e Prominent Hill mine.

This is an edited report that was supplied to Oz Minerals. 


OreFox was given eight terabytes of exploration data to find an economic mineral deposit in the Olympic Dam domain, South Australia. 

The objective of this project is to predict any economic mineralisation locations in the Mount Woods project area. OZ Minerals has been exploring for economic deposits in the Mt Woods target tenements for nearly a decade. During this time, they have invested significant resources into the area, but have not found a suitable deposit.



Methods of Analysis

OreFox targeted any economic mineralisation types in the target area.

OreFox primarily used a combination of three Artificial Intelligence (AI) systems in its unique approach during this project; ‘Prospector AI’, ‘Hunter AI’ and ‘Target AI’. A commercial-off-the-shelf data mining system, which provides actionable insights from previous exploration data, was administered to complement the Hunter AI and Target AI systems further.

Prospector AI is a data-driven, deterministic system, inspired by greenfields exploration work, for areas of low data density. Prospector AI uses a supervised deep learning algorithm that compares the data of economic-grade Australian gold and copper deposits against the data of areas to be explored.

Hunter AI is a hybrid probabilistic system which uses both deep learning and machine learning to refine targets generated from the Prospector AI. This works by applying various additional data, allowing the system to produce defined target locations for drilling. Hunter AI technology is capable of GIS-based, machine learning, multi-mineral prospectivity mapping. By using artificial neural networks, support vector machine, random forest and AdaBoost for any economic mineral possibilities, Hunter AI can produce a detailed visual representation of these target locations.

Map od Projects Generated for Oz Minerals Limited





The OreFox technology, Prospector AI, generated 108 targets for exploration using its proprietary machine and deep learning algorithms. Of these targets, 69 were found to have previously been drilled. These drilled targets were further examined, and all downgraded, with the exclusion of Joe’s Dam. The Joe’s Dam target, located North-West of the Prominent Hill deposit, produced data that warranted further research. However, after an in-depth investigation, it was removed as a potential top 10 target.

The Hunter AI system was then implemented to refine the remaining targets down to 10. These top 10 targets can be found in Table 1. Three targets of significant note are Nick’s Rings, Amy’s Target and OreFox Pipe.

OreFox explored the potential of Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) spectral bands for lithological and minerals detection over the Prominent Hill area before ground disturbance, this showed a possible alteration outline near the orebody. We think that more ground truthing would improve any remote sensing work.

Accurate digital and physical 3D model representations of geophysics and geology were found to be crucial in understanding surface and subsurface interactions.

Text mining was utilised in the rapid processing of the 160GB of PDF files supplied by OZ Minerals. OreFox believes that the geology of the region has been significantly well studied. However, without access for ground truthing, it must be acknowledged that information cannot be added to what is already known and published.

Training the Systems

Every mineral deposit, even when classified into a deposit model, has its unique characteristics and different expressions at different scales. OreFox’s algorithms are designed for a “mineral systems” approach to targeting, looking for not only the orebody but the components that make up an economic deposit. This approach utilises a combination of positive and negative training examples.

Data for known economic gold and copper deposits within Australia were employed as the positive training data for the Prospector AI system. Known economic deposits are locations defined as having been mined; are currently being mined; or holding justifiable authentication of a resource, such as a JORC report. This approach allows the Prospector AI to be unconstrained from the specificity of finite deposit models and avoid biases due to the locality, such as simply concentrating the search area for IOCG’s around Prominent Hill or Olympic Dam.

Numerous negative training examples were utilised. This data consisted of known uneconomic deposits rather than known unmineralized areas or randomly selected locations.

Testing of the system was completed by excluding specific datasets, containing known deposits, from the training data process and requesting Prospector AI to select deposits in this vicinity. OreFox technology easily processed the available data and highlighted locations where, unknown to the system, economic ore deposits had been previously confirmed. A map book of these economic deposits identified can be found HERE.

Training Data

Due to the nature of the system, if the training data utilised is imperfect, then the insights and information extracted will be imperfect. OreFox doesn’t use open file deposit databases, problems we have noticed with typical deposits databases are:

  • Existence of data - who has visited the site and verified it is an economic mineral deposit?
  • Quality of data:
    • Some data is from the 1800s - how accurate is it? Spatially as well as geologically?
    • Confusion of alluvial and hard rock deposits
  • Sub-economic deposits - is this introducing false positives?
  • Data inconsistencies - poor data transcribing, use of local grids etc.

