Evaluation of Sc concentrations in Ni-Co laterites using Al as a geochemical proxy

16 Scandium (Sc) is used in several modern industrial applications, including aluminium-based alloys and 17 solid oxide fuel cells. So far, Sc production remains marginal as the lack of reliable and cost-effective 18 production limits its widespread adoption by the industry. Recently, significant Sc concentrations (~100 19 ppm) were reported in some nickel-cobalt lateritic ores, where Sc may be valuably co-produced along 20 with Ni and Co. However, Sc is typically not included in routine analyses of Ni-Co ores, precluding the 21 assessment of Sc concentration and distribution in existing Ni-Co deposits. This contribution examines 22 the relevance of using routinely analysed elements as geochemical proxies for providing first-order 23 estimates of Sc concentration and distribution in already assayed Ni-Co deposits. Three Ni-Co lateritic 24 deposits from New Caledonia, developed after harzburgite (Ma-Oui deposit, Koniambo mine), 25 harzburgite-dunite (Coquette Red tenement, Cap Bocage mine) and lherzolite (East Alpha deposit, 26 Tiébaghi mine), were investigated. In each deposit, we demonstrate that Sc is globally well correlated 27 with Al 2 O 3 , providing that the parent lithology is homogeneous and relatively depleted in Al. Deposit-28


Introduction
Scandium (Sc) is mainly used as a hardening additive to aluminium to form Al-Sc alloys for aerospace industries and manufacture of high-quality sports equipment (Royset and Ryum, 2005;Toropova et al., 1998).In addition, Sc is notably used in high-temperature lights, lasers and ceramics manufacturing and finds promising application in the development of Solid Oxide Fuel Cells (SOFCs).The global Sc supply and consumption remain marginal (~15 to 25t/yr Sc2O3, US Geological Survey, 2022).Sc is solely recovered as a by-product through titanium, zirconium, uranium, cobalt and nickel process streams.
There, Sc enrichment is largely residual and results from the intense leaching of mobile cations during the lateritisation of the parent rock.Scandium is thus trapped and concentrated in neo-formed goethite, and moderate remobilisation can occur during repeated stages of goethite dissolutionrecrystallisation and transformation into hematite.The chemical speciation of Sc in laterites from Australia and the Philippines has been investigated using a combination of X-ray absorption near-edge structure (XANES) spectroscopy and sequential extractions (Chassé et al., 2017(Chassé et al., , 2019;;Qin et al., 2020).
These studies further support the preferential affinity of Sc for goethite compared to hematite or smectite.However, the relative importance of adsorption and incorporation processes and the impact of goethite crystallinity in the formation of Sc-rich goethite zones remain debated.They may vary from one deposit to another.In New Caledonia, significant Sc concentrations were reported in several Ni-Co lateritic oxide ores (Teitler et al., 2019;Ulrich et al., 2019).There, maximum Sc grades reach ~100 ppm in the yellow limonite horizon.Although such concentrations are too low to be economically attractive as primary resources, Sc could be a valuable by-product of Ni and Co processing, provided that Sc-rich zones sufficiently overlap Ni-and Co-rich zones (Teitler et al., 2019).However, Sc is typically not routinely analysed during exploration and resource estimation of Ni-Co laterites.Consequently, mining operators do not evaluate the potential Sc resources hosted in Ni-Co lateritic ores.
In contrast, Fe, Al and Cr are routinely analysed and could serve as geochemical proxies for inferring the distribution of Sc concentrations in lateritic Ni-Co ores as suggested by previous investigations (Teitler et al., 2019).Although Fe is the predominant constituent of lateritic oxide ores with up to 80 wt% Fe2O3, Teitler et al. (2019) suggested that Fe may be correlated to Sc only in the lower to intermediate sections of the lateritic profiles and that the Sc-Fe2O3 correlation is usually not valid in the Sc-rich, yellow limonite horizon.In contrast, Sc is better correlated with Al in all the facies of the lateritic profiles, except for the uppermost ferruginous duricrust.
Nevertheless, the Sc-Al2O3 correlation proposed by Teitler et al. (2019), based on the combined analysis of individual vertical profiles from several deposits, shows significant dispersion and does not allow to assess how reliable the Sc-Al2O3 proxy is at the deposit scale.In this contribution, we discuss the relevance of the Sc-Al2O3, Sc-Fe2O3 and Sc-Cr2O3 correlations at the deposit scale in three Ni-Co deposits from the Koniambo, Cap Bocage and Tiébaghi mines, New Caledonia, based on the geochemical analysis of spatially and lithologically representative samples from each deposit.Using laser ablation inductively coupled mass spectroscopy (LA-ICP-MS), we then examine the dependency of bedrock mineral compositions on the Sc-Al2O3 regression coefficients and the variability of Al and Sc contents in secondary-formed minerals.Third, we conduct selective chemical leaching on some Scbearing samples to provide further insights on the speciation of Sc in the investigated deposits.These results are used to discuss the potential of the Al proxy for the evaluation of Sc concentrations and distribution in peridotite-hosted laterites in the perspective of Ni-Co-Sc co-valorisation.

