Bruno Afonseca1, Alexander Chojno2, Richard Hague3
- Resource Geology Manager – Vale Base Metals – bruno.afonseca@vale.com
- Senior Specialist Geologist – Vale Base Metals - alexander.chojno@vale.com
- Specialist Geologist – Vale Base Metals - richard.hague@vale.com
Keywords: Geometallurgy, conditional simulation, geostatisics, model uncertainty
ABSTRACT
Geometallurgical variables are some of the most challenging attributes to model spatially in mining applications. Unlike grade variables, metallurgical responses usually rely on a complex combination of geochemical, physical properties, mineralogical factors, often linked through nonlinear relationships and supported in most cases by sparse testwork data. In practice, this makes metallurgical uncertainty much harder to quantify accurately than variability from pure statistical fluctuation of grades feeding the models. A common limitation in many applications is that uncertainty is propagated only through the input variables. When deterministic recovery equations are applied to simulated geological or geochemical realizations, the transfer function is effectively assumed to be error-free. This means that the resulting uncertainty reflects variability in the predictors, but not the uncertainty associated with model fit, missing variables, experimental error, and simplification in the response function. In practical terms, this can produce uncertainty ranges that are too narrow and give a misleading sense of risk in the outcomes. This case study proposes a practical workflow to address the two sources of uncertainty. The first source comes from the well-known geochemical drivers of metallurgical prediction and is handled by conventional geostatistical simulation on a multivariate framework. The second source from the imperfect recovery regression models is incorporated by adding a stochastic residual term in the model prediction. In this way, the workflow aims to capture not only variability in the inputs but also uncertainty in the response model itself. The methodology is being applied at the Onto Porphyry gold-copper deposit on Subawa Island, Indonesia. The expected outcome is a more realistic probability representation of metallurgical response at the local scale, with direct application to risk-based planning with a better understanding of where recovery uncertainty is most influential.