Controls, Expressions, and Discovery Potential of Gold Mineralization in the Central-Eastern Yilgarn Craton, Western Australia: New Insights from an Integrated Targeting Study

Abstract

This paper presents the results of an integrated targeting study covering the central-eastern Archean Yilgarn Craton of Western Australia, a region renowned for its substantial gold endowment (>40 Moz Au). The cornerstones of this study included custom-built geophysical and remote sensing targeting tools, a new lithostructural interpretation of the area, a targeting model based on the mineral systems approach, and a best-practice mineral potential modeling (MPM) workflow employing five complementary modeling techniques. The geophysical targeting tools were used to identify proximity, association, and abundance relationships between gold mineralization and gravity ridges or edges, as well as 95th-percentile K/Th radiometric and remotely sensed goethite–clay–iron feature depth index ratio anomalies. The lithostructural interpretation revealed structural trends oblique or orthogonal to the NNW-SSE-striking greenstone belts, likely representing important structural controls on gold mineralization. Fry analysis, used to assess the spatial distribution of geological point patterns, showed similar directions of maximum gold occurrence alignment. Together, these observations proved to be strong predictors of gold prospectivity in the MPM component of this targeting study. The MPM not only identified most known gold occurrences but also highlighted several underexplored areas with significant potential. The highest-priority MPM targets represent roughly an order-of-magnitude reduction in search space, the hallmark of a well-performing and practically useful targeting methodology.

1. Introduction

The central-eastern Yilgarn Craton of Western Australia (Figure 1), the focus of this study, hosts a gold endowment exceeding 40.0 Moz Au [1], mainly contained within eleven orogenic gold deposits or deposit clusters of Archean age. Given this significant endowment, the presence of 78 operating gold mines, and the potential for new discoveries, the area has attracted considerable interest in both gold exploration [2] and gold-related research [3,4,5,6,7,8,9,10,11,12,13].

Despite the region’s economic importance, published targeting studies for orogenic gold deposits within all or parts of the central-eastern Yilgarn Craton are scarce and largely limited to works of a few authors [14,15,16,17,18]. In 2000, Groves et al. [14] developed a comprehensive conceptual targeting model that informed a knowledge-driven, vectorial fuzzy logic approach [19] to mineral potential modeling (MPM) over an area of ~17,000 km² in the southern Kalgoorlie Terrane, an area with minimal spatial overlap with the present study. The model was based on only two input datasets: known gold occurrences and a lithostructural (solid geology) map. The resulting MPM outputs were interpreted to suggest that gold mineralization is controlled by a limited set of critical factors, including proximity to NW-SE-striking granitoid–greenstone contacts, crustal-scale faults (particularly NW-SE-striking fault segments), NW-SE lithological contacts, anticlinal hinges, and chemically reactive host rocks.

In contrast to Groves et al. [14], the 2010 study by Czarnota et al. [15], which covered both the Kalgoorlie and Kurnalpi terranes (~130,000 km²), benefited from a decade of advances in computing power, significant improvements in geographic information systems (GIS) and MPM, a greatly enhanced understanding of the geology of the Yilgarn Craton and its orogenic gold deposits, and the emergence of the mineral systems concept [20,21,22] This framework marked a shift from descriptive, deposit-centric targeting toward a broader, scale-integrated approach focused on the processes of mineral deposit formation and their geodynamic context. Unlike the earlier study by Groves et al. [14], Czarnota et al. [15] developed numerous geological, geochemical, and geophysical predictor maps believed to represent mappable features of orogenic gold systems. However, instead of applying mathematical algorithms (i.e., MPM) to integrate these predictors, the authors used what they described as a “simple additive GIS treatment.” This pseudo-MPM approach prevented the use of spatial statistics to assess the validity and performance of both the input predictor maps and the model. Instead, Czarnota et al. [15] defined an arbitrary buffer zone around areas of high gold prospectivity, which they claimed captured 75% of the known gold endowment within 5% of the modeled area. The buffer zone concept was also used to build potential occurrence maps along the Porcupine Fault, Abitibi gold province, Canada [23,24].

The most recent MPM study, by Witt et al. [16], produced gold prospectivity maps for both the Eastern Goldfields Superterrane (EGST; ~235,000 km²) and the entire Yilgarn Craton (~680,000 km²). The authors quantified spatial relationships between known gold endowment and a comprehensive set of regional geological, structural, geochemical, and geophysical criteria, from the best of which they generated prospectivity maps using a knowledge-driven fuzzy logic approach [25,26]. According to their spatial statistical analyses, the best regional-scale predictors of gold potential included proximity to mafic granite intrusions, areas of high fault density, regional fault bends, and greenstone belts.

This MPM-assisted targeting study, conducted over an area of ~27,150 km², differs from the earlier work in our key ways: (i) it is geophysically driven, using enhanced gravity and magnetic data to map the structural architecture, including cryptic basement features; (ii) it uses enhanced radiometric data and remotely sensed imagery to delineate prospective alteration systems; (iii) it establishes a new, internally consistent lithostructural framework that integrates those datasets into a single solid geology base for MPM; and (iv) it applies and cross-validates five complementary MPM techniques, spanning knowledge-driven, data-driven, and continuous methods. Together, these steps reduced the search space by an order of magnitude and revealed underexplored, gold prospective corridors largely invisible to traditional outcrop-based methods.

2. Geology

2.1. Eastern Goldfields Superterrane (EGST), Yilgarn Craton

2.1.1. Overview

The study area lies within the central-eastern Yilgarn Craton, a largely, poorly exposed Archean (~3730–2660 Ma) crustal block composed of extensive granite–greenstone and high-grade metamorphic gneiss terrains [27] (Figure 1). More specifically, it straddles the boundary between the Kalgoorlie and Kurnalpi terranes, which form part of the ~2960 to 2660 Ma EGST [7,12,28,29,30], one of the world’s most significant and highly endowed Late Archean metallogenic provinces [8]. Each terrane of the EGST (from west to east: Kalgoorlie, Kurnalpi, Burtville, and Yamarna) represents a distinct, structurally bounded entity distinguished by major differences in stratigraphic age as well as the whole rock and isotopic geochemistry of its volcanic and intrusive rocks [28,31]. These terranes are further subdivided into lithostructural domains characterized by internally consistent stratigraphy and structural histories, and bounded by interconnected fault systems interpreted largely from geophysical data [31].

2.1.2. Geology and Structure

The greenstone belts and major fault systems of the EGST generally strike N-S to NNW-SSE. Deep seismic reflection data show that the principal fault systems, traceable in gravity and magnetic data for up to 700 km along strike, are E-dipping, crustal-scale structures that intersect the Mohorovičić discontinuity. The supracrustal volcanic and sedimentary successions defining the greenstone belts are preserved in synformal basins, up to seven kilometers thick, with stratigraphies typically dipping and facing away from granite domes that flank or core the belts [7]. As summarized by [10], greenstone sequences across the EGST display a broadly coherent stratigraphy, typically comprising (from base to top) (i) a basal volcanic package dominated by ~2720 to 2690 Ma komatiite and basalt; (ii) a sedimentary package dominated by ~2690 to 2670 Ma deep marine siliciclastic and volcaniclastic sedimentary rocks; and (iii) a ~2665 to 2655 Ma ‘late-basin’ fill package of fluvial and deep marine sedimentary rocks, resting unconformably on the older greenstone successions [32,33]. Older (~2960–2770 Ma) greenstone successions, composed mainly of intermediate to felsic and mafic–ultramafic volcanic rocks, have been identified in the Burtville Terrane and show lithological and temporal affinities with the Youanmi Terrane in the western Yilgarn Craton [7].

Granitoids are abundant and widespread throughout the EGST and can be classified into five principal groups [7,8,12,34]: (i) high-Ca granites, the most abundant group (~60% of all EGST granitoids), occur both within and outside greenstone belts. They are typically granite, granodiorite, or trondhjemite, with emplacement ages clustering at ~2800 Ma, 2740 to 2650 Ma, and 2685 to 2655 Ma; (ii) high field strength element (HFSE)-enriched granites represent 5 to 10% of Yilgarn granites, typically granite or granodiorite, occurring mostly internal or marginal to greenstone belts. Emplacement ages cluster at >2720 to 2665 Ma and 2700 to 2680 Ma; (iii) mafic granites also comprise 5 to 10% of Yilgarn granites. This group includes granite, granodiorite, tonalite, trondhjemite, and diorite, occurring mainly internal or marginal to greenstone belts. They are commonly spatially associated with gold mineralization, with emplacement ages between >2720 and 2650 Ma, though younger examples likely exist; (iv) low-Ca granites, the second most abundant group (~20%), occur mainly external to greenstone belts. They are typically granitic or granodioritic, emplaced between 2655 and 2630 Ma, and are interpreted as products of partial melting of high-Ca granite source rocks; and (v) syenitic granites, the least abundant group (<5%), are typically internal or marginal to greenstone belts. They are syenite or quartz syenite in composition, spatially associated with gold mineralization, and emplaced at 2650 Ma, and 2655 to 2645 Ma.

2.1.3. Deformation History

According to [7], poor rock exposure and strain partitioning make it difficult to correlate structural information across the EGST. Consequently, its structural evolution remains poorly constrained and the subject of ongoing debate. To address these challenges, Blewett et al. [7] conducted an integrated structural framework study combining over 10,000 new structural measurements with stratigraphic, magmatic, metallogenic, and geophysical data. Based on this synthesis, the authors proposed a six-stage (D₁–D₆) deformation history (Figure 2):

• D₁ (~2720–2670 Ma): ENE-WSW-directed extension marked by rifting and greenstone deposition.
• D₂ (~2670–2665 Ma): ENE-WSW-directed shortening associated with cessation of volcanism, development of NNW-SSE-trending upright folds, and N-S-to NE-SW-striking dextral strike–slip and reverse faults.
• D₃ (~2665–2655 Ma): NE-SW-directed extension and extensional doming involving deep crustal exhumation and formation of late basins that record the first deposition of granite detritus in the EGST.
• D_4a (~2655–2650 Ma): ENE-WSW-directed shortening resulting in tightening of earlier folds, WSW-directed thrusting along NNW-SSW-striking faults, and generation of NNW-SSE-trending upright folds and reverse faults.
• D_4b (~2655–2650 Ma): WNW-ESE-directed shortening characterized by reactivation and sinistral transpression along earlier NNW-SSE-striking faults and generation of ENE-WSW-striking thrust faults recording NW and SE transport.
• D₅ (~2650–2635 Ma): NE-SW-directed shortening marked by dextral strike–slip movement along N-S- to NNE-SSW-striking faults and thrusting along NNW-SSE- to NW-SE-striking faults.
• D₆ (<2630 Ma): Low-strain vertical shortening and horizontal extension marked by crenulation development.

2.1.4. Metamorphic History

The EGST records metamorphic conditions ranging from sub-greenschist to granulite facies developed over ~130 Myr between ~2750 and 2620 Ma. A broadly conformable Neoarchean metamorphic history is indicated by regional isograds extending across terrane and domain boundaries [11,36]. Five regional metamorphic events have been identified across the EGST and temporally and spatially integrated with the structural framework of Blewett et al. [7] (Figure 2).

• M_a (>2750 Ma): Early, low-P/high-T upper-amphibolite to granulite facies assemblages are rare, restricted to magmatic arc-related ~2730 to 2810 Ma greenstone sequences in the western Burtville Terrane, and to HFSE granites and ~2675 to 2715 Ma greenstones of similar affinity in the Gindalbie Domain of the Kurnalpi Terrane.
• M₁ (~2750–2700 Ma): This event produced high-P/moderate-T assemblages preserved in narrow, upper amphibolite-grade zones along major, crustal-scale faults. The structural context, burial depth, and rapid exhumation suggest partial subduction and burial of buoyant magmatic arcs during arc accretion in subduction-like environments.
• M₂ (~2680–2670 Ma): A low-P/moderate-T event associated with contact metamorphism linked to emplacement of voluminous high-Ca granite melts into the upper crust, generated by partial melting of a subducted slab beneath the Kalgoorlie and Kurnalpi terranes. The M₂ event coincided with cessation of volcanism and D₂ crustal shortening.
• M₃ (~2665–2650 Ma): This event likely reflects lithospheric extension following subduction cessation with slab rollback and sag of the previously subducted plate causing extension of the overriding plate. The M₃ event coincided with D₃ metamorphic core complex formation.
• M₄ (~2650–2610 Ma): Low-P/high-T metamorphism was likely triggered by lower-crustal delamination, resulting in mantle upwelling and a thermal anomaly in the upper crust associated with widespread low-Ca granite magmatism.

2.1.5. Geodynamic Implications

Stratigraphic, structural, geochemical, and isotopic data support interpreting the Kalgoorlie, Kurnalpi, and Yamarna terranes of the EGST as failed intracontinental rift basins developed on older ‘proto-Yilgarn’ crust. These basins were later closed, uplifted, deformed, and metamorphosed during the ~2670 to 2630 Ma Kalgoorlie Orogeny [37,38].

2.2. Kalgoorlie–Kurnalpi Rift

The ~2720 to 2600 Ma Kalgoorlie–Kurnalpi Rift (Figure 1), identified by Witt et al. [12] based on stratigraphic, geochronological, and isotopic correlations across the Kalgoorlie and Kurnalpi terranes of the EGST, is interpretated to have formed in response to a major mantle input event that exploited weakened crust along the eastern margin of the Yilgarn proto-craton, comprising the South West, Narryer, and Youanmi terranes. According to [12], the genesis and evolution of the intracontinental Kalgoorlie–Kurnalpi Rift are consistent with spatially and temporally overlapping plume-related magmatism in the Kalgoorlie Terrane and west-directed subduction to the east of the Burtville Terrane. This geodynamic model, however, implies that plate tectonic processes comparable to modern-style plate tectonics operated as early as the Archean Eon, a contentious hypothesis that remains under active debate [39,40,41].

