b'AEGC 2023Short abstractsThe regional geological model was built from geologicalAs the industry increases their undercover exploration efforts, constraints (mapping, drill holes) but also in close integrationto help support the generation of new targets and ultimately with potential fields data, producing a viable starting model fornew discoveries (near-mine, brownfields and greenfields), new geologically constrained inversion to solve for rock propertytechnologies that integrate with more traditional approaches are variations within the various geological domains. When the modelbeing developed, adapted and/or evaluated.was submitted to geologically constrained inversion to reconcile unexplained response as property variations within thoseOne of these approaches is the use of Machine Learning (ML) in domains, sensible/stable property variations were recovered thatprospectivity analysis, which promises an objective approach could be interpreted in terms of alteration or exploration targets. that minimises human bias and the ability to rapidly interrogate large and disparate data sets. In real world applications, The VHMS exploration criteria were translated using the 3Dhowever, prospectivity analysis is subject to a number of other integrated model to create exploration vectors representativebiases that can be attributed to the data used, the complicated of the mineral system. In other words, these vectors wereand tenuous relationship between these data and the mineral numerical realisations of the various targeting criteria. All thesystem proxies they aim to map and the sometimes equally exploration vectors used in the final mineral prospectivitytenuous relationship between proxies and the mineral system analysis were grouped according to three conventionalprocesses of interest. Ill-informed or a perceived lack of links categories of exploration criteria: 1) Metal Source; 2)between data, proxy and process is a weakness that can be Architecture, effective conduits and circulation and; 3) Metalminimised with a nuanced combination of geoscience and data trapping (deposition). The prospectivity analysis at Jaguarscience.used a Machine Learning approach, namely Random Forests, to generate 3D mineral prospectivity maps based on differentBy combining advanced data processing, knowledge combinations of input exploration vectors. compilation, geological interpretation and ML, multiple approaches have been investigated to support the discovery of large porphyry copper deposits. Outcomes of case studies to Application of Time-of-Flight secondary ion massdate include:spectrometry to the geosciences.A productive collaborative approach allows for powerful William Rickard and Xiao Sun cross-fertilisation of ideas with no one specialist or specialist team driving outcomesCurtin UniversityEnhanced and new datasets, based on geological knowledge, Time-of-Flight Secondary Ion Mass Spectrometry (ToF-SIMS)are very important to outcomes achievedhas renowned capability in soft matter analysis for its abilityML-based Fusion models are an effective means of combining to analyse molecules at high lateral resolution with impressiveknowledge-based models that capture expert knowledge sensitivity. Increasingly the technique has been applied to hardand deep learning models that are data-driven and have been matter samples due to the capability for light and trace elementinstrumental in generating more optimal mineral potential/mapping at high lateral resolution as well as the ability forprospectivity models.depth profiling. However, isobaric interferences and inadequateAn overview of the approach and implications for modern sensitivity to ions of interest have limited many studies in themineral exploration are discussed.geosciences up until now.The latest generation of ToF-SIMS instruments have significantlyDeep learning based geological interpretation using increased the applicability of the technique to geoscience research. Improved mass resolution, higher primary ion beamgeophysics.currents, the availability of Cs/O sputter beams and a widerMichael Roddavariety of analytical modes have driven the extended capability.In this work we present case studies from a variety ofFor Round 5 of the Collaborative Exploration Initiative (CEI), geoscience applications using the Iontof M6 ToF-SIMS in theCaldera Analytics was engaged by Strategic Energy Resources John de Laeter Centre at Curtin University. (SER) to develop a deep learning model that performed lithological interpretation of basement geology using geophysics. SER has two main projects in the Mt Isa region that A combined approach to improve prospectivityare both located under significant amounts of younger cover, analysisintegrating geological knowledge andmaking geological interpretation a difficult task.machine learning. The machine learning model, which used gravity and magnetics Mark Rieuwers 1 Ben Jupp 1, Chris Woodfull1, Alex Tunnadine1,as an input, was trained to predict six key lithology groups, Peter Stuart-Smith1, Antoine Cat2, Viswanath Avasarala3, Timone of which was hydrothermal magnetite, a key vector for McMahon3 and Rebekah Roche1 iron-oxide-copper-gold (IOCG) deposits. The six groups were chosen based on a combination of the business value to SER, 1 SRK Consulting (Australasia) Pty Ltdgeophysical constraints and the resources available for data 2 SRK Consulting (Canada) Incquality assurance. The training dataset for the models was 3 DeepIQ created from historical drillholes and mapped solid geology in the exposed areas of the Mt Isa inlier.The challenges facing the mining industry are varied not the least being trends toward undercover and deeper explorationIn recent years self-supervised learning (SSL) has emerged as the rate of viable near-surface deposit discoveries declines,as a key research field in deep learning, where the goal is to especially in well-established and -endowed mining jurisdictions.boost performance of a model using unlabelled data. SSL 135 PREVIEW FEBRUARY 2023'