3. Procedures to Determine Offsets
The quality of all data used in the analyses was extensively controlled by the investigators responsible for the measurements. Therefore, significant care was taken to avoid suggesting erroneous corrections. This is why different and largely independent approaches were used to look for biases. Adjustments were recommended only if there were clear-cut and consistent differences of greater than 4 µmol/kg for DIC and 6 µmol/kg for TAlk. The cutoffs correspond to about twice the precision of the measurements performed on the cruises based on replicate and deep-water comparisons. Issues of data variability with depth and/or station-to-station variability were more frequent than clear-cut offsets. Sometimes the variability was large enough to question the integrity and utility of the cruise for the purposes of large-scale integration.
Several methods were used to assess the consistency among DIC and TAlk data from the cruises. For the first method, cruise-to-cruise comparisons of measurements were performed at crossover points for deep water (>1500 m). The assumption was that variability in deep water would be very small and that the anthropogenic signal, which would complicate the comparison of cruise data spanning eight years, would be negligible. The analysis was performed in density space to avoid biases due to movement of water masses. The method incorporated the implicit assumption that there were no systematic biases with density over the time period. Several recent investigations have shown changes in properties throughout the water column, but limited time-series work suggests that the changes are smaller than the level of agreement we are striving for (4 µmol/kg for DIC and 6 µmol/kg for TAlk). For instance, the time series station data at Bermuda show constancy in deep-water DIC values to within 3 µmol/kg (Fig. 4) (Nick Bates, personal communication).
In total, 53 crossovers were investigated. The breakdown of carbon parameters compared at the crossovers is shown in Table 3. As the table indicates, 20 of the 53 total crossovers had DIC as a common parameter for the junction of cruises; 16 had comparable TAlk and DIC data on both cruises. For 17 of the crossovers, three or four carbon system parameters were measured on both cruises.
The second method was used for cruises that covered a similar cruise track. It involved an MLR of DIC or TAlk with parameters known to influence DIC or TAlk levels and/or those that are known to regress with DIC or TAlk. The measurements for the overlapping cruises were combined for depths generally greater than 1500 m, and the regression coefficients were determined for DIC or TAlk with T, S, SiO2, apparent oxygen utilization (AOU), and NO3. The calculated DIC or TAlk values from this MLR were then compared with the measured values to determine systematic differences among cruises. This method of comparison was applied to ten cruises that overlapped in space.
The third method involved assessment of internal consistency for cruises in which three or more inorganic carbon system parameters were measured. In this approach, the thermodynamic relationships of Mehrbach et al. (1973) as refit by Dickson and Millero (1987) were used to calculate DIC from TAlk and fCO2 or pH (or to calculate TAlk from DIC and fCO2 or pH). The calculated values were then compared with the measured values. When significant differences are found between calculated and measured values, this approach does not a priori establish which of the three parameters is in error, but the results can be used with the other methods to identify the culprit. Ten cruises had three or four inorganic carbon system parameters for which this approach could be applied.
The fourth method used the same MLR technique as was used in the overlapping cruise comparisons, but it was applied to cruises in a particular region. The regression fit was determined by using the cruises that had met the crossover criteria in a particular region for depths greater than 1500 m. The relevant data from all cruises were separated into three regions with distinctly different fits for the regressions: (1) The subtropical and polar regions north of 15° N, (2) the equatorial region between 15° N and 15° S, and (3) the subtropical and polar region south of 15° S. The DIC and TAlk data for the individual cruises within each region were then compared with the calculated values from the corresponding regressions.
