Tracing Mantle Sources to Tectono–Magmatic Settings: Machine Learning Classification of Basalts Over the Past 1 Billion Years

Authors:
Open Access
Publisher website
Google Scholar
Abstract
The geochemical heterogeneity of the mantle, recorded by mantle‐derived basalts, offers crucial insights into the evolution of mantle sources. However, recognizing mantle end‐members of basalts is challenging, as traditional trace element proxies yield ambiguous overlapping results in older basalts. Here we employ machine learning techniques to classify basalts, by their major and trace element characteristics, derived from three primary mantle sources: Depleted Mantle, Enriched Mantle, and Hydrated Mantle. By training on Cenozoic basalts, our machine learning achieves a high accuracy of global test sets spanning Phanerozoic to late Proterozoic, proving both its spatial and temporal effectiveness. Additionally, comparing our source predictions with full‐plate paleogeography since 1 billion years ago and testing the method in complex tectono–magmatic settings further assures spatiotemporal robustness and general applicability. We thus provide an effective machine learning classifier for accurately tracing mantle sources of basalts globally farther back in geologic time.
Funding Information
  • National Natural Science Foundation of China (2021FYC0002)

This publication has 54 references indexed in Scilit: