Machine Learning Investigation of Clinopyroxene Compositions to Evaluate and Predict Mantle Metasomatism Worldwide. Issue 5 (23rd May 2022)

• Record Type: Journal Article
• Authors: Qin, Ben; Huang, Fang; Huang, Shichun; Python, Andre; Chen, Yunfeng; ZhangZhou, J
• Is Part Of: Journal of geophysical research. Volume 127:Issue 5(2022)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2022-05-23
• Subjects: machine learning -- mantle metasomatism -- classification -- prediction -- probability distribution; Geomagnetism -- Periodicals; Geochemistry -- Periodicals; Geophysics -- Periodicals; Earth sciences -- Periodicals; 551.1
• DOI: 10.1029/2021JB023614
• Languages: English
• ISSNs: 2169-9313

Abstract

Plain Language Summary: Clinopyroxene is a major mineral in Earth's upper mantle. Previous studies have attempted to discriminate between reactions modifying the mantle by plotting clinopyroxene major and trace element compositions in two‐dimensional (2‐D) diagrams. However, these 2‐D methods show poor accuracy when applied to global datasets. Therefore, we suggest a machine learning approach to evaluate clinopyroxene compositional data in higher dimensions. Our results demonstrate that machine learning can significantly improve the accuracy of clinopyroxene compositional predictions over classical methods utilizing elemental ratios. Furthermore, the application of our algorithm to a global clinopyroxene dataset suggests that mantle metasomatism is globally widespread. Abstract: Clinopyroxene major and trace element compositions document their physicochemical evolution and have been widely used to detect mantle metasomatism. Classical methods typically rely on one or several elemental ratios such as Ca/Al, Mg/Fe, La/Yb, and Ti/Eu to determine whether rocks or minerals have been metasomatized. These methods have proven useful at specific sites, but not globally. In this study, we used machine learning methods to classify the chemical compositions of clinopyroxenes from mantle xenoliths and examine their relationship with mantle metasomatism. We compiled major element data from 8, 713 clinopyroxene samples (21, 605 analyses) and trace element data from 1, 235 clinopyroxene samples (2, 967 analyses). Samples were labeled "positive" if clearly affected by patent metasomatism based on petrographic evidence, "negative" if clearly unaffected by metasomatism, or were left unlabeled if neither case applied. We then trained an XGBoost machine learning model, which achieved higher accuracy than traditional methods using a limited number of elemental ratios. Our results identify numerous locations with high mean probabilities of mantle metasomatism and show variability in the probability distributions observed across locations worldwide. These results indicate that metasomatism may be globally widespread, but the probability of metasomatism is not correlated with geophysical parameters such as crustal thickness, lithospheric thickness, or mantle S ‐wave velocity. Hence, the spatial distribution of metasomatism appears mainly driven by unobserved factors. Key Points: Mantle clinopyroxene major and trace element data compiled and evaluated with machine learning models Accuracy comparisons between low‐ and high‐dimensional dataspaces reveal the most important features for classification Machine learning models identify clusters of mantle metasomatism worldwide