2.3. Kalgoorlie Terrane

The Kalgoorlie Terrane (Figure 1 and Figure 3), the westernmost lithostructural element of the EGST, is bounded by the Ida–Waroonga fault system to the west and Ockerbury fault system to the east. The terrane is divided into ten domains, five of which (Boorara, Jundee, Moilers, Ora Banda, and Wiluna) occur within the study area. The terrane consists mainly of 2710 to 2660 Ma greenstones overlain by coarse clastic (‘late basin’) sequences dated at 2658 to 2655 Ma. Older (>2730 Ma) lithostratigraphic sequences appear restricted to the Norseman, Boorara, and Wiluna domains. Due to extensive cover, well-defined lithostratigraphic sequences have been established only in the southern Kalgoorlie Terrane, where greenstones are subdivided into the ~2710 to 2690 Ma Kambalda Sequence (tholeiitic and komatiitic mafic–ultramafic rocks) and the overlying, ~2690 to 2660 Ma Kalgoorlie Sequence (felsic volcaniclastic and epiclastic rocks with subordinate lavas). The late basin sequences consist mainly of poorly sorted, polymict conglomerates that grade upward into sandstone and siltstone. The Kalgoorlie Terrane greenstones form highly deformed, curvilinear belts separated by variably deformed and metamorphosed granitoid rocks, emplaced principally between 2760 and 2620 Ma [7,28,36,42,43].

2.4. Kurnalpi Terrane

The Kurnalpi Terrane (Figure 1 and Figure 3), adjoining the Kalgoorlie Terrane to the east, is bounded to the west by the Ockerburry and to the east by the Hootanui fault system. It comprises seven domains, five of which (Edjudina, Gindalbie, Laverton, Menangina, and Murrin) occur within the study area. Lithostratigraphically, the Kurnalpi Terrane greenstones are subdivided into four main sequences: (i) The ~2800 Ma Laverton sequence, consisting of mafic–ultramafic volcanic rocks and banded iron formation with minor clastic sedimentary rocks intruded by felsic porphyries; (ii) the ~2715 to 2705 Ma Kurnalpi Sequence, comprising mafic volcanic and intrusive rocks, intermediate calc–alkaline igneous rocks, and feldspathic sedimentary units, with no clear stratigraphic equivalent in the Kalgoorlie Terrane; (iii) the ~2695 to 2675 Ma Minerie Sequence, composed of mafic–ultramafic volcanic rocks correlated with the Kambalda Sequence in the Kalgoorlie Terrane; and (iv) an unnamed, ~2690 to 2680 Ma sequence of bimodal rhyolite–basalt and felsic to intermediate, calc–alkaline igneous rocks forming a linear belt along the western margin of the Gindalbie Domain. These greenstones are overlain by late basin sequences younger than ~2673 Ma, consisting mainly of turbidites with subordinary carbonaceous shale, sandstone, conglomerate, chert, and magnetic shale [7,28,42,43,44].

3. Gold Mineralization

3.1. Gold Endowment

In the EGST, gold was first discovered at Coolgardie in 1892 followed by Kalgoorlie in 1893. Since then, more than 20 deposits containing over one million ounces (Moz) of gold (Au) have been discovered in the region. Cumulative production exceeds 145 Moz Au, with the Golden Mile at Kalgoorlie accounting for 65 Moz Au (or ~45% of the total) [35,45]. Including unmined resources estimated at >125 Moz Au [1], the total endowment of the EGST exceeds 270 Moz Au. The study area contains >40.0 Moz Au of this endowment, largely within eleven major gold deposits and deposit clusters [1] (Figure 1 and Figure 3;

Table 1. Main gold deposits in the study area.

Deposit Name	Discovery (Year)	Endowment (Moz Au)	Absolute Age (Ma)	Geology and Mineralization
Paddington	1894	>11.7	Unk	Principal host rock(s): granophyric qtz-dolerite (greenschist facies); Mineralization style(s): closely spaced, 1 to 5 cm wide, subhorizontal, sheeted gold and sulfide (apy > py, sp > gn)-bearing qtz-dol-ank-ab veins, and a 3 m wide, steeply dipping, laminated, gold- and sulfide (apy > py, ccp, gn > sp)-bearing qtz-cb vein; Alteration type(s): carbonatization, chloritization, sericitization, silicification, sulfidation (apy, py, po); Metal association: Not reported (Au-As?); Ore control(s): D2 kinematics and associated brittle–ductile structures; interaction of key structural elements (synclinal fold structure, location along the crustal-scale Bardoc Tectonic Zone), strong competency contrast between dolerite and surrounding ultramafic and sedimentary rocks
Gwalia	1896	>8.2	2755	Principal host rock(s): mafic schist, basalt (± pillowed) (lower amphibolite facies); Mineralization style(s): variably deformed, millimeter- to meter-scale, laminated and typically tightly folded and boudinaged gold- and sulfide (py, po > ccp)-bearing qtz-cb veins; Alteration type(s): carbonatization, biotitization, sericitization, silicification, sulfidation (py); Metal association: Not reported; Ore control(s): D2 kinematics and associated ductile structures, interaction of key structural elements (mylonite zone, fold hinge of a large M-shaped fold, Poker/Gwalia Fault, bulge of the Raeside Batholith, proximity to crustal-scale Keith–Kilkenny fault system)
Mt Morgans	1896	>5.0	2650–2630	Principal host rock(s): banded iron formation (BIF) (greenschist facies); Mineralization style(s): structurally controlled, disseminated gold–sulfide (py > ccp) in BIF and along the margins of qtz-cb veins; Alteration type(s): silicification, carbonatization, sulfidation (py > po, ccp, sp); Metal association: Not reported; Ore control(s): D4/5 kinematics and brittle(–ductile) structures, interaction of key structural elements (fault intersections, fold hinges developed on overturned anticlinal fold structure, dilational jog, proximity to crustal-scale Celia fault system), chemically reactive rock type (mag replacement by sulfides)
Tarmoola/King of the Hills	1897	>4.4	2650–2630	Principal host rock(s): trondhjemite, komatiite (greenschist facies); Mineralization style(s): sets of conjugate, 20 cm to 2 m wide, gold-, telluride-, sulfide (py, ccp, sp, gn)- and ±scheelite-bearing, laminated qtz-cb veins and breccias; Alteration type(s): silicification, carbonatization, sericitization, chloritization, albitization, and sulfidation (py, ccp, sp, ga); Metal association: Au-Sb-Mo-W ± Bi; Ore control(s): D4/5 kinematics and associated brittle–ductile structures, interaction of key structural elements (proximity to local shear zones and crustal-scale Keith–Kilkenny fault system), strong competency contrast between trondhjemite and komatiite, fault–valve action; Note: Tarmoola/King of the Hills is the largest known granite-hosted gold deposit in the Yilgarn Craton
Thunderbox	1999	>4.4	Unk	Principal host rock(s): porphyritic dacite (upper greenschist facies); Mineralization styles: structurally controlled, disseminated gold–sulfide (apy, po > py, sp, gn) accumulations and mm to cm thick, boudinaged and folded gold- and sulfide (apy)-bearing qtz veins; Alteration styles: carbonatization, silicification, albitization and sulfidation (apy, po); Metal association: Not reported (Au-As?); Ore control(s): D4/5 kinematics and brittle–ductile structures, interaction of key structural elements (local fold axes, Thunderbox shear zone, proximity to crustal-scale Perseverance fault system), strong competency contrast between porphyritic dacite and enclosing sedimentary and mafic volcanic rocks
Apollo Hill	1986	>2.0	Unk	Principal host rock(s): basalt (pillowed), dolerite, felsic volcaniclastic rocks; Mineralization style(s): Four sets of mm to cm thick, sheeted and stockwork-type, gold- and sulfide (py > ccp, sp, gn, po)-bearing qtz-cb veins; Alteration type(s): carbonatization, chloritization, sericitization, silicification, pyritization; Metal association: Not reported (Au-Ag-Cu-Pb-Zn?); Ore control(s): D4/5(?) kinematics and associated brittle–ductile structures, interaction of key structural elements (Apollo-Ra shear zone, proximity to crustal-scale Keith–Kilkenny fault system), strong competency contrast, lithological contacts
Aphrodite	1996	>1.6	Unk	Principal host rock(s): volcaniclastic rocks, felsic to intermediate (dacitic) porphyries; Mineralization style(s): conjugate, mm to cm scale gold- and sulfide (py > apy)-bearing qtz veins and breccias; Alteration type(s): silicification, carbonatization, sericitization, biotitization and sulfidation (py > apy > gn, ccp, stb); Metal association: Not reported (Au-As-Sb?); Ore control(s): D2 kinematics and associated brittle–ductile structures; interaction of key structural elements (local fold axes and crenulations, location along the crustal-scale Bardoc Tectonic Zone), strong competency contrast, chemically reactive sedimentary rock
Ulysses	1993	>1.6	Unk	Principal host rock(s): qtz-dolerite (sills), basalt; Mineralization style(s): stacked, shear zone-hosted, gold- and sulfide-bearing qtz veins; Alteration type(s): silicification, carbonatization, sericitization, albitization, sulfidation (py, po > ccp) ± biotitization, chloritization; Metal association: Not reported; Ore control(s): D4/5(?) kinematics and associated brittle–ductile structures; interaction of key structural elements (fault intersections with dolerite sills), strong competency contrast
Menzies	1891	>1.4	Unk	Principal host rock(s): metasedimentary rock, basalt, amphibolite ± porphyritic granodiorite; Mineralization style(s): locally stacked, shear zone-hosted, gold- and sulfide (py > apy)-bearing qtz veins and zones of brecciation; Alteration type(s): biotitization, chloritization, sericitization, silicification, sulfidation (py, po) ± carbonatization; Metal association: Au-As; Ore control(s): D4/5(?) kinematics and associated brittle–ductile structures; interaction of key structural elements (shear fabric, proximity to Menzies shear zone, location along the crustal-scale Bardoc Tectonic Zone)
Wonder	1890s	>0.9	Unk	Principal host rock(s): syenogranite (Bundarra Batholith) with partially assimilated greenstone rafts (mafic roof pendants); Mineralization style(s): gold- and sulfide (py > ccp, gn)-bearing qtz veins; Alteration type(s): silicification, carbonatization, sericitization, propylitization (hem), sulfidation (py) ± chloritization; Metal association: Not reported; Ore control(s): D4/5(?) kinematics and associated brittle–ductile structures; interaction of key structural elements (local faults, granite margin, proximity to the crustal-scale Keith–Kilkenny fault system), strong competency contrast between granite and mafic greenstone rafts
Zoroastrian	1894	>0.6	Unk	Principal host rock(s): granophyric dolerite; Mineralization style(s): steeply dipping and flat-lying, gold- and sulfide (apy, py, po)-bearing qtz stockwork veins; Alteration type(s): silicification, carbonatization, sericitization, chloritization, sulfidation (apy, py, po); Metal association: Not reported; Ore control(s): D4/5(?) kinematics and associated brittle–ductile structures; interaction of key structural elements (narrow synclinal fold structure, constriction zone between two granite domes, location along the crustal-scale Bardoc Tectonic Zone), strong competency contrast between dolerite and surrounding sedimentary, mafic and ultramafic rocks

Sources: Paddington = [5], Gwalia = [9,46], Mt Morgans = [47], Tarmoola/King of the Hills [4,9], Thunderbox = [48], Apollo Hill = [49], Aphrodite = [50], Ulysses [51,52], Menzies = [53], Wonder = [54,55], Zoroastrian = [56,57]. Gold endowment data were compiled from the above sources and [1]. Mineral abbreviations [58,59]: ab = albite, ank = ankerite, apy = arsenopyrite, cb = carbonate mineral; ccp = chalcopyrite, dol = dolomite, gn = galena, hem = hematite, mag = magnetite, po = pyrrhotite, py = pyrite, qtz = quartz, sp = sphalerite. Other abbreviations: Unk = unknown.

).

3.2. Gold Deposit Styles

The EGST hosts a diverse range of gold deposit styles controlled by various structural settings and hosted in multiple rock types spanning metamorphic grades from sub-greenschist to amphibolite facies. Despite this diversity, their shared characteristics strongly suggest that gold deposits within the EGST, and broader Yilgarn Craton, formed as part of regionally extensive gold mineralizing systems active during different stages of the late Archean evolution of the craton [60,61].

According to Tripp et al. [62], gold mineralization occurs in structurally and/or lithologically controlled lodes, typically as quartz ± carbonate veins, breccias, or disseminations enveloped by gold-related wall-rock alteration. Gold is present in all rock types, though it occurs most commonly in basalt and dolerite, and least commonly in large granitoid bodies [16,62]. Structural control was the dominant factor influencing mineralization, across scales, governing both hydrothermal fluid flow and the architecture and localization of gold deposition [62].

