debate about whether the association is purely coincidental (Arndt 2011) or genetically related by complex sequences of events e.g. (Li et al. 2009; Yao and Mungall 2021). Taking a purely empirical view, Norilsk–Talnakh remains to this day the only example in a mafic magmatic system where a clear spatial relationship exists between mineralized intrusions and PGE-depleted lavas, although it is now generally agreed that the lavas in question are slightly older than the intrusions (Latyshev et al. 2020) and are geochemically not directly related (Latypov 2002). Elsewhere, it is very rarely possible to sample rocks that demonstrably represent the parent magmas to ore-hosting intrusions; either they are contaminated by wall rocks in the margins of the intrusions, or they have been eroded away. Hence, the applicability of chalcophile element depletion to fertility analysis of a particular terrane is limited. The plume-related Tertiary provinces of east and west Greenland may be exceptions, where intrusions can be recognized in the basement to a lava pile containing PGE-depleted basalts (Keays and Lightfoot 2007).

The applicability to mafic-related systems in orogenic belts is complicated by the possibility that mantle-derived magmas may be relatively low-degree partial melts that were sulfide-saturated at source and hence left PGEs in the mantle. This effect is seen in the tendency of lower-MgO basalts to have a wider spread of PGE contents towards lower values (Barnes et al. 2015, 2016a, b), and also in magmatic sulfide deposits in orogenic settings typically having significantly lower PGE contents over a similar range of Ni (Fig. 4), a feature that cannot be explained by lower R factors. Consequently, PGE depletion is not a reliable signal of ore formation in orogenic settings.

Finding hot magmas

It is a widely believed principle that magmatic sulfide ores are preferentially associated with high-MgO rocks and therefore by inference high-temperature magmas. This is intuitively obvious: high-temperature, high-MgO magmas have a greater potential to assimilate their country rocks, and hence to achieve the first stage of sulfide melt formation. However, applying this principle is complicated by the fact that in exploration contexts we only rarely (if ever) sample liquids, and most of the rocks we analyse are cumulates. The MgO content and the Mg/Fe ratio of an ultramafic cumulate rock is inevitably higher than that of the magma it formed from.

Major elements: olivine as an MgO/T proxy

In the present context of fertility, a better index of the temperature of the ore-forming magma is the Mg number [molar [Mg/(Mg + Fe)] of the most common associated cumulus silicate mineral in such systems, olivine.

A compilation of the range of measured olivine compositions (expressed as molar percent forsterite content, Fo, the same thing as the Mg number) is shown in Figure 5, for small intrusions with basaltic to high-Mg basaltic parent magmas. Fo in olivine shows a very wide range between high values around 90 (implying high-Mg basaltic parents) and low values of less than 50 implying relatively low-T, evolved magmas. One of the larger deposits in this class, Voisey's Bay, has among the least forsteritic olivines. On these empirical grounds, Mg numbers and olivine compositions do not appear to have a great deal of predictive or discriminant ability in Ni sulfide systems. It is probably true (although unproven) that within any given province the most Mg-rich olivines are most likely to be associated with ore, but this is not applicable between provinces. A detailed analysis of olivine compositions in relation to mineralization is provided by Barnes et al. (2023).

Major elements: MgO and FeO in parent magmas

Several studies over the years have suggested that a class of mafic to ultramafic magma called ‘ferropicrite’ may be preferentially associated with magmatic Ni–Cu deposits (e.g. Lu et al. 2019). This is undeniably true for the Pechenga deposits in northwestern Russia (Hanski et al. 2011) although, as we have seen, these are highly distinctive in their trace element chemistry compared with other mineralized magma suites. The definition of ferropicrites is not consistently agreed upon, but essentially they are magmas with moderate to high MgO (10–20%) and high FeO (>12%) (Fig. 6). Regardless of the definition, it is useful to investigate whether FeO as well as MgO contents of parent magmas may be predictive.

