The ocean oxygen (O2) inventory has declined in recent decades but the estimates of O2 trend is uncertain due to its sparse and irregular sampling. A refined estimate of deoxygenation rate is developed for the North Atlantic basin using machine learning techniques and biogeochemical Argo array. The source data includes 159 thousand historical shipboard (bottle and CTD-O2) profiles from 1965 to 2020 and 17 thousand Argo O2 profiles after 2005. Neural network and random forest algorithms were trained using 80% of this data using different hyperparameters and predictor variable sets. From a total of 240 trained algorithms, 12 high performing algorithms were selected based on their ability to accurately predict the 20% of oxygen data withheld from training. The final product includes gridded monthly O2 ensembles with similar skills (mean bias < 1mol/kg and R2 > 0.95). The reconstruction of basin-scale oxygen inventory shows a moderate increase before 1980 and steep decline after 1990 in agreement with a previous estimate using an optimal interpolation method. However, significant differences exist between reconstructions trained with only shipboard data and with both shipboard and Argo data. The gridded oxygen datasets using only shipboard measurements resulted in a wide spread of deoxygenation trends (0.8-2.7% per decade) during 1990-2010. When both shipboard and Argo were used, the resulting deoxygenation trends converged within a smaller spread (1.4-2.0% per decade). This study demonstrates the importance of new biogeochemical Argo arrays in combination with applications of machine learning techniques.

Earth System Models project a decline of dissolved oxygen in the oceans under warming climate. Observational studies suggest that the ratio of O2 inventory to ocean heat content (O2-OHC) is several fold larger than can be explained by solubility alone, but the ratio remains poorly understood. In this work, models of different complexity are used to understand the factors controlling the O2-OHC ratio during deep convection, with a focus on the Labrador Sea, a site of deep water formation in the North Atlantic Ocean. A simple one-dimensional convective adjustment model suggests two limit case scenarios. When the near-surface oxygen level is dominated by the entrainment of subsurface water, surface buoyancy forcing, air-sea gas exchange coefficient and vertical structure of sea water together affect the O2-OHC ratio. In contrast, vertical gradients of temperature and oxygen become important when the surface oxygen flux dominates. The former describes the O2-OHC ratio of individual convective event in agreement with model simulations of deep convection. The latter captures the O2-OHC ratio of interannual variability, where the pre-conditioning of interior ocean gradients dominates. The relative vertical gradients of temperature and oxygen, which in turn depend on the lateral transport and regional biological productivity, control the year-to-year variations of O2-OHC ratio. These theoretical predictions are tested against the output of a three-dimensional regional circulation and biogeochemistry model which captures the observed large-scale distribution of the O2-OHC ratio, and agrees broadly with the prediction by the simpler model.

Marine ecosystem is influenced by multiple factors under the global warming. Increasing temperature raises the matabolic demand for oxygen, however, its supply declines as the concentration of dissolved oxygen decreases. Metabolic index is defined as the oxygen supply to demand ratio for a marine organism, quantitatively combining the effects of ocean warming and deoxygenation. A subset of the earth system models participating in the Coupled Model Intercomparison Project phase 6 (CMIP6) are used to calculate the century-scale changes in the metabolic index under three different scenarios. Under the most aggressive warming scenario, metabolic index can decline over 50\% at northern extratropics including the US west coast. Overall magnitude of the change is dependent on the emission scenarios, whereas spatial patterns are model-dependent, in particular at low latitudes due to the large uncertainty in projected oxygen changes.

Phytoplankton growth in the Indian Ocean is limited by nitrogen and phosphorus in the north and by iron in the south. Increasing anthropogenic atmospheric deposition of nitrogen and dissolved iron (dFe) into the ocean can thus lead to significant responses from the Indian Ocean ecosystems. Previous modeling studies investigated the impacts of anthropogenic nutrient deposition on the ocean, but their results are uncertain due to incomplete representations of the Fe cycling. This study uses a state-of-the-art ocean ecosystem and Fe cycling model to evaluate the transient responses of ocean productivity and carbon uptake in the Indian Ocean, focusing on the centennial time scale. The model includes three major dFe sources and represents an internal Fe cycling modulated by scavenging, desorption, and complexation with multiple, spatially varying ligand classes. Sensitivity simulations show that after a century of anthropogenic deposition, increased dFe input stimulates diatom in the southern Indian Ocean poleward of 50S and the southeastern tropics. However, diatom decreases in the southern Arabian Sea due to the phosphorus limitation, and diatom is outcompeted there by coccolithophores and picoplankton, which have a lower phosphorus demand. These changes in diatom and coccolithophores productions alter the balance between the organic and carbonate pumps in the Indian Ocean, increasing the carbon uptake poleward of 50 S and the southeastern tropics while decreasing it in the Arabian Sea. Our results reveal the important role of ecosystem dynamics in controlling the sensitivity of carbon fluxes in the Indian Ocean under the impact of anthropogenic nutrient deposition.