Magma differentiation in dynamic mush domains from the perspective of multivariate statistics: open- vs closed-system evolution

Open-conduit conditions characterize several of the most hazardous and active volcanic systems of basaltic composition worldwide, persistently refilled by magmatic inputs. Eruptive products with similar bulk compositions, chemically buffered by continual mafic inputs, exhibit nevertheless heterogeneous glass compositions in response to variable magma mixing, crystallization, and differentiation processes within different parts of the plumbing system. Here we document how multivariate statistics and magma differentiation modeling based on a large data set of glass compositions can be combined to constrain magma differentiation and plumbing system dynamics. Major and trace elements of matrix glasses erupted at Stromboli volcano (Italy) over the last twenty years provide a benchmark against which to test our integrated petrological approach. Principal component analysis, K-means cluster analysis, and kernel density estimation reveal that trace elements define a multivariate space whose eigenvectors are more readily interpretable in terms of petrological processes than major elements, leading to improved clustering solutions. Comparison between open- and closed-system differentiation models outlines that steady state magma compositions at constantly replenished and erupting magmatic systems approximate simple fractional crystallization trends, due to short magma residence times. Open-system magma dynamics imply lower crystallinities pervade the magmatic storage than those associated with closed-system scenarios, allowing efficient crystal-melt separation toward the top of the reservoir, where eruptible melts continuously supply the ordinary activity at the volcano. Conversely, a mush-like environment constitutes the bottom of the reservoir, in which poorly evolved magmas result from mixing events between mush residual melts and primitive magmas injected from deeper crustal levels.