As previously noted, direct determination of parent magmas to intrusions is fraught with difficulty, but some estimates can be made where olivine cumulates of variable composition can be recognized within an intrusion. This method assumes that a suite of rocks exists which can be approximated as mixtures of olivine with constant composition and variable proportions of liquid in equilibrium with that olivine. (The model requires a number of steps, described in detail in Supplementary Appendix 1). This method was first used in the present context for the Jinchuan intrusion (Chai and Naldrett 1992) and has since been used in a number of studies, most recently by Taranovic et al. (2022b) for the host intrusions to the Nova–Bollinger deposits.

A summary of parent magma estimates for a variety of intrusion-hosted deposits and potentially associated mafic volcanic suites is shown in Figure 6. Generally, MgO contents of ore-forming magmas are towards the high end of the field for continental flood basalts in both MgO and FeO, but below the range of the distinctive ferropicrites assumed to be the parent magmas of the Pechenga orebodies. They all fall outside the main cluster for continental arc basalts. The natural range is wide, and several deposit (Norilsk–Talnakh, Savannah, Voisey's Bay, Huangshandong) have unexceptional parent magmas well within the typical range of continental plume magmas. However, there does appear to be a threshold MgO–FeO liquid composition curve (red dashed line in Fig. 6) below which none of the deposits represented here fall. This combination of high MgO and high FeO reflects a combination of high melting temperature and depth of generation; higher pressure produces high MgO for a given degree of partial melting and hence also higher FeO (Herzberg et al. 2007). This lends some support to the idea that deep-seated Fe- and Mg-rich plume magmas are ‘fertile’ and that FeO–MgO whole-rock relationships may be at least moderately diagnostic. A combination of liquid Fe enrichment for given MgO together with lithophile trace elements appears promising as a fertility indicator.

Recognizing deposition sites: identifying cumulate rocks

Favourable intrusions for magmatic sulfide deposits are marked by the presence of cumulate rocks: the solid products of fractional crystallization. Cumulates are igneous rocks formed by the accumulation of liquidus phases separated from their parent magmas, regardless of the process by which this accumulation occurs. Some cumulates probably form by mechanical accumulation driven by gravity, either by crystal settling or by deposition from gravity flows, while others form by in situ nucleation and growth (Wager et al. 1960; Campbell 1978; Lesher and Keays 2002; Latypov et al. 2017, 2020) or a combination of both (Mao et al. 2018). The accumulating components in a cumulate rock are referred to as cumulus phases. Magmatic Ni–Cu sulfide deposits exist where one of the cumulus phases is immiscible sulfide liquid, which is likely to have been transported initially as suspended droplets (Robertson et al. 2015) and deposited mechanically by processes related to magma flow dynamics (Barnes et al. 2016a; Yao et al. 2020; Yao and Mungall 2021). The presence of cumulate silicates, particularly cumulus olivine, pyroxene and spinel, is the distal footprint of these deposition sites, so detection of cumulate rocks is an important objective of lithogeochemistry.

Cumulate rocks can form in two distinct settings relevant to prospectivity: (1) closed-system differentiated bodies, where a body of magma is emplaced in a single event and undergoes fractional crystallization in place; and (2) where cumulates are deposited in a dynamic open system such as a feeder conduit, with continuous flux and replenishment (Fig. 7). Both situations can arise in large or small intrusions; e.g. closed-system differentiation in a large layered intrusion such as Skaergaard (Tegner et al. 2009) or Kiglapait (Morse 1996), or open-system replenishment in the Bushveld Complex (Cameron 1978) or in small ore-hosting conduits such as Jinchuan (Li and Ripley 2011) and Xiarihamu (Song et al. 2016). The open-system case is much more favourable for Ni–Cu sulfides. The hallmark of open systems is accumulation of a high proportion of uniform cumulates with a limited range of cumulus mineral compositions, reflecting a steady-state balance between crystallization and recharge.

