Error analysis of CO₂ and O₂ estimates from the long-term geochemical model GEOCARBSULF

Abstract

Long-term carbon and sulfur cycle models have helped shape our understanding of the Phanerozoic history of atmospheric CO₂ and O₂, but error analyses have been largely limited to testing only a subset of input parameters singly. As a result, the full ranges of probable CO₂ and O₂ are not quantitatively known. Here we investigate how variation in all 68 input parameters of the GEOCARBSULF model, both singly and in combination, affect estimated CO₂ and O₂. We improve formulations for land area, runoff, and continental temperature, the latter of which now excludes land area not experiencing chemical weathering. We find our resampled model CO₂ and O₂ estimates are well bounded and provide high confidence for a “double-hump” in CO₂ during the Phanerozoic, with high values during the early Paleozoic and Mesozoic, and low values during the late Paleozoic and late Mesozoic-to-Cenozoic. Our analyses also support a distinct atmospheric O₂ peak during the late Paleozoic (>30%) followed by low values near the Triassic-Jurassic boundary (∼10%). Most of the spread in CO₂ is contributed by three factors: climate sensitivity to CO₂-doubling and the plant-assisted chemical weathering factors LIFE and GYM. CO₂ estimates during the Paleozoic to early Mesozoic are highly concordant with independent records from proxies, but are offset to lower values during the globally warm late Mesozoic to early Cenozoic. The model-proxy mismatch for the late Mesozoic can be eliminated with a change in GYM within its plausible range, but no change within plausible ranges can resolve the early Cenozoic mismatch. Either the true value for one or more input parameters during this interval is outside our sampled range, or the model is missing one or more key processes.