Constraints and quantifying uncertainty on resource domain boundaries

Abstract

Identifying areas of mineralisation is a critical step of mineral resource estimation in the mining industry. Mineralisation zones are defined by domains used to determine the distribution of target minerals via established mineral resource estimation procedures. These domains can be modelled using manual interpretation, implicit modelling, and advanced geostatistical approaches. Manual domaining is the most widely used method in mining applications, however it is labour-intensive and prone to subjective judgement errors. Also, this method produces a largely deterministic output that overlooks the significant uncertainty associated with domain interpretation and boundary definitions. To address these issues, an objective framework that can automatically define mineral domains and quantify the associated uncertainty is required. This thesis presents the development of a Hybrid Domaining Framework (HDF) based on simulated quantitative geochemical variables and XGBoost, a machine learning classification technique trained on quantitative geochemical and qualitative geological variables and properties, to predict the domains of interest. Additionally, a noise filtering method is introduced as a preprocessing step to enhance XGBoost performance, particularly in cases where domain boundaries are difficult to predict owing to similar geological characteristics. Data from An Iron Oxide Copper Gold (IOCG) deposit was used as a case study to demonstrate the application of the developed method. The results demonstrate that the HDF can accurately quantify the uncertainty of domain boundaries and handle complex multiclass problems with imbalanced features. Furthermore, geometallurgical models of the Net Smelter Return and grinding time illustrate the effectiveness of the framework. A comparative study between HDF and other domaining approaches such as Indicator Kriging, Sequential Indicator Simulation, and PluriGaussian Simulation is presented in the context of acquiring new data. The performance of these approaches was evaluated using assays and geological data obtained from exploration and grade control drillholes. The results show that the HDF framework can produce results comparable to other state-of-the-art methods but incorporate quantitative covariates into the modelling process resulting in higher certainty in resource domain boundaries.