A third setting, related to the second, is mechanical accumulation of crystals and sulfide liquid from a flowing slurry injected into ‘dead end’ intrusions; such systems are not strictly open, but reflect a population of crystals and droplets that have accumulated from a relatively large volume of magma and are hence homogeneous in composition. Geochemically, these are indistinguishable from open-system conduits. This interpretation has been placed on the Nova–Bollinger deposit (Barnes et al. 2022; Taranovic et al. 2022b) and (controversially) on the ore-hosting Norilsk–Talnakh intrusions (Krivolutskaya et al. 2018; Yao and Mungall 2021).

The great majority of the known deposits in small mafic–ultramafic intrusions is associated with cumulus olivine, by itself or with cumulus pyroxene. Orthopyroxene is the predominant cumulus pyroxene, and is in some cases the dominant phase, e.g. at Ntaka Hill (Barnes et al. 2016b, 2019a), such that harzburgite or olivine orthopyroxenite are probably the most common host rocks in small intrusions. Kevitsa (Luolavirta et al. 2017) and the deposits of the Pechenga belt (Hanski et al. 2011) are unusual examples of deposits associated with olivine–clinopyroxene cumulates (wehrlites) with minor orthopyroxene. Pyroxenes in ore-related conduit intrusions tend to have complex trace-element zoning patterns (Schoneveld et al. 2020b). In more evolved systems plagioclase is also a cumulus phase, and several important deposits have cumulus assemblages of olivine–orthopyroxene–clinopyroxene–plagioclase (olivine gabbronorite) or olivine–clinopyroxene–plagioclase (olivine gabbro) as the dominant lithology; the major example of the latter is Voisey's Bay (Naldrett et al. 1996a; Li and Naldrett 1999). Olivine and olivine–plagioclase cumulates are a major component of the host sills to the Norilsk–Talnakh orebodies (Czamanske et al. 1995; Barnes et al. 2019b). Chromian spinel, usually chromite, is a very widespread accessory phase in most ore-related ultramafic cumulates, particularly olivine-rich ones, but tends to disappear when pyroxene becomes a cumulus phase. Hence, identification of olivine–chromite+/−orthopyroxene cumulates is an important objective for lithogeochemical exploration.

Whole-rock compositions of cumulates are determined by the identity and proportion of the cumulus phases and the proportion of parent liquid trapped between the cumulus grains. This can vary from almost zero in adcumulates to as much as 60% in orthocumulates. Small changes in liquid composition close to phase boundaries can generate large discontinuous changes in the mineralogy and whole-rock composition of cumulates, as illustrated in Figure 8. Such changes can be exploited in the recognition of cumulus rocks in geochemical databases.

The main messages from the phase diagram (Fig. 8) are these.

  1. Ultramafic rocks do not require ultramafic magmas. In fact, most ‘normal’ mantle-derived basalts can make ultramafic cumulates, provided that they have not evolved too far from their original (mantle melt) compositions.

  2. A very minor change in the chemistry of the magma can cause a big jump in the cumulate rock it produces, e.g. from a peridotite (olivine + pyroxene (ultramafic)) to a troctolite or an olivine gabbro (mafic). These jumps are commonly present as sharply bounded layers in ore-hosting intrusions (Latypov et al. 2020).

  3. A small change in the ‘starting composition’ can cause a big change in crystallization sequence: e.g. changing the starting composition from A to E in Figure 8c by a small addition of SiO2 causes the crystallization path to change from dunite–troctolite–norite along the path E-F-H-H′ to dunite–harzburgite–orthopyroxenite–norite along path A-G-H′. The harzburgites formed this way have a characteristic texture called ‘poikilitic’ where large grains of orthopyroxene enclose many smaller, partially dissolved crystal of olivine. This is probably the most widespread rock type associated with intrusion-hosted Ni–Cu–Co deposits.

  4. The further down the crystallization path, the more the solid cumulate product chemically resembles the magma it crystallizes from.

