Bias Evaluation for Sensor-based Dissolved Oxygen from CTD and Profiling Floats in the World Ocean Database

Sensor-based O 2 measurements in the ocean have become the dominant data source for in situ O 2 since 2010. We examine the overall quality of the shipborne conductivity-temperature-depth (CTD)- and Argo-measured O 2 in the World Ocean Database (WOD) and evaluate biases by comparing them to independently-measured reference bottle-sampled profiles. No significant O 2 difference is found between CTD and bottle data for the global ocean, suggesting high quality of CTD O 2 observations generally. A negative residual bias is found in both post-corrected delayed-mode (−1.69±5.15 μmol kg −1 , fitted mean±standard deviation) and real-time adjusted (−4.68±6.99 μmol kg −1 ) Argo O 2 . Residual biases have both spatial (basin-scale and vertical) and temporal variations, indicating the complexity of residual biases. We also examine Argo O 2 biases based on the type of calibration methods. For the Argo delayed-mode O 2 profiles calibrated using World Ocean Atlas (WOA), on average, are offset by −1.60±5.30 μmol kg −1 , while a larger bias (−3.29±4.86 μmol kg −1 ) is found for profiles calibrated using in-air measurements for the global ocean. The Argo delayed-mode data calibrated with O 2 profiles from the WOD have a slight positive bias (0.33±4.16 μmol kg −1 ). Further analysis demonstrates that the O 2 profiles without matched reference profiles also have negative bias similar to those with matched reference profiles when they are compared to WOA. This analysis suggests the negative residual biases in Argo post-corrected O 2 should be carefully considered when using Argo O 2 data to draw conclusive inferences about anthropogenic impacts on ocean oxygen concentrations and/or derived quantities in a changing ocean.