OreFox believes in using quality over quantity. For this reason, OreFox does not use deposit databases for training our systems and instead employs only high quality, economic grade, mineral deposit datasets. These deposits have been proven beyond doubt that they contain enough metals to be mined. Our analogy is, if you were training an image recognition system to find lions, you would not use kittens as the training data. Although they are both felines, they are vastly different.

The OreFox training dataset was carefully compiled over two years. All care has been taken to exclude any man-made features. As seen in Figures 1a, b and c, there are artefacts in the magnetic data of Olympic Dam which have nothing to do with the orebody.




artefacts in the magnetic data of Olympic Dam


OreFox built a dataroom and two databases for data management within our distributed team.

The database for the Prospector AI, comprised of unbiased and uniformly sampled data such as geophysics, radiometrics and remote sensing data.

A second database was built for the Hunter AI using the same geophysics, radiometrics and remote sensing data, as well as a combination of interpretive data such as geochemistry and structural data. This variation, used in the later stages of the Hunter AI due to the nature of the data creation process, lessened the impact of possible biases in target generation.

To allow for domain adaptation, the Prospector AI system was cloned, and transfer learning was incorporated to retrain on Olympic Dam, Prominent Hill, Cairn Hill, Carrapateena and Mount Gunson. Multiple testing and retraining stages were undertaken, due to the limited size of the training dataset provided.

By using the Prospector AI, OreFox was able to generate 108 targets for further assessment. As part of this stage, known deposits identified were Olympic Dam, Prominent Hill, Joe’s Dam, Triton, Mt Woods, Cairn Hill, Zeus, White Hill South, White Hill North, Neptune, Bluebird, PH South and Nichol Well. The system did not identify Carrapateena or Peculiar Knob during this target generation stage.    

These identified targets were then used as the input data for the Hunter AI system. The Hunter AI system evaluated the targets generated by the Prospector AI, established 69 targets had been previously drilled and downgraded them accordingly. The remaining 39 targets were then investigated using the additional interpretative geochemical and structural data, allowing Hunter AI to order the locations and highlight the top 10 targets by downgrading targets with unsuitable characteristics.

Target AI was utilised for the final stage in the process. Joe’s Dam, a well-known and previously drilled deposit, was evaluated using this system. Assessment of the assay data collected from Joe’s Dam considered the location to be uneconomical; however, due to the analysis and output from the OreFox technology, it was considered possible that the drilling program had potentially missed the deposit. However, regardless of the locations being highlighted as having high potential by the Prospector AI and Hunter AI, it was deemed too small to be economical by the Target AI. Due to time restrictions, OreFox was unable to adequately evaluate the opportunities available at Joe’s Dam and believe that this prospect could be a case study for continued investigation.





AI Target Generation

Due to the three-step nature of the OreFox system, the targets were generated and downgraded systematically.

  • 108 targets were generated by the Prospector AI system
  • 69 of these targets were downgraded by the Hunter AI system
  • Outliers were investigated by the Target AI system
  • The cloned and locally trained Prospector AI and Hunter AI systems were used to refine the remaining 39 targets down to 10


OreFox Pipe Target

Mapping shows that the OreFox Pipe target is on a gravity high which coincides with a central magnetic high. This could be evidence of a magnetite core zone and an annular magnetic low due to magnetite destruction within a surrounding zone.

In the outer zone the magnetic response shows some probable deformation with a north west trending shear.

Although four locations were highlighted in this area, the target of significant focus has been labelled the ‘Primary Target’ in Figure 3.

OreFox Pipe IOCG Target


The target is in the Taurus sub-domain - a narrow, northwest-southeast trending sub-domain is bounded to both the southwest and northeast by long strike length magnetic discontinuities, interpreted to represent major shear zones and faults. The magnetic high target sits directly under an interpreted magnetite skarn (Betts, 2008) - which is bounded to the north by gabbronorite and to the south by undifferentiated granites - and hosted by a non-magnetic quartz-feldspar-mica gneiss as in Figure 4.

 Geology of OreFox Pipe IOCG made with ArcGis Pro Watercolor Theme



Modelling suggests the targets also have the right density for a prospective IOCG target with a conceptual target size between Prominent Hill and Carrapateena, and with a predicted depth range/shape likely to be amenable to open cut mining. The Hunter AI suggests this target is geophysically similar to Prominent Hill.

