Geometallurgical Responses on Lithological Domains Modelled by a Hybrid Domaining Framework

Abstract

Identifying mineralization zones is a critical component of quantifying the distribution of target minerals using well-established mineral resource estimation techniques. Domains are used to define these zones and can be modelled using techniques such as manual interpretation, implicit modelling, and advanced geostatistical methods. In practise, domaining is commonly a manual exercise that is labour-intensive and prone to subjective judgement errors, resulting in a largely deterministic output that ignores the significant uncertainty associated with manual domain interpretation and boundary definitions. Addressing these issues requires an objective framework that can automatically define mineral domains and quantify the associated uncertainty. This paper presents a comparative study of PluriGaussian Simulation (PGS) and a Hybrid Domaining Framework (HDF) based on simulated assay grades and XGBoost, a machine-learning classification technique trained on lithological properties. The two domaining approaches are assessed on the basis of the domain boundaries produced using data from an Iron Oxide Copper Gold deposit. The results show that the proposed HDF domaining framework can quantify the uncertainty of domain boundaries and accommodate complex multiclass problems with imbalanced features. Geometallurgical models of the Net Smelter Return and grinding time are used to demonstrate the effectiveness of HDF. In addition, a preprocessing step involving a noise filtering method is used to improve the performance of the ML classification, especially in cases where domain boundaries are difficult to predict due to the similarity in geological characteristics and the inherent noise in the data.

Keywords

• domain modelling for resource estimation
• geological uncertainty
• geostatistical simulation
• machine learning
• classification
• noise filtering