Gold deposits estimation using indicator kriging

"Indicator kriging is one of the non-parametric geostatistical techniques, ideal for estimating the reserves of irregular mineralizations (e.g. gold, uranium, platinum, diamonds, etc.). Such mineralizations are characterized by variograms which are very sensitive to extreme values (outliers) and very difficult to model.The classical approach to this problem in the industry is to trim off these high grades using arbitrary empirical upper limits established either from practice or subjectively. Contrary to this simplistic approach, indicator kriging discretizes the histogram of the grades (usually log normal) in several classes and carries out interpolation separately for each class. This paper gives an overview of the differences between ordinary and indicator kriging along with the steps required to carry out indicator kriging. The paper concludes with a case study involving a gold deposit estimation.IntroductionParametric geostatistics (ordinary kriging) has given us satisfactory results over the years in estimating the reserves of well behaved deposits with coefficients of variation close to 1 (e.g. sedimentary and porphyry type deposits). However, numerous problems have been encountered in estimating irregular mineralizations (e.g. small vein deposits of precious metals) with high coefficients of variation in the range of 2 to 5. Such deposits are characterized by a very small proportion of extreme values (outliers) which create distortions on the variograms (pure nugget effect) and render them useless(1,2).The solution often followed by the mining industry today is to trim off these values based on empirical limits. These limits are derived either by comparison with production results (or bulk samples) or by geological interpretations. Needless to say that such manipulations can lead to serious errors in estimating ore reserves (usually underestimation of total metal quantity), especially when production results or bulk samples are absent.On the other hand, the solution followed by the geostatisticians in the past was to smooth out the data using transforms (e.g. normal-score or natural logarithms). Such transforms are nonlinear and they required non-linear estimation techniques such as disjunctive kriging (DK) or multivariate gaussian kriging (MG) which require normal related hypotheses, not necessarily true in geology.Nonparametric geostatistics (indicator and probability kriging) have been -developed in order to handle these problems related to highly skewed distributions."