Regional geology
The "Grande Terre" island of New Caledonia (Fig. 1A) consists of a 300 km long allochthonous peridotite ophiolite referred to as the "Peridotite Nappe".The peridotite Nappe, formed at ca. 35 Ma (Cluzel et al., 2001(Cluzel et al., , 2012;;Paquette and Cluzel, 2007), represents about 30% of the surface of Grande Terre and is exposed in the "Massif du Sud" and several isolated tectonic klippes, mainly aligned in the N140° direction along the west coast of the island (Koniambo, Tiebaghi, Poum massifs).The Peridotite Nappe is primarily composed of harzburgite locally interlayered with dunite, except in the northernmost klippes where lherzolite dominates (Ulrich et al., 2010).Notably, the extensive lateritisation of the peridotite ophiolite during or soon after obduction led to the development of a thick regolith cover and the formation of world-class Ni-Co(-Sc) lateritic resources representing about 10% of the world's nickel reserves (Maurizot et al., 2020).The development of Ni-Co(-Sc) laterites in New Caledonia results from (i) the leaching of most cations including Mg and Si after hydrolysis of olivine and pyroxene, (ii) the redistribution and concentration of Ni within the saprock as secondary hydrous silicates, (iii) the development of Ni-bearing saprolite, containing both silicates and oxides, at the expense of the saprock, and (iv) the development of oxide-rich, Ni-bearing limonite and duricrust at the expense of the saprolite (Butt and Cluzel, 2013;Cathelineau et al., 2016Cathelineau et al., , 2017;;Freyssinet et al., 2005;Golightly, 2010;Manceau et al., 2000;Trescases, 1975;Wells et al., 2009).The Ni-rich (> 2.0 wt% Ni) Ni-silicates ore reserves are rapidly being depleted as they have been actively mined since the late 19th century.Lower-grade (1.0-2.0%Ni) oxide ore ("limonite") reserves will therefore represent the bulk of Ni reserves of New Caledonia in the future.Also, Ni oxide ores often yield elevated Co (>2000 ppm) and Sc (60-100 ppm) concentrations adding significant value to the ore (Fig. 1B).Maximum Co concentrations (> 2000 ppm) typically occur at the interface between saprolite and limonite (referred to as transition zone), where Co is mainly associated with Mn-oxides (e.g.asbolane, lithiophorite).
Upwards in the limonitic horizons, Co concentrations decrease being partially scavenged and trapped in ochreous goethite (Dublet et al., 2017).Maximum Sc grades are reached in the yellow limonite horizon (Fig. 1B).Therefore, Ni-, Co-and Sc-rich zones in limonite may partly overlap each other, and the degree of such overlap is critical in the perspective of co-valorising Ni, Co and Sc from lateritic Ni ores.
Fig. 1: (A) Geological map of New Caledonia and location of the investigated mine sites.Modified from Maurizot and Vendé-Leclerc (2009).(B) Geochemical evolution along a typical Ni-lateritic profile in New Caledonia (modified from Bailly et al., 2014).

Sampling strategy
Sampling was conducted in multiple locations within the investigated deposits, along several drillcores and pit walls, encompassing the diversity of representative lithofacies.The objective was to check whether Sc-Al2O3 correlations may be generalised at the deposit scale.At the Ma-Oui deposit (Koniambo), 39 samples were collected along 6 drillcores and one pit wall profile (Fig. 2A).At the Coquette Red tenement (Cap Bocage), 27 samples were collected along 2 drillcores and one outcrop profile (Fig. 2B).At the East Alpha deposit (Tiébaghi), 48 samples were collected along 9 pit wall profiles, including one located in saprolite formed after gabbro (Fig. 2C).

Analytical strategy
Among the 114 samples collected for whole-rock geochemical analysis, 49 samples (batch 1) were analysed at the SARM analytical service of the CRPG (France), and 65 samples (batch 2) were analysed at the NILAB laboratory (New Caledonia).Sixteen thin polished sections were prepared and examined using reflected/transmitted light and scanning electron microscopy with a JEOL JSM7600F at Georessources Laboratory and SCMEM.In situ mineral chemistry analysis for major/minor elements (Mg, Al, Si, Ca, Cr, Mn, Fe, Co, Ni) was conducted using a CAMECA SX100 electron microprobe (EPMA, WDS analysis) at the SCMEM with typical beam conditions of 15 kV and 10 nA.In situ analysis for Sc and minor/trace elements was conducted using LA-ICP-MS (193 nm MicroLas ArF Excimer coupled with Agilent 7500c quadrupole ICP-MS).Sequential extractions were performed at the CEREGE laboratory on seven Sc-bearing samples from Cap Bocage and Tiébaghi.Four reagents were successively used to extract major and trace elements selectively: (i) ultrapure water for easily soluble elements, (ii) 0.1 mol.L -1 hydroxylamine hydrochloride NH2OH-HCl at pH = 3.5, which is particularly efficient and selective for manganiferous phases, (iii) 0.2 mol.L -1 ammonium oxalate (NH₄)₂C₂O₄ at pH = 3 for amorphous and poorly crystallised iron oxides and (iv) citrate-bicarbonate-dithionite Na2S2O4 (22% Nacitrate and 1 g Na-dithionite) for well-crystallised iron oxides (Bailly et al., 2014).Additional information on analytical procedures and PCA are given in Appendix A.