When compared in 3D to the Prominent Hill magnetic signature (Fig 6. And 7.), the OreFox Pipe target has a similar amplitude, shape and annular trough magnetic low.

Prominent Hill magnetic signature in 3D

Figure 6 Prominent Hill magnetic signature in 3D

OreFox Pipe Target in 3D

Figure 7 OreFox Pipe Target in 3D





Seismic interpretation of OreFox Pipe Target

Line 05, in Figure 9, with interpretation shows lots of thrusting in an area identified by the Hunter AI, the OreFox Pipe target, in the southeast of the OZ Minerals tenements. This work and the interpretations were done independently of the AI process.

Figure 9. Line05 seismic time section shows strong differentiation in seismic facies. The orange facies has low reflectivity and low frequency. Green facies has high reflectivity and medium frequency and negative polarity. Yellow facies has high frequency and high reflectivity.

The orange chaotic, low reflectivity package seems to extend down 1.8s, or 4500m @5000m/s along with some deep normal faulting, consistent with the tectonic reversal regime, Betts et al. 2016. With the sedimentary-style layers, green facies, thrusted and faulted above this "pipe", the southern end presents a good target for brecciated sediment-hosted deposits.

 Seismic interpretation of line 5 data, IOCG

Figure 9 Seismic interpretation of line 5



 Previous Exploration 

The target is surrounded by the Bellatrix, Taurus, and Blaze deposits, which have all been drilled showing basement at around 92m. Drilling has been carried out to the west of the target along a magnetic high ridge area. 

Extract from ENV06960

“Taurus is a NW-trending coincident gravity anomaly and high amplitude magnetic anomaly NE of the White Hill area interpreted as a fault splay containing abundant IOCG (magnetite-dominant) alteration.

Previous Work: Attempts by Normandy to aircore drill the target in 1996 failed to reach the basement. The prospect was revisited by Normandy in 1999 and successfully drilled with 10 vertical mud/diamond holes. Many of these holes intersected intensely magnetite-actinoliteserpentine altered and brecciated metasediments (meta-carbonate and meta-pelite), granitic gneisses and mafic intrusives (possibly Hiltaba-aged). Best results came from Hole 99DD253 which intersected 95.2m @ 0.12% Cu (incl. 36m @ 0.17% Cu) in jigsaw textured hydrothermal magnetite breccias.

TUR002 has a number of anomalous radiometric spikes, but is most notable for one particular wide interval of gamma response that could be considered worth pursuing. The interval 387.3 to 410.3 m (total 23 m) contains an average 0.0195% eU3O8* or 195 ppm. This includes a narrow 1.5 m thick interval (394-395.5 m) of average 450 ppm eU3O8*. The peak is 800 ppm. The consistency of this gamma response is typical of IOCG systems and would imply that this interval will contain anomalous copper and/or gold, based on sta16-ohmal analysis of Prominent Hill drill hole geochemistry. This interval is also associated with high plateau resistivity, consistent with intense haematite or magnetite alteration. Immediately above is an unusual interval (350-375 m) where point resistivity is high relative to 16-ohm formation resistivity, suggesting this is a zone of semi-massive sulphide.

The Taurus Prospect in Domain 5 may be related to a large intrusion lying to the east. Both Domains 5 and eight should be considered prospective for magnetite breccia systems similar to Joes Dam and Manxman.”


Nicks Rings Target

Our geophysicist, Nick Josephs, independently outlined two high-value targets in the same area that the Prospector AI and Hunter AI outlined a target (Fig 14).

The most promising combination of geophysics and structural geology occurred at the “Nick’s North Ring” primary target, which sits mostly in the ‘Devil’s Playground’ volcanics and is bounded by the Mulgathing Complex.


Nicks Ring Target

The prospectivity of Nick’s Ring targets are supported by structural faulting and by MT continuous, conductive flares (Fig 16.) from very deep (400km) to shallow (~1.2km, the noise/resolution limit). The primary target has even stronger evidence of an anomaly with a concentration of medium grey Worm lineaments (Fig 15.), a high Total Magnetic Intensity (TMI) and its proximity to a gravity crest.

This target also sits in the “Ceduna Ring” zone, that encompasses Olympic Dam, Carrapateena, Mt Gunson, Prominent Hill and Cairn Hill.

