loading page

E-MORB and OIB petrogenesis investigated with machine learning
  • +3
  • Zachary Eriksen,
  • Stein Jacobsen,
  • Charles Langmuir,
  • Junjie Dong,
  • Matthew Brennan,
  • Jesse Gu
Zachary Eriksen
Harvard University

Corresponding Author:[email protected]

Author Profile
Stein Jacobsen
Harvard University
Author Profile
Charles Langmuir
Harvard University
Author Profile
Junjie Dong
Harvard University
Author Profile
Matthew Brennan
Harvard University
Author Profile
Jesse Gu
Harvard University
Author Profile

Abstract

Oceanic basalts provide an invaluable window into evolutionary processes governing mantle spatial and temporal chemical heterogeneity. Ocean island basalts (OIBs) and enriched mid-ocean ridge basalts (E-MORBs) are powerful tracers of mantle melting and crust-mantle recycling processes. Whether the elemental and isotopic variations observed in both E-MORBs and OIBs are derived from similar mechanisms, however, remains under debate. Investigating compositional differences between E-MORBs and OIBs is a simple approach to constrain their origins, a technique for which machine learning classification algorithms are optimal. Here we implemented a novel machine learning approach complemented by mantle component mixing models to highlight compositional differences between E-MORBs and OIBs and further investigate their petrogenesis (data sourced from GEOROC database and Gale et al., 2013). Considering Random Forest-based Gini indexes, elements sensitive to pressure and degree of melting (FeO, TiO2, Lu, and Sr) were identified as the best discriminators between E-MORBs and OIBs. Our Gaussian process classification algorithm successfully classified OIBs and E-MORBs better than 97% of the time when considering 1) Sr & FeO and 2) TiO2 & Lu. The probabilistic nature of Gaussian process modeling permitted calculation of new quantitative discriminant diagrams rooted in probability (Sr vs. FeO and TiO2 vs. Lu). Complementary trace element modeling yielded compositionally similar E-MORB and OIB sources with moderately incompatible element enrichments in the OIB source due to the influence of recycled oceanic crust (Prytulak & Elliott, 2007). Our source compositions are consistent with a simple, joint model for E-MORB and OIB petrogenesis after Donnelley et al. (2014): low-degree partial melts of subducted slabs metasomatize the depleted mantle producing a re-fertilized mantle (RM). RM is randomly sampled at mid-ocean ridges to produce E-MORB, while upwelling plumes sample both RM and recycled oceanic crust, yielding OIB. References: Donnelly et al. (2004). Earth and Planet. Sci. Lett., 226(3–4), 347–366. Gale et al. (2013). Geochem., Geophys., Geosyst., 14(3), 489–518. Prytulak & Elliott (2007). Earth and Planet. Sci. Lett., 263(3–4), 388–403.