Lithofacies and mineral assemblages
Progressive weathering of peridotite involves hydration and hydrolysis of primary silicates and the formation of secondary silicates and oxides-oxyhydroxides. Therefore, the investigated Ni-Co deposits exhibit a continuum of alteration facies with specific mineral assemblages and textures.At Koniambo and Cap Bocage, moderately serpentinised harzburgite represents the predominant parent rock of Ni-Co laterites (Fig. 3A, 4).It is worth noting that the Coquette Red tenement investigated at Cap Bocage hosts significant volumes of dunite lenses so that the lateritic profiles are developed both on harzburgite and dunite.Serpentinisation occurs within the peridotite as mm to dm-large veins of serpentine (mostly lizardite) distributed as fracture networks and progressively replacing mantle silicates resulting in a typical mesh texture (Fig. 3A, 3C, 5A, 6A).At Tiébaghi, lherzolite dominates over harzburgite, although the intensive serpentinisation of the peridotite complicates its univocal recognition in the field.In the saprock, weathering initiates through the progression of the mantle silicate alteration front, marked by the onset of forsterite and enstatite hydrolysis (Fig. 5A, 6A) and the development of Ni-rich (up to 15-20 wt% Ni), greenish talc-like (kerolite) veins that root into the unweathered peridotite (Fig. 3C, 5A, 6A).Congruent hydrolysis of mantle silicate leads to the formation of 100-micron large pores with the development of skeletal goethite along silicate grain boundaries and crystallographic planes.Serpentine veins and mesh are preserved from dissolution in the saprock, although partly transformed into secondary, nickeliferous serpentine/talc-like (Fig. 5A, 5B).At Koniambo and Cap Bocage, local evidence for incongruent hydrolysis and epigenetic replacement of mantle silicate by smectite is also observed but is spatially restricted to the bedrock-saprock interface (Fig. 6A, 6C).There, epigenetic smectite is, in turn, readily replaced (e.g. by Mn-Co-Ni-rich lithiophorite, Fig. 6C) or dissolved and does not occur upwards in the profile.The evolution from the saprock to the saprolite, wherein weathering is more pronounced but remains isovolumetric, is characterised by the complete disappearance of mantle silicate and the partial replacement of serpentines by goethite (Fig. 3B, 5B, 6D).In contrast, at Tiébaghi, smectite extensively develops at the expense of both mantle silicates and serpentine (Fig. 3E, 4, 5C, 6B) and remains preserved throughout most of the saprolite until it is eventually replaced by ochreous goethite (Fig. 5D).The interface between the saprolite and the overlying limonite, referred to as the transition zone, is marked by the progressive disappearance of macroscopic primary structures and the accumulation of Mn-Co-Ni oxides (lithiophorite, asbolane) either as mm-to cm-thick veins or as diffuse accumulations in the iron-rich matrix (Fig. 3D).Upwards, the transition zone evolves into the yellow limonite wherein ochreous goethite predominates, characterised by the disappearance of all primary minerals except chromiferous spinel that remains mostly resistant to weathering throughout the entire profile.This horizon is marked by significant compaction.However, primary textures such as fragmented relics of serpentine veins or skeletons of mantle silicates, both replaced by goethite, may still be recognised at the microscopic scale (Fig. 5E,    6E).The yellow limonite then grades into the red limonite, wherein hematite forms at the expense of ochreous goethite resulting in a mineral assemblage dominated by goethite but containing significant amounts (> 5 vol%) of hematite (Fig. 5F).Lateritic profiles are capped by a ferruginous duricrust that is either directly developed after the underlying limonite or derived from the ferrugination of transported material (Fig. 5G, 5H).Following the terminology of Anand et al. (2002), the term duricrust describes regolith materials cemented by Fe, irrespective of the substrate origin.When it is residual, the iron-cemented material is called a lateritic residuum, and when it is formed and indurated in a transported cover, ferricrete.Lateritic residuum exhibits, at the microscopic scale, some locally preserved textural relics (Fig. 6F).There, both goethite and hematite are largely recrystallised into coarser crystallites.In contrast, ferricrete shows angular goethite and hematite nodules with goethite pisolitic envelopes.They are cemented by vitreous goethite (Fig. 6G).Subvertical dykes of amphibolitic gabbros locally exposed in the East Alpha deposit, Tiébaghi, alter to form a mineral assemblage composed of kaolinite-gibbsite-hematite (Fig. 3F, 6H, 6I).Precursor minerals are indicated in brackets.

Whole-rock geochemical correlations at the deposit scale
Whole-rock geochemical analysis of the investigated lateritic profiles shows that the lateritisation of unweathered peridotites is associated with the progressive enrichment of poorly mobile elements.In particular, Fe, Al, Cr, and Sc concentrations co-increase during weathering, exhibiting positive correlation trends (Fig. 7, Supplementary Table 1).In the Ma-Oui deposit, Al2O3, Fe2O3, Cr2O3 and Sc concentrations progressively increase from the bedrock to the oxide-rich horizons up to about 6 wt%, 80 wt%, 4.5 wt% and 80 ppm, respectively.The Sc-Al2O3, Sc-Fe2O3 and Sc-Cr2O3 linear regression models encompassing variable lithofacies (unweathered harzburgite, saprock, saprolite, Mn-Co-rich transition zone, yellow and red limonite) provide a particularly relevant fit of the data (Fig. 7A).Scandium concentrations may be approximatively estimated from the Al2O3, Fe2O3 and Cr2O3 concentrations from the bedrock to the red limonite as follows: (1) Sc (ppm) = 12.03* Al2O3 (wt%) + 2.42 ; R 2 = 0.93 (2) Sc (ppm) = 0.92* Fe2O3 (wt%) + 0.07 ; R 2 = 0.86 (3) Sc (ppm) = 16.27*Cr2O3 (wt%) + 0.31 ; R 2 = 0.91 The Sc-Fe 2 O 3 correlation trend shows low dispersion in the low-Sc concentration range (0-50 ppm) but significant dispersion in the higher-Sc concentration range (50-80 ppm).In addition, two samples collected in the transition and yellow limonite zone, respectively, fall well out of the Sc-Fe2O3 regression line.The Sc-Cr2O3 correlation trend is better defined, yet it presents moderate data dispersion.Comparatively, the best fit is obtained for the Sc-Al2O3 regression line, which exhibits the lowest dispersion of the data.Notably, the decrease in Al and Sc concentrations from the yellow to red limonite, as commonly identified by Teitler et al. (2019) in New Caledonian lateritic profiles, is not observed in the Ma-Oui geochemical dataset.In the Coquette Red tenement (Cap Bocage), Al2O3, Fe2O3, Cr2O3 and Sc concentrations increase to about 5 wt%, 80 wt%, 5.5 wt% and 65 ppm in the limonite, respectively.Like the Ma-Oui deposit, the Sc-Al 2 O 3 regression line efficiently fits the data from the bedrock to the red limonite, although the laterite is developed on dunite and harzburgite (Fig. 7B).
Analysed duricrust samples exhibit a large offset from the trend, with a substantial depletion in Sc relative to Al2O3.It is worth noting that the Sc-Al2O3 regression lines obtained at Ma-Oui and Coquette Red have similar regression coefficients (respectively 12.03 and 12.84; equations 1 and 4).
Nevertheless, significant dispersion from the regression line is also observed in the yellow and red limonite horizons compared to the Ma-Oui deposit and the Coquette Red tenement (Fig. 7C).Similar and sample scores (circles).See Fig. 7 for lithology colour coding.

