Radial Basis Functions And Kriging – A Gold Case Study

Recent advances in mining software have enabled the implementation of radial basis functions (RBFs) as interpolation and extrapolation algorithms for both continuous (grade) and categorical (geology) data.

RBFs approximate a specific type of kriging called Dual Kriging (Horowitz et al 1996). A recent paper by Stewart et al (2014) details some of the theory of RBFs and kriging and compares estimates using a simulated data set from a negatively skewed distribution (approximating low grade iron ore). RBFs are the underlying algorithms used in the Leapfrog software which is used for this case study.

This article presents a less theoretical and more empirical look at RBF estimates using a real gold data set for which both exploration drilling and grade control drilling data sets are available. This gold distribution is a positively skewed distribution which typically presents more challenges during estimation due to the sensitivity of estimate to the high grade tail of the distribution in comparison to normal or negatively skewed distributions.