One or two phase ultramafic cumulates (neglecting chromite) are generally easy to identify from geochemical data in that they form linear arrays on standard binary geochemical plots. However, as magmas become more evolved, the cumulates become harder to recognize (point 4 above). In the idealized phase diagram in Figure 8, they are identical at the ternary eutectic point E. In natural multicomponent systems they continue to evolve with the addition of further cumulus phases such as magnetite, ilmenite and, in extreme cases, apatite, but will still have the same broadly mafic mineralogy as the products of isochemical solidification of the starting magma. Practically this means that plagioclase-bearing cumulates, i.e. gabbronorites and olivine gabbros, can be difficult to distinguish on their major element chemistry from non-cumulate mafic rocks representing solidified liquids. This can be resolved by the use of two approaches: plots using (Table 2) whole-rock Mg number (Mg#, molar percent MgO/[MgO + FeO]) (Fig. 9), and molar ratio variation diagrams.

Figure 9 shows a way to distinguish a cumulate gabbro (indicating a deposition site: position 4, Fig. 8) from a mineralogically similar rock that simply represents the magma crystallizing to a solid of the same composition – such as might be found in a chilled margin, for example (position 3, Fig. 8). Using whole-rock data, a plot of Al2O3 wt% v. Mg# discriminates cumulate from non-cumulate rocks. This is because cumulates have higher Mg#, due to Fe–Mg minerals always having higher Mg# than the magmas they crystallize from. Ultramafic cumulates have high Mg# and low Al2O3 and plot along curved mixing lines representing mixtures of olivine (or pyroxene, or both) with trapped liquid. Gabbroic cumulates also have high Mg#, although typically slightly lower than ultramafic cumulates because they tend to crystallize from more evolved liquids as is evident from the phase diagram model, but they have much higher Al2O3 because of the presence of cumulus plagioclase (vertical vector, Fig. 9c).

Not all ore-hosting intrusions contain ultramafic cumulates, where the cumulus phases are combinations of olivine, pyroxene and (usually) minor chromite, but there are very few that do not. As can be seen in Figure 9, the ore-bearing intrusions in a number of prospective belts are strongly dominated by cumulate rocks compared with other mafic rocks in the same belt, a particularly clear example being the Nova intrusions in the Fraser Zone of Albany Fraser Orogen. The Hart Dolerite represents a very high-volume LIP almost completely devoid of cumulate rocks that has so far proved entirely barren for this deposit type. This would be typical of the signatures of unmineralized suites of mafic rocks.

There are two important caveats in the use of the Al2O3 w% v. Mg number plot. Firstly, in closed-system intrusions, such as the Skaergaard intrusion or differentiated dolerite sills like the Golden Mile Dolerite at Kalgoorlie, the Mg number of the cumulus phases can evolve to very low values – almost to zero in the Skaergaard case, where almost complete fractional crystallization took place (Nielsen et al. 2015). Such rocks would plot in the ‘non-cumulate’ field of the Al2O3 v. Mg number plot. These can be discriminated by plotting a highly incompatible element, such as Zr, against Mg# as in Figure 9d. Such elements are at much lower concentrations in cumulates than the parent liquid for the same Mg#, due to the presence of the (e.g.) Zr-free cumulus phases. Where this approach is used, the Zr–Mg# plot should be used first to filter for this category. The second caveat concerns komatiites, where non-cumulate komatiitic liquids can plot at high Mg#; these diagrams should be used only in provinces dominated by mafic magmas.

An alternative to the Al2O3 w% v. Mg number plot uses the mass ratio Al2O3/TiO2 v. Mg number. An example is shown in Figure 10. Mixing of ultramafic cumulus assemblages of olivine and/or pyroxene with liquid generates an approximately horizontal trend, because the Al and Ti contents of the solid phases are low. Addition of cumulus plagioclase causes a rapid increase in Al2O3/TiO2, generating the L-shaped trends shown. This plot is complicated by the effect of the liquid component becoming progressively enriched with Ti over Al as fractionation proceeds, up to the point of magnetite saturation, and also by the presence of high Al in cumulus pyroxenes in high-pressure cumulates such as those at Nova–Bollinger, but is nevertheless useful in discriminating relatively primitive ultramafic and mafic cumulate rocks.

