Classification of rock types using multivariate analysis in the Dezak area, SW Iran

Abstract

The geochemical multivariate techniques were applied to recognize different geological units and categorized rock types which accompanied by different penalties. Knowing changes and classification of raw materials with macroscopic and microscopic quantification have direct effect on economic and efficiency in a precipitated calcium carbonate plant. Based on lithological changes, the Darian Formation was divided into four rock units based on a total of 492 chip samples collected from the study area. The main aim objective of this study was to classify the rock types in this area using a multivariate analysis combining the R-mode factor analysis and K-Means method. Due to the opposite vector orientation of SiO2, Al2O3, Fe2O3, K2O, MgO, Na2O, MnO, TiO2, P2O5, SrO, and SO3 versus CaCO3, the lowest geochemical similarity and maximum separation distance were observed between CaCO3 and other parameters using the R-mode factor analysis, which has also been verified by the Ward’s hierarchical agglomerative clustering method. Complementary Q-mode analysis was applied to determine the position of the samples (eigenvalues) with respect to the variables (eigenvectors). In order to geochemical classification of the samples, the K-Means clustering method classified the samples taken from the study area into four geochemical groups. The second group, which comprised a total of 422 composite samples, contained high-concentration CaCO3 samples with an average concentration greater than 97%. The average SiO2 concentration in these samples was 1%, and the sum of Al2O3 and Fe2O3 was about 0.38%. Therefore, the materials falling in this group can be appropriately used as raw materials to feed the plant. Using clr-transformed data for R- and Q-mode analysis, the eigenvectors and eigenvalues show sub-compositionally coherent and closure effect is overcomes. The total variance which is explained by the first factor decreases from 78.6% in real R-mode analysis to 61.7% in clr transformation. Due to major occurrence of the first dimension on geochemical changes, the samples categorized as richest (the best data set), contaminated (the worst data set), and transmission zone.