Mineral Chemistry
The mineral chemistry data obtained from LA-ICP-MS analyses (Supplementary Table 2) allow us to assess the contribution of primary mantle silicates to the Al and Sc budget in the unweathered peridotites.Forsterite has a low Sc content in all of the investigated sites, about 3-6 ppm at Ma-Oui and Coquette Red and about 7 ppm at East Alpha (Fig. 9), together with Al below detection limits.
Enstatite yields higher Sc and Al concentrations, which vary from one site to another.Sc concentrations in enstatite from Ma-Oui and Coquette Red are similar (24 and 22.5 ppm, respectively).Al2O3 concentrations in enstatite are slightly higher at Ma-Oui (about 1.8 wt%) than at Coquette Red (about 1.4 wt%).In contrast, enstatite from East Alpha is enriched in Sc (about 32 ppm) and Al2O3 (about 3.4 to 3.8 wt%).Importantly, Sc-Al2O3 regression lines obtained from enstatite and forsterite mineral compositions in each investigated site (Fig. 9) closely match those obtained from whole-rock geochemical analysis throughout the weathering sequences (Fig. 7).More specifically, regression coefficients estimated from mantle silicate mineral chemistry versus global whole-rock geochemistry, respectively, are (i) 12.03 vs 11.02 at Ma-Oui, (ii) 13.18 vs 12.84 at Coquette Red, and (iii) 6.71 vs 6.56 at East Alpha.Further support for the intimate relationship between Sc-Al2O3 contents of peridotite mantle silicates and Sc-Al2O3 contents of peridotite-derived laterite samples is provided by the mineral chemistry data of secondary mineral phases along with laterite profiles as analysed at East Alpha (Fig. 10, Supplementary Table 2).Together with forsterite and enstatite, Sc and Al2O3 concentrations in smectite, ochreous goethite from the limonite zone, and goethite-hematite from the lateritic residuum appear to be strongly proportional, with a regression line closely similar to that obtained from wholerock geochemistry.Smectite in the smectitic saprolite and ochreous goethite in the limonite exhibit variable yet proportional Sc and Al2O3 concentrations.Indeed, Sc and Al2O3 concentrations in smectite vary from 20 to 80 ppm and from 2.5 to 10 wt%, respectively, while Sc and Al2O3 concentrations in limonitic, ochreous goethite vary from 55 to 122 ppm and from 7 to 16 wt%, respectively.Goethitehematite from the lateritic residuum is, comparatively, depleted both in Sc and Al2O3, yet it falls on or close to the regression line.In contrast, goethite and hematite nodules and pisolitic cortexes from the ferricrete appear further depleted in Sc to fall below the regression line.

Sequential extractions
The sequential extractions for Fe, Al, Ni and Sc are given in Fig. 11.The ultrapure water did not permit solubilising these elements in any investigated samples during the first extraction step.With regards to Fe, the hydroxylamine hydrochloride extraction also proved unsuccessful.In iron-rich samples (earthy saprolite, yellow and red limonite), about 10-15% of the total Fe mass was extracted during the subsequent extraction step involving ammonium oxalate, while about 50-60% was removed during the last extraction step involving citrate-bicarbonate-dithionite (CBD).In iron-rich lithologies, the overall procedure therefore permits the extraction of 60 to 80% of the total Fe content, and CBD appears the most effective reagent to solubilise Fe.In contrast, the general approach only resulted in the extraction of about 25% of the total Fe in the smectitic saprolite sample (TIEA-08) from East Alpha, mainly through the action of CBD and, to a lesser extent, of ammonium oxalate.With regards to Al, the overall procedure succeeded in solubilising 40-50% and 10% of the total Al (Fig. 10A) in iron-rich and smectite-rich samples, respectively.The relative amount of extracted Al is higher in iron-rich samples from Coquette Red (PZ1B-06, PZ1B-09, PZ1B-13 and PZ1B-17) compared to their equivalents from East Alpha (TIEA-07 and TIEA-12).Among this fraction, a small portion (less than 5% of the total Al extracted) was solubilised using hydroxylamine hydrochloride.

Discussion and implications for the evaluation of Sc in Ni-Co laterites
This study aims to assess in which conditions Al may be reliably used as a geochemical proxy for a firstorder evaluation of Sc grades and distribution in Ni-Co lateritic ores.In the following, we discuss the causes of the Al-Sc co-variation regarding the mineralogical evolution in lateritic profiles and the relevance of using deposit-scale Sc-Al 2 O 3 correlations to assess the distribution of Sc in Ni-Co laterites.