Recognizing ultramafic cumulate rocks is generally fairly straightforward: they are high in Mg, Cr and Ni and low in components such as Al and Ti that are not concentrated in these minerals. However, ultramafic rocks are very susceptible to modification of chemistry by alteration. Orthopyroxene is a particularly useful indicator, in that most mantle-derived magmas do not crystallize much of it. The presence of orthopyroxene cumulates is a good indication that magmas have been contaminated with silica-rich country rocks (Jenkins and Mungall 2018), causing the shift from composition E to A in Figure 8c, which is another positive indicator for fertility.

Figure 11 shows a geochemical technique that allows the recognition of olivine and orthopyroxene cumulates from whole-rock geochemistry, to be used in conjunction with Figures 9 10. The mineralized Ntaka Hill intrusion in Tanzania (Barnes et al. 2019a, b) shows up clearly as an intrusion with abundant orthopyroxene cumulates (Fig. 4). This method requires reliable SiO2 analyses, which are not provided in some commercial laboratory element suites. Silica is such an important component that it is worth the additional cost to analyse it using inductively coupled plasma mass spectrometry or X-ray fluorescence.

A further advantage of this approach, and the Al/Ti v. Mg# plot, is that element ratios are unaffected by dilution by volatiles (H2O, CO2) during alteration. In the case of ultramafic rocks alteration can be accompanied by introduction of up to 20% of these components, causing reduction of major element oxides due to closure (analyses must sum to 100%). Use of element ratios and triangular plots mitigates this effect.

At a prospect scale, the focus moves from target selection to vectoring towards ore and eventually (hopefully) to orebody definition. Geochemical proxies are used at this scale to identify the distal signals of ore-forming processes. Within prospective intrusions, mapping out ultramafic and gabbroic cumulates from non-cumulate chilled liquid rocks using spatially constrained geochemical datasets, as described above, provides a powerful tool for unravelling the internal structure of a potentially fertile magmatic system. More obviously, recognizing subtle signals of sulfide deposition and fractional extraction can potentially serve as direct vectoring tools.

Ni in olivine and whole rock

Nickel contents of olivine have been widely used as petrogenetic indicators and as fertility indicators for magmatic sulfide potential of mafic–ultramafic intrusions, mainly predicated on the assumption that olivines crystallized from magmas that had equilibrated with sulfide liquid should be relatively depleted in Ni compared with sulfide-free baseline (Häkli 1971; Duke and Naldrett 1978). This has given rise to a large accumulation of data on volcanic and intrusive rocks. Results are discussed in detail by Barnes et al. (2023) and a brief summary follows.

Ni content of olivine, at given Fo content, is subject to wide range of controls, not all attributable to sulfide interaction. Baselines for Ni in olivine in relation to Fo content are somewhat lower in orogenic belt settings relative to intrusions in continental LIPs. No clear, universal discrimination is evident in Ni in olivine between ore-bearing, weakly mineralized and barren intrusions even when tectonic setting is considered. However, sulfide-related signals can be picked up at intrusion scale in many cases. Low-R factor, low-tenor sulfides are associated with low-Ni olivines in a number of examples such as Kabanga (Maier et al. 2011) and these cases stand out clearly. Anomalously high-Ni olivines are a feature of some mineralized intrusions. In these cases, enrichment may be due to crystallization of trapped liquid in orthocumulates or re-entrainment of ‘primitive’ Ni-rich sulfide by a more evolved Fe-rich magma, driving the olivine to become Ni-enriched due to a Fe–Ni exchange reaction between sulfide and olivine. Wide variability of both Fo and Ni within and between related intrusions at regional scale may be a useful prospectivity indicator. In general, the use of Ni–olivine as a fertility tool is more likely to generate false negatives than false positives, but both are possible.

Nickel in olivine is an important control on whole-rock Ni contents of sulfide-free cumulate rocks, but in many cases the whole-rock Ni signal is overwhelmed by the variation in the modal proportions of the different cumulus phases in the rock (Fig. 12). This is important for establishing baseline levels of silicate Ni in unmineralized rocks and hence for recognizing the presence of minor components of sulfide. Because of the uncertainties involved, Ni background levels in multi-phase cumulates are widely variable (Fig. 12), such that it is generally more reliable to use the Cu content rather than the Ni content as a proxy for presence of minor traces of magmatic sulfide. This will apply as long as the rocks are not excessively altered such that Cu may be mobile.

