The locations of minerals and mineral-forming environments, despite being of great scientific importance and economic interest, are often difficult to predict due to the complex nature of natural systems. In this work, we embrace the complexity and inherent "messiness" of our planet's intertwined geological, chemical, and biological systems by employing machine learning to characterize patterns embedded in the multidimensionality of mineral occurrence and associations. These patterns are a product of, and therefore offer insight into, the Earth's dynamic evolutionary history. Mineral association analysis quantifies high-dimensional multicorrelations in mineral localities across the globe, enabling the identification of previously unknown mineral occurrences, as well as mineral assemblages and their associated paragenetic modes. In this study, we have predicted (i) the previously unknown mineral inventory of the Mars analogue site, Tecopa Basin, (ii) new locations of uranium minerals, particularly those important to understanding the oxidation-hydration history of uraninite, (iii) new deposits of critical minerals, specifically rare earth element (REE)- and Li-bearing phases, and (iv) changes in mineralization and mineral associations through deep time, including a discussion of possible biases in mineralogical data and sampling; furthermore, we have (v) tested and confirmed several of these mineral occurrence predictions in nature, thereby providing ground truth of the predictive method. Mineral association analysis is a predictive method that will enhance our understanding of mineralization and mineralizing environments on Earth, across our solar system, and through deep time.
Keywords: association analysis; association rules; machine learning; mineral deposits; mineralogy.
© The Author(s) 2023. Published by Oxford University Press on behalf of National Academy of Sciences.