Limiting the Influence of Extreme Grades in Ordinary Kriged Estimates

"The management of outlier grades in positively skewed gold distributions is a contentious issue. Incorporating outliers in standard ordinary kriging (OK) estimation procedures in a way that honours the data without smearing extreme grades into surrounding areas has been problematic. Cutting or capping of outliers to mitigate their influence in estimation techniques is common practice, while methods that manipulate the OK system of equations fail to honour the data. We propose a method of post-processing of kriging weights that provides realistic OK estimates and mitigates smearing without manipulating kriging equations or changing the original grades. The method requires that the data is not clustered, is approximately equally spaced, and is of the same support. Positively skewed data is ordered on attribute grade and nonlinearly transformed to a Gaussian histogram of categorical bins whose frequency is based on their likelihood of occurrence and location in the sample distributions. Factors that restrict kriging weights are calculated by dividing the percentage frequency of data in each bin by the percentage frequency of data in the bin with the highest frequency. Restriction factors applied to the kriging weights in the OK estimation restrict the range of influence in proportion to their probability of occurrence in the distribution. Smear reduction post-processing is easy to implement and addresses issues arising from negative kriging weights while considering the spatial location of samples, the sample grades, and their probability of occurrence. The method mitigates both smoothing and conditional bias.IntroductionOutliers are sample observations that deviate considerably from the standard or expected (Hawkins, 1980). These deviations can be high or low in grade depending on the distribution. The presence of outliers in a mine sampling campaign could provoke mixed reactions. Mining engineers might view high-grade outliers as a promise of achieving higher grades, while the geostatisticians view outliers as affecting the accuracy and precision of the overall estimate, among other issues. Having identified outliers, the geostatistician must decide how to account for their presence and treat them so that true underlying estimates are not distorted. Advanced knowledge of the deposit allows the geostatistician to identify and treat outliers acceptably, especially where samples are limited in number and the true grade distribution is uncertain. If ignored, or treated incorrectly, outliers are likely to lead to smearing of extreme grades into the surrounding estimates, thereby steering regions towards potential under- or over-estimation."