Model based Paleozoic atmospheric oxygen estimates: a revisit to GEOCARBSULF

Abstract

Geological redox proxies increasingly point towards low atmospheric oxygen concentrations during the early Paleozoic Era, with a subsequent protracted rise towards present-day levels. However, these proxies currently only provide qualitative estimates of atmospheric O2 levels. Global biogeochemical models, in contrast, are commonly employed to generate quantitative estimates for atmospheric O2 levels through Earth’s history. Estimates for Paleozoic pO2 generated by GEOCARBSULF, one of the most widely implemented carbon and sulfur cycle models, have historically suggested high atmospheric O2 levels throughout the Paleozoic, in direct contradiction to competing models. In this study, we evaluate whether GEOCARBSULF can predict relatively low Paleozoic O2 levels. We first update GEOCARBSULF by adopting the recent compilation of the δ13C value of marine buried carbonate and replacing the old formulation of the sulfur isotope fractionation factor with empirical sulfur isotope records. Following this we construct various O2 evolution scenarios (with low O2 levels in the early Paleozoic) and examine whether GEOCARBSULF can reproduce these scenarios by varying the weathering/degassing fluxes of carbon and sulfur, or carbonate δ13C. We show that GEOCARBSULF can, in fact, maintain low-O2 (even 1–5% atm) levels through the early Paleozoic by only varying the carbonate δ13C within 2 standard deviation (SD) bounds permitted by the geological record. In addition, it can generate a middle–late Paleozoic rise in O2 concentration, coincident with the diversification of land plants. However, we also argue that tracking atmospheric O2 levels with GEOCARBSULF is highly dependent on carbonate carbon isotope evolution, and more accurate predictions will come from an improved C isotope record.

Publication
American Journal of Science
Avatar
Shuang Zhang
Postdoc Fellow

Shuang Zhang loves exploring the global carbon cycle and climate change through data analysis and machine learning.