What can quartz trace elements tell us? A machine learning approach for magmatic systems
L. M. Fonseca Teixeira 1, E. Schoonover2, M. Ackerson1
Affiliations: 1 Smithsonian National Museum of Natural History, Washington, DC, USA 2 The Pennsylvania State University, University Park, PA, USA
Presentation type: Poster
Presentation time: Thursday 16:30 - 18:30, Room Poster Hall
Poster Board Number: 219
Programme No: 1.9.26
Abstract
In recent years, the development of machine learning algorithms has provided scientists of different fields with a multitude of new possibilities to analyse and draw information from large, pre-existing datasets. In igneous petrology and volcanology, machine learning studies have found widespread use in reinterpreting trace element datasets, offering fresh insights into chemical evolution. In this study, we apply machine learning to investigate the trace element composition of quartz crystals from a diverse range of magmatic rocks (granitoids, lavas, ignimbrites, pegmatites, porphyry deposits, hydrothermal veins, and greisens) with varying chemical signatures (I-, A-, and S-types). Our results show that Al, Ti, Ge, and Li can be particularly useful for predicting quartz crystallisation environments. Through hierarchical clustering, an unsupervised machine learning algorithm, our model can classify quartz crystals into magmatic, magmatic-hydrothermal, and hydrothermal groups, which can be particularly valuable for uncovering chemical evolution in systems that have been overprinted by hydrothermal activity (e.g. porphyry-Cu systems). To validate our model, we tested it on the Tava Sandstone (Pikes Peak Batholith, CO, USA), a sedimentary unit with well-established quartz provenance, yielding successful results. These findings demonstrate that our model holds potential not only for deconvoluting overprinted/multi-stage complex chemical histories but also for supporting provenance studies in sedimentary geology.