Figure 4 investigates the magnitude of the AAIW variability found in the Argo float observations in more detail. In this figure the 171 floats are sorted according to the standard deviation of their associated WOA-based time series. While the WOA-derived time series show a slow increase of the standard deviations from near-zero to about 0.005 for the first 140 floats followed by a rapid increase to 0.03, the Argoderived standard deviations are more uniform and closer to 0.006 before following the steep rise and show several particularly large values. It should be noted that values below 0.003 are below the resolution of the salinity algorithm. The low values for the WOA-derived time series are the result of floats passing through a highly smoothed salinity field. Most Argo-derived values, on the other hand, are clearly larger than the sensor resolution and do not represent mere instrumental noise. The Argo team guarantees an accuracy of 0.01 for delayed mode quality controlled data, despite the fact that the instrument stability is often closer to 0.003 (Wijffels, personal communication), but Oka (2005) estimates sensor drift at as low as 0.004 per year, based on recalibration of recovered floats. About half the floats used here have a standard deviation less than or close to the accepted accuracy if the larger Argo accuracy of 0.01 is used as a yardstick