18 
3.6 Calculation of biases for historical data-set 
Whereas biases for the reference cruises were obtained by solving (2), biases for historical 
cruises were calculated relative to the reference cruises, corrected for systematic errors. 
Year 
60 
40 
20 
0 
-20 
-40 
-60 
-80 
1920 1930 1940 1950 1960 1970 1980 1990 2000 
Year 
Fig. 11 Cruise biases versus time (red - reference cruises, blue - historical cruises). 
Much fewer deep water samples are typically available for the historical data. In order to get a 
sufficient number of samples for statistically meaningful estimates, the maximum station 
separation was increased to 400 km (compared with 300 km for the high-quality data set) with 
the minimum of 10 station pairs, required for the offset estimate. For each historical cruise 
property biases were calculated as the average of individual offsets relative to reference 
stations within the respective crossover area. For all parameters historical cruise biases 
exceed considerably those for the cruises of the reference data-set (Fig.10).