Horizontal maps To describe the spreading of water masses within the Southern Ocean, distributions of potential temperature, salinity, neutral density, isopycnal depth, oxygen, nitrate, phosphate and silicate are shown along a small number of density surfaces and depth levels. No plots for CFC-11, total CO2, alkalinity, δ3He, tritium or Δ14C are included because of inadequate sampling. The number of maps that can be presented in this printed atlas is necessarily fewer than required to fully describe Southern Ocean properties, even within the major water masses. Because the important water masses differ from one ocean to another, the choice of levels is not always consistent among the atlas volumes. Depth levels for the Southern Ocean are 50 m, 200 m, 800 m, 1500 m, 2500 m, 3500 m and within 200 m of the sea floor for water depths greater than 3500 m.
Five isopycnal surfaces were selected to portray the characteristic water masses in the Southern Ocean. The 27.40 kg/m3, 27.84 kg/m3 and 28.05 kg/m3 isopycnals correspond to the salinity minimum in the Antarctic Intermediate Water, the oxygen minimum in the Upper circumpolar Deep Water and the salinity maximum in the Lower Circumpolar Deep Water, respectively. The underlying 28.20 kg/m3 and 28.27 kg/m3 isopycnals span the Lower Circumpolar Deep Water and the top of the Antarctic Bottom Water. Additional level and isopycnal maps are available from the on-line Southern Ocean Atlas (http:/woceSOatlas.tamu.edu).
Color breaks on horizontal maps are chosen to show clearly the spreading of waters along the different levels. Color ranges are given in the individual plates. A Polar Lambert Azimuthal Equal-Area projection (Snyder, 1989) is used for the Southern Ocean Atlas. The horizontal maps include all WOCE data (which are the most reliable) but these alone are spatially too sparse to provide the distribution needed.
Over 122,000 hydrographic stations in the region south of 25°S were individually screened against WOCE standards. They were compiled over the years from a number of data sources, but the vast majority were obtained from the National Oceanographic Data Centre (see http:/www.nodc.noaa.gov/), specifically their World Ocean Database (WOA98; WOA2001; Conkright et al., 2002).
The quality control process began with the examination of WOCE property profiles, supplemented in some regions with high quality historical data. Fifty control regions, generally on a basin or sub-basin scale, were identified as having distinct property differences in deep water. Properties of stations in the control regions were averaged on neutral density surfaces and standard deviation envelopes constructed. The first examination was done in Θ-S space for stations within 5-degree squares. Stations (samples) whose deep Θ -S points fell outside +/-2 standard deviations from the deep-water control region mean were rejected (flagged). Roughly 25% of the stations did not pass this first stage.
In the second stage, oxygen and nutrient profiles were examined on a cruise-by-cruise basis, using the appropriate control area data to judge acceptability. Several thousand stations have had entire profiles of oxygen, phosphate, silicate and nitrate flagged because of poor quality. For the acceptable stations, about 0.5% of the 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. The last step in the quality control process was to correct or assign missing water depths to about a quarter of the stations using a 5-minute grid composite of the Southern Ocean bathymetry from satellite radar altimetry data (Smith and Sandwell, 1997) and the GEBCO-1997 (IOC et al., 1997) digital isobaths.
A polar equal-area grid with 501 by 501 points about 24 km apart was designed to optimally estimate the initial property fields shown in this atlas. The mapping of tens of thousands of oceanographic data points onto hundreds of thousands of grid points without any loss of meaningful information requires specifying correlation lengths as a function of location that are tightly related to the underlying ocean bathymetry. Such control in the optimal estimation relies heavily on the notion that smaller scale variability, such as mesoscale rings and frontal fluctuations, tend to occur at places where energetic ocean currents interact with relatively shallow topography. This is evident in the Southern Ocean where the Antarctic Slope Current interacts with the irregular bathymetry of the continental shelf-slope regime and where the Antarctic Circumpolar Current interacts with the various ocean ridges found along its path.
Because the currents within the interior of the Southern Ocean are mainly zonal, the ellipses of influence were aligned in that direction. A fixed 2:1 anisotropy in the zonal:meridional correlation lengths was adopted at each grid point.
The reliability of optimal estimation depends on the spatial density of the available observations. Ellipses of influence that are suitable for temperature or salinity contain too few data points to adequately map sparsely sampled nutrient data. For non-nutrient data, ellipse sizes decrease from 666 km by 333 km at grid points in the oceanic regime (bottom depths greater than 4000 m), to 444 km by 222 km over the ocean ridges (1000-4000 m depths), to a minimum of 222 km by 111 km within the continental slope regime (depths shallower than 1000 m). For nutrient maps, ellipse sizes are double these values.
Property distributions are portrayed in two ways. Each of the eight shades of color occupy roughly one-eighth of the geographical area covered by the maps, providing a visualization of how the properties vary in space. The proportionality is nearly exact for shallow maps, and less precise for deeper maps where large regions are masked out where depth and density maps intersect the bathymetry. The color distribution is supplemented by labelled contour lines at more standard intervals.
Unlike the vertical sections, which consist entirely of high-quality WOCE stations, the property maps have not been edited by hand. Most of the poor historical stations were eliminated in the initial quality control, and it is not feasible (or in many cases, possible) to identify anomalies among 94,000 stations and determine whether they are due to natural variability or measurement error. Therefore, the maps are presented as contoured by the computer. Although some cosmetic corrections have been applied, such as removing contours poorly supported by the data distribution, other less obvious discrepancies remain, such as the shape of contours that are seemingly unsupported by data.