Detecting trace sulfides: Cu and Cu/Zr ratios in fresh cumulate rocks

Copper behaves as an incompatible element during crystallization of cumulus silicate minerals from a sulfide-undersaturated magma, such that the ratio of Cu to other incompatible elements such as Zr remains constant. Furthermore, it retains the value in the original magma and indeed in the original mantle source (assuming no sulfide was left in the mantle restite). Extraction of a cumulus sulfide component causes Cu to be depleted relative to Zr due to the strongly chalcophile character of Cu (Kiseeva and Wood 2015), such that this ratio can be used as a proxy for ore-forming processes (Sun et al. 1991; Maier et al. 1998). Values of Cu/Zr significantly lower than the expected PM value of c. 5 (McDonough and Sun 1995) are potentially indicative of magmas that have experienced extraction of sulfide liquid. Conversely, rocks that contain even a small component of accumulated sulfide liquid should have Cu/Zr greater than the PM value.

This principle is tested using three extensive data sets, the Fraser Zone, the Halls Creek Orogen (Table 1) and the Hart Dolerites (Fig. 13).

The Fraser Zone data show a very clear distinction between the mineralized Nova–Bollinger complex and the ‘background’ Fraser Gabbros and other intrusions and meta-dolerites. Presence of cumulus sulfides is clearly defined by Cu/Zr values on the Cu-rich side of the mantle line while the background grouping straddles the line and shows a strong mode with high Zr and distinctly depleted Cu. These compositions are plausibly interpreted as magmas that fractionated sulfide liquid, some of which may have been picked up and transported into the orebodies. A similar although less well-defined pattern is observed in the Halls Creek dataset, although here the mineralized samples appear to contain both Cu-enriched and Cu-depleted components. The mainly unmineralized background straddles the mantle line and extends to distinctly Cu-depleted compositions. The Hart Dolerite set is strongly clustered on the Cu-depleted side of the mantle line with no samples indicating Cu-enrichment. The apparent depletion in this case is so consistent that it probably represents a source characteristic, i.e. the assumption of a mantle source with Cu/Zr = 5 is not correct. More significant is the fact that the unmineralized suite has a consistent, tightly clustered distribution with no enrichment, in contrast to the mineralized belts showing a wide spread of both depleted and enriched samples.

The benefit of the Cu–Zr approach is that these elements are commonly available in regional pre-competitive datasets such as geological survey databases. They are therefore amenable to data mining in a way that other potential discriminants such as PGEs are not. The potential drawback is the high mobility of Cu during hydrothermal alteration, such that the approach is only reliable where the rocks are, for the most part, reasonably pristine.

Detecting contamination using variably incompatible lithophile elements

A key component of the standard genetic model is that ore-forming magmas need to interact with the country rocks in order to assimilate sulfide (or sulfate) and this process should lead to distinct signals of crustal contamination. This approach has been widely applied to komatiitic systems, where it appears to be useful as a belt-scale discriminant but not so much on a local scale, due to the complexities of flushed and recharged magma channels (Lesher and Arndt 1995; Lesher et al. 2001; Barnes et al. 2007, 2013; Barnes and Fiorentini 2012) and the effects of variable time-scales for component processes in ore formation (Barnes and Robertson 2019). However, the approach has generally proved less successful in mafic-hosted systems.

We have already seen geochemical effects of contamination in the Th/Yb v. Nb/Yb plots (Figs 2, 3). Mineralized intrusions tend to follow the crustal contamination trend of steeply increasing Th/Yb over limited Nb/Yb, which follows from the high abundance of Th relative to Nb and Yb in most crustal rocks. This trend is essentially the same as that seen in Archean komatiite–basalt sequences. However, within individual provinces such as the Frazer Zone and Halls Creek Orogen, there is no particular preference for crustal contamination trends to be present in the ore-bearing intrusions as opposed to the regional unmineralized or weakly mineralized intrusions. It is likely that crustal contamination is so widespread in continental settings that proxies for it generate far too many false positives to be useful at the scale of individual intrusions.