The OreFox Hunter AI outlined a target where a gravity high sits on an interpreted, subtle, NE trending, magnetic low feature, with a N-NW fault system terminated at a longer NW trending fault system.


Magnetotelluric continuous conductive flares from very deep 400km to shallow

Magnetotellurics continuous, conductive flares from very deep (400km) to shallow

Nicks Ring target on 12.5 Worm map


Figure 13 Nicks Ring target on 12.5 Worm map



Amy’s Target

The OreFox Prospector AI highlighted an area of interest that was then reprocessed by the OreFox Hunter AI. It outlined a single target of interest in this area; in fact, all three AI systems highlighted the exact same area (Fig 15.)




Amys Target map

Figure 15 Amy's Target map



Maps show a gravity anomaly and the magnetics in SA_TMI_LP800_VRTP_TILT show some interesting small NE trending magnetic anomalies at the intersection of two major structural features (Figs 18 and 19)


Amys Target magnetics in 3D

Figure 18 Amy's Target magnetics in 3D

3D magnetics with iso lines to highlight anomalies

Figure 19 3D magnetics with isolines to highlight anomalies




Geological information for this area is sparse, with maps showing mainly transported regolith. ASTER shows an intense iron enrichment, possibly gossanous, but without ground truthing there is a likelihood of it being transported enrichment.

An interpreted Archaean to Mesoproterozoic NE trending fault system conforms to the large NE trending magnetic anomaly.

Looking at O’Driscoll lineaments, the target is within the Gz corridor, and close to the Olympic Dam (G9s) corridor (Fig 21). This target conforms to Dr. O’Driscoll’s hypothesis that “major Australian ore deposits hosted by a particular lithology occur at or within 4 km of the position where their host lithology is crossed and disturbed by a conspicuous, relatively narrow, regional lineament exceeding 80 km in length. A further auspicious characteristic is the presence of a second, approximately orthogonal, lineament sharing the intersection”.

Some historical drilling has taken place near this target, but no reports were found with any useful information.

The OreFox Target AI highlighted two areas to drill, both on steep gravity gradients.

OreFox’s geologists suggest drilling an inexpensive RC hole to test the structure of the anomaly.

Drilling to basement here has the potential to be deep, with an estimated depth to the basement being 300m-400m.

It is possible that the NE magnetic highs visible in the SA_TMI_LP800_VRTP_TILT are remnants of a NW trending magnetic ridge that has been cross-cut with a faulting structure; allowing strong alteration to occur resulting in magnetite destruction, with the possibility that the low magnetic zones are the fault structures. (Fig. 21)

Target magnetic image with gravity contours

Target magnetic image with gravity contours

Amys Target on Gravity map with lineaments

Figure 20 Amy's Target on Gravity map with lineaments




Kudzai’s Target


The Prospector AI and Hunter AI systems outlined two targets with affinities to the Cairn Hill deposit. These were labelled the Kudzai North and Kudzai South targets.

For the Kudzai North target, two gravity highs are located on the edges of a North-West (NW) trending magnetic high, which geology maps show as a magnetic quartzo-feldspathic gneiss, which runs parallel to a large fault system. An interpreted NW fault runs along with the magnetic high, with a second northern fault terminating near the gravity high.

The Kudzai North target was determined the most prospective. It can be seen in Figure 23 as the ‘Primary Target’.

Drilling here is predicted to be within 200m-300m.

Kudzais Target







Sheree’s Target

The OreFox Prospector AI highlighted a gravity anomaly which coincided with a magnetic high ridge, bisected by an interpreted fault structure and an interpreted fold axis.

Geology shows linear magnetite-iron formations in non-magnetic pelites and quartzo-feldspathic gneiss.

From this information, three targets were determined as locations of interest. Of these, the southernmost target was considered the most viable. This is labelled as the ‘Primary Target in Figure 24.

The AI suggests that all three of these targets have affinities to the Cairn Hill deposit.

Drilling depth to target would be estimated at 100m-300m.


Sherees Target






Alan’s Target

The Prospector AI highlighted this location as an area of interest. Reprocessing by the Hunter AI outlined three primary targets, all of which have an overall gravity signature similar to the Mt Gunson deposit.

