Confidence Intervals for Local Estimation: Application to the Witwatersrand Basin

One advantage of kriging is that it provides both an estimate of the block or point grade, and an indication of the local precision via the kriging variance. If one assumes that the estimation errors are normally distributed, then confidence intervals for the grades can be calculated. But in mining, grades are rarely normally distributed; in general, they are skew. This is particularly true for precious metals. In this paper we study confidence intervals for variables with non-normal distributions. Using a non linear geostatistical technique, the grades are transformed into their normally distributed equivalents. Confidence intervals are obtained for these normal variables, and back transformed to give the corresponding confidence intervals for the grades, for either points or blocks. This approach is applied to gold data from the Witwatersrand basin in South Africa.