Some Practical Aspects Of The Use Of Lognormal Models For Confidence Limits And Block Distributions In South African Gold Mines

Krige, D. G.
Organization: The Southern African Institute of Mining and Metallurgy
Pages: 6
Publication Date: Jan 1, 2004
For the purpose of determining confidence limits for ore grade estimates, an extensive simulated database was used, which included 1728 ore blocks with ?actual? grades and the corresponding estimates based on ordinary and simple kriging. The practical results show that conditional unbiasedness is essential and that acceptable confidence limits can then be set using Normal theory for the simulated error distribution based on the logarithms of the ratios of ?actuals?/estimates with a variance related to the kriging variance. The mean of the data used to prepare the variogram and the estimated block grade are to be used properly in this approach. A practical analysis was also done of the actual point and block data from a very large mined out Carbon Leader area. The distributions of the actual grades for two levels of block sizes with different levels of variability were fitted using three models, ie. two three and compound lognormal. The trends of tonnages, grades and relative profits for a range of pay limits, as well as the usual tonnage-grade curves, show a practical insensitivity for the choice between the three lognormal models considered.
Full Article Download:
(1236 kb)