Document Type

Presentation - Poster - Presentation

Identifying Mantle Reservoirs in Hawaiian Ocean Island Basalts: A Machine Learning Approach with Trace Element Ratios as Isotopic Proxies

Authors

Daniel J. O'Hare, Loma Linda UniversityFollow
Ana M. Martínez Ardila, Loma Linda UniversityFollow
Benjamin L. Clausen, Geoscience Research InstituteFollow
Germán H. Alférez, Southern Adventist UniversityFollow

Department

Computing

Date of Activity

9-25-2024

Abstract

A vast amount of geochemical data from ocean island basalts (OIBs) have become available and have been widely used to identify contributions from various mantle reservoirs, including Depleted Mantle (DM), Enriched Mantle I (EM-I), Enriched Mantle-II (EM-II), and HIMU (high µ, or high ²³⁸U/²⁰⁴Pb). Isotopic ratios of Pb, Sr, Nd, and Hf are particularly valuable for identifying these reservoirs, due to minimal effects by secondary alteration, minimal fractionation during melting and crystallization, and the distinctive isotopic histories of the reservoirs. However, the high cost of isotopic analysis motivates the development of alternative, cost-effective approaches.

We applied two machine learning (ML) methods, hierarchical clustering and k-means clustering, to trace element and isotopic data from Hawaiian OIBs to investigate whether specific trace element ratios could act as proxies for isotopic signatures.

Optimal cluster numbers for Pb, Sr, and Nd isotopes, as well as trace element ratios, were identified using hierarchical clustering and the Silhouette Coefficient method. K-means clustering, an unsupervised ML technique, revealed distinct groupings in isotopic and trace element data. Rb/Ba and Rb/Sr effectively differentiated between Nd and Sr isotopic clusters, with higher values corresponding to a stronger EM influence. Rb emerged as a reliable discriminant for these clusters. Ratios such as Rb/Ba, Ba/Th, Th/U, and Rb/Sr successfully discriminated between ²⁰⁸Pb/²⁰⁴Pb and ²⁰⁶Pb/²⁰⁴Pb clusters.

Using ²⁰⁸Pb/²⁰⁴Pb and ²⁰⁶Pb/²⁰⁴Pb data, we also tested various trace element ratios for the ability to differentiate between the two established geochemical trends of Hawaiian OIBs: the enriched Loa-trend and the more-depleted Kea trend. Rb/Ba, Th/U, Rb/Sr, and U/Pb ratios successfully detected the two trends, whereas Rb/K, Ba/La, and Th/Rb ratios did not.

Our findings suggest that certain trace element data can reliably act as proxies for isotopic signatures, offering cost-effective identification of DM and EM sources. Future research will focus on expanding the trace element dataset and refining proxy relationships, providing deeper insights into mantle domains and advancing our understanding of mantle geochemistry and geodynamics.

Recommended Citation

O'Hare, Daniel J.; Martínez Ardila, Ana M.; Clausen, Benjamin L.; and Alférez, Germán H., "Identifying Mantle Reservoirs in Hawaiian Ocean Island Basalts: A Machine Learning Approach with Trace Element Ratios as Isotopic Proxies" (2024). Achieve. 2220.
https://knowledge.e.southern.edu/achieve/2220