Coherent Modelling of Mineral Grades and Zones by Coupling Cokriging and Support Vector Machine

Abstract

The theories of geostatistics and machine learning are originated from statistics, however rooted in different applications. They are sometimes looked as competing theories, sometimes as complementary. The latter perspective is a foundation of this article, which tries to use them jointly to cover their shortages in mineral resources modelling. The theory of geostatistics provides methods for a robust spatial modelling of mineral resources out of univariate and multivariate datasets. However, it demands much effort and experience to generate a coherent model if the dataset contains many categories, either with a single categorical variable or because of crossing two or more categorical variables. This limitation could be covered by machine learning to establish the classification rules between continuous variables and the categorical variables in the space of drillholes. The proposed workflow is applied to a synthetic porphyry copper dataset to verify and illustrate its performance in a multivariate application. The dataset consisting of five mineral grades within five mineral zones. The concluding remark is that the choice of geostatistical interpolator should be done with cautious not to alter the statistical distribution (variance and dimension-support change) of the input core data in the drillhole space while interpolating to the block space.