As seen in Figure 25, three gravitational anomalies – coinciding with a magnetic high ridge – formed on the northernmost limb of a magnetite-iron formation, synclinal structure, which is bisected by interpreted fault structures and folds and hosted in non-magnetic pelites.


Alans Target magnetics in 3D

Figure 20 Alans Target magnetics in 3D



The proximity of the Mount Woods and Mount Woods South deposits supports the presence of mineralizing events in this area.

Shallow drilling was carried out in 1964 near the North West end of the target. However no report was found with results of this program – DELHI ENV.422 (SITE 6, PEG 617).

The southernmost target was considered the most viable. This target can be seen as the ‘Primary Target’ in Figure 21.

Alans Target

Figure 21 Alan's Target




Ashleigh’s Target

This target is located in the ‘Devil’s Playground’ volcanics. It is a gravity anomaly on the edge of a magnetic high ridge that could be affected by an EW trending fault, and a NW trending fault that is interpreted to be terminated by the EW fault. There is potential for an unmapped NE trending structural feature that would make the target area the junction of major structural features.

The proximity of the Halifax Hill and Torch deposits, and Fe-Oxide breccias to the north of the target, also support the hypothesis of mineralising events occurring within this area.

Of the three target points established within this area, the easternmost location has been deemed the most viable. This can be seen as the ‘Primary Target’ in Figure 22.

 Ashleighs Target

Figure 21 Ashleigh's Target







Prospectivity Targets




A more conventional Prospectivity Algroitum was used to generate 3 targets using drill data and leveraging from our deep crustal fault research.

Targets showed persistent, high concentrations of copper and sit in areas about magnetic and gravitational highs, in the vicinity of deep faults.

This also allowed an estimate of mineralization depths for the other targets.


Prospectivity Targets


 The Drumstick prospectivity zone 

The Drumstick prospectivity zone (Fig 29) sits mostly out of the OZ Minerals tenement area and has been drilled on its western end. The area is mostly Calcsilicate-qtz-fld-mica gneiss, with a magnetic quartzo-feldspathic gneiss, and a small intrusion of undifferentiated granite.

Two areas of interest sit within the target area, being two areas of coincident magnetics, steep gravity gradients, and faulting. They sit at a depth of between 200 and 300m


 Drumstick Target

Arrowhead target

The Arrowhead target (Fig 30) sits on a magnetic high, with some small gravity high anomalies, and is bisected by faulting. Target depth is between 200 and 300m.

The area is dominated by low magnetic gabbros and gabbronorites, has a small granite plug to the north, and is close to a magnetite skarn.


Arrowhead Target map






Jackknife target

The Jackknife target (Fig. 31) sits on the edge of a gravity low and magnetic high, with some minor faulting. Target depth is between 200 and 300m

The target sits in non-magnetic quartz-feldspar-mica gneiss with a magnetic linear quartzo-feldspathic gneiss nearby.








Concentrations of Cu, Ag, Zn, and Au in parts per million (ppm) from drill hole assays were regressed on depth (average of the DEPTH_TO and DEPTH_FROM fields), and the principal components of distance to the nearest deep faults, Total Magnetic Intensity and 1st vertical derivative of gravity measurements using a Random Forest. The rationale for these choices were that TMI and gravity are geophysical measurements that should respond to mineralisation, and that proximity to deep faults provides a structural basis and pathway for fluid movement and mineralisation. The depth of the assay measurements was used as a covariate because (1) it would allow us to make spatial predictions that varied meaningfully with depth, and (2) depth is an important factor for mineralisation, and, more importantly, extraction.

The chosen covariates were highly correlated. This was especially true of the TMI data, for which there are two sources of correlation. The first source of correlation is the red, green and blue bands which make up the image. Each band from each image was included as a potential covariate. The second source of correlation is the three different magnetic intensity products which are measuring the same thing but with different corrections applied to the raw data. They contain different information, but much of it is redundant. The principle components were calculated to decompose these raw, collinear data into 14 mutually uncorrelated dimensions. We used all the principle components as covariates because there were few, and our goal was not dimension reduction.

