Mineralization-alteration footprints in the Olympic Dam IOCG district, South Australia: The Acropolis prospect

https://doi.org/10.1016/j.gexplo.2019.106333Get rights and content

Highlights

  • k-means clustering is appropriate to define geochemically meaningful domains

  • W-Sn-Mo-U, Sb, Bi and REE are characteristic for the hematite-dominant assemblage

  • Cusingle bondAu deficient magnetite veins are overprinted by Cu-bearing hematite assemblage

  • Fe-V-Ni-Co-bearing magnetite in veins typifies early IOCG mineralization in Acropolis

  • The ‘hematite’ signature unequivocally correlates Acropolis with Olympic Dam

Abstract

The Acropolis prospect, 20 km southwest from the Olympic Dam Cu-U-Au-Ag deposit, South Australia, is a vein-style magnetite (±apatite ±hematite) system. A whole-rock dataset comprising 4864 core samples from 14 drillholes was analysed using multivariate statistical analyses to understand and identify geochemical signatures of mineralization, as well as the expressions and extents of hydrothermal alteration. Statistical analysis included unsupervised (principal component analysis, hierarchical and k-means clustering) and supervised (random forests) machine learning algorithms. The geology of the Acropolis prospect is presented as a 3D geological model complemented by cross-sections. The results of statistical analyses are overlaid and interpreted relative to the geological model, and encompass a projection of sodic and propylitic alteration as PC3, and mineralization signature as PC1.

Although the mineralization footprint of the Acropolis prospect partially overlaps with a Hiltaba Suite granite, it is not centred on the granite body. A distinct ‘magnetite’ signature of Fe-V-Ni-Co is developed in the southwestern part of Acropolis and represents samples containing >60 wt% Fe. In contrast, the ‘hematite’ signature displays an association of REE, W, Sn, Sb, U, Th, Ca and P, and is present throughout the Acropolis prospect with the exception of drillhole ACD5, which is non-mineralized. Interpolated values of Cu (> 200 ppm) indicate an offset from Fe-rich veins, thus supporting a genetic model in which Cu-bearing mineralization overprints Cu-Au-deficient magnetite-dominant veins. The results obtained provide insights into the evolution from magnetite to hematite-dominant IOCG systems and may provide a proxy for exploration of shallow and economically significant IOCG deposits in the eastern Gawler Craton.

Introduction

Iron-oxide copper gold (IOCG) deposits encompass a diverse group of mineralization styles found in a range of geological settings, and formed from the Archean to Tertiary (Groves et al., 2010). In the last three decades, IOCG systems spanning a continuum from magnetite- to hematite-rich end-members have been subject to many studies aimed at defining common mineralogical and geochemical characteristics and ore genesis (Hitzman et al., 1992; Hitzman, 2000; Williams et al., 2005; Corriveau, 2005; Barton, 2014 and references therein). These deposits are well known for their broad alteration haloes – particularly alkali-(calcic) alteration, and also for their marked deposit-scale zoning, as seen for the IOCG systems at Olympic Dam deposit, South Australia (Ehrig et al., 2012 and references therein; Mauger et al., 2016; Dmitrijeva et al., 2019), or Ernest Henry, Queensland (Mark et al., 2006).

Conceptualization of IOCG-style mineralization and ore genesis has become increasingly problematic, however, particularly since diverse end-members have been added to this group. These include Cu-Au-poor ore systems, such as the Kiruna and El Laco types, which were initially considered as magnetite-apatite deposits of magmatic origin (Geijer, 1931; Frietsch, 1978; Park, 1961), albeit of very different ages (1888 ± 6 Ma, Romer et al., 1994, vs. 2.1 ± 0.1 Ma, Maksaev et al., 1988). As more Kiruna-type magnetite-apatite deposits were identified, notably in the Chilean Iron Belt (El Algarrobo, Ruiz et al., 1968; El Romeral, Bookstrom, 1977), and as detailed geochemical studies allowed improved discrimination of magmatic and hydrothermal generations of Fe-oxides (Dare et al., 2014; Knipping et al., 2015a), such deposits were considered kin to IOCG systems, and were redefined as a distinct subgroup, the iron-oxide apatite (IOA) type.

