11:00 AM - 1:00 PM
[MGI34-P07] Extracting major element compositions of magmatic end-member components using non-negative matrix factorization
Keywords:volcanic rock, major element, end-member magma, Machine learning
The chemical evolution of magma is driven by the mixing between different end-member magmas and the differentiation of specific mineral species. The determination of magma end-member components is essential for discussing the magma evolution process in the crust and the transition in eruptive sequence due to the involvement of different magmas. Analysis based on trends in chemical composition is widely used to determine end-member magmas. However, compositional data should be carefully treated as it is represented by weight percent (i.e., the constant-sum constraint). For example, given the wt% values are simply used for trend analysis, addition and and subtraction may not be distinguishable.
We applied a statistical method called non-negative matrix factorization (NMF) to analyze the major element compositions of a series of magmas. NMF is a statistical method that decomposes a matrix into two matrices of non-negative values. For whole-rock chemical composition data consisting of multiple (m) elements and multiple samples, this corresponds to the operation of expressing the composition of an individual sample as a mixture ratio of a small number (n<m) of end-member components. Yoshida et al. (2018, JMG) showed that the compositional variations of a series of metamorphic rocks in the Sambagawa metamorphic belt could be explained by four end-member components using NMF.
In this study, NMF was applied to a dataset of major element compositions of Quaternary volcanic lavas sampled from 17 different volcanoes in a volcanic group called the Sengan region, Northeastern Japan Arc. The results show that the major element compositional variation of magmas from basalt to rhyolite in the Sengan region can be expressed by a small number of petrologically interpretative end-member components. It was found that the end-member compositions obtained by NMF correspond to felsic and mafic end-member magmas and various crystallization processes. This means that the compositional variations in major element can be statistically decomposed using NMF into magma mixings and fractional crystallization. We will discuss the magma evolution in the arc crust based on the end-member components determined by the NMF analysis.
We applied a statistical method called non-negative matrix factorization (NMF) to analyze the major element compositions of a series of magmas. NMF is a statistical method that decomposes a matrix into two matrices of non-negative values. For whole-rock chemical composition data consisting of multiple (m) elements and multiple samples, this corresponds to the operation of expressing the composition of an individual sample as a mixture ratio of a small number (n<m) of end-member components. Yoshida et al. (2018, JMG) showed that the compositional variations of a series of metamorphic rocks in the Sambagawa metamorphic belt could be explained by four end-member components using NMF.
In this study, NMF was applied to a dataset of major element compositions of Quaternary volcanic lavas sampled from 17 different volcanoes in a volcanic group called the Sengan region, Northeastern Japan Arc. The results show that the major element compositional variation of magmas from basalt to rhyolite in the Sengan region can be expressed by a small number of petrologically interpretative end-member components. It was found that the end-member compositions obtained by NMF correspond to felsic and mafic end-member magmas and various crystallization processes. This means that the compositional variations in major element can be statistically decomposed using NMF into magma mixings and fractional crystallization. We will discuss the magma evolution in the arc crust based on the end-member components determined by the NMF analysis.