Co-evolution of Al and Sc concentration and speciation through weathering
Analytical data documented in the present contribution reveal significant correlations, at the deposit scale, between Sc and Al2O3 concentrations in some lateritic Ni deposits of New Caledonia.The high proportionality between Sc and Al2O3, observed in the investigated lateritic deposits from the bedrock to the red limonite, is identified from whole-rock geochemistry (scatterplots of untransformed data and PCA of centred log-ratio-transformed data) and mineral chemistry data (Fig. 7,8,10).Such In the unweathered mantle silicates, Ni is exclusively bound to forsterite.In contrast, virtually all of the Al is hosted in pyroxenes (enstatite and diopside), and to a lesser extent, in chromiferous spinel, as it is highly incompatible in forsterite.Sc is also predominantly hosted in pyroxenes, though forsterite incorporates Sc up to a few ppm.Diopside contains significant concentrations of Sc (up to about 60 ppm; Teitler et al., 2019;Ulrich et al., 2019) but is not abundant (Table 1) and has a minor influence on the bulk Sc content of unweathered lherzolite.Teitler et al. (2019) proposed that serpentinisation of the peridotite most likely has a marginal effect on Sc concentrations in the bedrock.Therefore, the global Sc and Al 2 O 3 contents in unweathered peridotite mainly depend on the relative proportion of enstatite and its chemical composition.
In the lower part of the weathering sequence (saprock and smectitic saprolite), Sc and Al concentrations co-increase with Fe and Cr concentrations, suggesting that Sc and Al enrichments are essentially residual.Co-variations, along a quite extensive concentration range, of Al, Sc, Fe and Cr in smectitic saprolite samples from East Alpha (Fig. 7) result from inherent co-variation of smectite composition, as evidenced by mineral chemistry data (Fig. 10).The sequential extraction procedure proved relatively ineffective in solubilising Fe, Ni and Al and from the smectitic saprolite.These elements are typically present in octahedral and tetrahedral positions in nickeliferous smectite (Mano et al., 2014;Ratié et al., 2018) and are not leached easily from smectite.However, the significant amounts of Sc extracted from the smectite-rich saprolite using ammonium oxalate raise questions regarding the speciation of Sc in this horizon.Together with the elevated Sc concentrations measured in smectite from LA-ICP-MS analysis, the lack of Fe extraction by ammonium oxalate rules out any significant contribution of amorphous iron oxides/oxyhydroxides to the Sc budget in the smectite-rich zone.Based on the scandium K-edge XANES analysis, Chassé et al. (2019) proposed that Sc in the plasmic horizon of the Syerston-Flemington Sc laterite (Australia) is efficiently trapped in smectite through incorporation into octahedral sites.Such interpretation seems unable to account for the high quantities of Sc extracted using ammonium oxalate.Further investigations remain necessary to evaluate the Sc speciation in the smectitic saprolite from the East Alpha deposit.
Contrasting with the Fe-Al-Sc-Cr collinearity identified in the lower portion of the weathering sequences, Sc and Al exhibit a relative enrichment compared to Fe in the earthy saprolite/yellow limonite and a relative depletion in the red limonite.This distribution pattern, observed at Coquette Red and East Alpha (Fig. 7B, 7C) as well as in several New Caledonian Ni-Co laterites (Teitler et al., 2019), but not at Ma-Oui (Fig. 7A), supports specific mobility of Sc and Al.It is suggested that these elements are, to some extent, remobilised from the red limonite and accumulated downwards in the yellow limonite.The remobilisation of Sc and Al possibly result from the release of Sc and Al after the dissolution/recrystallisation of increasingly crystallised goethite during the maturation of the lateritic profile (Dublet et al., 2015;Teitler et al., 2019).Such a model is commonly accepted for explaining the distribution pattern of Ni concentrations, which generally decrease gradually from the earthy saprolite upwards in conjunction with an increase of the mean coherent domain (MCD) size of the goethite crystallites (Dublet et al., 2012(Dublet et al., , 2015)).This typical Ni distribution pattern is, for instance, wellevidenced at Coquette Red, where a gradual decrease of Ni concentrations occurs from the earthy saprolite (1.5 wt% Ni) to the red limonite (0.5 wt% Ni).There, Sc and Al distribute differently from Ni, as Sc and Al concentrations are higher in the yellow limonite than in the earthy saprolite.Although the release of Sc during dissolution/recrystallisation of goethite has been proposed as a relevant model for explaining some elevated Sc concentrations in the yellow limonite (Chassé et al., 2019;Teitler et al., 2019), the mobility of Sc released during this process must therefore be lower than that of Ni.In addition, the formation of hematite at the expense of goethite in the red limonite likely results in the downward redistribution of Sc and Al, as both Sc and Al substitute more easily for Fe in goethite than in hematite (Levard et al., 2018;Schwertmann and Latham, 1986;Trolard et al., 1995), thus contributing to Sc enrichment in the yellow limonite.
Sequential extraction provides further insights on the speciation of Sc and its association with other elements in the earthy saprolite, yellow and red limonite.It is worth noting that the extraction procedure here applied differs from the method used by Chassé et al. (2019) and Qin et al. (2020) on ultramafic-derived laterites from the Syerston-Flemington (Australia) and the Berong (Philippines) deposits, respectively.In these two studies, the extraction procedure is adapted from Hall et al. (1996) and Sanematsu et al. (2011).It includes sodium acetate as the first reagent to assess the adsorbed and exchangeable species.In contrast, we used ultrapure water, which only allows us to extract the easily exchangeable species and therefore does not estimate the adsorbed species.(Qin et al., 2020), the amounts of Sc adsorbed on goethite appear more significant when estimated from XANES spectra (24-49%) than from sequential extraction (below 5%).
Both of these studies then use hydroxylamine hydrochloride to extract amorphous iron oxides.In contrast, we used hydroxylamine hydrochloride at lower concentrations to extract manganiferous species.No Fe nor Sc, and only a very slight fraction of Ni and Al, are extracted using hydroxylamine hydrochloride, highlighting the marginal presence of Mn oxides in the investigated samples and the absence of detectable Sc in Mn oxides.Following hydroxylamine hydrochloride treatment, we used ammonium oxalate to extract amorphous iron oxides.Ammonium oxalate has been reported as an efficient and selective dissolving agent for amorphous and poorly crystalline ferric oxides/oxyhydroxides, without significant dissolution of crystalline goethite and hematite, nor attacking silicates (Leermakers et al., 2019;Poulton and Canfield, 2005).Chromiferous spinel, which is residually enriched in these horizons, contains about 15 wt% Al (Teitler et al., 2019).Therefore, even in minor amounts, the presence of extraction-resistant Cr-spinel implies that a non-negligible portion of the whole-rock Al content is hosted in Cr-spinel and cannot be extracted through the used procedure.However, such quantities of unextractable Al are likely insufficient to account for the apparent lower extractability of Al, nor for the differences in Al extraction rates between Coquette Red and East Alpha, as both sites have similar Cr contents.
Alternatively, although not observed in any of the peridotite-derived limonite samples, the presence of trace amounts of kaolinite in the lherzolite-derived limonite cannot be ruled out at East Alpha.
Similar to Cr-spinel, the used extraction procedure is ineffective in solubilising kaolinite, whose potential presence in the East Alpha limonite may explain the lower extraction rate for Al.Regarding Sc, total extraction rates in the Fe-rich horizons are similar to that of Fe (and therefore higher than that of Al), supporting the strong association of Sc with iron oxides/oxyhydroxides, and in particular with crystalline goethite.Nevertheless, ammonium oxalate appears relatively more efficient in extracting Sc (about 20 to 35% of total Sc) than Fe (about 10% of the total Fe).The proportion of Sc extracted from amorphous iron oxides in the present study is therefore slightly higher than that obtained by Chassé et al. (2019), that is about 15-25%, and significantly higher than that obtained by Qin et al. (2020), that is below 3%.This discrepancy possibly results from a better efficiency of ammonium oxalate to extract Sc from amorphous iron oxides/oxyhydroxides than hydroxylamine hydrochloride.
Our results suggest that Sc may have a higher affinity for amorphous iron oxides/oxyhydroxides than crystalline goethite.Significant amounts of Sc can be thus concentrated in amorphous iron oxides/oxyhydroxides despite the predominance of crystalline goethite throughout the lateritic sequence.The speciation of Sc in oxide-rich facies differs probably in part from that of Al.The latter is mainly incorporated in the lattice of crystalline goethite and preserved in weathering-resistant Cr- spinel.Yet, despite second-order differences between the speciation of Sc and Al2O3, these elements show strong proportionality from the bedrock to the red limonite.As these elements (i) are both mainly immobile during the weathering of peridotite, (ii) have a preferential affinity for goethite, and  2019), possibly results from the contribution of allochthonous material to the duricrust.Indeed, in situ Sc analysis of goethite-hematite in duricrust, were identified as lateritic residuum, shows a good fit with the Sc-Al2O3 regression line obtained on saprolitic and limonitic minerals.On the opposite, nodular and pisolitic goethite and hematite from the ferricrete exhibit a relative depletion in Sc compared to Al, similar to the offsets identified in the whole-rock geochemical dataset.In the East Alpha deposit, the offset of gabbro-derived saprolite from the Sc-Al2O3 correlation trend also results from the allochthonous nature of the gabbro compared to the peridotite-derived Ni-Co laterite, together with the specific mineral assemblage of the gabbro-derived saprolite.The formation of kaolinite during the weathering of gabbro is related to its substantial Al content (Schwertmann et al., 2000;Teitler et al., 2019;Trolard and Tardy, 1989).As Sc is typically poorly concentrated into kaolinite (Chassé et al., 2017;Teitler et al., 2019;Ulrich et al., 2019), the weathering of gabbros may lead to Sc remobilisation and trapping into nearby yellow limonite, while Al remains concentrated in saprolitised gabbro as kaolinite.Consequently, Sc and Al2O3 may be positively correlated in lateritic deposits if kaolinite, or other Al-bearing phases such as gibbsite, are mostly absent from the lateritic profiles.The formation of kaolinite (or gibbsite) during the lateritisation process requires that the parent rock contains significant amounts of Al, so that Sc-Al2O3 correlation trends may only be observed in ultramafic-derived laterites, wherein the Al content is low.

