Machine learning for reconstructing the primary carbon contents of mid-ocean ridge basalts

Abstract

One of the primary locations of mafic magma production on Earth is the global mid-ocean ridge system, with basalts erupted from these ridges serving as valuable probes for assessing the compositional variability of the upper mantle and exploring the deep carbon cycle. However, directly measuring carbon contents in mid-ocean ridge basalts (MORBs) has proven challenging due to degassing during magma ascent. Early investigations indicate that some incompatible-trace-element- depleted and -enriched MORBs avoid heavy degassing, and show a narrow range of CO2/Ba ratios, which was generally applied to reconstruct the primitive CO2 content of global MORBs. However, increasing studies reveal significant variability in the CO2/Ba ratios of MORBs. Here, we compiled a dataset including the geochemical compositions of MORB glasses and melt inclusions for which studies supported no significant degassing. Based on it, we constructed a supervised machine learning (ML) model capable of accurately predicting CO2 contents in individual samples using the selected elemental contents. Applying our model to a global MORB database reveals that CO2 contents and CO2/Ba ratios of global MORBs are highly variable, highlighting the significance of mantle heterogeneity, which can be attributed to the interactions with deep-sourced plume, or the recycled components associated with the big subduction zone. Our findings underscore the potential of ML as a powerful tool for uncovering hidden structural patterns in complex geological data, shedding light on the intricate interplay between carbon, mantle composition, and Earth's long-term geological processes.