Presented By: Earth and Environmental Sciences
Smith Lecture - Dr. Andrew Thomson, University College London
Coupling Machine Learning with Geochemistry: New Evidence for Mantle Carbon Recycling from Sub-lithospheric Diamonds
The H/C ratio in earth’s exosphere is higher than it is in the source region of primitive basalts, suggesting the presence of an enriched carbon reservoir in the mantle [1]. A plausible explanation is that subduction of carbon over time may have enriched the mantle in recycled material over time. Whilst recent estimates of carbon fluxes between surface and interior reservoirs suggest very little carbon reaches the asthenospheric mantle [2], natural sub-lithospheric diamonds provide direct evidence for mantle carbon.
Sub-lithospheric diamonds from kimberlites in the Juina regions of Brazil contain trapped silicate inclusion assemblages reminiscent of portions of mafic eclogitic lithologies at depths of 300-800 km beneath the surface. Combined with isotopic measurements demonstrating a crustal signature [3] and extreme enrichment of trace elements [4], these samples appear consistent with production from melts of subducted crust. Petrological experiments have demonstrated that the carbonated oceanic crust produces low degree melts capable of generating the diamond-hosted assemblages observed during interaction with the overlying mantle [5].
Ultimately, verification of this model would benefit from quantitative evidence that the sub-lithospheric diamonds and/or their inclusions originate from transition zone depths. Compared with other types of inclusions in these diamonds, majoritic garnets provide the best opportunity to accurately estimate the depth of diamond formation because their chemistry is known to vary as a function of pressure [e.g. 6]. However, the available empirical barometers cannot be confidently applied to diamond-hosted inclusions which are observed as lone inclusions without any coexisting phase assemblages. Adoption of a machine learning approach has provided a novel solution to this problem, allowing accurate predictions of diamond formation conditions [7].
[1] Hirschmann, M. & Dasgupta, R. (2009) Chem. Geol. 262, 4–16. [2] Kelemen, P. & Manning, C. (2015) PNAS 112, E3997-E4006. [3] Burnham et al. (2015) Earth. Planet. Sci. Lett. 432, 374. [4] Thomson, A. et al. (2016) Lithos 265, 108-124. [5] Thomson, A. et al (2016). Nature 529, 76-79. [6] Akaogi, M., and S. Akimoto (1977), PEPI, 15, 90–106. [7] Thomson et al. (2021) JGR: Solid Earth 126, e2020JB020604.
Sub-lithospheric diamonds from kimberlites in the Juina regions of Brazil contain trapped silicate inclusion assemblages reminiscent of portions of mafic eclogitic lithologies at depths of 300-800 km beneath the surface. Combined with isotopic measurements demonstrating a crustal signature [3] and extreme enrichment of trace elements [4], these samples appear consistent with production from melts of subducted crust. Petrological experiments have demonstrated that the carbonated oceanic crust produces low degree melts capable of generating the diamond-hosted assemblages observed during interaction with the overlying mantle [5].
Ultimately, verification of this model would benefit from quantitative evidence that the sub-lithospheric diamonds and/or their inclusions originate from transition zone depths. Compared with other types of inclusions in these diamonds, majoritic garnets provide the best opportunity to accurately estimate the depth of diamond formation because their chemistry is known to vary as a function of pressure [e.g. 6]. However, the available empirical barometers cannot be confidently applied to diamond-hosted inclusions which are observed as lone inclusions without any coexisting phase assemblages. Adoption of a machine learning approach has provided a novel solution to this problem, allowing accurate predictions of diamond formation conditions [7].
[1] Hirschmann, M. & Dasgupta, R. (2009) Chem. Geol. 262, 4–16. [2] Kelemen, P. & Manning, C. (2015) PNAS 112, E3997-E4006. [3] Burnham et al. (2015) Earth. Planet. Sci. Lett. 432, 374. [4] Thomson, A. et al. (2016) Lithos 265, 108-124. [5] Thomson, A. et al (2016). Nature 529, 76-79. [6] Akaogi, M., and S. Akimoto (1977), PEPI, 15, 90–106. [7] Thomson et al. (2021) JGR: Solid Earth 126, e2020JB020604.
Explore Similar Events
-
Loading Similar Events...
