Overall consistency of Delayed Mode corrections

We have checked 578 floats processed in delayed mode (DM) in the North Atlantic, North of 30°N.  Among the 578 floats, 392 do not show a significant salinity drift or bias according to the PI decision and were not corrected in DM, the other 186 floats have been corrected. 

For each float, we have run a slightly modified OW method. Compared to the OW original method, our configuration better take into account the interannual variability, that was shown to induce spurious corrections with the standard OW method settings. Particularly, a Gaussian decay was added with a time scale of 2 years when computing the covariance matrix that is used to estimate the large scale field at the float profile position. Thus it is given greater weight to contemporary reference data. The original OW method takes into account the temporal variability but only when the small scale field is estimated and the large scale field is assumed to be constant. Moreover, the modified OW method provides an improved estimate of the error bars. The modified OW method is described in more details here . 

For each float, we have compared the original correction made by the PI and the result of the slightly modified OW method. We found 26 floats among 578 (see table below) for which the salinity correction proposed by the PI differs significantly from our results. For those floats, we think it may be necessary to revise the original correction. For plots showing the comparison between the original DM correction and our results for these 26 floats, see report here

Anomalies detected at NA-ARC Oct 2014