Modelling gold potential in the Granites-Tanami Orogen, NT, Australia: A comparative study using continuous and data-driven techniques

Abstract

The Granites-Tanami Orogen (GTO), a poorly exposed component of the North Australian Craton, is host to significant Paleoproterozoic orogenic gold mineralization amounting to a total endowment of greater 20 Moz Au. This study adopted a multi-technique approach to mineral potential modelling (MPM) of orogenic gold deposits in the GTO, using four different techniques: Continuous fuzzy gamma, geometric average and data-driven index overlay and data-driven random forest methods. The predictive models generated in this study delivered the first public-domain gold prospectivity maps for the Northern Territory (NT)-portion of the GTO. The predictor maps at the core of these models represent mappable proxies of mineralizing processes believed to have been critical in the formation of the orogenic gold deposits, and were generated in the framework of a mineral systems approach. Normalized density, prediction-area and area under the receiver operating characteristic curve methods were used to test the robustness of these predictors and determine the most effective. The performance of the gold potential models generated in this study was assessed using the improved prediction-area plot procedure, with the data-driven random forest approach having delivered the best performing model. The multi-technique approach taken in this study, in which multiple mineral potential models are generated for the same study area and using the same predictor maps, is a relatively new and, as of yet, underutilized approach to MPM that, in contrast to the standard procedure (i.e., generation of a single data- or knowledge-driven model) ensures optimal utilization of the available conceptual and empirical information, helps to better understand and minimize stochastic and systemic uncertainties, facilitates cross-validation of the resulting prospectivity models, and allows the resulting models to be compared and contrasted. From a practical, exploration point of view, the models served as decision-support tools that helped to improve confidence in targeting underexplored, gold-prospective areas in the GTO more effectively and identify new search areas previously not recognized as areas of high gold potential.