15 
(expert) quality control. In areas with sparse data coverage systematic errors often manifest 
themselves as strong local features, which are identified by a proper control of property dis 
tribution maps or temperature-parameter diagrams. Substantial improvements in the 
observational techniques and methods reduced significantly measurement errors, but, on the 
other hand, stressed the importance of estimating systematic differences between the data 
when merging data of different origin. Differences in observational techniques and methods 
are the main cause for systematic errors (offsets) in the data, as soon as a composite data 
base is considered. CTD-systems have increasingly replaced Nansen bottles since the mid- 
1960s, accompanied by a replacement of the titration method by conductive salinometers. 
Manual methods for nutrient determinations have been replaced by automated methods, and 
the method for titrating oxygen samples changed in the late 1950s. A brief description of the 
possible causes for inter-cruise offsets is given below. 
Salinity. Among commonly measured oceanographic parameters salinity is the only parameter 
whose measurement can be referenced to a common standard, i.e. IAPSO Standard Seawater 
(SSW). However, a number of factors lead to systematic errors in salinity measurements such 
as (1) different bottle types, (2) time lag between water sampling and salinity determination, (3) 
offsets between different batches of IAPSO Standard Sea Water, (4) salinometer response shift, 
(5) differences between up and down cast values because of the hysteresis in pressure, 
temperature or conductivity sensors. 
Oxygen. The method for titrating oxygen samples changed in the late 1950s. Though some 
corrections to the older data have been proposed (Worthington, 1976; Gordon and Molinelli, 
1982) problems with oxygen data forced Lozier et al. (1995) to eliminate pre-1960 data 
completely from the database. Culberson et al. (1991) compared results of four scientific groups 
and noted two main sources of errors in oxygen determination: (1) the concentration of dissolved 
oxygen in the reagents and (2) the value of the seawater contribution to the blank. The analytical 
methods used to determine dissolved oxygen employ volumetric techniques and give the 
amount of oxygen per unit volume of seawater. Transfer to the weight concentration requires 
knowledge of the temperature of the seawater at the time of sampling and it was not routinely 
measured in the past. According to Culberson et al (1991) a 25°C difference between the 
sampling and assumed temperatures may result in 0.5% error in the weight oxygen 
concentration. 
Nutrients. Problems with the unsatisfactory quality of nutrient data are also well known. Nutrient 
scatter diagrams usually exhibit rather dispersed clouds of points, where other more precisely 
measured parameters reveal tight temperature-parameter relationships. One of the major 
concerns with respect to the nutrient data is whether measurements made using manual 
methods may be combined with measurements made using automated methods. Another 
source of discrepancies was found to be due to problems in the standardisation procedures 
used by different laboratories. A thorough treatment of random errors and biases in nutrient 
determinations was given by Holley (1998), who divides factors affecting precision and accuracy 
of nutrient measurements into four problem areas: 
(1) Instrument, mechanical and chemical factors. These include refractive index problems, 
bubble production and its effect on mixing, the contamination of a sample, drift within a run 
due to changes in the reagent, and differences in the sample wash time. 
(2) Standardisation. The working calibration standards differ between laboratories by type, 
supplier, batch, quality, and preparation technique. Standards prepared in low nutrient sea 
water are less stable than in distilled or artificial sea water due to the organisms present. 
3.4. Calculation of inter-cruise offsets 
A very important improvement relative to the WOA94 and WOD98 and WOA01 climatologies 
is the treatment of systematic errors in the data. Such error (or biases) were determined 
through the analysis of the inter-cruise property offsets. The method was successfully