Non-Supervised Machine Learning applied to the Geochemical Classification of Brazilian A-type Granites
A-type granites; Geochemistry; Data-Driven Discovery; K-means algorithm.
A-type granites are typically characterized by their alkaline trend, high concentrations of HFSE, high values for Ga/Al and HF/H2O ratios, and are mostly associated with post-collisional extensional and anorogenic environments. Over the last few years, however, a series of works have demonstrated that these rocks are much more complex both in terms of their geochemistry, associated tectonic environments and petrogenetic processes involved. In Brazil, A-type granites are described both in Archean to Paleo-Mesoproterozoic cratonic areas and in Neoproterozoic mobile belts, with several occurrences associated with important mineral deposits (Sn, In, W, among others). Many of these occurrences are described and interpreted in a local or regional context, so that an integrated and comparative analysis on a broader scale is still absent in the literature. In this context, this dissertation presents the results of data-driven discovery, by applying data science techniques to manipulate, process and analyze (with the aid of unsupervised machine learning with the K-means algorithm) an extensive database of rock-total geochemical data (>1,900 samples) of A-type granitoids from Brazil in various geotectonic contexts and of various ages. The results show that A-type magmatism in Brazil can be separated into three subgroups named A1, A2 and A3, with distinct geochemical (and partly petrographic) characteristics. The first two show correspondence with the A1 and A2 subtypes already established in the literature, while the third defines a new subgroup, characterized by less evolved lithotypes. These subgroups are individualized in several binary and ternary diagrams proposed in this work from combinations of chemical variables (major and trace elements) and normative minerals. The A1 subgroup encompasses alkali to alkali-alkalic rocks, of peralkaline to metalluminous character. The A2 granites are alkali to calcium-alkaline, peraluminous to weakly metalluminous; while the A3 subgroup includes alkali, alkali-calcic, calcium-alkaline to calcic, metalluminous to peraluminous (to a lesser extent) rocks. Although they do not have preferences for a particular tectonic context or range of occurrence over geologic time, the A1, A2 and A3 subgroups can occur together in the same body, with seven possible combinations, suggesting that they may represent different facies related to each other by magmatic differentiation processes. The association of the subgroups with a specific geotectonic context is not always direct. However, in general terms, plutons formed exclusively from A1-type granites relate more strongly to anorogenic to post-orogenic environments, whereas A2 and A3 subgroups are linked to syn- and post-collisional to late-collisional contexts.