The magnetite-apatite association is, however, also part of early, alkali-calcic alteration within IOCG systems and has been well documented in mineralogical studies from deposits of the Olympic Cusingle bondAu province, South Australia (e.g., Ismail et al., 2014; Krneta et al., 2016, Krneta et al., 2017a, Krneta et al., 2017b). Therefore, magnetite-apatite is not as discriminative between IOA and IOCG, as is the lack of alkalic (Nasingle bondK) alteration in the deposit footprint reported for some IOA deposits (e.g., Los Colorados, Chile; Knipping et al., 2015b).

One of the best endowed IOCG terranes on Earth is the Mesoproterozoic Olympic Cusingle bondAu province, Eastern Gawler Craton, South Australia, in which mineralization relates to ~1.6 Ga magmatism comprising both bimodal volcanics (Gawler Range Volcanics; GRV) and igneous rocks (Hiltaba Suite granitoids) (Fig. 1; Skirrow et al., 2007).

Drilling of geophysical targets over the past 40 years has led to the discovery of multiple satellite prospects within a 5–100 km-radius from Olympic Dam deposit (Fig. 2). Among these, the Acropolis prospect represents a geophysical target at least 3 times the size of Olympic Dam. This prospect represents a relatively Cu-Au-poor example of magnetite-dominant mineralization that is mainly hosted within GRV. The Acropolis prospect is broadly considered contemporaneous with Olympic Dam based upon comparable mineral assemblages and geochronology (Krneta et al., 2017b; Cherry et al., 2018).

In this contribution, we present a 3D geological model and apply compositional data analysis to the whole-rock data set (4864 samples) to characterize the mineralization and alteration styles of the Acropolis prospect. Machine learning techniques are increasingly used in geochemistry to obtain data-driven geological insights from large datasets (e.g., Grunsky, 2010). Given the increased volume and multi-element character of modern data sets, a quantitative and reproducible approach to data analysis is required.

Application of statistical analysis to deposit-scale whole-rock datasets from Olympic Dam (Dmitrijeva et al., 2019) allowed recognition of a characteristic IOCG signature which, apart from common elements Fe, Cu and Au, also comprises indicator elements such as W, Mo, As and Sn. In the present contribution, an analogous approach is taken by applying unsupervised (principal component analysis, hierarchical and k-means clustering), and supervised (random forests, RF) machine learning algorithms to whole-rock data from Acropolis. Coupled with an implicit geological model, these techniques allow visualization of the associations of key elements, alterations and mineralization within three-dimensional space, also facilitating direct comparison with the Olympic Dam deposit.

Our overarching aim is to understand the genetic relationships between Olympic Dam and nearby prospects using Acropolis as an example to track the evolution of IOCG systems from magnetite- to hematite-rich end-members.

Section snippets

Geological background and rationale

The Olympic Cu-Au(single bondU) province is a 700-km-long N-S striking metallogenic belt along the eastern margin of the Gawler Craton (Fig. 1). Cu-Au(single bondU) metallogeny is represented by IOCG-style mineralization associated with a major magmatic event, the Gawler Silicic Large Igneous Province (SLIP), at ~1.6 Ga (Fig. 1B) that generated Hiltaba Suite intrusive rocks and their cogenetic equivalents, the Gawler Range Volcanics (GRV) (Blissett et al., 1993). Host lithologies for IOCG systems vary from Hiltaba

Methods

The work-flow diagram (Fig. 4) illustrates the successive stages of data preparation, clustering, classification, validation and interpretation.

Variation matrix and hierarchical clustering

The obtained variation matrix comprised of the variances of the pair-wise log-rations τij is given in Table 2. The extended variation matrix is in Appendix A. The variation matrix identifies the key geochemical associations within the data, whereas the small τij values correspond to a high geochemical affinity among elements and vice versa. For instance, the smallest τij values characterize element pairs which are predominantly hosted within silicate minerals such as K-feldspar, albite and

IOCG signatures at Acropolis prospect: a comparison with Olympic Dam

The kriged values of PC1 display a mineralized domain in the western part of the prospect (ACD1, ACD7 and ACD9), and within Donington Granitoids from the southeastern corner of the prospect (ACD3 and ACD12) (Figs. 6; 10). Similar to the IOCG footprint in the Olympic Dam deposit, the multi-element signature of mineralization at Acropolis encompasses a range of indicator trace elements alongside the IOCG-defining metals Fe and Cu (Table 4; Fig. 5). The major distinction between mineralization