Implications for the assessment of Sc in Ni-laterites
The relevance of using Al as a geochemical proxy to conduct a first-order estimation of Sc Interestingly, the alternance of harzburgite and dunite in the bedrock does not seem to significantly affect the Sc-Al2O3 regression coefficient nor the dispersion of the data, providing that the composition of enstatite remains similar in both facies.Nevertheless, lithological heterogeneities involving a change in the composition of enstatite, elevated amounts of diopside and plagioclase (e.g. in some lherzolite facies), or the occurrence of mafic intrusive dykes, may cause significant variations of the Sc/Al2O3 concentration ratio both in the parent rocks and in their weathered derivatives.These potential variations of the parent rock lithology at the deposit scale may be tested by geological characterisation and geochemical assay data analysis.In addition, alteration facies containing significant amounts of Alrich phases (e.g.kaolinite, gibbsite) are commonly characterised by a substantial deviation from the Sc-Al2O3 correlation trend, established from smectite-and Fe-oxide-dominated lithologies.The possible occurrence of Al-rich phases, typically poor in Sc, must be examined to prevent overestimation of the Sc content.Al-rich phases are commonly associated with weathered intrusive rocks such as gabbros.Still, they may also form during the weathering of some peridotites that yield significant Al concentrations.In particular, some plagioclase-bearing lherzolites (not investigated in the present study) can exhibit Al2O3 contents up to ~4 wt% (Marchesi et al., 2009;Ulrich et al., 2010) and may consequently alter to kaolinite or gibbsite.
Geochemical homogeneity and low Al-content of the parent rock are, therefore, necessary conditions for the occurrence of a reliable Sc-Al2O3 correlation trend at the deposit scale.Determination of a reliable, deposit-scale Sc-Al2O3 correlation requires assessing the range of Al2O3 concentrations for which Sc is well correlated with Al2O3 through an adequate sampling strategy that encompasses the whole range of Al2O3 concentrations and the mineralogical diversity throughout the deposit.Finally, the inherent data dispersion in Sc-Al2O3 biplots may be variable depending on the investigated deposit.
Therefore, the number of samples used for establishing the Sc-Al2O3 correlation must be adapted to the inherent data dispersion to develop a reliable correlation with a good correlation coefficient.Once verified, deposit-scale Sc-Al2O3 correlations may first prove useful to estimate Sc concentrations on surface outcrops or pit walls using portable devices such as pXRF.Indeed, the relatively low Sc concentrations observed in Ni-laterites (<100 ppm) limit the use of pXRF to directly assess Sc (Lacroix et al., 2021) whereas Al concentrations, typically about a few wt%, can be confidently estimated using such a device.The slightly lower accuracy obtained on Al analysis from pXRF than from a classical assay analysis would only marginally affect the estimation of Sc.More importantly, applying this approach to deposits where Al is routinely assayed could provide a first-order Sc concentration and distribution

