Insights Into Magma Storage Beneath a Frequently Erupting Arc Volcano (Villarrica, Chile) From Unsupervised Machine Learning Analysis of Mineral Compositions

A key method to investigate magma dynamics is the analysis of crystal cargoes carried by erupted magmas. These may comprise crystals that crystallize in different parts magmatic system (throughout crust) and/or at times. While an individual eruption likely provides a partial view subvolcanic plumbing system, compiling data from multiple eruptions can build picture whole system. In this study, we use machine learning techniques analyze large (>2,000) compilation mineral compositions highly active arc volcano: Villarrica, Chile. Villarrica's postglacial eruptive activity (14 ka–present) displays variation style (mafic ignimbrites Hawaiian effusive eruptions), yet its products have near constant basalt-basaltic andesite bulk-rock composition. What therefore, driving explosive Villarrica and differences storage be related style? Here, hierarchical cluster detect previously unseen structure composition olivine, plagioclase clinopyroxene revealing presence compositionally distinct clusters within each population. Using rhyolite-MELTS thermodynamic modeling, these intensive variables: temperature, pressure, water content, oxygen fugacity. Our results provide evidence for existence discrete (spatial temporal) reservoirs beneath where melts differentiate mix with incoming more primitive magma. The compositional diversity cargo strongly correlates intensity, postulate mixing between differentiated drives Villarrica.