The Random Forests were then used to predict elemental concentrations across the entire tenement domain, with depth slices at intervals of 25m between depths of 25 to 1000m. The predictions were in ppm. The Au and Ag predictions were kept on this scale because ppm is equivalent to grams per tonne. The Cu and Zn predictions were converted to weight percent before mapping. The Cu predictions are most likely to be reliable based on the validation statistics (Table 2, Fig. 32), with vastly improved Root Mean Square Error (RMSE) compared to the natural variation present in Cu concentrations and correspondingly high explanatory power (~71% goodness of fit). Au and Zn predictions represent slight improvements over a null model for their distribution over the tenement area but should only be taken as indicative at best. Ag concentrations are very poorly predicted, most likely due to small observed concentrations of silver in all but a few drills.

Table 2 Validation statistics for Random Forests per element (Cu, Au, Ag, Zn). RMSE is the root mean squared error. R2 is the coefficient of determination or the proportion of variation in the data explained by the model.

Element Number of training observations The standard deviation of the variable Out-of-bag R2 Out-of-bag RMSE
Copper (Cu) 55,434 5332.566 0.709 2875.270
Gold (Au) 41,196 58.760 0.244 51.089
Silver (Ag) 20,817 7.787 -0.295 8.862
Zinc (Zn) 94,687 347.573 0.186 313.424

The most meaningful predictions appeared to be for Cu, so these were mapped alongside the drill holes, the OZ Minerals tenements, and targets identified with the Prospector AI.

Relationship between the observed

Figure 26 Relationship between the observed concentrations of Cu and the predicted concentrations.

The red line is the line for 1:1 correspondence between the predictions and observations (i.e. the ideal case with perfect predictions)

Geological Structure

Lack of outcrop in the target area makes the possibility of deep structural interpretations problematic, and there is a high likelihood for inaccurate interpretation.

We feel that major deep structures tend to get overlooked in many prospectivity systems.

The importance of high-permeability conduits for the ascent of metal-bearing, deeply derived magmas and fluids cannot be overstated. See Fig. 30

McCuaig & Hronsky (2014) write in “The Mineral System Concept: The Key to Exploration Targeting”:

It is proposed that all mineral systems comprise four critical elements that must combine in nested scales in space and time.

These include whole lithosphere architecture, transient favourable geodynamics, fertility, and preservation of the primary depositional zone. Giant mineral deposits have an association with large, long-lived, deeply penetrating, and steeply dipping structures that commonly juxtapose distinctly different basement domains. These structures are vertically accretive in nature, often having limited or subtle expressions at or above the level of ore deposition.”

Motta et al. 2019 writes in “Proxies for Basement Structure and Its Implications for Mesoproterozoic Metallogenic Provinces in the Gawler Craton”:

“these provinces are part of a Mesoproterozoic mineral system with an extensive hydrothermal alteration footprint, which formed during complicated tectonic mode switches. We show that both types of mineralization are in proximity to crustal-scale structures that appear to connect deep crustal fragments”


There were two surveys that crossed into the OZ Minerals Tenements according to the SARIG and OZ Minerals databases; the 1985 “C.R.A.E.P.E.L. 85” Seismic Survey targeting basement and petroleum plays, and the 2012 “Mount Woods 2D” Seismic Survey targeting structure relating to the Prominent Hill Mine orebody.

The data set used was the five lines shot in the Mount Woods 2D Seismic Survey as the other data set was unavailable, paper only, and covers little of the OZ Minerals Tenements. The seismic data shows large-scale structural features including crustal normal faults and many thrust faults. Several of these were interpreted on sections and generally aligns with the supplied 2008 Betts Regional Geology data, giving gravitas to it away from the seismic lines.

Looking at the magnetic trends and structural boundaries, the seismic line that most represents the Prominent Hill orebody is to the South-East. There seems to be a stand-out seismic facies based on frequency and amplitude differences compared to the surrounding rock. One particular seismic facies is also at the approximate depth of the Prominent Hill orebody, using 4000-5000m/s velocity. As there is no ground-truthing on this line, and the magnetic signature is weaker and more dispersed here, this particular seismic facies couldn’t be used to hunt for similar frequency/signal anomalies. Line 04 (Figure 39) also shows the Bulgunnia Fault Zone. This is seen as two steeply dipping divergent faults throwing in opposite directions, North- East and South-West. This agrees with the dips from gravity and aeromagnetic expressions previously interpreted by Harris et al 2013, but not their seismic fault dip-directions.

This previous structural interpretation based on seismic, magnetic and gravity indicates that Prominent Hill sits over a ‘thin-skinned’ tectonic environment caused by multiple thrusted units. The area is complex structurally with events such as momentous tectonic force reversals detailed by Betts et al 2016, but these same structural elements allow for migration pathways for deep fluids and concentrating structures such as fault brecciation and thrust-fault packages.

