Multivariate block simulations of a lateritic nickel deposit and post-processing of a representative subset
J Deraisme, O Bertoli, P Epinoux

Societe le Nickel (SLN) exploits the Dome lateritic nickel orebody at its Tiébaghi operations in New Caledonia. The site geology has an obvious bearing on recovery and metallurgical performance, and the controls are extremely complex at all scales: lithology, oxidation-reduction conditions, mineralogy, and multivariate geochemistry. In that context, establishing an adapted recoverable resource estimation method that is efficient and transparent proves an interesting challenge. The method must address the key notions of:
* Support effect: the exploration data-set can only warrant the estimation of large panels 20×20×3 m3, which are much larger than the selective mining units (SMUs) of 5×5×3 m3
* Information effect: the selection at production stage uses estimates based on information much denser than that available for mine planning.
The problem is rendered more complex in this type of deposit by the fact that the additive variables used for any estimation process (metal accumulation and ore tonnage) are not the ones used to establish the selection at mining stage (i.e. the nickel grade of SMUs, which is the ratio of accumulation on tonnage). Eventually, the selection criteria involve not only Ni grades but also other constituents such as Al2O3, Fe2O3, MgO, and SiO2.
The solution presented in this paper is the construction of a platform of SMU multivariate simulations in the saprolitic horizon, together with efficient post-processing aimed at producing the multivariate recoverable resource estimates. The efficiency constraints imposed by the number of blocks to be simulated (1.5 million) lead to resorting to direct block simulations. The underlying multivariate discrete Gaussian model (DGM), the validity of which is tested beforehand, is put to good use to mimic the selection process at the mining stage, by offering the ability to simulate in each block a composite value located at random within the block.
Finally, the paper presents the application of a scenario reduction algorithm to pick a representative subset of a few simulations to help appraise the risk attached to the downstream phases (reserve optimization, mine sequencing) of the project. The implementation presented here is for the first generation of the Scenario Reduction plugin built in Isatis software, where measuring the distance between the initial set of scenarios and the reduced set is based on Ni quantities only. Research is under way to adapt the measure to the true multivariate nature of the problem.
Keywords: direct block simulations, discrete Gaussian model, multivariate, scenario reduction.