Machine learning prediction of quartz forming-environments

Abstract

Trace elements of quartz document the physical-chemical evolutions of quartz growth, which has been a great and most applied tool in the study of geological settings in quartz-forming environments. A classic method is using graphic diagram plots visualizing the quartz trace element discriminations and trends, examples including the Al-Ti diagram (Rusk, 2012, https://doi.org/10.1007/978-3-642-22161-3_14) and the Ti-Al-Ge diagram (Schrön et al., 1988, https://www.researchgate.net/publication/236149159_Geochemische_Untersuchungen_an_Pegmatitquarzen). However, those diagrams are limited to two dimensions and cannot show the information in a higher dimension. In the study, we thus used a machine learning-based approach to evaluate quartz trace elements, and visualized them for the first time in the high-dimensional diagrams. We revisited 1,626 quartz samples from nine geological environments from previous studies, and applied a support vector machine to characterize values of the contained trace elements, including Al, Ti, Li, Ge, and Sr. We demonstrate that support vector machines can identify the crystallization environment of quartz with a significantly higher accuracy than the traditional plotting methods. Our work can massively improve the confidence on distinguishing quartz origin from different geological environments with a high efficiency. The method may also be applicable for other minerals, and we anticipate our research is a starting point for investigating mineral trace elements with machine learning techniques. Our quartz classifier can be accessed via https://quartz-classifier.herokuapp.com.