Fig. 2 :
Fig. 2: Outline of the investigated deposits and location of collected samples.(A) Ma-Oui deposit

Fig. 3 :
Fig. 3: Field photographs of representative lithofaciès as observed in the Koniambo, Cap Bocage and

Figure 10 :
Figure 10: Sc-Al2O3 scatterplot and regression line obtained for silicates, smectite and iron oxides in

Figure 11 :
Figure 11: Results of sequential extractions.Cumulative histograms showing the total quantities of Fe proportionality of whole-rock Sc and Al2O3 concentrations have already been documented byTeitler et al. (2019), although the relevance of the Sc-Al2O3 correlation was not investigated at the deposit scale.Also, the study conducted bySantoro et al. (2022) on trace element concentrations in goethite from various Ni-laterite deposits showed a strong association between the Sc and Al contents of goethite from the Wingellina deposit (Australia).The authors interpreted such association as being primarily controlled by the composition of the parent rock, highlighting the influence of Al-and Scbearing pyroxenite lenses on the composition of their goethite-bearing, weathered derivatives.The Sc-Al2O3 proportionality, documented in the present contribution throughout the weathering sequences of various Ni-laterites from New Caledonia (i.e. from the bedrock to the red limonite), argues for a similar behaviour of Al and Sc during weathering.
Nevertheless, the extracted Sc amounts obtained byChassé et al. (2019) andQin et al. (2020) from sodium acetate treatment are low(10-15%)  in the Syerston-Flemington laterites to very low (<5%) in the Borong laterites.These two studies combine sequential extraction with XANES analysis.Chassé et al. (2019) observed that the low proportion of exchangeable goethite-hosted Sc obtained from sequential extraction (10-15%) is at odds with the results of XANES analysis.The latter is interpreted to reflect an elevated contribution of adsorbed Sc (up to 80%) in the global Sc budget of iron oxides/oxyhydroxides and only a small proportion of Sc being substituted in iron oxides or oxyhydroxides.The authors favour better reliability of the XANES spectra interpretation than the sequential extraction results and propose reconciling these data by arguing for the high stability of the Sc adsorption complex on goethite, thus preventing the extraction of adsorbed Sc through sodium acetate treatment.Similarly, in the Gorong deposit Chassé et al. (2019) andQin et al. (2020) did not pursue sequential extraction further than the amorphous iron oxides/oxyhydroxides extraction step, hypothesizing that the residue is mostly representative of crystalline iron oxides.In the present study, the use of CBD as the last extraction step unambiguously demonstrates that crystalline iron oxides/oxyhydroxides predominates over amorphous iron oxides/oxyhydroxides in oxide-rich horizons.Along with Fe, the elevated Ni, Al and Sc fractions extracted using CBD confirm that these elements are mainly bound to crystalline goethite.At Coquette Red, a slight decrease in the proportion of total extracted Fe is observed from the earthy saprolite to the red limonite, from about 80 to 60% (Fig.11).Such decrease, resulting from a slight lowering of both CBD and ammonium oxalate extraction efficiencies (from about 60 to 50% and 15 to 8%, respectively), may indicate the upwards increase in goethite crystallinity.Compared to Fe, Al extraction appears globally less effective, especially at East Alpha, wherein the Al contents of oxide-rich facies are higher than at Coquette Red.Such a discrepancy between the extraction efficiencies of Al and Fe possibly results from the presence of extraction-resistant, Al-bearing minerals in the oxide-rich facies.

