Publications

Machine learning and 3D modeling techniques were applied to a high-resolution magneticradiometric dataset collected over the Fletchers Awl copper-gold project in Central Queensland to map the 2D and 3D geology of the project area. Two supervised machine learning methods (Random Forests, Support Vector Machines) were used, both requiring training data to classify the input data into lithologies. After tuning the input data, training points, and model parameters, these methods produced maps that closely resembled the government geology map. An unsupervised method (ISO Cluster Classification) was also tested, which segments the input maps into classes without the use of training data. This produced a reasonable approximation of the mapped geology, but distinct units were grouped together. We found that the best use for the machine learning outputs was as guides for the more traditional geological interpretation of the geophysics data that was undertaken. The interpretation was extended into 3D using wireframing and implicit modeling. A magnetic inversion, drilling data, and gravity maps were also incorporated to guide the 3D mapping. The resulting 2D and 3D maps are great improvements on the existing mapping and have improved the understanding of the mineral system model for the area, allowing more effective exploration targeting.

Finding new metal deposits has become more difficult due to exploration maturity and information and data overloads. This means that traditional subjective exploration targeting is less effective. New computer based exploration targeting techniques, including machine learning, should be used more often by the Exploration Industry, to address the issues of data overload. However, the Exploration Industry rarely uses the new targeting techniques in real world exploration. This appears to be due to a lack of trust in the results from these systems and a lack of understanding of how the results from mineral potential modeling can be used to help support decision making in exploration and mine development. Mineral potential modeling was used as a decision support tool in the acquisition and development of the Greenfields Bundarra copper, silver and gold porphyry system in Central Queensland and also to help constrain resource estimation at the Tampia Gold mine in Western Australia. These case studies are examples of how mineral potential modeling can be used at either end of the exploration and mining value chain and provide ideas on how mineral potential modeling can be integrated into exploration and mining decision support systems from Greenfields exploration targeting through resource development to mining.

With the increasing use of machine learning for big data analytics, several methods have been implemented for the purpose of exploration targeting using mineral potential mapping in a GIS environment. Random forests (RF) have been successfully applied to data-driven mineral potential mapping using relatively small numbers of input maps that have typically been pre-classified by a geologist familiar with the mineral system being targeted. However, it is useful to understand how well RF perform for mineral potential mapping when a large number of multi-class categorical or non-thresholded numeric input maps are used in the classification or when weighted or ranked training data are used. Four different implementations of RF are presented to examine how the results vary depending on the degree of intervention from an expert in the modeling process. A case study has been devised using data from the eastern Lachlan Orogen in New South Wales (Australia) for the purposes of targeting porphyry Cu–Au mineralization related to the Macquarie Arc. The results demonstrate that the use of a large number of multi-class categorical or non-thresholded numeric predictive input maps results in a poor mineral potential map outcome. An expert review to determine reclassifications or thresholds that produce geologically meaningful maps as proxies for the mineral system being targeted results in more effective RF-based mineral potential maps being produced. Weighting or ranking the deposits used as training data produces more narrowly defined prospective areas that may assist with targeting tier-one economic deposits. Comparison of the RF results to a standard weights of evidence analysis highlighted some significant differences in which predictive maps should be considered important for modeling, and in the extent of prospective area delineated from each output mineral potential map.