Remote Sensing

CSIRO Satellite ASTER Geoscience images for the target area were obtained, and processed with Esri’s ArcGIS Pro.

ASTER is good for mapping hydrothermal alteration minerals and to better discriminate geological structural features. Image transformation techniques such as band rationing and Principal Component Analysis were used to delineate lithological units and alteration minerals.

Images were processed with unsupervised k-means classifier; maximum likelihood classification and supervised object-based classification with random trees was applied to detect differences between indicator alteration minerals associated with known deposit areas.

CSIRO’s ASTER mineral maps “Ferrous Iron Index” and “Ferric Iron Composition” showed some potential alteration at the Prominent Hill orebody in ASTER images taken before ground disturbance. Machine learning image analysis was attempted to classify and use supervised learning to delineate new ASTER based targets; no new targets or validation of our targets was achieved. Further investigation and research is recommended.

A Landsat 1 image from 1972, of the Olympic Dam area (pre disturbance), was retrieved from the Landsat archive. It was processed and interrogated using Esri’s ArcGIS Pro and the AI for any signs of the Olympic Dam orebody. During our literature review for the project, we could find no similar studies and knowing the nature of the Olympic Dam orebody we expected to find nothing, but we carried out the exercise to prove the hypothesis that it wasn’t possible.


The Australian Lithosphere Architecture Magnetotelluric Project (AusLAMP) is an ongoing initiative of the Next Generation Copper Discovery Program. It is a collaboration between several parties, and the Magnetotellurics (MT) [Figure100 of Aus-LAMP] is open data hosted by the South Australian government’s SARIG. The AusLAMP project aimed to cover most of Australia and does provide evidence for Olympic Dam and the Carrapateena mine in the economically attractive Gawler Craton, particularly in the Olympic Fe‐Oxide Copper‐Gold Province. MT provides crustal scale resistivity or conductivity measurements, and this is crucial for providing the magmatic source for our final economic mineral concentrations.

The MT shows highly conductive areas that intersect deep normal faults that terminate below/inside OZ Minerals tenements; setting up fluid pathways from the rich deep magma to more shallow depths. The AusLAMP MT data, Betts 2008 structural map, and the greater resolution of the seismic data provided substantial evidence for some of OreFox’s Hunter AI targets.

The most promising combination of geophysics and structural geology occurred at the “Nick’s Ring” primary target.

All of the Nick’s Ring targets are backed by MT continuous, conductive flares (Figure 40) from very deep (400km) to shallow (~1.2km, the noise/resolution limit) and structural faulting.

Magnetotellurics data was projected in 3D to allow a better spatial understanding of the subsurface architecture (Figure 41a and b

Data available from sarig.sa.gov.au

Data from the unearthed portal

O'Driscoll Lineament Maps of Australia, Geoscience Australia

Claoue-Long, J.C. (Author)

Geoscience Australia

Queensland Department of Natural Resources and Mines (Qld IOCG)

SA_TMI is an unfiltered total magnetic intensity grid of South Australia, with an 80m grid cell size.

SA_TMI_LP800 is a low pass filtered (800m, cut-off rate 1) TMI

SA_TMI_VRTP is a variable reduction to pole (RTP) grid of SA_TMI generated by a Fast Fourier Transform (FFT)

with IGRF field data: 01/01/1995; -30; 135, altitude 150m

SA_TMI_LP800_VRTP is a low pass filtered (800m, cut-off rate 1) SA_TMI_VRTP

SA_TMI_LP800_AS is a low pass filtered (800m, cut-off rate 1) Analytic Signal of SA_TMI. Analytic Signal is the

square root of the sum of the squares of horizontal and vertical gradients. It is effective in mapping the distribution of

shallow magnetisations independent of their magnetisation direction.

