First screening was for T-S in 5-degree squares to eliminate duplicate stations in historical database. Stations (samples) whose deep T-S points fell outside ±2 standard deviation envelopes were rejected (flagged). About 25% of stations failed this first stage, and about 30,000 T-S outliers were flagged.

Oxygen and nutrients were examined by cruise, using control area data as guides. Several thousand stations had entire profiles of oxygen or nutrients flagged for poor quality. For the acceptable stations, about 10,000 individual samples were flagged as outliers. A few hundred cruises were identified that had systematic offsets in at least one property, and additive or multiplicative adjustments were applied to the nearly 11,000 stations affected.

An example of quality improvement in the Southern Ocean Atlas Database is shown for the eastern Atlantic and western Indian sectors of the Southern Ocean.

Station positions on the map (any color) represent the 2,492 stations in the World Ocean Database of 1998 (WOA98) that NODC has flagged acceptable for T-S.

Red stations passed our quality control. They include 77% of the WOA98 stations, but add 266 stations that were flagged bad by NODC. WOCE data and stations in our local database bring the total to 1,922 red stations.

The pairs of scatter plots shown below correspond to WOA98 and Southern Ocean Database data. Plus and minus one-standard deviations around the density-averaged mean temperature-salinity, oxygen, silicate, phosphate and nitrate profiles are shown as red curves.

Cyan stations in WOA98 are the WOCE Pacific S4 data with incorrect longitudes. Black dots are stations in the Weddell and Scotia seas and Bransfield Strait, also with incorrect longitudes. The green stations are CTD data that we found were inaccurately calibrated to bottle salinity samples. The deep data are saltier than any measured in the region. T-S points near -1° are clearly anomalous to the Enderby Basin, but the others look remarkably "Indian-like", and might be difficult to detect statistically.

Standard deviations for the Southern Ocean Database scatter plots are generally 2-3 times smaller than for the WOA98 data. Although outliers remain in our data, the mean property curves begin to show maxima and minima whereas the WOA98 curves are generally shapeless.