The large Yandi channel iron deposit (CID) is located in the north-west of Western Australia and is currently mined via open pit methods. Drilling at Yandi is widely spaced and current estimates are mostly linear averages of the surrounding local samples. Linear estimates such as ordinary kriging (OK) based on block sizes that are significantly smaller than the dimensions of the drilling grid are less variable than the unknown reality (Armstrong and Champigny, 1989).Conditional simulation (CS) provides a series of plausible realisations that represent the actual grade variability at a small scale (eg 1 m_x000D_
1 m_x000D_
1 m). These simulations are very useful for a number of purposes, eg: optimising grade control, stockpiling and blending selection of optimal drill hole spacing assessing resource risk on contained tonnage and metal._x000D_
At Yandi the major constituents of the ore (Fe, P2O5, SiO2, Al2O3 and LOI) are correlated to varying degrees. Independent CS of each assay variable will not reproduce these correlations. Reproducing the correlations (and the shape of scatterplots) is important if the simulations are used for the abovementioned purposes. This paper examines and compares various methods of CS that aim to reproduce the correlations between the assay variables.The CS methods evaluated in this paper are conditional co-simulation (CCS) using the linear model of co-regionalisation and two other, alternative multivariate simulation methods, which use min/max autocorrelation factors (MAF) or the stepwise conditional transform (SCT) to de-correlate assay variables before performing independent CS. It was found that CCS best reproduced the characteristics (histograms, variogram, scatterplots, etc) of the declustered drill hole data (Figure 1). Whilst simulation using MAF generally produced acceptable results, the method did not adequately reproduce the variance of the assay variables. The univariate statistics, histograms and experimental variograms were acceptable for simulations generated from the SCT. Bivariate statistics, cross-variograms and scatterplot reproduction for the SCT were unacceptable. The latter method produced inferior results because there were insufficient samples to obtain a reliable transform of the third assay variable in the transformation sequence.None of the abovementioned simulation methods provides good reproduction of all the scatterplots. This is due to the presence of clay pods, which cause some scatterplots to display two trends. These clay pods contain higher percentages of SiO2 and Al2O3 than the surrounding CID. The clay pods are generally smaller than the drill-grid dimensions. Thus their locations are typically unknown and they cannot be defined by manual interpretation, eg via wireframes. The locations of the clay pods were simulated via indicator CS. CCS was then performed separately for the clay pods and the CID. This significantly improved the reproduction of the correlation coefficients, the scatterplots and visually the simulation was more geologically realistic. Simulating the location of clay pods followed by CCS of assay variables is the recommended approach for simulation at the Yandi CID.CITATION:De-Vitry, C, 2013. Simulation of correlated assay variables - a case study from the Yandi Channel iron deposit, in Proceedings Iron Ore 2013, pp 137-148 (The Australasian Institute of Mining and Metallurgy: Melbourne).