Isatis for resource classification and confidence intervals
Classification of the resources in different categories, measured, inferred, or indicated, implies reliable calculations of the confidence level in the grade estimates, at a local scale (SMU) and at greater scales (quaterly or yearly exploitation areas).
ISATIS provides several tools to calculate reliable and realistic Confidence Intervals of these different estimates. Among them, the exclusive and rapid application Confidence Intervals.
The kriging variance is certainly not the best way to classify your deposit’s resources. There are two main reasons for this. The error distribution is unlikely to be gaussian and the kriging variance is data-independent (though greater uncertainty is expected in high grade areas than in low ones). The distribution of mining data is, by nature, skewed; standard linear Confidence Intervals are, consequently, not relevant since they are not consistent with the data.
A better approach would be to obtain the error distributions from simulations. But this is extremely time-consuming, only to get two numerical values. A shortcut involving similar hypothesis is proposed in Isatis.
This original methodology was first applied in 1996 by Roth and Armstrong (from the Centre of Geostatistics of the “Ecole des Mines de Paris”) on gold grades of the Witwatersrand basin.
A more recent application has been succesfully made by Rio Tinto and has reduced drastically the time required for updating reserve classifications.
The basic idea is to use the Discrete Gaussian Model to calculate the Confidence Intervals directly from the gaussian grade estimates.
The figures above compare these Confidence Intervals to standard linear Confidence Intervals and to those derived from stochastic simulations.