Resource blocks estimated within a particular domain should only be informed by sample points from within that domain. If this fundamental principle of mineral resource estimation is not adhered to it may severely compromise the quality of the final resource estimate. Nonetheless, the repeated reminder of resource downgrades or even complete project devaluations as the consequence of poor domaining practice suggests that this principle is still not well-entrenched in industry practice. As many practitioners have warned and as documented in books, course materials and online blogs over the years: geology is important. It is good practice to base domains primarily on geological information derived from geological logging of drill core or chips, underpinned by an understanding of the structural geometry of the ore body and a well-understood genetic model of mineralisation. However, a sound statistical analysis of geochemical data that are accurate and precise is also required to evaluate the domains. Multi-element geochemistry from laboratory or portable XRF instruments combined with multivariate data analysis and machine-learning (ML), as well as core scanners and down-hole optical televiewers, and other recent technological advances are powerful tools that enable the proper delineation of domains. However, these tools are often underutilised as they require additional investment at a time in a project when their value can be hard to demonstrate. Unfortunately, the use of geological information to create domains is the exception rather than the rule. Our review of mineral resource estimates published since 2017 suggests that more than half of all estimation-constraining wireframes are built using grade cut-offs and are not informed by any primary geological information such as lithology or alteration. While using a grade cut-off may seem perfectly logical to delineate domains when detailed geological information is not available, if not treated with caution, this can lead to poor domain integrity. Books and courses on resource estimation clearly express the importance of domaining but offer few practical solutions or rules of thumb. There appears to be a lack of clear standards, a lack of a framework to distinguish good from bad domaining practice and there is a perpetuation of bad practice masked as common practice. Here we offer some recommendations to raise domaining standards across the industry and present a rules-based approach to improve domaining practices at the individual level.

An understanding of the modelled mineral system and high-quality data that accurately map this system are prerequisites for producing geologically meaningful mineral potential maps. Critical to this is the translation of the targeted mineral system components into mappable targeting criteria and their spatial proxies. This paper presents a workflow that illustrates this translation process, also highlighting that, if done well, mineral potential mapping can produce statistically valid, geologically meaningful, and practically useful results that not only predict the location of known mineralisation but also identify new target areas. An important ingredient of the workflow described herein is what we call a mineral systems atlas. This compendium includes (1) a spatial data table detailing the translation of the modelled mineral system, (2) the predictive maps that capture the mappable components of the targeted mineral system, and (3) the final mineral potential maps. A case study implementing and illustrating this workflow is presented for intrusion-related Au and Sn +/- W mineral systems in the Southern New England Orogen, New South Wales, Australia. Importantly, the mineral potential maps generated as part of this study succeeded in identifying areas of known intrusion-related Au and Sn +/- W mineralisation and new areas with high potential for discovery.

A 3D mineral potential model was developed for the Tampia Gold Project in Western Australia to help constrain resource estimation, understand the distribution of gold grades from the resource estimation techniques with respect to geological and physiochemical continuity, and predict the location of new gold mineralisation for future exploration drilling to expand the gold resource at Tampia. The 3D mineral potential model was generated using predictive maps based on a local granulite-facies orogenic gold mineral system model. These were generated from regional scale data and data collected during a 40 m by 40 m resource drilling programme, and included lithology, structure, rock property data and geochemical data. The predictive capacity of each map was tested for the spatial correlation with training data from high grade gold drill intersections, using the weights of evidence technique. There were 44 predictive maps created that can be used as proxies to map the physical and chemical processes active in the orogenic mineral system at Tampia. Of these, 11 were chosen for the final model that had the highest spatial correlation with the training data and did not duplicate map patterns. A closely spaced infill drilling programme was subsequently undertaken over an area where the post probability results indicated high and continuous probability for gold mineralisation, while the resource model estimated less continuous and lower grade gold mineralisation. This infill drilling aimed to compare the gold continuity at a 10 m by 10 m drill spacing with the resource estimate gold grades and post probability distribution developed from 40 m by 40 m spaced resource drilling. The results from the 10 m by 10 m spaced drilling were thereby used to test the performance of both the resource and prospectivity models, and assess the utility of mineral potential modelling for use in developing geological domains to constrain resource estimation. Results from the first phase of infill drilling, which only covers 4% of the total model area, confirm the continuity of the post probability values and suggests that the mineral potential model predicts the location and distribution of gold mineralisation within the area drilled. The results were also better and more continuous than predicted by the resource estimate. Importantly, these results confirm that geological and physiochemical controls on gold mineralisation can be numerically measured and mapped at the scale of an orebody. This allows mineral potential modelling to be considered as an option to constrain and help inform the results of geostatistical techniques used in resource estimation.