Effect of high grades on the geostatistical estimation of gold deposits

The geostatistical estimation of the reserves of an ore deposit is highly dependent on the quality and stability of the variograms of the sample grades. Gold deposits commonly show a skewed grade distribution. The few high grades at the tail end of the distribution produce unstable variograms that are difficult to model. If these grades are cut to a lower grade value or removed, more stable variograms are obtained. However, outlier grades may make a deposit economically viable, and their cutting or removal should be treated with caution. To reduce (but not to eliminate) the effect of high grades, a transformation of the raw data can be performed. Most commonly used transformations are: logarithm, gaussian anamorphosis, rank and indicator. Variograms of the transformed grades are resistant to high grades and present a clear structure. The weak point of data transformation appears when grades are estimated by kriging; the estimates are very sensitive to slight variations in the parameters of the back-transformation. More robust variogram estimation procedures can also be used, but often these procedures are not suitable for skewed grade distributions as they are based on an assumption of a contaminated normal distribution. The Porgera gold deposit in Papua New Guinea has a highly skewed grade distribution, with grades ranging from 0.01 to 3500 g/t and a coefficient of variation greater than 5. For the geostatistical estimation of the high grade zone of the deposit, the influence of the high grades was quantified by calculating and comparing the following variograms: raw grades, raw grades with the highest grades removed and transformed grades. The study confirmed that there is no general method for handling high grades in a geostatistical reserve study. Several methods should be tried to quantify the influence of high grades.