3.3. Gold Depositional Events

The EGST experienced multiple, protracted gold mineralization events during the late Archean, associated with both extensional and compressional deformation episodes within a ~60 Myr timeframe between ~2670 and 2630 Ma [5,7,8,9,12,15,62]. Although gold deposition occurred during most EGST deformation events [7], the formation of the largest deposits is attributed to three key stages (Figure 2) [7,8,15]:

• D₃ (~2665–2655 Ma): Development of the EGST crustal architecture during NE-SW-directed extension and metamorphic core complex formation. Major crustal-penetrating fault systems were established at this stage, linking the upper crust to a metasomatized mantle, as evidenced by the first emplacement of mafic and syenitic granites. This mantle connection likely introduced significant heat at the crustal–mantle boundary, promoting mantle-to-crust metal transfer. An important D₃ deposit in the study area is Gwalia (>8.2 Moz Au).
• D₄ (~2655–2650 Ma): The extensional crustal framework formed during D₃ was inverted during D₄, a brittle–ductile deformation event comprising an initial phase of ENE-WSW-directed shortening (D_4a) followed by WNW-ESE-directed shortening (D_4b). This phase coincided with a shift from high- to low-Ca granite magmatism driven by crustal melting and represents the most significant gold mineralization event in the EGST. Strike–slip deformation and reactivation of pre-existing structural heterogeneities during D_4b served as highly effective fluid-focusing mechanisms. The most prominent D₄ deposit is the Golden Mile (>65 Moz Au), located ~35 km south of the study area.
• D₅ (~2650–2635 Ma): A subsequent stress field reorientation induced NE-SW-directed shortening accompanied by dextral strike–slip, thrusting, and low-Ca granite emplacement. Gold mineralization was primarily controlled by brittle structures. A key D₅ deposit is Sunrise Dam (>10.3 Moz Au), located ~20 km east of the study area.

4. Materials and Methods

4.1. Study Background

The study described in this paper was conducted in late 2021. It was commissioned by the Australian mining company St Barbara Limited, which also defined the layout of the study area. At that time, the central-eastern Yilgarn Craton hosted St Barbara’s Leonora operations, comprising the Gwalia underground mine and a 1.4 Mtpa processing plant, along with nearby gold development and exploration projects. In 2023, the Leonora operations were sold to another Australian gold miner, Genesis Minerals Limited [63].

4.2. Methodology

This study employed a systematic, six-step workflow (Figure 4) that integrates geological understanding with advanced geospatial analysis for gold exploration targeting. The first step involved reviewing orogenic gold deposit models within the study area and the broader Yilgarn Craton to establish the conceptual basis for a mineral systems targeting framework [20,21,22,64,65,66], based on the critical processes in orogenic gold formation and their mappable expressions. Geophysical and remote sensing datasets were then filtered and enhanced to extract structural information and hydrothermal alteration signatures [67,68], which were incorporated into a new lithostructural (solid geology) interpretation. Building on this framework, mineral potential modeling (MPM) was conducted using a multi-technique approach that combined continuous, data-driven, and knowledge-driven methods, enabling cross-validation and comparison of the resulting prospectivity maps [69]. The final stage synthesized these results to identify and prioritize gold exploration targets across the central-eastern Yilgarn Craton.

4.3. Geoscience and Exploration Data

The geoscience and exploration data used in this study came from two principal sources (

Table 2. Types and sources of datasets used in this study.

Category	Data Type/Name	Source	Comments
Gold occurrences	Mines and mineral deposits (MINEDEX)	GSWA	Data available from DEMIRS Data and Software Centre: https://dasc.dmirs.wa.gov.au/
	MINEDEX operating mines map
Geology	1:100,000 state interpreted bedrock geology of Western Australia
	1:500,000 interpreted bedrock geology of Western Australia
	1:100,000 geological series maps
	1:500,000 state regolith geology
	In-house Eastern Yilgarn Craton geology map	SBM	Confidential dataset
	pmd*CRC 1:100,000 solid geology map, eastern Yilgarn Craton	[15]	Data or data links provided in quoted references
	Yilgarn Craton metamorphic facies map	[11,36]
Geochemistry	Yilgarn Craton εNd (juvenile crust) map	[34]
Drilling	Mineral exploration drill holes (open file)	GSWA	Confidential dataset
	Leonora drill hole database	SBM	https://dasc.dmirs.wa.gov.au/
Geophysics	400 m Bouguer gravity merged grid of Western Australia 2020 version 1	GSWA	Data available from MAGIX Online: https://geodownloads.dmp.wa.gov.au/downloads/geophysics/72203/, https://geodownloads.dmp.wa.gov.au/downloads/geophysics/72204/ and https://geodownloads.dmp.wa.gov.au/downloads/geophysics/72205/
	40 m reduced to the pole (RTP) magnetic merged grid of Western Australia 2021 version 1
	Radiometric grids (80 m) of Western Australia
Remote Sensing	ALOS World 3D—30 m (AW3D30) ALOS Global Digital Surface Model	OpenTopography	Data available from https://opentopography.org/
	Sentinel-2 (blue, green, red, and near-infrared (NIR) bands at 10 m and other bands at 20 m resolution)	European Space Agency	Data available from https://dataspace.copernicus.eu

Key to abbreviations: DEMIRS = Department of Energy, Mines, Industry Regulation and Safety; GSWA = Geological Survey of Western Australia; pmd*CRC = Predictive Mineral Discovery Cooperative Research Centre; SBM = St Barbara Limited. The data repositories listed here were last accessed on 8 October 2025.

): open-access repositories maintained by the Geological Survey of Western Australia (GSWA) and proprietary company data.

4.4. Processing and Interpretation of Geophysical and Remote Sensing Data

4.4.1. Geophysical Data and Processing

Geophysical data were downloaded in grid format (Table 2) and windowed to the study area, allowing for adequate buffer zones. Minimal processing was required, as the datasets were available as merged products with the appropriate data reduction already applied, providing complete coverage of the study area at various spatial resolutions (i.e., cell sizes). Access to objective, ‘blanket-coverage’ geophysical data of sufficient spatial resolution is essential for exploration targeting in areas where prospective bedrock is concealed by significant cover, such as in the central-eastern Yilgarn Craton. Moreover, geophysical data are well suited to mapping the expressions of mineralizing processes and are therefore a fundamental component of a mineral systems-type targeting approach.

Three types of geophysical data were sourced and processed for inclusion in this targeting study [70]: (i) Gravity data, which respond to variations in rock density and reflect subsurface density contrasts. Gravity surveys are therefore considered ‘depth penetrative’ methods. (ii) Magnetic data, which respond to variations in rock magnetism, primarily controlled by magnetic susceptibility. Like gravity data, they reflect subsurface contrasts. (iii) Radiometric data, which measure surface variations in naturally occurring gamma radiation, primarily emitted by radioactive isotopes of potassium (K), thorium (Th), and uranium (U).

In the EGST, gravity gradients show a clear association with faults that host gold deposits [71]. The extraction of gravity gradients from gridded gravity data has been performed in various ways since the method was first introduced in the late 1970s [72,73]. In this study, we adapted the phase congruency approach of Kovesi [74], using the differential upward continuation operator of Jacobson [75] instead of log-Gabor filters. This resulted in a multi-scale feature detection filter for potential field (gravity and magnetic) data. Topological features [76] extractable from gridded potential field data at multiple scales are as follows: (i) edges (i.e., maximum gradient curves), representing traces of physical property boundaries across gravity or magnetic data grids, typically arising from structural breaks such as faults, shear zones, or lithological contacts. (ii) Ridges (i.e., surface maxima or peak curves), corresponding to ridgeline features in gravity and magnetic data grids. Gravity ridges, for example, indicate peak gravity responses. In granite–greenstone terranes, these ridges often coincide with the thickest, dense, mafic–ultramafic rock units, effectively mapping structurally thickened sequences in the central parts of inverted greenstone basins. (iii) Valleys (i.e., surface minima), corresponding to valley-line features in gravity or magnetic data grids. Magnetic valleys often coincide with faults or fault corridors where primary magnetite has been partially or completely destroyed by fluid–rock interaction.

Radiometric data, by contrast, are particularly effective for detecting and mapping outcropping hydrothermal alteration systems. Studies by de Quadros et al. [77], Herbert et al. [78], and Shebl et al. [79] have demonstrated the utility of this technique for identifying zones of gold-related white mica (sericite) alteration through computations involving normalized radiometric data and their ratios.

4.4.2. Remote Sensing Data and Processing

ALOS digital elevation data [80] were downloaded in grid format (Table 2) and processed similarly to the geophysical data described above. The topographic data represent the Earth’s surface morphology, where elevation highs often correspond to units more resistant to weathering. Thus, even in ancient peneplained cratonic settings such as the Yilgarn Craton, topographic data can aid in mapping outcropping or subcropping bedrock, structural trends in exposed rock, and areas affected by past or present mining. Surface topography may also reflect geological processes relevant to exploration targeting, especially in well-exposed neotectonic terrains. In this study, the topographic data provided important input for both the lithostructural interpretation (mapping of outcropping and subcropping bedrock and structural trends) and the interpretation of the radiometric and remote sensing data (matching of anomalies with areas of outcropping and subcropping bedrock, and mine workings).

Multi-spectral Sentinel-2 data were downloaded as multiband grids (Table 2) and clipped to the study area. Several salt lakes within the area were manually delineated and removed to avoid false positives during processing. Spectral indexes were then calculated to highlight goethite and other ferric-iron-bearing phases (Band4/Band2) and clay minerals (Band11/Band12). The iron feature depth (IFD), a remote sensing index used to map the presence and abundance of iron-bearing minerals, was computed to emphasize potential weathering sulfides in the regolith [81]. The calculation of IFD involved determining the expected value at Band 8A of the Sentinel-2 data by performing a linear interpolation between Band 6 and Band 11. The ratio was then computed between the expected Band 8A value and the measured Band 8A value. This process measures absorption on Band 8A, which is associated with the presence of iron in minerals such as jarosite. The three indexes (goethite, clay, and IFD) were combined into a cyan–magenta–yellow (CMY) ternary image, which was converted to CIELab, a color space defined by the Commission Internationale de l’Eclairage (CIE), to calculate the Euclidean distance to black [82]. Black areas in the CMY image correspond to zones where all three indexes are high, indicating probable clay alteration and weathering sulfides. The distance-to-black values were then inverted so that higher values represent areas where all three indexes are relatively high. The resulting inverse distance grid was thresholded and polygonized.

4.4.3. Lithostructural (Solid Geology) Map

The workflow used to interpret the geophysical data and produce the solid geology map followed the approach outlined by [83]. It involved (i) constructing form lines, (ii) identifying magnetic rock units, (iii) characterizing geophysical domains, (iv) integrating geophysical interpretations with supporting data (e.g., geological maps), (v) defining structural elements, (vi) mapping and interpretating lithology, and (vii) developing an integrated lithostructural framework.

4.5. Mineral Systems Concept

The targeting model developed in this study was based on the mineral systems approach [20,21,22,64,84] and implemented following the methodologies described in [64,68,85,86]. In essence, the mineral systems concept views mineral deposits as localized expressions of a sequence of geological processes operating across multiple temporal and spatial scales: (i) source processes extract essential mineral deposit components (melts and/or fluids, metals, and ligands) from crustal or mantle sources; (ii) transport processes move these components from source to trap regions via melts and/or fluids; (iii) trap processes focus melt and/or fluid flow into physically and/or chemically receptive, deposit-scale sites; (iv) deposition processes enable efficient metal extraction from melts and/or fluids passing within the traps; and (v) preservation processes maintain the accumulated metals over time.

Where one or more of these critical processes fail to operate, a mineral deposit cannot form or be preserved. The probabilistic nature of the mineral systems framework aligns well with MPM [64,85,86], which relies on mathematical algorithms. A further strength of the mineral systems approach is its capacity to provide a robust yet flexible framework for developing a holistic, process-based targeting model and for identifying, mapping, or querying expressions of key mineralizing processes within available geoscience datasets [22,87].

4.6. Fry Analysis

Fry analysis [88] is a straightforward geometrical method for visualizing spatial patterns in point data. It records the distance and bearing from each point in a dataset to every other point, either manually or computationally. For n points, this yields n(n − 1) spatial relationships. These point translations are displayed in autocorrelation (Fry) plots and Rose diagrams, which enhance subtle spatial trends and reveal directions of maximum continuity in mineral deposit alignments [89,90,91]. In this study, we used DotProc 1.0, a Fry analysis freeware programmed by A. Kuskov, A. Mikhailov, and P. Dirks.

4.7. Background to Mineral Potential Modeling (MPM)

MPM, first introduced in the late 1980s alongside the rise of Geographic Information Systems (GIS) [92,93], has since evolved into a robust, cost- and time-efficient exploration targeting tool with strong big data analytics capabilities. MPM can concurrently handle, integrate, process, and model diverse and often large datasets, including geological, geophysical, geochemical, remote sensing, and drilling data. It is particularly well suited for effective screening and target generation across large search areas, from individual mineral districts or belts to entire countries or continents. In general, MPM comprises four main stages [21,87,94,95,96,97]: (i) the genetic model stage identifies the key geological processes responsible for the formation of the targeted deposit type to establish a conceptual model; (ii) the targeting model stage translates the genetic model into a targeting model in which critical ore-forming processes are expressed by mappable criteria (also referred to as spatial proxies, predictor maps, or targeting elements); (iii) the mathematical model stage integrates the predictor maps using weighted mathematical algorithms; and (iv) the target identification and prioritization stage maps and ranks the most prospective areas.

Mathematical modeling approaches are generally categorized as knowledge-driven, data-driven, hybrid (data- and knowledge-driven), or continuous methods employing logistic functions [26,94,98]. The choice of weighting methods depends largely on data availability, specifically the number of known mineral occurrences supporting the targeting model (prospect locations). Data-rich ‘brownfield’ regions, typically well explored around known mineral deposits, favor data-driven weighing methods. In contrast, data-poor ‘frontier’, ‘grassroots’, or ‘greenfields’ regions often require knowledge-driven approaches reliant on expert judgement [93,94]. Continuous weighting methods, by comparison, require neither prospect locations nor expert input, using continuous spatial evidence to achieve a superior predictive performance relative to discretized data [98].

In this study, we applied a multi-technique MPM framework [69] incorporating continuous (fuzzy gamma [99]; geometric average [100]; improved index overlay [101]), data-driven (random forest, RF [102]), and knowledge-driven (best-worst method-sum additive weighting, BWM-SAW [103,104,105]) modeling techniques.

