Olivine in picrites from continental flood basalt provinces classified using machine learning

Abstract

Picrites, dominantly composed of highly forsteritic olivine, can serve as important constraints on primary magma composition and eruption dynamic processes in global Continental Flood Basalt (CFB) provinces.Picrites are commonly divided into high-Ti and low-Ti groups based on whole-rock TiO2 content or Ti/Y ratio. Here, we use an Artificial Neural Network (ANN) to classify the individual olivine in picrites from global CFB provinces according to whether their parental magma is high-Ti or low-Ti to better understand the primary origin and magmatic processes.After training the ANN on one thousand olivine major element compositions data points, the network was able to differentiate chemical patterns for high-Ti and low-Ti olivine, and classify olivine into correct types with an accuracy of >95 %. Moreover, we find that two types of olivine mix in some single samples from Etendeka, Emeishan, and Karoo CFB provinces. Combining the results with chemical markers of source This is the peer-reviewed, final accepted version for American Mineralogist, published by the Mineralogical Society of America.The published version is subject to change. Cite as Authors (Year) Title. American Mineralogist, in press.

Insights on Source Lithology and Pressure‐Temperature Conditions of Basalt Generation Using Machine Learning

“…(2020), Bergen, Johnson, de Hoop, and Beroza (2019), Cheng et al. (2022), Caricchi et al. (2020), Hazen (2014), Hazen et al.…”

“…In contrast to earlier methods for identifying source lithology, machine learning (ML) models have the advantage of sidestepping the need to solve complex problems using linear relationships based on compositional attributes like Fe/Mg and/or FCMS. Instead, ML models adopt a data-driven approach, as highlighted by Ardabili 2020), Bergen, Johnson, de Hoop, and Beroza (2019), Cheng et al (2022), Caricchi et al (2020), Hazen (2014), Hazen et al (2019), Morrison et al (2017), andPetrelli et al (2020). ML models excel in unraveling non-linear relationships present in extensive data sets, leveraging what is often referred to as a "learning process" (Shai & Shai, 2014).…”

“…Machine learning may provide a solution to those complex problem which convert experience into expertise or knowledge during learning (Ardabili et al, 2020;Bergen, Johnson, de Hoop, & Beroza, 2019;Bergen, Johnson, Maarten, & Beroza, 2019;Maros et al, 2020). Many high performance models have been built in Earth science studies (e.g., Boschetty et al, 2022;Cheng et al, 2022;Hazen, 2014;Higgins et al, 2022;Li & Zhang, 2022;Morrison et al, 2017;Petrelli & Perugini, 2016;Ueki et al, 2018). Guo et al (2021) built a binary machine learning model to reveal mafic and peridotitic source based on the database from Yang et al (2019) using Support Vector Machine and Deep Neural Networks.…”