Underestimation of global O ₂ loss in optimally interpolated historical ocean observations

Abstract. The global ocean’s oxygen content has declined significantly over the past several decades and is expected to continue decreasing under global warming with far reaching impacts on marine ecosystems and biogeochemical cycling. Determining the oxygen trend, its spatial pattern and uncertainties from observations is fundamental to our understanding of the changing ocean environment. This study uses a suite of CMIP6 Earth System Models to evaluate the biases and uncertainties in oxygen distribution and trends due to sampling sparseness. Model outputs are sub-sampled according to the spatial and temporal distribution of the historical shipboard measurements, and an optimal interpolation method is applied to fill data gaps. Sub-sampled results are compared to full model output, revealing the biases in global and basin-wise oxygen content trends. The optimal interpolation underestimates the modeled global deoxygenation trends, capturing approximately two-thirds of the full model trends. North Atlantic and Subpolar North Pacific are relatively well sampled, and the optimal interpolation is capable of reconstructing more than 80 % of the oxygen trend. In contrast, pronounced biases are found in the equatorial oceans and the Southern Ocean, where the sampling density is relatively low. Optimal interpolation of the historical dataset estimated the global oxygen loss of 1.5 % over the past 50 years. However, the ratio of global oxygen trend between the subsampled and full model output, increases the estimated loss rate to 1.7 to 3.1 % over the past 50 years, which partially overlaps with previous studies. The approach taken in this study can provide a framework for the intercomparison of different statistical gap-fill methods to estimate oxygen content trends and its uncertainties due to sampling sparseness.