5. Data Integration and Examination

5.1. Insights from the Enhancement Filtering of Geophysical and Remote Sensing Data

Enhancement filtering of gravity data revealed two critical insights into the proximity, association, and abundance relationships between gravity features and gold occurrences. Belt-parallel gravity ridges act as proxies for greenstone basin centers or zones of thicker mafic–ultramafic material along greenstone keels and root zones (Figure 5). A confidential, continent-scale gravity enhancement study for Australia demonstrated that a large proportion of Yilgarn gold (and nickel) deposits show strong proximity, association, and abundance relationships with gravity ridges. A similar pattern is evident in the study area, where most gold occurrences, except those wholly within intrusive rocks, exhibit comparable relationships with gravity ridges observable at multiple scales (Figure 6). Gravity edges represent discontinuities such as faults, shear zones, or major lithological boundaries. In the study area, the Thunderbox, King of the Hills, Gwalia, and Ulysses gold deposits all occur along a curvilinear, belt-parallel gravity edge that can be traced for >230 km along its strike, extending north beyond the study area boundary. Notably, this gravity linear coincides only locally with the Perseverance–Ockerbury–McClure fault system but shows strong spatial correspondence with the basement granite–greenstone contact (Figure 6). Similar, semi-parallel lineaments elsewhere in the study area may represent attractive first-pass exploration targets.

Figure 5. Schematic model for gold targeting in greenstone belt-hosted systems (modified from [106]). In this model, large, multi-million-ounce gold camps form in the central parts of now-inverted ‘greenstone basins’, particularly where these are intersected by cross-structures. Structurally thickened mafic–ultramafic rock packages in the central basin areas are mapped as gravity ridges (Figure 6), whereas strongly deformed basin margins, typically associated with smaller gold deposits, are mapped as gravity edges (Figure 7).

Figure 5. Schematic model for gold targeting in greenstone belt-hosted systems (modified from [106]). In this model, large, multi-million-ounce gold camps form in the central parts of now-inverted ‘greenstone basins’, particularly where these are intersected by cross-structures. Structurally thickened mafic–ultramafic rock packages in the central basin areas are mapped as gravity ridges (Figure 6), whereas strongly deformed basin margins, typically associated with smaller gold deposits, are mapped as gravity edges (Figure 7).

Figure 6. Gravity ridges (800 m scale) serve as proxies for centers of now-inverted greenstone basins or zones of thickened mafic material along greenstone keels and root zones. (a) Belt-parallel gravity ridges. (b) Cross-belt gravity ridges. High values (red colors) indicate high data confidence whilst low values (blue colors) indicate low data confidence. See text for details.

Figure 6. Gravity ridges (800 m scale) serve as proxies for centers of now-inverted greenstone basins or zones of thickened mafic material along greenstone keels and root zones. (a) Belt-parallel gravity ridges. (b) Cross-belt gravity ridges. High values (red colors) indicate high data confidence whilst low values (blue colors) indicate low data confidence. See text for details.

Figure 7. Gravity edges (3200 m scale) represent major faults or lithological boundaries. (a) Belt-parallel gravity edges. (b) Cross-belt gravity edges. Label #1 marks a curvilinear, belt-parallel gravity edge traceable for over 230 km and likely extending beyond the study area. The Thunderbox, King of the Hills, Gwalia, and Ulysses gold deposits all occur along this feature. Notably, this gravity edge does not coincide spatially with the nearby Ockerbury–McClure–Perseverance fault system but instead aligns with a granite–greenstone contact. Semi-parallel gravity edges, particularly those labeled #2 and #3, may represent promising exploration targets. High values (red colors) indicate high data confidence whilst low values (blue colors) indicate low data confidence. See text for details.

Figure 7. Gravity edges (3200 m scale) represent major faults or lithological boundaries. (a) Belt-parallel gravity edges. (b) Cross-belt gravity edges. Label #1 marks a curvilinear, belt-parallel gravity edge traceable for over 230 km and likely extending beyond the study area. The Thunderbox, King of the Hills, Gwalia, and Ulysses gold deposits all occur along this feature. Notably, this gravity edge does not coincide spatially with the nearby Ockerbury–McClure–Perseverance fault system but instead aligns with a granite–greenstone contact. Semi-parallel gravity edges, particularly those labeled #2 and #3, may represent promising exploration targets. High values (red colors) indicate high data confidence whilst low values (blue colors) indicate low data confidence. See text for details.

Magnetic edges typically mark faults, shear zones, or lithological boundaries. Within the study area, magnetic edge enhancement filtering highlights (i) proterozoic dolerite dykes, (ii) certain intrusive contacts, and (iii) segments of major and subsidiary shear and/or fault zones. Notably, all larger gold deposits coincide with magnetic edges (Figure 8).

Where exposed at the surface or in open-cut mines, orogenic gold systems in the study area show spatial proximity, association, and abundance relationships with (i) domains exhibiting high (≥95th percentile) K/Th ratios derived from radiometric data, and (ii) goethite–clay–iron feature depth index (IFD) anomalies extracted from remote sensing imagery.

Overall, the integration of advanced geophysical and remote sensing data modeling enabled the mapping of potential ore-controlling structures and ore signatures, significantly enhancing the toolkit for gold targeting in the central-eastern Yilgarn Craton. The semi-automated filtering methods applied to these datasets helped reduce bias and inconsistencies inherent in purely human-based interpretations.

5.2. Lithostructural (Solid Geology) Interpretation

The main objective of the lithostructural interpretation was to produce a coherent base map for MPM and target generation. Priority was given to developing a new structural framework that integrates the results of geophysical enhancement and filtering. Whilst the interpretation incorporated and respected elements of previous solid geology maps produced by the Geological Survey of Western Australia, pmd*CRC, and St Barbara Limited (Table 2), particularly regarding geological boundaries, the overall structural framework was reconstructed from the ground up, guided by the newly derived geophysical evidence.

In the study area, form lines (Figure 9a) delineate stratigraphy and schistosity in greenstones, compositional layering in granitoids, and schistosity in gneissic units. Although form line orientations vary considerably, NNW-SSE trends predominate, reflecting the principal stratigraphic and tectonic grain of the greenstone sequences. Local variations are mainly attributable to folding on the kilometer to multi-kilometer scale.

The major, first- and second-order faults within the study area (Figure 9b–d), likely the products of D₃ deformation and subordinate to earlier, N-S-striking D₁ master faults (cf. Figure 3 in [7]), include the (i) the Menzies–Boorara fault system separating the Ora Banda and Boorara domains of the Kalgoorlie Terrane; (ii) the Ockerburry–McClure–Perseverance fault system dividing the Kalgoorlie and Kurnalpi terranes; (iii) the Melita–Emu fault system separating the Gindalbie and Menangina domains of the Kurnalpi Terrane; (iv) the Keith–Kilkenny fault system dividing the Menangina and Murrin domains of the Kurnalpi Terrane; and (v) the Celia fault system separating the Murrin and Edjudina/Laverton domains of the Kurnalpi Terrane.

The subsidiary third- and fourth-order fault and fracture network defines several structural corridors with dominant N-S, E-W, NE-SW, and NW-SE orientations (Figure 9e,f). These corridors, oblique or orthogonal to the general NNW-SSE trend of the greenstone belts, likely played a key role in focusing fluid flow and controlling the gold mineralization. This interpretation is supported by Fry analysis results, which indicated preferred gold occurrence alignments along NNW-SSE, N-S to NNE-SSW, NE-SW, and NW-SE trends (Figure 10).

Lithologically, the study area can be divided into two broad Archean domains (Figure 11), namely felsic to intermediate granite–gneiss, and greenstone terrain. Intrusive rocks dominate, covering ~62% of the study area (plan view). These include large regional and deformed gneissic (‘external’ or ‘basement’) granitoid plutons, typically of monzogranitic composition, often delineated by well-defined gravity lows. Smaller ‘internal’ plutons intruding the greenstone belts range in composition from granitic to gabbroic.

The greenstones, generally associated with gravity highs, are classified into four broad lithological groups: (i) Late basin sequences comprising polymictic to oligomictic conglomerate and sandstone. (ii) Sedimentary siliciclastic sequences consisting of felsic volcaniclastic sandstone, siltstone ± conglomerate, chemical sedimentary rocks, and interleaved felsic volcanic rocks. (iii) Felsic to intermediate volcanic sequences comprising rhyolite to rhyodacite, andesite, felsic volcaniclastic, and siliciclastic rocks ± felsic and mafic volcanic rocks, interleaved with coeval basalt and dolerite. (iv) Mafic to ultramafic volcanic sequences consisting of basalt, komatiite, peridotite, serpentinite, and mafic–ultramafic chlorite, tremolite, and fuchsite-andalusite schists ± coeval dolerite and gabbro. Within all the above sequences are narrow, curvilinear units marked by strong positive magnetic anomalies, typically comprising banded iron formation (BIF), basalt, dolerite, gabbro, komatiite, peridotite, or ultramafic schist.

The study area is also traversed by numerous E-W- to ENE-WSW-striking dolerite intrusions belonging to the Paleoproterozoic (~2408–2401 Ma) Widgiemooltha dyke swarm [107,108] (Figure 11). These dykes are easily identifiable in magnetic datasets due to their strong magnetic contrast and linear geometry. Two groups are distinguished: (i) 2408 ± 3 Ma dykes that produce positive magnetic anomalies and typically strike ~075°, and (ii) 2401 ± 1 Ma dykes that produce negative magnetic anomalies and generally strike ~085°. The contrasting magnetic polarities reflect emplacement during different paleomagnetic regimes. Composed of solidified mantle-derived melts, the dykes are interpreted to have exploited pre-existing crustal heterogeneities, marking deep-seated, mantle-tapping structures of Archean origin that were reactivated during the Proterozoic. Notably, many gold occurrences, particularly the larger deposits, are situated along or within 5 km of the dyke swarm corridors.

6. Targeting Model

Developing an effective targeting model requires careful consideration of several interrelated factors [21,68,109,110]:

• The geological processes responsible for forming the targeted mineral deposit type [20,65,84,111,112].
• The mappable expressions of these processes in available geoscience and exploration datasets [84,113].
• The spatial and temporal scales at which these processes occurred and can be mapped [113,114].
• Geological uncertainty and human bias [65,115].

In this study, we adopted a mineral systems approach to exploration targeting [21], structuring our model around the key processes controlling orogenic gold formation and their corresponding mappable expressions, as summarized in

Table 3. Orogenic gold deposit targeting model.

