A key method to investigate magma dynamics is the analysis of the crystal cargoes carried by erupted magmas. These cargoes may comprise crystals that crystallize in different parts of the magmatic system (throughout the crust) and/or at different times. While an individual eruption likely provides a partial view of the sub-volcanic plumbing system, compiling data from multiple eruptions can build a picture of a whole magmatic system. In this study we use machine learning techniques to analyze a large (>2000) compilation of mineral compositions from a highly active arc volcano: Villarrica, Chile. Villarrica’s post-glacial eruptive activity (14 ka–present) displays large variation in eruptive style (mafic ignimbrites to Hawaiian style effusive eruptions) yet its eruptive products have a near constant basalt-basaltic andesite bulk-rock composition. What therefore, is driving explosive eruptions at Villarrica and can differences in storage dynamics be related to eruptive style? Here we use hierarchical cluster analysis to detect previously unseen structure in the composition of olivine, plagioclase and clinopyroxene crystals erupted at Villarrica, revealing the presence of compositionally distinct clusters within each crystal population. Using rhyolite-MELTS thermodynamic modelling we related these clusters to intensive magmatic variables: temperature, pressure, water content and oxygen fugacity. Our results provide evidence for the existence of multiple discrete (spatial and temporal) magma reservoirs beneath Villarrica where melts differentiate and mix with incoming more primitive magma. The compositional diversity within an erupted crystal cargo strongly correlates with eruptive intensity, and we postulate that mixing between primitive and differentiated magma drives explosive activity at Villarrica.
- Unsupervised Machine Learning
- Crystal Cargoes
- Thermodynamic Modeling
- Magma mixing
- Large Mafic Ignimbrites