Approaching Simultaneous Local and Global Accuracy

When estimating block grades for mining purposes, the currently available methods allow us to maximize the accuracy of either global grade and tonnage curve prediction or local block selection but not both at once. Locally accurate block estimates provide the best result during actual selection and mining but can give highly distorted global grades and tonnages at cutoffs above zero. Globally accurate block estimates provide good prediction of grade and tonnage curves but perform badly during actual selection giving much higher misclassification rates leading to serious degradation of value of the material selected for processing. These statements hold true in varying degrees for all scales and combinations of sample spacing and block size. This paper puts forward a method that retains the properties of accurate global estimation while simultaneously approaching maximum local accuracy. The process is a simple application of rank and replace combining two estimates, one that targets local block accuracy and one that targets actual block variability. The method is empirically demonstrated using a case study using real data. The conclusion, for this data set, is that local selection accuracy can be greatly improved (but not maximized), in comparison to existing methods, while maintaining grade and tonnage curve accuracy that results from true block variability. Comparisons with ordinary kriging, sequential Gaussian simulation, turning bands, local uniform conditioning, and ordinary kriging with reduced sample numbers are presented.