Critical Processes	Constituent Processes	Targeting Criteria (Proxies)	Rationale for Proxies	Proxies Used for MPM
Source	Availability of energy to drive and sustain the mineral system	Source processes related to orogenic Au systems are cryptic in nature: – They operate at the broad regional scale, commonly involve the lower crust and upper mantle, and, thus, are often theorical in nature as are their proxies	Broad consensus exists in terms of orogenic Au systems of the Yilgarn Craton being formed in convergent margin settings, particularly in accretionary orogens, which, if mineralized, involve the following ingredients [34,116,117]: – High-heat flow environment that was sustained by mantle upwelling and associated, deep-seated magmatic processes – Fertile metasomatized upper lithospheric mantle that provided a prolific Au source – Translithospheric plumbing systems that focused magmas and fluids into the upper crust – Geodynamic switches from rifting to arc/back-arc accretion to craton collision that triggered regional extension or shortening events – Chemically important ligands for Au transport, such as hydrosulfide or chloride complexes that were liberated by metamorphic or magmatic devolatilization, respectively – Juvenile input linked to crustal thinning that allowed magmatism to tap younger, mantle-derived sources Ground-preparation is equally important, with mineralized orogens typically comprising the following [11,34,65]: – Lithospheric architecture, including lithospheric discontinuities – Attenuated greenstone belt terrain – Heterogenous strain distribution and steep metamorphic gradients – Large domains of sub-greenschist to lower amphibolite facies of metamorphic grade	Proximity to the following: – Domains of juvenile crust (εNd-values of −0.2 to 2.4) – Regional gravity highs – Greenstone belts – Domains of favorable metamorphic grade (sub-greenschist to lower amphibolite facies)
	Availability of fertile Au source region
	Availability of melts and fluids to extract Au from source region
	Availability of ligands to enhance Au solubility
	Favorable geodynamic/tectonic (“ground-preparation”) history
Transport	Fundamental translithospheric structures	First-order fault systems	– Serve as melt/fluid pathways – Facilitate Au transfer from the mantle/lower crust into the upper crust	Proximity to the following: – Principal faults – Subsidiary faults – Proterozoic dolerite dykes – Fold hinges – Late basin conglomerates – Flanks of granitoid bodies – Basement granitoids – Granitoid domes – Lithological contacts – Areas of demagnetization – Remotely sensed alteration systems – Domains of high K/Th values (≥95th percentile) – Mafic–ultramafic volcanic rocks – Felsic to intermediate volcanic rocks – Siliciclastic and sedimentary rocks – Internal granitoids – ‘High mag units’ – Known gold occurrences Density of the following: – Principal faults – Principal fault intersections – Proterozoic dolerite dykes – Lithological contacts – NNW-SSE- and ENE-WSW-striking gravity ridges – NNW-SSE- and ENE-WSW-striking gravity edges
	Crustal structures	Second-order fault systems	– Serve as melt/fluid pathways – Facilitate Au transfer from the lower into the upper crust
		Regional folds	– Play an important role as hinge-parallel fluid pathways in fold–thrust belts, in particular anticlinal folds [118]
		Domes	– Permeability is generally enhanced on flanks, apex, and nose regions of granite-cored domes and provides a focusing mechanism for Au mineralizing fluids
		Greenstone constriction zones	– Present areas where fault or shear zones converge and greenstone volumes are structurally attenuated
		Late basins	– Late basins are filled with (highly) permeable conglomerate units – They are typically developed in the hangingwalls of major fault systems
		Proterozoic dolerite dyke swarms	– Can be traced for hundreds of kilometers and mark deep-seated, mantle-tapping structures, that facilitated the ascent of mantle melts – May represent long-lived, reactivated Archean discontinuities
Trap	Transient catastrophic rock failure and concomitant structurally controlled, and highly focused fluid flow	Second- and higher-order faults	– Fry analysis indicates that Au deposits have proximity, association, and abundance relationships with NNW-SSE-, N-S- to NNE-SSW-, NE-SW-, and NW-SE-striking faults
		Fault irregularities – Dilational or contractional bends/jogs – Fault splays, tips, and wings – En-echelon fracture zones	– Fluid flow within fault zones is controlled by gradients in permeability and hydraulic head, which are highest at fault splays, bends, jogs, and intersections, localities commonly mineralized within orogenic Au systems [119]
		Structural intersections and intersection density	– Common association between high fracture density and Au mineralization throughout terranes and geological time indicates a fundamental underlying ore control [120] – High-density fracturing is commonly accompanied by enhanced fluid flow during fault–fracture mesh development, producing regional-scale fluid pressure gradients that focus hydrothermal fluids into preferentially fractured areas
		Fold structures	– Common association between folds and Au mineralization with the location of orogenic Au deposits often controlled by plunging fold hinges, fold noses, or limbs [121] – Truncated folds are particularly favorable structural settings for Au mineralization with many large Au deposits hosted in such settings (e.g., Golden Mile, Timmins, Damang, Callie, Bendigo)
		Ductile structures – Boudinage – Stretching lineation – Deflections in schistosity	– Boudinage, stretching lineations, and deflections in schistosity have little or no expressions in the available geophysical data and are generally not (comprehensively) recorded in the readily available 1:100,000 or 1:250,000-scale government mapping
		Competency contrasts	– May give rise to local zones of dilation and permeability, focusing fluid flow at or close to lithological contacts [122] – In the Yilgarn Craton, strong competency contrasts typically exist between greenstone units and internal granitoid/diorite/dolerite dykes, plugs, and stocks
		Lithological complexity	– Areas of lithological heterogeneity are characterized by geological diversity, which may be expressed, for example, by complex stratigraphic architecture, presence of small stocks, porphyries, or dykes, or occurrence of mantle magmas
Deposition	Physicochemical destabilization of Au-bearing fluids	Phase separation	– Linked to sudden fluid pressure drops as predicted by the fault–valve model [123]
		Fluid–rock interaction	– Orogenic Au systems typically form from reduced, near-neutral, low-salinity, CO2-rich, H2S-bearing fluids in which Au is transported in the form of Au(HS)2− complexes – Au deposition involves destabilization of these complexes due to fluids interacting with iron-rich rocks to form (arseno-)pyrite [124]
		Fluid mixing	– Orogenic Au deposits form along geochemical gradients as two or more end-member fluids, at least one of which is Au-bearing, which interact with one another [8,15]
Preservation	Geodynamics	Tectonic setting, crustal depth, and timing of Au deposit formation and post-Au deformation history	– The long-term preservation potential of orogenic Au systems is high because of their late orogenic timing, formation at crustal depths of typically >5 km, and within stable, thick, and buoyant lithospheric blocks [121]	Not used in this study: – The abundance and widespread distribution of Au occurrences is taken to indicate a good degree of preservation across the entire study area
	Peneplanation and climate	Peneplained, tectonically stable cratonic environments in (semi-) arid climate zones	– Peneplained cratonic blocks in (semi-)arid climate zones like the Yilgarn Craton are characterized by relatively low erosional activity, a factor that promotes the preservation of Au deposits – However, deep weathering (typically 100 to <300 m deep in the Yilgarn Craton) leads to oxidation, creating an upper oxidized and leached zone and a thicker, underlying saprolitic, clay-rich zone [125]

.

7. Mineral Potential Modeling (MPM)

7.1. Statistical Evaluation of Predictor Maps

The predictive performance of each predictor map was statistically evaluated using a reference dataset comprising 1773 gold occurrences, referred to as mineral deposit locations (MDLs), and 1773 unmineralized sites, referred to as non-deposit locations (NDLs). The MDLs include gold deposits, occurrences, and mineralized drill intercepts. The NDLs were selected based on the following criteria: (i) spatial randomization; (ii) sufficient distance from known gold occurrences, and (iii) location outside of the gold-permissive tract.

Two statistical methods were applied to assess predictor map performance: (i) the normalized density index (N_d) [126,127], illustrated in prediction-area plots (Figure S1), and (ii) the area under the receiver operating characteristic curve (AUC) [127,128] (Figure S2). Predictor maps meeting the thresholds of N_d > 1.00 [126] and AUC > 0.50 [128] were considered suitable for inclusion in the MPM process (

Table 4. Statistical parameters for the predictor maps as derived from prediction-area plots and receiver operating characteristic curves.

Predictor Map	Pr (%)	Oa (%)	Nd	AUC
Proximity to known gold occurrences	100	0	Infinity	1.000
Proximity to greenstone belts	69	31	2.230	0.965
Proximity to domains of favorable metamorphic grade	69	31	2.230	0.950
Proximity to felsic to intermediate volcanic rocks	59	41	1.440	0.948
Proximity to regional gravity highs	71	29	2.450	0.931
Proximity to mafic–ultramafic volcanic rocks	69	31	2.230	0.922
Proximity to basement granitoids	71	29	2.450	0.858
Proximity to ‘high mag units’	66	34	1.940	0.852
Proximity to areas of demagnetization	62	38	1.630	0.851
Proximity to subsidiary faults	66	34	1.940	0.833
Proximity to remotely sensed alteration systems	72	28	2.570	0.824
Proximity to lithological contacts	68	32	2.120	0.800
Proximity to fold hinges	59	41	1.440	0.794
Proximity to internal granitoids	64	36	1.770	0.788
Density of principal faults	61	39	1.560	0.786
Proximity to principal faults	64	36	1.770	0.778
Density of lithological contacts	58	42	1.380	0.758
Density of ENE-WSW-striking gravity ridges	63	37	1.700	0.746
Density of NNW-SSE-striking gravity ridges	59	41	1.440	0.715
Density of principal fault intersections	59	41	1.440	0.700
Proximity to siliciclastic and sedimentary rocks	64	36	1.770	0.691
Density of ENE-WSW-striking gravity lineaments	59	41	1.440	0.669
Proximity to flanks of granitoid bodies	59	41	1.440	0.639
Density of NNW-SSE-striking gravity lineaments	56	44	1.270	0.611
Proximity to domains of juvenile crust (εNd-values of −0.2 to 2.4)	52	48	1.080	0.611
Proximity to domains of high K/Th values (≥95th percentile)	55	45	1.220	0.560
Proximity to Proterozoic dolerite dykes	52	48	1.080	0.537
Density of Proterozoic dolerite dykes	51	49	1.040	0.515

Key to abbreviations: P_r = prediction rate, O_a = prediction area, N_d = Pr/Oa, AUC = area under the receiver operating characteristic curve.

).

7.2. Continuously Weighted Mineral Potential Models

Gold prospectivity in the Leonora district was modeled using three continuous integration techniques: (i) fuzzy gamma [99], (ii) geometric average [100], and (iii) improved index overlay [101]. Predictor maps were weighted using logistic [98] and small fuzzification functions [129,130], producing fuzzy membership scores within the interval [0, 1] for the proxy variables (Figures S5–S36).

The MPM outcomes for the Leonora district, generated from these integration functions, are shown in Figure 12a–c.

7.2.1. Continuous Fuzzy Gamma Model

Continuously weighted fuzzy spatial predictors can be integrated using fuzzy operators [26]. In this study, the fuzzy gamma operator, which combines the SUM and PRODUCT operators, was applied to achieve fine-tuned calibration of various inputs [131]:

$${\mu }_C=[1-{\displaystyle \prod _{i=1}^n(1-{\mu }_i){]}^{\gamma }\times [{\displaystyle \prod _{i=1}^n{\mu }_i{]}^{1-\gamma }}}$$

(1)

For each cell, ${\mu }_i$ represents the fuzzy score of the ith input spatial predictor, and ${\mu }_C$ denotes the resulting potential score obtained through the combination procedure, with the parameter constrained to the interval $(0\le \gamma \le 1)$. A gamma value of 0.9 was adopted in this analysis.

7.2.2. Geometric Average Technique

The geometric average provides an effective multi-criteria decision-making (MCDM) framework for coherently integrating continuously weighted predictor maps in MPM. Following the mathematical formulation in Equation (2), the geometric average (G_A) for each spatial cell is computed as the nth root of the product of the corresponding proxy values [100]:

$$G_A(F_1,F_2,\dots ,F_n)=\sqrt[n]{{\displaystyle \prod _{i=1}^n{F}_i}}=\sqrt[n]{F_1{F}_2\dots F_n}$$

(2)

For each spatial cell, n denotes the total number of predictor maps, while F_i represents the fuzzy weight corresponding to the ith predictor map. Equation (2) was adapted to the specific conditions of the study area as follows:

$$G_{Gold}(F_1,F_2,\dots ,F_{28})=\sqrt[28]{{\displaystyle \prod _{i=1}^{28}{F}_i}}=\sqrt[28]{F_1{F}_2\dots F_{28}}$$

(3)

In this formulation, $F_1,F_2,\dots ,F_{28}$ denote the fuzzy scores assigned to the evidential values within their respective predictor maps. After calculating the gold potential value (G_Gold) for each unit cell across the study area, the resulting outputs were spatially mapped to construct the geometric average gold potential model.

7.2.3. Improved Index Overlay Model

The Improved Index Overlay is an enhanced form of the data-driven index overlay method applied in both brownfield and greenfield MPM [101]. Unlike conventional or earlier data-driven approaches, this technique assigns weights to each predictor map using a modified Shannon’s entropy algorithm, thereby removing dependence on expert judgement or known mineral occurrences. The detailed workflow of this weighting procedure is described in [101]. Using this approach, predictor map weights were objectively determined (

Table 5. Predictor map weights as determined by the modified Shannon’s entropy method.

Predictor Map	Entropy (e)	Normalized Entropy Value (h)	Weight (W)
Density of lithological contacts	6095	0.0106	0.9894
Proximity to basement granitoids	7645	0.0133	0.9867
Proximity to felsic to intermediate volcanic rocks	8618	0.0150	0.9850
Density of principal fault intersections	14,307	0.0249	0.9750
Density of Proterozoic dolerite dykes	15,560	0.0271	0.9729
Density of principal faults	15,964	0.0278	0.9722
Proximity to greenstone belts	16,756	0.0292	0.9708
Proximity to domains of favorable metamorphic grade	17,192	0.0300	0.9700
Proximity to regional gravity highs	17,325	0.0302	0.9698
Density of NNW-SSE-striking gravity lineaments	18,651	0.0325	0.9675
Density of ENE-WSW-striking gravity ridges	20,723	0.0361	0.9639
Density of ENE-WSW-striking gravity lineaments	20,805	0.0363	0.9637
Density of NNW-SSE-striking gravity ridges	21,253	0.0371	0.9629
Proximity to mafic–ultramafic volcanic rocks	21,418	0.0373	0.9627
Proximity to fold hinges	23,135	0.0403	0.9597
Proximity to siliciclastic and sedimentary rocks	23,344	0.0407	0.9593
Proximity to domains of high K/Th values (≥95th percentile)	23,608	0.0412	0.9588
Proximity to domains of juvenile crust (εNd-values of −0.2 to 2.4)	23,831	0.0415	0.9585
Proximity to internal granitoids	24,255	0.0423	0.9577
Proximity to lithological contacts	24,480	0.0427	0.9573
Proximity to remotely sensed alteration systems	25,084	0.0437	0.9563
Proximity to ‘high mag units’	25,400	0.0443	0.9557
Proximity to known gold occurrences	25,649	0.0447	0.9553
Proximity to flanks of granitoid bodies	25,836	0.0450	0.9550
Proximity to areas of demagnetization	25,900	0.0452	0.9548
Proximity to Proterozoic dolerite dykes	26,630	0.0464	0.9536
Proximity to subsidiary faults	26,876	0.0469	0.9531
Proximity to principal faults	27,249	0.0475	0.9525

), showing that lithological contact density is the most influential predictor map, whereas proximity to principal faults has the lowest relative significance.

After calculating the predictor map weights via the modified Shannon’s entropy algorithm, the maps were integrated using the Improved Index Overlay framework as follows:

$$I_{IO}={\displaystyle \frac{{\displaystyle \sum _i^n{W}_{vi}{W}_i}}{{\displaystyle \sum _i^n{W}_i}}}$$

(4)

where for all cells within the study area, $I_{IO}$ denotes the Improved Index Overlay score, $W_i$ represents the weight of the ith predictor map assigned by the entropy algorithm, and $W_{vi}$ corresponds to the continuous cell value of the ith predictor map derived from logistic or small fuzzification functions. In this study, Equation (4) was formulated as follows:

$$I_{IO}={\displaystyle \frac{W{V}_1{W}_1+W{V}_2{W}_2+\dots +W{V}_{28}{W}_{28}}{W_1+W_2+\dots +W_{28}}}$$

(5)

Here, $W_1,W_2,\dots ,W_{28}$ denote the entropy-based weights of the predictor maps, representing the continuous evidential values assigned to each cell within the respective predictor maps.

