Multivariate geochemical classification of chromitite layers in the Bushveld Complex, South Africa

The Bushveld Complex, the largest layered mafic-ultramafic intrusion worldwide, is host of numerous, laterally continuous and chemically similar chromitite layers. Based on their stratigraphic position the layers are subdivided into a lower, middle and upper group (LG, MG and UG). Within these groups the layers are numbered successively - from the base to the top of each group. Attempts of discriminating between single layers based on their composition have failed - mainly due to the significant overlap of compositional fields, e.g. of chromitite mineral assemblages and chromite mineral chemistry between (neighboured) layers. In this contribution a tailored and easy to use multivariate classification scheme for the chromitite layers is proposed, based on a comprehensive classification routine for the LG and MG chromitites. This routine allows a clear attribution with known uncertainty of eight distinct chromitite layers. The study was carried out at the Thaba Mine, a chromite mine located on the western limb of the Bushveld Complex. The classification is based on a large geochemical database (N = 1205) from Thaba Mine. It comprises of a hierarchical discrimination approach relying on linear discriminant analysis and involves five distinct steps. Using default homogeneous prior probabilities, classification results are excellent for the first discrimination steps (LGs vs. MGs, 97 %; LG-6 vs. LG-6A, 94 %) and very good for the following steps (MG-1/2 vs. MG-3/4, 86 %; MG-1 vs. MG-2, 92 %; MG-3 vs. MG-4, 93 %; MG-4 vs. MG-4Z, 97 %; MG-4 vs. MG-4A, 88 %). The classification scheme was tested using the same sample set as a training set with unknown composition. Overall classification results for unknown samples belonging to one of the layers are 81 %. Hence, the classification scheme is at least valid for the Thaba mine. The approach may, however, be extended across the entire Bushveld, provided that an appropriate geochemical data set is available.