Mineral Prospectivity as an Inference Question Towards the Unification of Knowledge and Data Driven Approaches

<div>Mineral prospectivity studies seek to map evidence of mineral system activity, with the aim of informing mineral exploration decisions and guiding exploration in the face of uncertainty. These studies leverage the growing volumes of information that are available to characterise the lithosphere by compiling covariate (or feature) grids that represent key mineral system ingredients. Previous studies have been categorised as either “knowledge-driven” or “data-driven” approaches depending on whether these grids are integrated via expert elicitation or by the empirical relationship to known mineralisation, respectively. However, to our knowledge, the underlying modelling framework and assumptions have not been systematically reviewed to understand how choices in the approach to the problem influence modelling outcomes. Here we show the broad mathematical equivalence in these approaches and highlight the limitations inherent when optimising to minimise misfit in potentially under-determined problems. We argue that advances in mineral prospectivity are more likely to be driven by careful consideration of the model selection problem. Focusing effort on model selection will not only drive more robust mineral prospectivity predictions but may also simultaneously refine our understanding of key mineral system processes. To build on these results, we present the Mineral Potential Toolkit; a software repository to facilitate feature engineering, statistical appraisal, and quantitative prospectivity modelling. The toolkit enables a novel approach that combines the best aspects of previous methods. Abstract presented to the 26th World Mining Congress 2023 (https://wmc2023.org/)