Source: Geochemistry, Geophysics, Geosystems
Some volcanoes erupt frequently but change their behavior from explosive to effusive. Effective civil defense requires methods to anticipate not only when these volcanoes will erupt but also how explosively.
Boschetty et al.  successfully apply an unsupervised machine learning technique (hierarchical clustering) to analyze of the crystal cargo of numerous eruptions from a single volcano over time. Their approach detected populations of crystal compositions in multidimensional space that were not apparent in previous visualization methods. The results are combined with thermodynamic modeling of magmatic fractionation to identify likely conditions where each cluster of mineral compositions formed. The authors then use the patterns of occurrence of crystals from each cluster in each eruption to reconstruct details of the plumbing system over time and to distinguish products of explosive and non-explosive eruption events.
Generalization of this strategy to other volcanic systems will help forecasters anticipate how those volcanoes may erupt.
Citation: Boschetty, F. O., Ferguson, D. J., Cortés, J. A., Morgado, E., Ebmeier, S. K., Morgan, D. J., et al. (2022). Insights into magma storage beneath a frequently erupting arc volcano (Villarrica, Chile) from unsupervised machine learning analysis of mineral compositions. Geochemistry, Geophysics, Geosystems, 23, e2022GC010333. https://doi.org/10.1029/2022GC010333
—Paul Asimow, Editor, Geochemistry, Geophysics, Geosystems