(
iii) are only moderately remobilised following recrystallisation and goethite replacement by hematite, the composition of the parent rock remains the first-order control on their concentrations throughout the weathering sequences.Thus, the Sc-Al2O3 regression lines obtained from forsterite-enstatite mineral compositions are close to those obtained from weathering-related mineral compositions and whole-rock geochemistry along weathering profiles.The slightly lower regression coefficients obtained from whole-rock geochemistry than those obtained from mantle silicate mineral composition may result from the second-order contribution of Cr-spinel to the global Al budget in whole-rock geochemical compositions.These results indicate that (i) the Sc-Al2O3 content of enstatite in a given peridotite bedrock drives the Sc-Al2O3 regression line within its weathered derivatives, and (ii) the relative proportion of enstatite together with its Sc content primarily control the maximum Sc concentrations reached in the yellow limonite.Contrasting with the proportionality observed between Sc and Al 2 O 3 from the bedrock up to the red limonite, samples from the duricrust are out of the Sc-Al2O3 whole-rock correlation trends.Such offset, previously documented byTeitler et al. ( concentration and distribution in a given Ni-Co deposit depends both on the specific reliability of the Sc-Al2O3 correlation at the deposit scale and on the method used to characterise Al concentration and distribution.The reliability of the Sc-Al2O3 correlation may be influenced by (i) potential heterogeneities in the parent rock lithology, (ii) the occurrence of alteration facies containing Al-rich phases such as kaolinite or gibbsite and (iii) the inherent data dispersion in Sc-Al 2 O 3 scatterplots.
estimate.The reliability of such estimation would depend on the reliability of the Sc-Al2O3 correlation and the reliability of the Al distribution model.Indeed, block models are developed firstly to estimate the resource and distribution of Ni both in saprolitic and limonitic facies.They rely on statistical variograms set explicitly for several metals (e.g.Si, Mg, Fe, Ni) but not necessarily Al.In such a case, a specific evaluation of the variability of Al in the limonitic facies may be beneficial.ConclusionThis contribution examines the relevance of using Al as a geochemical proxy for first-order estimates of Sc distribution and concentration in some peridotite-hosted, Ni-Co laterites from New Caledonia.Apparent correlations are identified at the deposit scale between Sc and Al2O3 concentrations from the bedrock to the red limonite.These correlations put forward the similar behaviour of Al and Sc during the peridotite weathering, which concentrations are primarily issued from the residual enrichment.Local remobilisation from the uppermost horizons is shown.Al and Sc are both predominantly hosted in crystalline goethite, but Sc has a relatively higher affinity for amorphous iron oxides compared to Al.In all investigated deposits, the Sc-Al2O3 regression coefficient remarkably depends on the Sc content in enstatite.Providing that the parent lithology is homogeneous and relatively depleted in Al, reliable Sc-Al2O3 correlations may thus be determined at the deposit scale after analysing a limited number of spatially and chemically representative samples.An adequate sampling strategy is required to cover the range of Sc and Al2O3 concentrations throughout the deposit and to take in account potential occurrences of specific alteration facies that may affect the relevance of the deposit-scale correlation (ferricrete, weathered intrusive rocks, kaolinite-or gibbsite-bearing alterite).Although the zones of maximum Sc enrichment are situated above the zones of maximum Ni and Co enrichment, Sc-rich limonite may overlap the Co-rich transition zone, so that the base of the limonite may yield elevated Co and Sc concentrations together with sub-economic Ni grades.In such conditions, lateritic ores may be valuably exploited for Ni, Co and Sc assuming cost-effective covalorisation of these metals.
.e. about three times more elevated than in the harzburgite±dunite-derived limonite at Koniambo and Cap Bocage.In comparison, gabbro-derived saprolite yields Al2O3 concentrations up to about 30 wt%.Maximum Fe2O3 and Cr2O3 concentrations at East Alpha are similar to those observed in the other investigated sites, at 80 and 4 wt%, respectively.The Sc-Al2O3 regression line modelled for East Alpha samples fits relatively well the data, except duricrust and gabbro samples that fall well out of the correlation trend.The regression coefficient of the Sc-Al2O3 line is about twice lower at East Alpha than at Ma-Oui and Coquette Red (equation 7).It is worth noting that smectitic saprolite samples from East and Coquette Red.Also, Al2O3 concentrations reach about 15 wt% in the lherzolite-derived limonite, i of the total Ni extracted) yet higher than that observed for Al.The efficiency of Ni extraction by ammonium oxalate appears similar to that of Fe (about 5 to 15% of the total Ni extracted) and slightly lower than Al.As for Fe and Al, Ni is best extracted through the last extraction step using CBD.nor the hydroxylamine hydrochloride extraction proved successful in extracting Sc from any of the samples.Sc extraction using ammonium oxalate, which accounts for about 20 to 30% of the total Sc contained, appears significantly more efficient than Fe, Al and Ni in iron-rich samples.Nevertheless, in these samples, CBD remains the most efficient reagent for Sc extraction.In contrast, in the smectitic saprolite sample from East Alpha, Sc is almost exclusively extracted using ammonium oxalate.However, the total amount of extracted Sc from this sample remains moderate (about 50% of the bulk Sc content).To summarise, it should be noted that, in oxide-rich facies, (i) the extraction procedure is globally efficient for solubilising Fe, Al, Ni and Sc, though significant differences in global extraction efficiencies are observed depending on the considered element, and the sample provenance (ii) only Al and Ni can be slightly extracted using hydroxylamine hydrochloride, (iii) Fe, Al, Ni and Sc are best solubilised using CBD and, to a lesser extent, ammonium oxalate, (iii) the efficiency of Sc extraction using ammonium oxalate is significantly greater than that of Fe, Al and Ni.In contrast, The relative amount of Al removed by ammonium oxalate is similar to that of Fe, about 10% or less.CBD remains the most efficient reagent for Al extraction but seems less effective for solubilising Al than Fe.In addition, Al appears more efficiently extracted by ammonium oxalate in iron-rich samples from Coquette Red (about 10% of the total Al or 15-30% of the whole extracted Al) than in their equivalent from East Alpha (about 5% of the total Al or 10% of the extracted Al).Bulk Ni extraction is similar or higher to Fe extraction.It is, in particular, higher in iron-rich samples from Coquette Red than in their equivalents from East Alpha.More specifically, two samples from Coquette Red (PZ1B-09 and PZ1B-17) show almost complete Ni extraction, while Ni extraction in East Alpha does not exceed 60%.In contrast, Ni is weakly extracted from the smectitic saprolite sample from East Alpha.The relative proportion of Ni removed by hydroxylamine hydrochloride remains low (5 to 15% the overall procedure is less well suited for the sequential extraction in smectitic saprolite, so the greater extraction efficiency of ammonium oxalate relative to other reagents must be considered cautiously.