40 
8. Error estimates for the observed and gridded data 
8.1. Observed data 
WOCE hydrographic data comprise the main part of the reference dataset used in this 
study. For the first time a uniformly accurate dataset was obtained for the Global Ocean, with 
the quality standards for water samples set by the WOCE Hydrographic Program (WOCE 
Operations Manual, 1991). An objective method was applied to the almost complete set of 
WOCE hydrographic cruises to calculate inter-cruise offsets (Gouretski and Jancke, 2001). It 
was shown, that quality requirements for the WOCE Hydrographic Programme have been 
obviously fulfilled (Table 2). 
Table 2. WHP one-time survey standards for water samples 
Property 
WOCE standard 
(accuracy / precision) 
(from WHP Manual, 1994) 
Average WOCE 
inter-cruise offset 
(Gouretski&Jancke, 2001) 
Temperature 
0.005/0.002 °C 
Salinity 
0.002/0.001 
0.0019 
o 2 
~1% / 0.1% 
-1.4% 
N0 3 
~1% / 0.2% 
-1.6% 
'n|- 
O 
0. 
-1-2% / 0.4% 
-1.9% 
Si0 2 
-1-3% /0.2% 
-2.6% 
Accuracy of historical hydrographic data varies strongly between the cruises with possible 
errors being generally an order of magnitude higher compared with the reference data. 
8.2. Gridded data 
A formal squared absolute objective interpolation error is : 
2 2 >2 
8 = C 8 
where c 2 is a property variance and s' is the relative error of the objective interpolation. 
Absolute formal errors are largest in the upper layers, where property variability is strong, 
and decrease rapidly with depth as shown in the Fig. 27. We note that the error estimate is 
inaccurate, because the "true" covariance function is unknown (a useful discussion is given 
by Sokolov and Rintoul, 1999)