7.3. Best–Worst Simple Additive Weighting (BWM–SAW) Mineral Potential Model

Among knowledge-driven MPM techniques, multi-criteria decision-making (MCDM) approaches are particularly valued for their efficiency [97,132,133,134,135]. In these methods, weights are typically assigned to predictor maps using either matrix-based or comparison-based procedures. Notably, the simple additive weighting (SAW) method [103] and the best–worst method (BWM) [105] are well-established examples of these approaches, both of which have been successfully applied in MPM studies [97,132,133,134,135].

In this study, a hybrid MCDM framework, namely BWM–SAW [134], was employed in conjunction with the overall performance (O_p) index [136,137,138]. The O_p index is derived from an improved prediction-area (P–A) plot incorporating three fundamental criteria: (i) occupied area curve, (ii) NDL prediction rate curve, and (iii) MDL prediction rate curve [136]. The BWM technique was used to objectively quantify the weights of the decision criteria represented by the predictor maps (Table 4). The O_p index then served to determine the weights of the predictor maps and to identify the best and worst predictors (Figure S33,

Table 6. Improved prediction-area plot parameters for the effective predictor maps.

Effective Predictor Maps	Parameters
	Pm	Pn	100 − Pm	100 − Pn	TPr	FPr	Op
0to known gold occurrences (DC1)	100	46	0	54	1	0.46	0.54
Proximity to regional gravity highs (DC2)	71	43	29	57	0.71	0.43	0.28
Proximity to basement granitoids (DC3)	71	43	29	57	0.71	0.43	0.28
Proximity to greenstone belts (DC4)	69	42	31	58	0.69	0.42	0.27
Proximity to domains of favorable metamorphic grade (DC5)	69	43	31	57	0.69	0.43	0.26
Proximity to mafic–ultramafic volcanic rocks (DC6)	69	44	31	56	0.69	0.44	0.25
Proximity to subsidiary faults (DC7)	66	42	34	58	0.66	0.42	0.24
Proximity to remotely sensed alteration systems (DC8)	72	48	28	52	0.72	0.48	0.24
Proximity to ‘high mag units’ (DC9)	66	43	34	57	0.66	0.43	0.23
Proximity to principal faults (DC10)	64	43	36	57	0.64	0.43	0.21
Proximity to lithological contacts (DC11)	68	47	32	53	0.68	0.47	0.21
Density of ENE-WSW-striking gravity ridges (DC12)	63	45	37	55	0.63	0.45	0.18
Proximity to internal granitoids (DC13)	64	47	36	53	0.64	0.47	0.17
Density of principal faults (DC14)	61	45	39	55	0.61	0.45	0.16
Proximity to siliciclastic and sedimentary rocks (DC15)	64	48	36	52	0.64	0.48	0.16
Proximity to areas of demagnetization (DC16)	62	46	38	54	0.62	0.46	0.16
Proximity to fold hinges (DC17)	59	47	41	53	0.59	0.47	0.12
Proximity to felsic to intermediate volcanic rocks (DC18)	59	48	41	52	0.59	0.48	0.11
Density of NNW-SSE-striking gravity ridges (DC19)	59	48	41	52	0.59	0.48	0.11
Density of ENE-WSW-striking gravity lineaments (DC20)	59	48	41	52	0.59	0.48	0.11
Density of principal fault intersections (DC21)	59	48	41	52	0.59	0.48	0.11
Proximity to flanks of granitoid bodies (DC22)	59	48	41	52	0.59	0.48	0.11
Density of lithological contacts (DC23)	58	49	42	51	0.58	0.49	0.09
Density of NNW-SSE-striking gravity lineaments (DC24)	56	49	44	51	0.56	0.49	0.07
Proximity to domains of juvenile crust (DC25)	52	47	48	53	0.52	0.47	0.05
Proximity to domains of high K/Th values (DC26)	55	50	45	50	0.55	0.5	0.05
Proximity to Proterozoic dolerite dykes (DC27)	52	50	48	50	0.52	0.5	0.02
Density of Proterozoic dolerite dykes (DC28)	51	51	49	49	0.51	0.51	0.00

Key to abbreviations: P_m = hits, P_n = false alarms, 100-P_m = misses, 100-P_n = correct rejections, TP_r = true positive rate, FP_r = false positive rate, O_p = overall performance, DC = decision criterion.

). Based on these results, the others-to-worst (OW) and best-to-others (BO) vectors were established (Tables S1 and S2) according to the O_p values listed in Table 6. The optimal weights $(W_1^{\ast },W_2^{\ast },\dots ,W_n^{\ast })$ and ${\xi }^{\ast }$ were obtained by solving the following problem [134]:

$$\begin{array}{l}\min \xi \\ \mathrm{s}.\mathrm{t}.\\ \left|W_B-a_{Bj}{W}_j\right|\le \xi ,\hspace{0.17em}\hspace{0.17em}\mathrm{for}\hspace{0.17em}\mathrm{all}\hspace{0.17em}j\\ \left|W_j-{\mathrm{a}}_{jW}{W}_W\right|\le \xi ,\hspace{0.17em}\hspace{0.17em}\mathrm{for}\hspace{0.17em}\mathrm{all}\hspace{0.17em}j\\ {\displaystyle \sum _j{W}_j=1}\\ W_j\ge 0,\hspace{0.17em}\hspace{0.17em}\mathrm{for}\hspace{0.17em}\mathrm{all}\hspace{0.17em}j\end{array}$$

(6)

where $a_{Bj}$ represents the precedence of the best decision criterion B over the decision criterion j, $a_{jW}$ demonstrates the precedence of the decision criterion j over the worst decision criterion W, and ${\xi }^{\ast }$ is the optimal consistency value. For this study, Problem (6) was formulated as shown in Equation (7):

$$\begin{array}{l}\min \xi \\ \mathrm{s}.\mathrm{t}.\\ \left|W_1-2W_2\right|\le \xi ,\hspace{0.17em}\mathrm{for}\hspace{0.17em}\mathrm{all}\hspace{0.17em}j\\ \left|W_1-2W_3\right|\le \xi ,\hspace{0.17em}\mathrm{for}\hspace{0.17em}\mathrm{all}\hspace{0.17em}j\\ \left|W_1-2W_4\right|\le \xi ,\hspace{0.17em}\mathrm{for}\hspace{0.17em}\mathrm{all}\hspace{0.17em}j\\ \hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}⋮\\ \left|W_1-9W_{28}\right|\le \xi ,\hspace{0.17em}\mathrm{for}\hspace{0.17em}\mathrm{all}\hspace{0.17em}j\\ \left|W_2-5W_{28}\right|\le \xi ,\hspace{0.17em}\mathrm{for}\hspace{0.17em}\mathrm{all}\hspace{0.17em}j\\ \left|W_3-5W_{28}\right|\le \xi ,\hspace{0.17em}\mathrm{for}\hspace{0.17em}\mathrm{all}\hspace{0.17em}j\\ \left|W_4-5W_{28}\right|\le \xi ,\hspace{0.17em}\mathrm{for}\hspace{0.17em}\mathrm{all}\hspace{0.17em}j\\ \hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}⋮\\ \left|W_{27}-2W_{28}\right|\le \xi ,\hspace{0.17em}\mathrm{for}\hspace{0.17em}\mathrm{all}\hspace{0.17em}j\\ {\displaystyle \sum _j{W}_j=1}\\ W_j\ge 0,\hspace{0.17em}\mathrm{for}\hspace{0.17em}\mathrm{all}\hspace{0.17em}j\end{array}$$

(7)

Optimal weights $(W_B^{\ast },W_1^{\ast },\dots ,W_W^{\ast })$ and ${\xi }^{\ast }$ were obtained by solving Equation (7) (Table S3). For $a_{BW}=a_{13}=9,$ the resulting consistency index was 5.23 [134], and the consistency ratio $0.009/5.25=0.002,$ indicating excellent consistency. Following the assignment of predictor weights, the SAW method was applied to rank all decision alternatives. A decision matrix $B_{1201497\hspace{0.17em}\times \hspace{0.17em}28}$ was built comprising 1,201,497 decision alternatives, each corresponding to a cell within the predictor maps, and 28 decision criteria. Alternatives were ranked via the step-by-step SAW procedure described by [134]. The resulting BWM-SAW gold potential model for the Leonora district is shown in Figure 12d.

7.4. Random Forest (RF) Mineral Potential Model

An increasing number of recent studies have demonstrated the effectiveness of RF-based MPM, consistently showing its superior performance compared to a wide range of previously applied supervised machine learning techniques [66,69,86,138,139,140,141,142,143]. The enhanced performance of RF is largely due to its ability to mitigate overfitting and improve predictive accuracy through the bootstrap aggregation (bagging) approach [102]. RF is an ensemble-based machine learning algorithm that generates multiple random subsets of training data, each used to construct an unpruned decision tree [102]. In this framework, bootstrapping with replacement is applied to the spatial proxy values at mineral deposit and non-deposit locations (labeled data). Approximately two-thirds of the labeled data (in-bag samples) are used to train each tree, while the remaining out-of-bag (OOB) samples are used to estimate model error and impurity. In summary, the RF model represents an ensemble of unpruned decision trees, each built from distinct training subsets that capture different underlying patterns (existent patterns).

The implementation of RF requires specifying two main hyperparameters: (i) the number of decision trees (n) in the ensemble and (ii) the number of predictor variables (m) considered at each node split [102]. In this study, n = 1000 and m = 9 were adopted following the procedure outlined in [139]. The relative importance of predictor variables was quantified using the mean decrease in the Gini impurity index (M_G) and the mean decrease in accuracy (M_a). The M_G index measures each variable’s contribution to improving node homogeneity and, consequently, the model’s overall discriminative power. M_a, on the other hand, is derived from OOB error, and reflects accuracy-based predictor relevance. As noted by [102], higher M_a and M_G values indicate greater predictor importance. Figure S34 shows the relative significance of predictor variables in the RF model: Proximity to volcanic rocks of felsic to intermediate composition is the least significant predictor, whereas proximity to juvenile crust (based on M_a) and proximity to basement granitoids (based on M_G) are the most significant.

Figure 12e presents the RF-based gold potential model generated from 27 validated fuzzified predictor maps (Figures S5b–S31b), excluding the “proximity to known gold occurrences” map to minimize exploration bias. The associated training error curve (Figure S35) shows a steady decline in model error with an increasing tree number, from an initial mean squared error of ~0.065 for the first tree to about 0.028 at the 1000th iteration.

8. Discussion

8.1. Geological and Structural Implications

One of the key geological observations in this study is that gold occurrences display proximity, association, and abundance relationships with structures oriented orthogonally or at high angles to the dominant NNW-SSW grain of the terrane- and domain-bounding fault systems and regional schistosity within the greenstone belts.

Notably, a spatial relationship was observed between gold occurrences, particularly the larger deposits, and E-W-striking magnetic edges, which correspond to geological features such as Proterozoic dolerite dyke corridors (Figure 8). A similar relationship was reported by Isles et al. [144] for the Yilgarn Craton as a whole, where most major gold districts exhibit “intriguing spatial associations” with Proterozoic dykes. Isles et al. interpreted this association as evidence that the dykes occupy much older planes of weakness formed prior to, or contemporaneous with, Archean gold mineralization. We concur with this interpretation and suggest that Proterozoic dyke swarms, which can be easily traced in magnetic and gravity data, delineate important yet cryptic basement structures, some of which, or segments thereof, likely played a critical role in localizing gold mineralization. This interpretation is supported by spatial statistical analyses conducted in this study, which yielded AUC > 0.52 and N_d >1.04 for predictor maps representing Proterozoic dolerite dyke proximity and density (Table 4). Following published criteria [126,128], any predictors with AUC > 0.50 and N_d > 1.00 are considered statistically valid, confirming a meaningful spatial relationship between gold occurrences and Proterozoic dolerite dykes, or, more precisely, the structures that host them.

Secondly, we identified N-S, E-W, NE-SW, and NW-SE structural trends, represented by subsidiary, late-tectonic (likely D_4/5), low-displacement faults and fractures that are oblique or orthogonal to the dominant NNW-SSW structural grain. Fry analysis revealed similar directions of maximum gold occurrence alignment, suggesting that these structures exerted significant control on gold localization. Some coincide with NE-SW- and NW-SE-striking low-displacement faults and fractures interpreted by [145], termed Perkins Discontinuities. These planar boundaries mark abrupt terminations of mineralization continuity, typically observed at hand-sample to deposit scales. According to [145], the Perkins Discontinuities identified at the Aphrodite deposit and surrounding area predate mineralization, resulting in an asymmetric gold distribution across these typically unmineralized faults and fractures. A detailed lithostructural interpretation of the Bardoc area by [146] showed that Aphrodite lies within a ~5 km wide, NE-SW-striking corridor defined by a series of fractures and low-displacement (<100 to 300 m) dextral faults (Figure 13). These structures, together with less common NW-SE-striking ones, cut obliquely across the dominant NNW-SSE trend of the Bardoc Tectonic Zone. Deposit-scale studies at Aphrodite reported cross-structures of similar orientation, including the NW-SE-striking, moderately NE-dipping Epsilon orebody [50] and numerous, weakly mineralized ENE-WSW (070°) and WNW-ESE (130°)-striking quartz–carbonate ± pyrite veins [50,147]. We infer that the intersection of the dominant NNW-SSE trend with subordinate NE-SW (±22.5°) and NW-SE (±22.5°) structures created a high-permeability network that strongly influenced gold localization, not only at Aphrodite but also in other areas where similar structural interactions occur. Further detailed studies across multiple scales and sites are required to better constrain the role of these cross-structures in controlling gold mineralization and ore geometry.