Use of preserved element ratios

When interpreting lithogeochemical data on highly altered or moderately weathered rocks, it is important to recognize that some otherwise informative elements such as Mg, Si, S and Cu might be highly mobile and hence useless. A solution to this problem is to use ratios of relatively immobile elements whose relative proportions are insensitive to alteration and mild to moderate degrees of weathering. This category includes such useful elements as Ni, Cr, Ti, Zr and the REEs, and use of ratios between these elements has been applied to mapping of variably weathered komatiites and basalts in lateritic terranes. (Barnes et al. 2014). Triangular plots of combinations of these elements are particularly reliable and informative (Fig. 14).

The primary purpose of this paper is to de-mystify some of the principles of igneous petrology that underpin geochemical variability, and to translate those into easily usable proxies that can add value to large bodies of data acquired during exploration programs. Some of the plots shown here are also applicable to data-mining legacy datasets.

Based on an extensive data compilation, exemplified by the regional datasets we have presented here, there are several distinct proxies that, when taken together, can be used to prioritizetargets.

  1. Mineralized terrains tend to have incompatible trace element patterns indicative of mixing of magmas from primitive or mildly deleted mantle sources with an overprint of contamination by continental crust. These can be identified on plots of Th/Nb v. Nb/Yb.

  2. Mineralized intrusions in almost all cases contain cumulate rocks, with a strong preponderance of olivine-bearing cumulates. Orthopyroxene cumulates are favourable indicators in some terranes but are not universally present. These can be identified using a number of different plots involving whole-rock analyses of Mg, Fe, Al and Zr, and triangular plots using whole-rock Ni, Cr and Ti.

  3. Terrains where mafic intrusions are dominated by non-cumulate rocks tend to have low prospectivity.

  4. Indicators of high-Mg magmas such as high forsterite contents in olivine do not appear to have useful predictive value.

  5. Chalcophile element enrichments and depletions at terrane and intrusion scale are positive indicators, with Cu v. Zr being a useful discriminant in all but highly altered rocks. PGEs are of limited use owing to the very wide variability in parent magmas.

  6. Widely variable Ni content in olivine and pyroxene for similar Mg# is a strong positive indicator, but requires high-precision microprobe analyses.

It is important to emphasize that these approaches should be combined with all other available datasets in a weights-of-evidence approach. There are no geochemical magic bullets. Importantly, geochemical datasets are self-evidently only applicable to the rocks that were sampled. That said, it is hoped that the tools and techniques presented here will allow explorers to apply geochemical proxies for ore-forming processes, improve rock type identification and in other ways add value to the large volumes of geochemical data that already exist and continue to be collected.

*See Supplementary Appendix 1 for splitting of whole rock FeO, Fe2O3 and sulfide-associated Fe. The approximation potentially introduces a small error, up to about relative 3%, in cumulate rocks.

Thanks to Steve Beresford for critiquing an early version of the manuscript, Mike Lesher and Jim Mungall for helpful reviews, IGO Ltd and Panoramic Resources for financial support for recent studies of Nova–Bollinger and the Halls Creek Orogen and multiple industry supporters for financial support and access to data over several decades.

SJB: conceptualization (lead), data curation (lead), investigation (lead), methodology (lead), writing – original draft (lead), writing – review & editing (lead)

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

All data are from previously published sources or public domain datasets and are fully referenced in the text. A zip file containing diagram templates for use in IoGas software is available at https://research.csiro.au/magnico/workshops-and-resources/.

Data for selected figures is compiled and available from the CSIRO data Access Portal, https://data.csiro.au/collection/csiro:56131.

This is an Open Access article distributed under the terms of the Creative Commons Attribution 4.0 License (http://creativecommons.org/licenses/by/4.0/)