Backward elimination procedure for a predictive model of gold concentration

In mineral exploration campaigns, concentrations of the element(s) sought and pathfinder elements are determined in order to delimit targets that may lead to the discovery of hidden mineral deposits. The pathfinder elements for epithermal Au deposits commonly include Ag, As, Sb, and Hg. Traditionally, contours of anomalous concentrations of these pathfinder elements are superimposed in map form as a means of assessing the Au potential of a target area. However, the information generated is more qualitative than quantitative. This paper proposes a statistical modeling approach that quantitatively predicts Au concentration from the known concentrations of the aforementioned pathfinder elements. Element concentration data from the Florida Canyon deposit, Nevada (USA), a large disseminated epithermal gold deposit, were used to develop and examine the validity of this predictive model. A geologically constrained data set of 115 data points and 575 elemental concentrations obtained from a total of 28 vertical and inclined drill holes located in a restricted area of the deposit formed the basis for model development and validation. A backward elimination procedure was pursued starting with a cubic regression model excluding all higher order (>2) interaction terms to arrive at a suitable predictive model. After seven steps of factor elimination, a quadratic model including several two-factor interaction terms was obtained with an adjusted R-2 value of 0.994 and F-ratio of 1665.5. The model was validated with another 112 geologically constrained data points (560 elemental concentrations) obtained from 13 additional drill holes in the same area of the deposit. Further analysis highlighted the effects of individual and combined pathfinder element values on predicted Au concentrations in the Florida Canyon deposit. Although the model developed herein is deposit specific, the statistical approach can be applied to other disseminated epithermal Au deposits and deposit types where the model constraints are satisfied (e.g., geological homogeneous domain). Such an approach may be utilized in mineral exploration programs to predict gold concentrations in undersampled areas of a deposit, optimize the number of pathfinder elements required for the efficient detection of the element of interest, and evaluate geochemical data sets generated by different laboratories or analytical techniques. (c) 2007 Elsevier B.V All rights reserved.