Oceanic oxygen: a diffusion model fills the gaps in climate models using only temperature and salinity

Linus Vogt and Laure Zanna generate oxygen fields for models that do not calculate them, and suggest that biases in the oxygen-poor region of the Pacific are overestimated.

Dissolved oxygen conditions marine ecosystems and major biogeochemical cycles, but models of the Earth system diverge significantly on the quantity present in the ocean, both today and in the future. Some of these models do not simulate oxygen at all.

Learn the relationship between physics and oxygen

Linus Vogt and Laure Zanna trained a generative conditional diffusion model on the outputs of a set of climate models. It learns the distribution of oxygen in the upper layers of the ocean from physical variables such as temperature and salinity. The model performs well in the Atlantic and Southern Oceans, produces realistic fields under conditions not included in training, and has been validated on observation sets.

The authors use it to reconstruct oxygen from models that do not provide any. Notable result: by broadening the range of ocean states taken into account, the biases of the models in the tropical Pacific oxygen minimum zone could be weaker than previously thought. The work was submitted to arXiv on October 6.