SA_TMI_LP800_VRTP_1VDis the 1st Vertical Derivative of low pass filtered (800m, cut-off rate 1) Variable RTP of


SA_TMI_LP800_1VD is the 1st Vertical Derivative of variable RTP of SA_TMI

SA_TMI_VRTP_LP800_2VD is the 2nd Vertical Derivative of Low-pass filtered (800m, cut-off rate 1) Variable RTP


SA_TMI_VRTP_UC1000_Residual is the Upward Continued 1000m Residual of Variable RTP of SA_TMI

SA_TMI_VRTP_PseudoGravity is the Pseudo Gravity of Variable RTP of SA_TMI. The ideal relationship between

gravity and magnetic fields which would exist for an ideal relationship between density and magnetisation allows the

prediction of gravity field variations from magnetic field measurements (Garland 1951, Baranov 1957, Bott and Ingles

1972). This is achieved by suitable transform of the magnetic field data, namely a reduction to pole and integration

known as the pseudogravity transform.

SA_TMI_VRTP_LP800_TILT is the Low pass filtered (800m cutoff, cut-off rate 1) VRTP TMI TILT Angle. The tilt

angle is derived from the ratio of vertical and horizontal gradients transformed to an angle (range -90° to +90°) using

the arc-tangent function (Miller and Singh, 1994). This ratio is independent of the magnitude of the gradients and is

everywhere defined, which means that it is subject to noise across regions of low gradient.


U.S. Geological Survey

Landsat 8







Austin, J., Geuna, S., Clark, D., & Hillan, D. (2014). Remanence, self-demagnetization and their ramifications for magnetic modelling of iron oxide copper-gold deposits: an example from Candelaria, Chile. Journal of Applied Geophysics109, 242-255. https://doi.org/10.1016/j.jappgeo.2014.08.002

Betts, Peter & Armit, Robin & Venn, Caroline & Ailleres, Laurent. (2016). Tectonic switches during the Palaeo-Mesoproterozoic transition: implications for mineral systems.

Boucher, R. K., (1997) Lineament tectonic data, South Australia. With a focus on the Cooper Basin. Department of Mines and Energy South Australia. Report Book 97/39. 76pp.

Freeman, H. and Tomkinson, M., (2011) Geological Setting of Iron Oxide Related Mineralisation in the Southern Mount Woods Domain, South Australia; in Porter, T.M., (ed.), Hydrothermal Iron Oxide Copper-Gold & Related Deposits: A Global Perspective, v. 3 - Advances in the Understanding of IOCG Deposits; PGC Publishing, Adelaide, 171-190.

Harris, Thomas & Funk, Charles & Murphy, Finbarr & Betts, Peter. (2013). Mt Woods 2D Seismic Reflection Survey, Gawler Craton, South Australia: An Integrated Minerals Exploration Case Study. ASEG Extended Abstracts. 2013. 1. 10.1071/ASEG2013ab247.

Hronsky, J. (2013). Understanding Major Trans-Lithospheric Structures, their Evolution and Relationship to Ore Deposits. CCFS Lithosphere Dynamics Workshop Perth, WA

Judd, Deane B.; Wyszecki, Günter (1975). Color in Business, Science and Industry. Wiley Series in Pure and Applied Optics (third ed.). New York: Wiley-Interscience. p. 388. ISBN 978-0-471-45212-6.

McCuaig, T & Hronsky, J. (2014). The mineral system concept: The key to exploration targeting. Society of Economic Geologists Special Publication. 18. 153-175. 10.1080/03717453.2017.1306274.

Motta, João & Betts, Peter & Roberto de Souza Filho, Carlos & Thiel, Stephan & Curtis, Stacey & Armit, Robin. (2019). Proxies for Basement Structure and Its Implications for Mesoproterozoic Metallogenic Provinces in the Gawler Craton. Journal of Geophysical Research: Solid Earth. 10.1029/2018JB016829

Robinson, Larry. (2007). The Spatial and Temporal Distribution of the Metal Mineralisation in Eastern Australia and the Relationship of the Observed Patterns to Giant Ore Deposits.

Skirrow, Roger & Davidson, Garry. (2007). A Special Issue Devoted to Proterozoic Iron Oxide Cu-Au-(U) and Gold Mineral Systems of the Gawler Craton: Preface. Economic Geology. 102. 1373-1375. 10.2113/gsecongeo.102.8.1373.

Thomas M. Harris, Finbarr C. Murphy, Charles W. Funk & Peter G. Betts (2013) Mt Woods 2D Seismic Reflection Survey, Gawler Craton, South Australia: An Integrated Minerals Exploration Case Study, ASEG Extended Abstracts, 2013:1, 1-4, DOI: 10.1071/ ASEG2013ab247