Quantifying porphyry copper deposit fertility from zircon geochemistry using predictive modelling: from theory to applications

Abstract

The demand for copper in the next 20 years is expected to be equal to its total historic production due to copper's fundamental role in the transition to green energy. Porphyry copper deposits are the source of 75% of the world's copper but the rate of discovery of new deposits has declined over the last decade. Zircon is an accessory mineral that is resistant to chemical weathering and physical abrasion. It is common in siliceous igneous rocks, including those associated with porphyry copper deposits. Zircon's chemical composition can provide information about the magma it crystallizes from and potentially be used to distinguish ore-forming porphyry magmas from barren magma systems. This thesis covers three projects aimed at understanding how zircon geochemistry can be used in porphyry Cu exploration. The first presents an empirical method, based on Onuma diagrams and the lattice strain theory, to calculate missing rare earth elements (REE) and Y data in legacy datasets in which some elements were not analysed. The results have been calibrated against known concentrations, and lattice strain theory estimates, using a dataset of ~1,500 zircons with no missing REE + Y concentrations. The results show that REEs can be confidently imputed with as few as five REEs and allow the estimation of La, Pr and Ce* in zircons with greater confidence than traditional methods. These methods enable the reconciliation of old and new REE + Y data. The second project is a data-driven analysis of zircon geochemistry that compares those that formed in magmas associated with porphyry copper deposits with those that formed in barren magmatism. This study compiles a database of more than 23,000 zircons from > 30 porphyry copper deposits, barren sources and rivers and uses machine learning for classification and comparison with traditional geochemical thresholds. The study confirms some of the zircon fertility indicators proposed in the literature (Eu/Eu*, Dy/Yb, Ce/Nd), discard those that are not diagnostic (Hf, Th/U), and proposes two new fertility indicators (P and shape of the REE pattern). It also shows that random forest models outperform traditional geochemical methods by correctly identifying up to 20% more fertile zircons. Shallow machine learning algorithms outperform the traditional geochemical discriminators and provide insights into characteristics that have not previously been considered when evaluating porphyry copper fertility using zircon geochemistry. The last project is a test case of zircon fertility classification using random forest in the world's most prolific porphyry copper region: the Loa River drainage basin, Northern Chile, that includes the Chuquicamata, El Abra, Collahuasi, Spence and Sierra Gorda porphyry districts. Nearly 1,600 zircon grains, collected along the Loa River basin, were analysed for trace elements, geochronology, and their class membership probability (fertile or barren) calculated using predictive models. The results indicate that periods with higher zircon frequency coincide with the ages of the magmatic arcs in the area. Furthermore, the age of the deposits matches the end of these periods and is accompanied by an increase in the number of zircons classified as fertile. There is also an increase in the fraction of fertile zircons in samples collected close to deposits, which reach their maximum near the El Abra deposit. Show more