Univariate Geostatistical Outlier Detection Methods Based on Variogram Pairs, Case Study: Sarigunay Gold Deposit, Iran

Statistically, outliers are data points that are dissimilar to the whole dataset beyond stated limits. These existing outliers may give rise to misinterpretations in statistical and geostatistical analyses. To detect outliers two methods of (1) boxplot as a representative of statistical methods and (2) a combination of Mahalanobis Distance (MD) and network graph as a representative of geostatistical methods are applied. Variograms are the basis of geostatisticsal analysis which evaluate spatial variability. After application of variograms, pairs of data points are taken to draw H-scatter plots. In the H-scatter plots, data are illustrated through specific distances. In this case, lags of 20 meters are applied to the h-scatter plot. Then, mahalanobis distance and 97.5% confidence interval, taken from chi-square distribution, are applied to the h-scatter plot to detect pairs of outliers. In order to consider geospatial relation of each pair, a network graph is designed which counts the number of edges for each node. The number of edges demonstrates the outliers and their neighbouring nodes which the outlier detection is based on. The mentioned process, applied to the oxide zone of Sarigunay epithermal gold deposit in Iran, results in 286 data points detected as outliers throughout an 11945 sample dataset in which the ratio of outliers to raw data is 2.39%. The boxplot drawn for the raw data indicates the cut-off assay of 10 ppm Au. Substantially, combination of statistical and geostatistical outlier detection methods leads to robust variograms and more precise estimation.