Articles | Volume 1, issue 1
23 Dec 2009
 | 23 Dec 2009

CARINA oxygen data in the Atlantic Ocean

I. Stendardo, N. Gruber, and A. Körtzinger

Abstract. In the CARINA (Carbon dioxide in the Atlantic Ocean) project, a new dataset with many previously unpublished hydrographic data from the Atlantic, Arctic and Southern Ocean was assembled and subjected to careful quality control (QC) procedures. Here, we present the dissolved oxygen measurements in the Atlantic region of the dataset and describe in detail the secondary QC procedures that aim to ensure that the data are internally consistent. This is achieved by a cross-over analysis, i.e. the comparison of deep ocean data at places that were sampled by different cruises at different times. Initial adjustments to the individual cruises were then determined by an inverse procedure that computes a set of adjustments that requires the minimum amount of adjustment and at the same time reduces the offsets in an optimal manner. The initial adjustments were then reviewed by the CARINA members, and only those that passed the following two criteria were adopted: (i) the region is not subject to substantial temporal variability, and (ii) the adjustment must be based on at least three stations from each cruise. No adjustment was recommended for cruises that did not fit these criteria. The final CARINA-Oxygen dataset has 103414 oxygen samples from 9491 stations obtained during 98 cruises covering three decades. The sampling density of the oxygen data is particularly good in the North Atlantic north of about 40° N especially after 1987. In contrast, the sample density in the South Atlantic is much lower. Some cruises appear to have poor data quality, and were subsequently omitted from the adjusted dataset. Of the data included in the adjusted dataset, 20% were adjusted with a mean adjustment of 2%. Due to the achieved internal consistency, the resulting product is well suited to produce an improved climatology or to study long-term changes in the oxygen content of the ocean. However, the adjusted dataset is not necessarily better suited than the unadjusted data to address questions that require a high level of accuracy, such as the computation of the saturation state.