GEOCHEMICALLY “FINGERPRINTING” THE IDAHO BATHOLITH; MACHINE-LEARNING APPLICATIONS IN PROVENANCE RESEARCH

Methods in provenance research are continuously evolving as analytical technologies improve and as our understanding of the connections between mineral chemistry and tectonics expands. Multiproxy detrital mineral studies provide more contextual geologic information than the use of zircon alone (O’Sullivan et al., 2016; Gaschnig, 2019; Chew et al., 2020). Therefore, the combination of accessory mineral geochemical data and multivariate data analysis techniques should be able to provide a more robust assessment of the relationships between detrital minerals and their source rocks. To test this hypothesis, we use both agglomerative and hierarchical clustering methods to characterize geochronological and geochemical relationships in magmatic zircon, monazite, and titanite from the well-studied Cretaceous-Paleogene Idaho batholith. The clustering methods, which are unsupervised machine learning algorithms, identify characteristic relationships within the mineral samples, providing a ‘fingerprint’ that is then used to evaluate detrital minerals in modern sand from rivers that drain the batholith. Two different clustering methods are used to allow for an assessment of the consistency of the results.

Results from both clustering methods show sample groupings of magmatic minerals that generally align with suite ages and episodes of magmatic evolution. Additionally, when age control is removed, zircon clusters vary by Hf concentration, having inverse relationships with Gd/Yb and Th/U and direct relationships with Nb/Yb. Monazite and titanite grains tend to group according to periods of peraluminous vs. metaluminous magmatic activity and have opposing trace element trends that change over time. Detrital minerals seem to aggregate based on the strongest controlling variables, such as age and Th/U in monazite and Nb/Ta in titanite. Parallels between the magmatic and detrital mineral groupings identified in the clustering analysis demonstrate that provenance research can be strengthened by using these methods to identify the characteristic relationships present in both accessory mineral datasets.