Compositions and Classification of Fractionated Boninite Series Melts from the Izu–Bonin– Mariana Arc: A Machine Learning Approach Matthew J. Valetich 1 *, Charles Le Losq 12 , Richard J. Arculus 1 , Susumu Umino 3 and John Mavrogenes 1 1 Research School of Earth Sciences, The Australian National University, Canberra, ACT 2601, Australia; 2 Institut de Physique du Globe de Paris, Universite ´ de Paris, 75005 Paris, France; 3 Faculty of Geosciences and Civil Engineering, Kanazawa University, Kanazawa 920-1192, Japan *Corresponding author. E-mail: valetich@anu.edu.au Received 9 May 2019; Accepted 27 January 2021 ABSTRACT Much of the boninite magmatism in the Izu–Bonin–Mariana arc is preserved as evolved boninite series compositions wherein extensive fractional crystallization of pyroxene and spinel have obscured the diagnostic geochemical indicators of boninite parentage, such as high Mg and low Ti at intermediate silica contents. As a result, the usual geochemical discriminants used for the classi- fication of the broad range of parental boninites are inapplicable to such highly fractionated melts. These issues are compounded by the mixing of demonstrably different whole-rock and glass analy- ses in classification schemes and petrological interpretations based thereon. Whole-rock composi- tions are compromised by entrainment of variable proportions of crystalline phases resulting in in- consistent differences from corresponding in situ glass analyses, which arguably better reflect prior melt compositions. To circumvent such issues, we herein present a robust method for the classification of highly fractionated boninite series glasses. This new classification leverages the analysis of trace elements, which are much more sensitive to evolutionary processes than major elements, and benefits from the use of unsupervised machine learning as a classification tool. The results show that the most fractionated boninite series melts preserve geochemical indicators of their parentage, and highlight the pitfalls of interpreting whole-rock and glass analyses interchangeably. Key words: boninite; Chichijima; classification; glass; machine learning; Mukojima INTRODUCTION The Izu–Bonin–Mariana (IBM) arc of the western Pacific, which extends from Japan southwards for about 2500 km, is one of the largest and most thoroughly studied, nominally intra-oceanic subduction systems in the world (Fig. 1). Extensive dredging and drilling has yielded considerable information on the initiation and maturation of the arc, including early and voluminous generation of boninite series rocks (e.g. Umino, 1985; Arculus et al., 1992; Ishizuka et al., 2006, 2011; Reagan et al., 2010, 2019; Kanayama et al., 2012; Ishizuka et al., 2020). These boninite rocks include many composed of 52–65 wt% SiO 2 with >8 wt% MgO and <05 wt% TiO 2 , consistent with the definition endorsed by the IUGS (Le Bas, 2000). The consensus is that boninite is generated in the upper mantle at relatively high temperatures (1100–1450  C) and low pressures (<1 GPa) via hydrous partial melting of refractory (10–30 % prior melt extrac- tion) peridotite (Crawford et al., 1989; Parkinson & Pearce, 1998; Falloon & Danyushevsky, 2000; Dobson et al., 2006; Kushiro, 2007; Umino et al., 2015, 2018). The Bonin (Ogasawara) Ridge (Kikuchi, 1890; Petersen, V C The Author(s) 2021. Published by Oxford University Press. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/ open_access/funder_policies/chorus/standard_publication_model) 1 J OURNAL OF P ETROLOGY Journal of Petrology, 2021, 1–20 doi: 10.1093/petrology/egab013 Advance Access Publication Date: 5 February 2021 Original Manuscript Downloaded from https://academic.oup.com/petrology/article/62/2/egab013/6128870 by guest on 23 July 2022