Conclusions

This contribution demonstrates that application of supervised and unsupervised machine learning techniques is effective for investigation of mineralization and alteration signatures of a large data set, and for understanding the key geochemical associations among component elements in the absence of detailed petrographic and mineralogical information. Reinforced by the RF model, a categorical variable ‘Cluster’, obtained from k-means algorithm, is appropriate for characterization of a deposit

Acknowledgements

This work is a contribution to the FOX project ‘Trace elements in iron-oxides: deportment, distribution and application in ore genesis, geochronology, exploration and mineral processing’, supported by BHP Olympic Dam and the South Australian Mining and Petroleum Services Centre of Excellence. N.J.C. acknowledges additional support from the ARC Research Hub for Australian Copper-Uranium (Grant IH130200033). Two anonymous reviewers are thanked for their comments that greatly improved the

References (75)

  • S. Krneta et al.

    Apatite at Olympic Dam, South Australia: a petrogenetic tool

    Lithos

    (2016)
  • S. Krneta et al.

    The Wirrda Well and Acropolis prospects, Gawler Craton, South Australia: Insights into evolving fluid conditions through apatite chemistry

    J. Geochem. Explor.

    (2017)
  • J.J. O'Brien et al.

    Using Random Forests to distinguish gahnite compositions as an exploration guide to Broken Hill-type Pb–Zn–Ag deposits in the Broken Hill domain, Australia

    J. Geochem. Explor.

    (2015)
  • J.C. Ordóñez-Calderón et al.

    Machine learning strategies for classification and prediction of alteration facies: examples from the Rosemont Cu-Mo-Ag skarn deposit, SE Tucson Arizona

    J. Geochem. Explor.

    (2018)
  • J. Palarea-Albaladejo et al.

    zCompositions—R package for multivariate imputation of left-censored data under a compositional approach

    Chemometr. Intell. Lab. Syst.

    (2015)
  • P.J. Rousseeuw

    Silhouettes: a graphical aid to the interpretation and validation of cluster analysis

    J. Comput. Appl. Math.

    (1987)
  • M.R. Verdugo-Ihl et al.

    Textures and U-W-Sn-Mo signatures in hematite from the Cu-U-Au-Ag orebody at Olympic Dam, South Australia: defining the archetype for IOCG deposits

    Ore Geol. Rev.

    (2017)
  • J. Aitchison

    The statistical analysis of compositional data

  • O.B. Apukhtina et al.

    Early, deep magnetite-fluorapatite mineralization at the Olympic Dam Cu-U-Au-Ag deposit, South Australia

    Econ. Geol.

    (2017)
  • A.H. Blissett et al.

    Gawler range volcanics

  • A. Bookstrom

    The Magnetite Deposits of E1 Romeral, Chile

    Econ. Geol.

    (1977)
  • L. Breiman

    Random forests

    Mach. Learn.

    (2001)
  • L. Breiman et al.

    Classification and Regression Trees. Wadsworth, Belmont

    (1984)
  • N. Caciagli

    Multielement geochemical modelling for mine planning: case study from an epithermal gold deposit

  • A.R. Cherry et al.

    Tectonothermal events in the Olympic IOCG Province constrained by apatite and REE-phosphate geochronology

    Austral. J. Earth Sci.

    (2018)
  • C.L. Ciobanu et al.

    Silician magnetite: Si–Fe-nanoprecipitates and other mineral inclusions in magnetite from the Olympic Dam Deposit, South Australia

    Minerals

    (2019)
  • C.C.H. Conor et al.

    Alteration and mineralisation in the Moonta-Wallaroo copper-gold mining field region, Olympic Domain, South Australia

  • Cook, N.J., Ciobanu, C.L., Ehrig, K., Slattery, A., Verdugo-Ihl, M.R., Courtney-Davies, L., Gao, W., 2017. Advances and...
  • L. Corriveau

    Iron oxide copper-gold (± Ag ± Nb ± P ± REE ± U) deposits: a Canadian Perspective: Geological Association of Canada, Mineral Deposits Division

    Special Publication

    (2005)
  • L. Corriveau et al.

