Exploring the composition and mineralogy of the mantle with stable isotopes and machine learning

Abstract

Since its origin more than 4.5 billion years ago, the Earth has undergone a dynamic evolution driven by the slow cooling of its interior. Two-way mass exchange between its interior (i.e., the mantle) and exterior reservoirs (i.e., the crust), facilitated by volcanic activity and crustal recycling processes, has largely determined the geochemical trajectory of the Earth. The physical and chemical processes that shape the Earth’s evolution leave behind geochemical “fingerprints” within the mantle. These fingerprints manifest as complex mineralogical, elemental, and isotopic variations on a multitude of scales, collectively known as mantle heterogeneity. Understanding the origin and nature of mantle heterogeneity is the key to unraveling the complex processes that have shaped the Earth over the past 4.5 billion years. Oceanic basalts, including ocean island basalts (OIBs) and mid-ocean ridge basalts (MORBs), provide a comprehensive sampling of mantle heterogeneity. This thesis aims to advance our understanding of mantle heterogeneity and its connection to the geochemical evolution of the Earth by developing innovative analytical techniques for analyzing oceanic basalts. Chapters 2 and 3 demonstrate the potential of high precision stable calcium (Ca) isotope measurements for constraining the evolution of the mantle, while Chapter 4 explores the origin of enriched mantle reservoirs by utilizing machine learning classification algorithms. Lastly, Chapter 5 briefly describes a local geological study that was instrumental in developing laboratory and computational skills that were crucial throughout my Ph.D. Chapter 2 surveys the Ca isotope compositions of oceanic basalts. The distinctly light Ca isotope signatures of OIBs relative to MORBs are most easily explained by equilibrium Ca isotope fractionation during partial melting of the mantle. The Ca isotope signatures of MORBs are relatively invariable and conform to partial melting models with spinel-peridotite like source mineralogies, while the more variable Ca isotope compositions of OIBs require melt-mixing between a MORB-like melt and a low-degree melt with an isotopically light Ca isotope signature. Such a melt can only be generated from source lithologies enriched in pyroxene and garnet, suggesting that lithological heterogeneity may play an important role in the mantle. Chapter 3 builds on the insights from Chapter 2 to construct a comprehensive model that explains the global Ca isotope systematics of oceanic basalts. Island-averaged Ca isotope compositions exhibit negative relationships with lithospheric thickness and primitive TiO2 concentrations that are best explained in the context of decompression melting beneath oceanic lithosphere. Low degrees of partial melting beneath thick oceanic lithosphere preferentially sample fertile, recycling-related source lithologies rich in pyroxene and garnet, generating melts with isotopically light Ca isotope signatures. Conversely, extensive degrees of melting beneath thin oceanic lithosphere dilute the influence of recycling-related source lithologies, yielding melts with Ca isotope compositions more characteristic of typical peridotite sources. Additionally, Chapter 3 investigates the relationships between Ca isotopes and radiogenic isotopes. HIMU- and DMM-flavored oceanic basalts appear to trace sources with bulk silicate Earth-like (BSE) Ca isotope compositions, while EM-flavored OIBs require sources with Ca isotope signatures that are substantially lighter than BSE. Such a light Ca isotope signature is most easily generated by the incorporation of recycled lithospheric materials into a BSE-like source. Lastly, Chapter 4 implements Random Forest and Gaussian Process classification algorithms to explore the origin of enriched mantle reservoirs. Compositional differences between OIBs and enriched-MORBs (E-MORBs) suggest that they trace sources with different histories of enrichment. Joint application of Random Forest and Gaussian Process classification algorithms is a promising and versatile tool for geochemical exploration that can be applied to a wide range of datasets.