Skip to main content

Pedras: Modal mineralogy approximations from assay data using Bayesian inference

Mineralogy evaluation is critical to understand a deposit’s mineralogical variability, and inform decisions associated with ore beneficiation. Elemental-to-mineral conversion techniques (EMC) are a popular method to rapidly estimate a sample’s modal mineralogy from an assay dataset. EMC techniques are based in the principle that the bulk chemistry of a sample is proportional to the product of its modal mineralogy and the mineral’s elemental composition.

In this work, we present a Baeysian framework to infer modal mineralogy from compositional data, Pedras. It builds upon the works of Escolme et al. (2019) and Berry et al. (2011), an EMC method that uses linear programming to minimize coefficients representing the energy required to generate a given mineral assemblage. The minimization process implies thermodynamic equilibrium, which is rarely the case for hydrothermal environments. Instead, the thermodynamic coefficients are defined as a logistic probability distribution function, centred at the mineral assemblage’s equilibrium, which relaxes the thermodynamic coefficient’s minimization.

The framework is tested on synthetic alteration assemblages within a porphyry copper deposit and applied to a geochemical dataset from the Wainaulo porphyry copper deposit (Fiji). The results show that by relaxing thermodynamic minimization constraint, accurate modal mineralogy can be approximated at different stages of hydrothermal alteration in porphyry copper deposit systems. Major mineralogical domains are obtained from mineral approximations, reflecting different lithotypes, alteration and mineralization patterns. The modal mineralogy outputs also provide invaluable insights of mineral associations that vector towards mineralization, The method’s accuracy is enhanced when prior knowledge is objectively included in the modelling stage.


Angela Afonso Rodrigues1, Lachlan Grose1, Laurent Ailleres1, Scott Halley2, Angela Escolme3, Robin Armit1, Mehrtash Harandi1, Matthew Cracknell3
1Monash University, Australia; 2Mineral Mapping Pty; 3University of Tasmania
GeoBerlin 2023