    Alteration facies linkages among iron oxide copper-gold, iron oxide-apatite, and affiliated deposits in the Great Bear Magmatic Zone, Northwest Territories, Canada

    Econ. Geol.

    (2016)
  • L. Courtney-Davies et al.

    Matrix-matched iron-oxide laser ablation ICP-MS U-Pb geochronology using mixed solution standards

    Minerals

    (2016)
  • R.A. Creaser

    The Geology and Petrology of Middle Proterozoic Felsic Magmatism of the Stuart Shelf, South Australia

    (1989)
  • K.C. Cross

    Acropolis and Wirrda Well

  • S. Dare et al.

    Did the massive magnetite “lava flows” of El Laco (Chile) form by magmatic or hydrothermal processes? New constraints from magnetite composition by LA-ICP-MS

    Min Dep

    (2014)
  • M. Dentith

    Geophysical signatures of South Australian mineral deposits: miscellaneous and minor deposits

    ASEG Extended Abstracts

    (2003)
  • A. Edfelt et al.

    The Tjårrojåkka Fe-oxide Cu (–Au) occurrence, Kiruna area, northern Sweden

  • J.J. Egozcue et al.

    Groups of parts and their balances in compositional data analysis

    Math. Geol.

    (2005)
  • Cited by (16)

    • Geochemical characterization of the Central Mineral Belt U ± Cu ± Mo ± V mineralization, Labrador, Canada: Application of unsupervised machine-learning for evaluation of IOCG and affiliated mineral potential

      2022, Journal of Geochemical Exploration
      Citation Excerpt :

      Corriveau et al. (in press b) and Hofstra et al. (2021) have furthered expanded the classification to include deposits where iron precipitates in iron silicates, iron sulfides and iron carbonates, forming metasomatic iron and alkali-calcic systems (see also Skirrow, 2022). The use of unsupervised machine-learning to extract otherwise cryptic information from large geochemical datasets has been implemented with success in the study of IOCG and affiliated albitite-hosted U deposits (e.g., Montreuil et al., 2013, 2015; Dmitrijeva et al., 2019a, 2019b). The combination of these methods with detailed lithological, petrological, and mineralogical information is a powerful tool to identify correlations between trace elements that partition into ore and gangue minerals in specific alteration types and map their spatial distribution within the deposits and host ore systems (Montreuil et al., 2013; Corriveau et al., 2016; Corriveau et al., in press a-a; Dmitrijeva et al., 2019a, 2019b).

    • Textural re-equilibration, hydrothermal alteration and element redistribution in Fe-Ti oxide pods, Singhbhum Shear Zone, eastern India

      2021, Geochemistry
      Citation Excerpt :

      Pending further studies on the timing of hematitization of magnetite, the actual influence of hydrothermal alteration and mineralizing events on hematite chemistry cannot be deciphered at the moment. Enrichment of ‘granitophile elements’ in hematite has been reported from Olympic Dam and linked to granite-derived fluid or leaching from felsic rock (Courtney-Davies et al., 2019b; Dmitrijeva et al., 2019). No granite magmatism, time equivalent of hydrothermal alteration and mineralization in this part of the SSZ, has so far been reported.

    • ~1760 Ma magnetite-bearing protoliths in the Olympic Dam deposit, South Australia: Implications for ore genesis and regional metallogeny

      2020, Ore Geology Reviews
      Citation Excerpt :

      Mineralization across the province also differs in terms of style, ore grade and structural features. For example, vein-style mineralization is dominant in the Acropolis prospect (Krneta et al., 2017; Courtney-Davies et al., 2019a; Dmitrijeva et al., 2019a) whereas the Wirrda Well prospect features structurally-controlled breccia pipe mineralization (Krneta et al., 2017; Courtney-Davies et al., 2019a). Calcic skarns are observed in the Punt Hill area (e.g., in the Groundhog Zn-Cu skarn; Reid et al., 2011), and Cu-Au skarns are present at Hillside, Yorke Peninsula (Conor et al., 2010; Ismail et al., 2014).

    View all citing articles on Scopus
    View full text