Multivariate Geostatistical Simulation of the Gole Gohar Iron Ore Deposit, Iran

"The quantification of mineral resources and evaluation of process performance in mining operations at Gole Gohar iron ore deposit requires a precise model of the spatial variability of three variables (Fe, P, and S) which must be determined. According to statistical analysis there are complex multivariate relationships between these variables such as stoichiometric constraints, nonlinearity, and heteroskedasticity. Due to the impact of these complexities in decision-making, they should be reproduced in geostatistical models. First of all, in order to maintain the compositional and stoichiometric constraints, additive log-ratio (alr) transformation has been applied. In the next step cosimulation, using stepwise conditional transformation (SCT) and sequential Gaussian simulation (SGS) has been used to simulate multivariate data. Through statistical and geostatistical validations it is shown that the algorithms were able to reproduce complex relationships between variables, both locally and globally. IntroductionThe key steps in mining projects are the quantification of mineral resources, definition of mining reserves, and production scheduling. They rely on the construction of a block model that is used to represent basically the spatial distribution of ore grades (Montoya et al., 2012). The determination of grades and tonnages affects risk assessment and economic evaluation of mining projects. Evaluation of process performance in mining operations requires geostatistical modelling of many related variables (Barnett and Deutsch, 2012). Iron ore quality is characterized by multiple variables: not only the iron grade but also the contaminants that interfere in the subsequent steel manufacturing processes. Consequently, the spatial variability of multiple variables must be determined. Key variables are frequently correlated, and such correlations must be honoured during estimation and simulation. Data from iron ore deposits constitutes compositional data; furthermore, relationships the between assay data are often heteroskedastic."