Multi-model ocean oxygen fields predicted by conditional diffusion models

Dissolved oxygen is important for the ocean's ecosystems and biogeochemical cycles. Yet, Earth System Models (ESMs) vary in their simulations of the present-day and future ocean oxygen inventory. To narrow down the uncertainty in estimates of ocean oxygen content, we train a conditional generative diffusion model on outputs of a multi-model ESM ensemble to learn the conditional distribution of upper-ocean oxygen given physical input variables such as temperature and salinity. This generative model has considerable skill in the Atlantic and Southern Oceans, and can generate realistic oxygen sam…