Unsupervised Machine Learning for Determining Exploration Areas of Valuable Elements and Potential Toxicology Elements: A Case Study of the Bowen Basin Coal, Australia

Global coal production and demand have increased anually. In addition to its potential as an alternative source of critical elements, coal also has environmental risks through toxicology elements. Australia is the world’s second-largest producer of rare earth elements (REEs) and critical elements, making coal exploration a key focus of the country’s mining strategy. An unsupervised Machine learning algorithm was applied to 56 coal samples from three pits in Bowen Basin, e.g., Blake Central Pit, Blake West Pit, and Bowen No. 2 Pit, to correlate trace elements with the geochemical characteristics of coal, such as proximate and major oxides. Blake West Pit is enriched in Ba, Br, and Sr, which associated with inherent moisture and phosphor (P), extending SE-trend. Blake Central Pit and Blake West Pit are enriched in Hf, Mo, Ta, Th, Y, and REY, which are associated with ash and major elements such as Si, Al, Ti, and K, with a trend of potential exploration towards N-NW. However, both pits show the risk of contamination from the toxic element Zn, which is associated with volatile matter, and major elements e.g., Fe, Mg, and Mn, with a trend of distribution towards S-SW. Based on the correlation analysis and regional geology, trace element enrichment in Bowen Basin is controlled by two main factors: 1) the transgressive phase during Early-Late Permian, which enriched inherent moisture, P, Ba, Br, and Sr, and 2) volcanic activity during Early Permian, which enriched silicate minerals and elements such as Hf, Ta, Th, W, and REY. Unsupervised machine learning has proven effective for preliminary coal characterization to support further exploration.