8.2. Geophysical and Remote Sensing Implications

Magnetic and gravity data are the most widely used and available geophysical datasets for regional targeting of orogenic gold mineralization in the Yilgarn Craton. Greenstones typically exhibit strong magnetic and density contrasts relative to the surrounding granites. High-resolution magnetic data effectively map greenstone belt fabrics, with fabric-parallel features commonly representing faults, shear zones, or stratigraphic boundaries. In contrast, ‘cross-fabric’ features, widespread across the craton, most often correspond to cross-cutting fault–fracture systems or mafic dykes. Gravity data, though generally of coarser resolution, remain useful for delineating greenstone distribution, relative thicknesses, and major cross-structures. The main objective of the semi-automated geophysical filtering in this study was to highlight geological features in a consistent and relatively unbiased manner. This was achieved is by extracting features of specific orientations (Figure 6, Figure 7 and Figure 8), as different orientations often correspond to distinct deformation styles or events. The phase congruency algorithm extracts features in eight different orientations (0°, 22.5°, 45°, 67.5°, 90°, etc.), which can then be combined to highlight particular structural trends. For example, the fabric-parallel features in Figure 6a, Figure 7a and Figure 8a were produced by combining the 0° and 167.5° orientations to highlight NNW-SSE to N-S trends, typically corresponding to belt-parallel fault or shear zones and stratigraphic contacts. Gold-bearing structures are generally parallel the host greenstone fabric, whereas cross-structures influence grade continuity and distribution. Isolating these structural types separately enhanced targeting effectiveness. Importantly, the phase congruency method is relatively amplitude-independent, allowing subtle features to be detected, which is critical since orogenic gold systems are not always associated with strongly magnetic bodies. Rather, their geophysical signatures vary widely due to differences in host rocks, structural settings, mineralization styles, ore minerals, and alteration assemblages.

Direct detection of unweathered orogenic gold systems using Sentinel-2 data is challenging. Pyrite, a common alteration mineral phase, cannot typically be detected through satellite remote sensing. However, Sentinel-2 can detect its weathering products, such as goethite and jarosite [81,148]. Pyrite oxidation generates acidic conditions that promote clay formation during silicate weathering [149]. Given the extensive weathering and oxidation of the Yilgarn Craton [150], any exposed gold mineralization is expected to be associated with ferric iron phases and clays. In the study area, zones showing coincident highs in goethite, clay, and IFD indices correlate well with known mineralization. While single remote sensing products often yields many false positives, this issue was reduced using a CMY ternary combination of the goethite, clay, and IFD indices. This approach efficiently and accurately highlights areas with coincident anomalies while automated extraction minimizes interpreter bias or inconsistency. The main limitation of this method, however, is that Sentinel-2 can only detect surface mineralization. Any significant cover will obscure signals from underlying rocks.

8.3. Mineral Potential Modeling (MPM) Implications

8.3.1. Predictor Map Performance

The significance of predictor maps in MPM can be viewed from two complimentary perspectives. In the pre-modeling stage, emphasis is placed on identifying and selecting the most effective predictor maps for integration into a potential model (Table 4). In the post-modeling stage, predictor maps are ranked according to their relative contribution to the prediction of potential values (Figure S34). Careful selection of high-performing predictor maps prior to modeling minimizes systemic uncertainty and ensures that the model is based on robust and geologically meaningful variables. Conversely, post-modeling evaluation allows the exclusion of predictors with limited or negligible predictive capacity, refining subsequent model iterations. This iterative process enhances the overall efficiency and robustness of the modeling framework while enabling a comparative assessment of successive models to evaluate the sensitivity of results to predictor map selection.

It is strongly recommended to use a combination of statistical analysis tools to identify the most effective predictor maps and quantify their contribution to potential value prediction. Using multiple performance metrics enables a more accurate comparison, differentiation, and informed decisions regarding the retention or exclusion of predictor maps. For example, both the “proximity to fold hinges” (predictor M1) and “proximity to flanks of granitoid bodies” (predictor M2) maps achieved identical N_d scores of 1.440. However, their AUC scores distinguish them, with M1 scoring ~1.24 times higher than M2. Relying solely on the N_d index would not have revealed this difference, whereas incorporating AUC clearly demonstrated the superiority of M1. Similarly, as shown in Figure S34, the predictor map for “proximity to areas of demagnetization” ranks 21st by M_a, but 9th by M_G. To achieve more precise and robust comparisons, it is advisable to integrate additional performance metrics, such as entropy or information gain.

8.3.2. Multi-Technique Approach

The robustness of MPM is highly dependent on the quality of the underlying genetic model and the accurate translation of key genetic processes into mappable targeting criteria. Since the conceptual model forms the core of the MPM process, most stochastic uncertainty arises from inappropriate selection of predictor maps. Although these issues can be reduced through mineral systems modeling, both systemic and stochastic uncertainties can be further mitigated by adopting a multi-technique approach to MPM. This approach (i) ensures optimal use of available empirical and conceptual information, (ii) allows comparison, contrast, and cross-validation of MPM results, and (iii) better constrains exploration targets by integrating outputs from diverse numerical models [69,114]. As demonstrated in [69,86,97], a multi-technique MPM strategy not only yields more robust exploration targets but also provides insights unattainable through any single modeling technique, while contributing to the development and calibration of new analytical tools and techniques.

In addition to these advantages, combining multiple techniques with the development of diverse gold potential models (Figure 12) also enables the assessment of model performance. To this end, we applied the improved P-A plot procedure of Roshanravan et al. [136]. Based on Figure S36 and the performance statistics reported in

Table 7. Improved prediction-area (P-A) plot parameters for the different gold potential models.

	Fuzzy Gamma	Geometric Average	Improved Index Overlay	BWM-SAW	RF
Pm (Hits)	77	78	79	88	88
Pn (False Alarms)	42	43	42	41	36
100−Pm (Misses)	23	22	21	12	12
100−Pn (Correct Rejection)	58	57	58	59	64
True Positive Rate (TPr)	0.77	0.78	0.79	0.88	0.88
False Positive Rate (FPr)	0.42	0.43	0.42	0.41	0.36
Overall Performance (Op)	0.35	0.35	0.37	0.47	0.52

, the RF algorithm to MPM delivered the best-performing gold potential model for the study area (overall performance, O_p = 0.52) (Figure 12e). Notably, the RF-derived potential model exhibits distinct spatial patterns compared to those produced by the other techniques (Figure 12). This variation, which enhances prediction reliability in this study, reflects an intrinsic property of the RF algorithm: namely, its ensemble of regression trees generated through bootstrap sampling, with final predictions derived from the average outputs, thereby reducing variance and improving robustness.

8.3.3. Predictor Map Sensitivity

Several predictor maps used in this study have the potential to introduce bias into the MPM because they (i) represent ‘self-fulfilling prophecies’ (e.g., proximity to gold occurrences) that artificially enhance prospectivity near known deposits; (ii) show interdependencies with other predictors (e.g., proximity to greenstone belts or granitoid flanks) due to spatial overlap; (iii) are ineffective in areas of cover (e.g., proximity to remotely sensed alteration systems or domains of high K/Th values) where signals fail to penetrate the regolith; and/or (iv) do not uniformly or objectively represent the entire study area (e.g., proximity to fold hinges) owing to inconsistent mapping scales or data resolutions.

Given that such biases can significantly affect outcomes [114], we assessed model sensitivity to selected predictor maps. Specifically, we used the best-performing RF approach to generate two alternative gold potential maps for the study area by varying the number and type of input predictors: Alternative RF Model 1, excluding proximity to (a) gold occurrences, (b) greenstone belts, (c) granitoid flanks, (d) fold hinges, (e) remotely sensed alteration systems, and (f) high K/Th domains; and Alternative RF Model 2, excluding only (a) to (c). Both models achieved an excellent overall performance (O_p) ranging from 0.49 (Alternative RF Model 1) to 0.54 (Alternative RF Model 2), comparable with or slightly exceeding that of the original RF model (O_p = 0.52). The results are not only statistically but also visually similar, with minimal discernible differences between the original and alternative RF-derived gold prospectivity maps (Figure 14).

Figure 14. Alternative random forest (RF) gold prospectivity models generated to test sensitivity and performance with respect to selected predictor maps. (a) Alternative RF Model 1 excludes proximity to known gold deposits, greenstone rocks, remotely sensed alteration systems, fold hinges, granitoid flanks, and domains with high K/Th. (b) Alternative RF Model 2 proximity to known gold deposits, greenstone rocks, and granitoid flanks. Overall performance (O_p) values of the alternative models are comparable to, and for (b), even higher than the O_p of the original RF model. See text for details and Figure 12e for a color legend.

Figure 14. Alternative random forest (RF) gold prospectivity models generated to test sensitivity and performance with respect to selected predictor maps. (a) Alternative RF Model 1 excludes proximity to known gold deposits, greenstone rocks, remotely sensed alteration systems, fold hinges, granitoid flanks, and domains with high K/Th. (b) Alternative RF Model 2 proximity to known gold deposits, greenstone rocks, and granitoid flanks. Overall performance (O_p) values of the alternative models are comparable to, and for (b), even higher than the O_p of the original RF model. See text for details and Figure 12e for a color legend.

8.3.4. Using Fractal Thresholding for Target Generation

The concentration-area (C-A) fractal technique [151], a well-established method for classifying continuous spatial data [66,101,127,134,152,153,154], was applied to prioritize areas of gold potential derived from the best-performing RF model. Two thresholds (Figure S37), representing major slope changes and corresponding to higher potential scores and data populations more strongly associated with mineralization, were identified at 0.830 (2nd-order target areas) and 0.958 (1st-order target areas). These thresholds were used to generate a ternary-class exploration targeting map, categorizing model outputs into domains of high, moderate, and low gold potential (Figure 15). The delineated target areas show a strong spatial correlation with known gold occurrences and key predictors in the underlying genetic and targeting models for orogenic gold systems. As shown in Figure 15, the prioritized high-potential (1st order) target areas represent an order-of-magnitude (>13 times) reduction in search space, a hallmark of a robust, practically effective targeting method [114].

Beyond delineating high-priority targets, the MPM results served three additional purposes: (i) providing context and constraints for prioritizing future gold exploration, (ii) identifying the most prospective ground held by competitors to inform potential investment, acquisition, or joint-venture opportunities, and (iii) supporting the monitoring of mineral tenement status and identification of new vacant ground for acquisition.

8.4. Targeting Implications

8.4.1. Role of MPM as a Targeting Tool

Since its inception in the late 1960s and early 1970s [113], MPM has evolved significantly, both methodologically and as an effective targeting tool. According to [85], while methodological and technological advances have been comprehensively addressed and widely published, practical applications of MPM remain limited in the literature, aside from the presentation of mineral potential maps. Notable case studies where modeling results were used for further analysis or where comprehensive targeting models and quality targets were developed include [69,85,86,95,155,156,157,158,159].

As summarized by [69], the 2020 discovery of the Gonneville Ni-Cu-PGE deposit, located ~70 km northeast of Perth, Western Australia (Figure 1), represents the first world-class discovery that is, at least partly, attributable to MPM, validating its effectiveness as a targeting tool and providing proof-of-concept for this technology. The greenfield discovery, made by Chalice Mining Limited [160], was guided by a Ni-Cu-PGE prospectivity map produced by Geoscience Australia [161], which also opened up an entirely new search space for such orthomagmatic sulfide deposits along the >1000 km long, underexplored western margin of the Archean Yilgarn Craton.

Despite their capabilities, MPM-generated prospectivity maps do not guarantee discovery success and should not be viewed as ‘treasure maps’. Instead, they serve as starting points for ‘treasure hunts’. MPM is best regarded as another tool within the ‘exploration toolbox’, specifically, a decision support tool for delineating, ranking, and prioritizing exploration targets based on modeled prospectivity. It is also important to recognize that prospectivity maps represent a snapshot in time, reflecting both the current conceptual understanding of the targeted mineral deposit type and the quality and availability of supporting data at the time of modeling. As new or improved data become available, results will invariably change, typically for the better [66].

8.4.2. Assessment of MPM Target Areas

The RF model, the best-performing MPM approach in this study, identified all large (>1 Moz) gold deposits and deposit clusters in the study area, including Paddington, Gwalia, Mt Morgans, Tarmoola/King of the Hills, Thunderbox, Apollo Hill, Aphrodite, Ulysses, and Menzies (Figure 14 and Figure 15). It also correctly delineated most smaller gold occurrences, with 83.93% of deposits within anomalies covering only 7.69% of the total area [86,140,162]. This demonstrates the excellent predictive performance of the RF model.

Areas modeled as having high gold potential, but lacking known occurrences and exhibiting the same mappable features as mineralized zones, clearly warrant further investigation, particularly where underexplored. One way to assess exploration maturity is to examine existing exploration data, especially drilling records as drilling provides the most definitive subsurface test. Reverse circulation and diamond core holes are particularly informative. Areas with a high drilling density typically indicate intensive exploration and a good understanding of subsurface geology, making the presence of undiscovered deposits unlikely. In contrast, areas with sparse or shallow drilling (<100 m) may remain underexplored.

Open-file drilling data were obtained primarily from the Geological Survey of Western Australia (GSWA) and supplemented by confidential data provided by St Barbara Limited (SBM) (Table 2). The GSWA database includes over 231,000 records, while the SBM dataset comprises more than 77,000 records (

Table 8. Summary statistics for the open-file Geological Survey of Western Australia (GSWA) and confidential St Barbara Limited (SBM) drill hole databases.

