Geostatistical conditional simulation is a method to assess variability and risk in mineral deposits and the tool has wide potential applications in the iron ore industry. Correct reproduction of multivariate relationships is important in iron ore simulation, especially between Fe and impurities such as Al_2O_3, SiO_2 and P. The turning bands simulation method, using a full model of coregionalisation for multiple attributes, is the main multivariate conditional simulation algorithm used in the Western Australian iron ore industry. This paper discusses the results of a more recent approach using minimum/maximum autocorrelation factors (MAF) to transform and decorrelate the multivariate data prior to independent sequential Gaussian simulation (SGS). MAF-SGS results are compared to those of the turning bands approach in the Yandicoogina channel iron ore deposit (CID), and both methods performed well in simulating Fe, SiO_2, Al_2O_3 and P distributions. Extensive checking of simulations showed that both approaches could reproduce multivariate statistics and spatial continuity of the original conditioning data. Both approaches reproduced the histograms and variography of input sample composites. While the MAF-SGS approach requires additional transformations when compared with the single normal scores transformation required for the turning bands method, MAF requires only that the direct semivariograms be modelled. In contrast, using turning bands with multiple attributes requires modelling a full linear model of coregionalisation, which can be difficult to model because of the need to ensure the model is positive semi-definite. Turning bands may be preferred if there are a modest number of correlated variables allowing construction of a linear model of coregionalisation that adequately models the multivariate behaviour. MAF-SGS is preferable for larger numbers of correlated variables. Both MAF-SGS and turning bands methods performed well in conditional simulation of correlated variables at Yandicoogina.
展开▼
