10 
3. Data Quality Control 
3.1 Cruise identification 
The quality analysis procedure used in the study requires a cruise identification of the data. 
Unfortunately, in many cases the NODC collections still do not provide information to link the 
data to a particular oceanographic cruise occupied by one and the same ship. The historical 
profiles of the composite data set, selected for the analysis are distributed among about 10000 
NODC archive codes, which often are linked to the data from many different cruises of one and 
the same ship. Within each NODC archive cruise number all profiles were ordered by time, 
and a time demarcation between a pair of new ’’cruises” was set as soon as the time span 
between two consecutive stations exceeded 7 days. This time separation criterion was set 
rather arbitrarily, and it does not exclude the possibility when the time separation of 7 or 
more days occurred within one and the same hydrographic cruise, or when the gap between 
the two cruises was less than a week. As a result, all selected hydrographic profiles were 
ascribed to a total of 41757 new cruises. 
3.2 Random errors 
Evolution of measuring technique and methods along with very different quality standards 
caused a high degree of inhomogeneity of the historical hydrographic data set and invoked a 
large literature devoted to the problems of quality control of oceanographic data. Most of the 
quality control procedures (Levitus et al., 1994; Olbers et al., 1992; Curry, 1996; Gouretski and 
Jancke, 1999) were aimed to identify random errors in the data. 
The quality evaluation of the composite dataset used in this study benefited from the fact that 
all of the source data had already been validated to a certain degree. A description of the 
validation procedure applied to the WOA01 data was given by Levitus et al. (1994) and 
Conkright et al. (1994). All WOCE data have been checked for their quality both by 
respective principal investigators and (in many cases) by independent experts. However, the 
investigation of the historical hydrographic data quality for the South Atlantic (Gouretski and 
Jancke, 1995) and for the North Atlantic (Lozier et al., 1995) showed that some highly 
questionable data from the WOD98 database obviously passed through the quality checks 
implying the necessity of a more rigorous quality control. 
In order to further validate the data we used a method developed by Gouretski and Jancke 
(1999) and tested for the South Pacific historical data set. Quality checking is done in the 
density-parameter space. The method is based on the experimental fact that relations 
between potential temperature (or density) and other parameters are locally well defined in 
the World Ocean and are relatively tight below the thermocline level. The density-parameter 
curves for a group of neighbour stations is approximated by vertical subdivision into small 
density bins and connection of mean points for each bin with straight lines. Then mean 
values and standard deviations of parameters from the mean curve within each bin are 
computed and any value differing by more than some prescribed number of standard 
deviations (2.5 in our case) from the mean curve is rejected (flagged).