New Approach to Recoverable Resource Modelling: The Multivariate Case at Olympic Dam

Abstract

Traditional estimation techniques significantly under-call the true monetary value of the resource on which mine plans and operations base their business. At the Olympic Dam, this is worth billions of dollars. Realising this value requires mine planning engineers to be supplied with an accurate recoverable resource model that correctly estimates the tonnes and grade for a specified support and timescale, at the time of mining.

Models estimated using linear methods and wide-spaced drilling typically fail to accurately predict recoverable resources, mainly because of incorrectly accounting for the change of support and information effect. The unavoidable smoothing property of weighted averages is also a significant obstacle. These failures are more significant in underground mining scenarios where higher cut-offs (with respect to the average grades of mineralisation) are applied. This paper discusses a different approach to recoverable resource estimation based on conditional simulation methods.

The Olympic Dam deposit is one the world’s largest polymetallic deposits. The resource estimation practices at the Olympic Dam are comprised of a combination of linear and non-linear techniques to estimate 16 different grade variables critical to mine planning. Measured resources are supported by 20 m-spaced underground drilling fans where Kriged estimates perform well in terms of mine to mill reconciliation. However, this not the case for resources classified as Indicated and Inferred. Until infill drilling is undertaken, the accurate estimation of tonnes and grade to the mill is not possible with the Kriged model. This has a significant impact on life-of-mine economic valuations and ore reserve estimates of the Olympic Dam.

Conditional simulation has been used to generate a recoverable resource estimate from a single realisation. This conditional simulation model takes into account both the change of support and the information effect, without the undesired smoothing effect that classic methods introduce. This paper describes the significant challenges faced in applying this approach, including issues such as which realisation to choose, data conditioning in areas with little information, ensuring that the multivariate relationships among variables are respected at a block level, software and hardware challenges and defining benchmarks for ensuring that the “correct” grade-tonnage curves are reproduced. These challenges have to be overcome whilst ensuring that the resulting estimate is a JORC compliant and is also acceptable under BHP Billiton’s corporate governance standards.