Global ocean gridded dataset and hypoxic zone expansion: reconstructed dissolved oxygen (1960–2021) based on machine learning technique

Due to global warming and the excessive input of nutrients resulting from human activities, ocean deoxygenation is gradually intensifying. However, the sparse spatiotemporal distribution of in situ global dissolved oxygen (DO) observations poses a significant challenge to understanding the spatiotemporal variations of DO in the world’s oceans. In this study, we employed a SOM-FFNN approach to reconstruct a global monthly dissolved oxygen dataset from 1960 to 2021, with high spatial (1° × 1°) and vertical (to 2000 m) resolution. Compared to existing products, our model exhibits a lower RMSE (15.36 μmol/kg), a higher R² (0.96), and strong agreement with long-term observations, outperforming other available datasets. Analysis of this dataset shows that the global ocean has been experiencing a steady loss of total oxygen content since the late 1980s, with the rate of depletion over the past decade nearly twice as high as that during 1980–2010. Further investigation reveals that low-oxygen zones have expanded both horizontally and vertically, with this expansion also intensifying in the past decade. These findings highlight the accelerating nature of ocean deoxygenation and the growing extent of low-oxygen habitats.