Parameters	GSWA Database	SBM Database	Comments
Number of drill holes	231,760	77,675
Main hole type
RC	84,078 (36%)	19,874 (26%)	Only 39% of all drill holes completed in the study area are RC or DD holes, whereas 56% represent geochemical drill holes comprising RAB, AC, and AUG holes.
RAB	62,377 (27%)	43,180 (56%)
AC	52,905 (23%)	8875 (11%)
AUG	14,276 (6%)	889 (1%)
DD	7043 (3%)	1827 (2%)
Other	11,081 (5%)	3030 (4%)
Hole depth—all drill holes
Min	0.0 m	0.0 m	The median value demonstrates that 50% of all drill holes completed in the study area have hole lengths of only 39 m or less.
Max	2895.6 m	2895.6 m
Median	39.0 m	36.0 m
Mean	50.5 m	53.9 m
Hole depth—RC holes
Min	0.0 m	0.0 m	Of the 80,078 RC holes in the GSWA database, 31,713 (~40%) targeted Au whilst 32,666 (~41%) targeted Ni ± Co; the remaining holes targeted mostly base metals ± Au.
Max	1043.1 m	624.6 m
Median	41.0 m	69.0 m
Mean	53.1 m	81.3 m
Hole depth—DD holes
Min	0.0 m	6.0 m	Of the 7043 DD holes in the GSWA database, 3511 (~50%) targeted Au ± Ag, Ni; the remaining holes targeted mostly base metals ± Au.
Max	2895.6 m	2895.6 m
Median	211.9 m	220.0 m
Mean	292.8 m	403.0 m

Key to abbreviations: AC = aircore; AUG = auger; DD = diamond core; RAB = rotary air blast; RC = reverse circulation.

), with some overlap between them. The GSWA records are open-file only, meaning confidential holes are excluded. Despite these limitations, the combined datasets provide a reasonably complete picture of drilling within the study area. Although, the greenstone belts may appear well tested at first glance (Figure 16b), this impression changes when only reverse circulation and diamond drill collars are plotted. The resulting map reveals that large portions of the greenstone belts remain untested at depth (Figure 16c).

Our review of previous drilling (Figure 17) identified six broad areas of high gold potential with relatively low drill hole densities, particularly when considering only reverse circulation and diamond core drill holes ≥100 m in length. These areas exhibit many of the targeting criteria outlined in Table 4 and, therefore, represent attractive exploration targets.

A manual analysis [95,163] delineated eight discrete targets within the broader prioritized MPM target areas (

Table 9. ‘Manual targets’ generated from RF-derived MPM targets. Targets are labelled according to their ranking ranging from #1 (highest-ranked) to #8 (lowest-ranked).

Target ID	Name	Rationale	Exploration and Ownership
#1	Dingo	Lithostructural target comprising a cluster of poorly tested intrusions of the McAuliffe Well Syenite; partially covered by Lake Raeside; hosts Dingo and Bull Terrier Au occurrences; proximal to 1st-order Keith–Kilkenny fault system; located along an NNW-SSE-trending gravity ridge	Shallow saprolite drilling only although open-file drill hole data appear to be incomplete; best historic drill intercept: 1.00 m @ 12.28 g/t Au; disjointed ownership
#2	Westralia North	Lithostructural target comprising BIF units and syenite intrusions; Korong and Akicia Au occurrences; proximal to 1st-order Celia fault system; located along an NNW-SSE gravity ridge; along strike from the Mt Morgans Au deposit	No deep drilling >150 m vertical; best historic drill intercept: 6.70 m @ 13.15 g/t Au; disjointed ownership
#3	Mt Boyce	Lithostructural target; largely soil covered; no reported Au occurrences in 2021; proximal to 1st-order Keith–Kilkenny fault system; located along an NNW-SSE-trending gravity ridge	No deep drilling >100 m vertical; best historic drill intercept: 2.00 m @ 34.50 g/t Au
#4	Mt Redcastle	Lithostructural target in ‘nose region’ of a large granite dome and comprising internal granitoids; hosts several known Au occurrences; proximal to unnamed 2nd-order fault system; located along NW-SE-trending gravity ridge	No deep drilling >100 m; disjointed ownership
#5	Mt Remarkable	Lithostructural target covering part of the Pig Well Basin; no reported Au occurrences in 2021; proximal to 1st-order Keith–Kilkenny fault system; located along an NNW-SSE-trending gravity ridge	Drilling is mostly associated with the Marvellous Au occurrence; best historic drill intercept: 82.00 m @ 0.83 g/t Au; disjointed ownership; partly located within an extensive registered site of Aboriginal cultural heritage
#6	Twenty Six Well	Lithostructural target; largely soil covered; no reported Au occurrences in 2021; proximal to 1st-order Keith–Kilkenny fault system; located along an NNW-SSE-trending gravity ridge	Minimal drilling; disjointed ownership
#7	Malcolm	Lithostructural target comprising BIF units; hosts numerous Au occurrences over a strike length of 10 km; located in between the 1st-order Keith–Kilkenny and Melita-Emu fault systems; located along an NNW-SSE-trending gravity ridge	Limited drilling; best historic drill intercept: 11.00 m @ 1.75 g/t Au + 10.00 m @ 1.26 g/t Au; disjointed ownership
#8	Lake Raeside	Lithostructural target covering part of the Pig Well Basin; largely covered by lake Raeside; proximal to 1st-order Keith–Kilkenny fault system; located along an NNW-SSE-trending gravity ridge	Minimal drilling; disjointed ownership; partly located within an extensive registered site of Aboriginal cultural heritage

Key to abbreviations: BIF = banded iron formation.

; Figure 17). With one exception, all manual targets are located along NNW-SSE-trending, ‘800 m scale’ gravity ridges, features that map thicker portions of greenstone belts and commonly coincide with significant gold deposits. As shown in Figure 6, many large gold deposits occur on or near these 800 m scale NNW-SSE-trending ridges.

The two highest ranked manual targets are particularly noteworthy:

• Target #1: This target covers a cluster of poorly tested intrusions of the McAuliffe Well Syenite, which intruded mafic volcanic-dominated greenstone sequences, abut the first-order Keith–Kilkenny fault system, and are crosscut by E-W- to NW-SW-striking dolerite dykes. Shallow rotary air blast (RAB) drilling by Saracen Gold Mines Proprietary Limited (maximum hole depth of 45 m) returned intercepts of up to 1.00 m @ 12.28 g/t Au from 9.00 m (hole YER143), 1.00 m @ 6.94 g/t Au from 10.00 m (hole YER158), and 2.00 m @ 2.20 g/t Au from 9.00 m (hole YER159), defining surficial, saprolite-hosted gold mineralization over an area of ~400 × 300 m at the Dingo prospect. The nearby Bull Terrier prospect returned intercepts of 16.00 m @ 2.59 g/t Au from 105.00 m (hole YBD-2) and 16.00 m @ 1.52 g/t Au from 60.00 m, including 1.00 m @ 211.70 g/t Au from 65.00 m (hole YRC-63) [164,165]. No deeper or more systematic exploration drilling appears to have been undertaken across the McAuliffe Well Syenite intrusive cluster. Moreover, the area is held by several exploration companies, and this fragmented ownership, apparently persisting for decades, has likely hindered a more integrated exploration approach.
• Target #2: This target comprises prospective mafic and felsic volcanic-dominated greenstone sequences, containing chemically reactive banded iron formations (BIFs) and syenite intrusions, located along strike from the Mt Morgans gold production center. Previous drilling returned intersections of up to 6.70 m @ 13.15 g/t Au from 95.00 m (hole MRC036), 5.90 m @ 7.24 g/t from 79.00 m (hole MRC003), and 2.90 m @ 5.41 g/t Au from 112.00 m (hole MRC028), defining a modest resource of >150 koz Au @ 1.40 g/t Au at the Korong-Waihi prospect. The mineralization remains open along strike and at depth, with no drilling below a vertical depth of 150 m [166,167]. As with target #1, fragmented ownership currently hinders a more integrated exploration approach.

8.5. Post-Study Target Validation

The MPM and targeting study described in this paper was completed in late 2021. A brief review of subsequent exploration activity revealed three 2024 to 2025 drilling campaigns whose results independently validate several of this study’s predicted targets.

Saturn Metals Limited conducted aircore (AC) drilling that returned 13.0 m @ 1.32 g/t Au from 56.0 m (including 4.0 m @ 4.31 g/t Au from 56.0 m; hole AHAC04641) and 12.0 m @ 0.72 g/t Au from 40.0 m (including 4.0 m @ 1.57 g/t Au from 44.0 m; hole AHAC2616) from two parallel gold-bearing structures, Aquarius West and Aquarius East [168]. These coincide with priority MPM target zones and manual Target #5 (Figure 18). The Aquarius gold trend extends along a ~26 km NNW-SSE corridor, defined by additional mineralized intercepts such as 3.0 m @ 8.51 g/t Au from 102.0 m (hole MBAC0629) and 36.0 m @ 0.48 g/t Au from 24.0 m (including 3.0 m @ 1.65 g/t Au from 57.0 m; hole MBRC049). The trend follows an NNW-SSE gravity edge parallel to the Keith–Kilkenny fault zone and aligned with a complex granite–greenstone contact.

Arika Resources Limited completed reverse circulation (RC) and diamond core (DD) drilling at its Yundamindra project (Figure 19), where ~97% of historic holes were <50 m deep. The new drilling defined mineralized zones at the open-ended Pennyweight Point (e.g., 14.0 m @ 15.48 g/t Au from 46.0 m, hole YMRC077; 30.0 m @ 3.86 g/t Au from 89.0 m, hole YMRC069; 36.0 m @ 2.14 g/t Au from 104.0 m, hole 25YMD001) and Landed at Last prospects (e.g., 4.0 m @ 41.56 g/t Au from 52.0 m, hole 25AYRC007; 30.0 m @ 2.26 g/t Au from 26.0 m, hole YMRC050; 14.8 m @ 3.10 g/t Au from 87.0 m, hole 25YMD003) [169]. These prospects lie entirely within first-order MPM target zones, coinciding with a complexly faulted greenstone constriction wedged between two granitic batholiths. They also align with NNW-SSE and NNE-SSW gravity edges and coincide with distinct, remotely sensed goethite–clay–iron feature depth index (IFD) anomalies.

Genesis Minerals Limited is conducting infill and extensional drilling at the Admiral deposit (0.3 Moz Au; Figure 20) to confirm strike and depth continuity among several shallow open-pits, aiming to define a single large system. Recent RC drill results include 17.0 m @ 2.98 g/t Au from 154.0 m (hole 25USRC1685), 17.0 m @ 1.84 g/t Au from 1.0 m (hole 25USRC1661), 16.0 m @ 1.88 g/t Au from 135.0 m (hole 25USRC1683), 9.0 m @ 3.30 g/t Au from 175.0 m (hole 25USRC1677), and 11.0 m @ 2.73 g/t Au from 177.0 m (hole 25USRC1681) [170]. The Admiral open-pits occur within a first-order MPM target zone and coincide with belt-parallel and orthogonal gravity ridges, as well as K/Th radiometric and remotely sensed goethite–clay–iron feature depth index anomalies.

9. Summary and Conclusions

This study of Archean orogenic gold systems in the central-eastern Yilgarn Craton illustrates that a geophysically informed lithostructural framework, when combined with a mineral systems-based targeting model and a suite of complementary prospectivity algorithms, provides a powerful tool for gold exploration.

Enhancement filtering of gravity, magnetic, radiometric, and remote sensing data revealed cryptic structures and alteration systems that correlate with known gold deposits and extend into underexplored areas.

The solid geology map produced in this study formed the foundation for a multi-technique mineral potential modeling (MPM) workflow incorporating five complementary prospectivity methods spanning knowledge-driven, data-driven, and continuous approaches. The best-performing random forest (RF) model accurately predicted most known gold occurrences while significantly reducing and prioritizing the search space for new targets.

The key strengths of this approach are its reproducibility, explicit integration of deep-seated and cross-structure architectures, and capacity to test multiple weighting schemes to minimize bias. Overall, the results confirm that combining a detailed geophysical interpretation with mineral systems-based prospectivity modeling is an effective strategy for identifying concealed gold corridors and reducing exploration risk. The methodology is readily transferable to other commodities and geological provinces worldwide.

In summary, the following conclusions can be drawn from this study:

• Structural controls: Cross-cutting N-S, NE-SW, and NW-SE fault and fracture corridors overprint the greenstone belt-parallel fabric and consequently aligned with clusters of gold occurrences. Supported by Fry analysis, these oblique structures likely represent key fluid-focusing zones that controlled mineralization.
• Geophysical insights: Gravity and magnetic data reveal concealed structural controls. Belt-parallel gravity ridges correspond to thickened mafic–ultramafic roots of inverted greenstone basins, the most prospective settings for multi-million-ounce systems. Gravity edges and magnetic edges delineate fault zones and granite–greenstone contacts. Major deposits such as Thunderbox, King of the Hills, Gwalia, Ulysses, and Paddington occur along these edges, indicating previously unrecognized cryptic crustal boundaries.
• Integrated geological framework: Enhanced gravity, magnetic, radiometric, and remote sensing filters clarified subtle but continuous fault zones, alteration halos, and greenstone basin keels not visible in published maps. These datasets were synthetized into a new, geophysically constrained lithostructural framework for mineral targeting.
• Mineral potential modeling (MPM): Five complementary techniques were applied, with the random forest (RF) model performing best and detecting ~84% of known deposits while reducing the prospective search area to <8% of the total. Key predictive variables included proximity to (i) gravity ridges and edges, (ii) cross-structure density, (iii) alteration indices (high K/Th, Fe-oxide-clay ratios), (iv) greenstone belts, and (v) internal granitoid margins. Several underexplored, highly prospective corridors were delineated, particularly along concealed gravity-defined trends.
• Exploration implications: Deep-seated trans-lithospheric and cross-belt faults are confirmed as primary ore controls in the Yilgarn Craton. Integrating geophysical enhancement, structural reconstruction, and mineral systems-based modeling provides a quantitative, reproducible framework for regional-scale targeting. This approach effectively maps cryptic crustal architecture, reducing both the exploration risk and search space.