3. RESULTS AND DISCUSSION

3.1 Statistical Methods
Tables 1 and 2 summarize the crossover sites and parameters measured, and Tables 3, 4, 5, and 6 are summaries of the statistical data for each parameter at the crossover locations. Eleven laboratories from two countries participated in this comparison study that examines crossovers in both the North and South Pacific. At some of the crossover locations, the site was occupied on more than one occasion [i.e., the crossover at 170?W and 10?S was frequented by NOAA on three different cruises (CGC90, EqS92, and P15S), as well as by the Institute of Ocean
Science (IOS) (P15N) and the University of Hawaii (UH) (P31)]. A total of 30 crossover locations were studied in this analysis and 41 individual crossover comparisons were made. Individual plots of each carbon parameter, along with salinity and O2, were first created for every crossover against  using data from the entire water column (Appendix A). Only data sets that showed good agreement in both salinity and O2 data were used for the comparisons. An expanded area within the plot was examined further based on the region of reasonable agreement of the  vs salinity plot. In most cases,  > 27.0 was used in the expanded regions.

A curve-fitting routine was applied to the expanded plots (Appendix A) using a second-order polynomial fit (unless otherwise noted in Tables 3, 4, 5, and 6). The difference between each region of crossover was calculated based on evenly distributed intervals on the  axis; the intervals chosen were, on average, 0.04  units apart. In the case where more than one station on a given cruise was computed at a particular crossover location, averages of the resulting fits of the two or more stations for that cruise were determined, and the total mean of the differences over the entire  range was compared. This procedure was performed for every carbon parameter measured (Tables 3, 4, 5, and 6). The mean and standard deviation of the differences were computed, along with the mean and standard deviation of the absolute value of the differences. For the DIC data, the results were calculated both uncorrected and corrected using the CRMs as a basis for the corrections.

3.2 Cruise Results
The most detailed carbon parameter results are for DIC, as this parameter was measured on all of the cruises (Table 3). The next most frequently measured parameter was fCO2, followed by TAlk and pH (Tables 4, 5, and 6), respectively.
DIC CRMs were available to the investigators for almost every cruise during the survey. In general, there is excellent agreement between DIC data sets at the crossover locations. At the beginning of the program, the goal was to obtain
agreements between cruises that were less than 4.0 µmol/kg. On 31 of 41 crossover comparisons the uncorrected DIC differences were less than this value, and on 24 of the comparisons the differences were less than 2.0 µmol/kg.

Most of the cruises that did not meet this criteria occurred at the beginning of the program when methods were still being developed, and one comparison was during a strong El Niño event where the upper water column hydrography was
significantly different from normal (Feely et al. 1995). When the DIC data were corrected for CRMs, 36 of the 41 comparisons were less than 4.0 µmol/kg, and 31 comparisons were less than 2.0 µmol/kg. The mean of the absolute value of the differences was 2.4 ± 2.8 µmol/kg for the uncorrected data and 1.9 ± 2.3 µmol/kg
for the corrected data (Fig. 2). For a mean DIC concentration of approximately 2260 µmol/kg in the deep Pacific, this difference is equivalent to an uncertainty of approximately 0.08%. The excellent agreement of the DIC data was
likely due primarily to the use of the coulometer (UIC, Inc.) coupled with a SOMMA (Single Operator Multiparameter Metabolic Analyzer) inlet system developed by Ken Johnson (Johnson et al. 1985, 1987, 1993; Johnson 1992) of Brookhaven National Laboratory (BNL), as well as the use of CRMs as secondary standards during the cruises. The spirit of cooperation and close interactions among the scientists and technicians who were responsible for the measurements also
contributed to the outstanding quality of the data set.

The crossover comparison of fCO2 in seawater is not as straightforward as the comparison of the other carbon parameters because the measurement temperature for fCO2 differs for different cruises. The comparison thus requires a temperature normalization, which is performed by using the carbonate dissociation constants, and measured DIC. For comparison purposes, all values were normalized to 20癈 in this report. The normalization is dependent on the dissociation constant used. In this comparison, we used the constants of Mehrbach et al. (1973) as refitted by Dickson and Millero (1987). An example of the effect of constants on the final comparison is given in Table 7 in which we use typical deep-sea DIC and fCO2 values as found in the southeastern Pacific. Also included in the table are the fCO2@20癈/DIC values in µatm/(µmol/kg to illustrate the sensitivity of discrete fCO2 measurements relative to DIC in deep waters.

We analyzed 16 crossover comparisons for fCO2, and observed differences ranging between -28.7 and 34 µatm, excluding the large difference during the 1992 El Niño at 5?N, 110?W. The mean of the absolute value of the difference was 17.6 ± 16.3 µatm. In deep water 10 atm of fCO2 measured at 20癈 is approximately equivalent to an uncertainty of 1.5 µmol/kg DIC. Thus, with the possible exception of two or three crossover locations, the systematic differences in the fCO2 data corresponded to a similar uncertainty to that of the majority of the DIC results. Since there were no CRMs available for fCO2 during the Pacific expeditions, the analysts used their own compressed gas standards for the measurements. Some of the differences between the data sets may have resulted from systematic differences etween standards and/or differences between methods employed.

The agreement of the TAlk data between the 15 crossover locations is not quite as good as the DIC results. The differences between cruises ranged from -11.5 to 7.8 µmol/kg; generally, the smallest differences correspond to the excellent agreement by the same laboratory on different cruises. As with DIC and fCO2, the largest offsets generally occur during the strong El Nino event in 1992. The mean of the absolute value of the difference was 5.7 ± 3.3 µmol/kg; this corresponds to a mean uncertainty of approximately 0.2%. CRMs were available for TAlk where crossover comparisons were made for this report, and all data have been normalized to the certified values. Three laboratories performed pH analyses, and as a result, only five crossover locations were available to compare the pH results. All comparisons were made on the total seawater scale. The differences ranged from -0.0005 to 0.0062 and the mean of the absolute value of the difference was 0.0023 ± 0.0025. In the deep Pacific, an uncertainty of 1 µmol/kg DIC is equivalent to approximately 0.003 pH units. These results suggest that the limited amount of pH data in the Pacific were in excellent agreement with each other.

The summary data in Tables 3, 4, 5, and 6 should be viewed as one of several indicators of the overall quality of the carbon data from the Pacific. In addition to these results, there also are the shore-based analyses of replicate DIC samples taken during each of the cruises (Guenther et al. 1994) and the interlaboratory analyses of the CRMs (Dickson 1992). These three pieces of information should be used together with thermodynamic models in the process of evaluating the overall quality of the database. In several cases, particularly with respect to the NOAA data sets, three or four carbon parameters were measured during the cruises. In these situations, the internal consistency of the individual parameters in the data sets can be checked using an appropriate thermodynamic model (Millero et al. 1993; Byrne et al., in press; Wanninkhof et al., 1999). In this way, two parameters may be used to check the validity of the third and, in some cases, fourth parameter. For example, very precise and accurate DIC and pH data may be used to validate the fCO2 and TAlk data. We recommend that individual data sets be evaluated in this manner before they are used in physical and biogeochemical models. In addition, it is our recommendation that DIC data are reported to the database manager as both uncorrected and corrected with respect to CRMs, and that the CRM results are appended in a "meta" file. This file should contain at minimum CRM batch number, number of CRMs run, the given value and observed values, along with the standard deviation and number of CRM results rejected. The method of correction of the data should be clearly described, including if the correction was applied per cell, per cruise, using a longer-term mean, or if the correction was an additive or a ratio. In order to obtain a coherent data set of DIC from this program, it is imperative that the data be corrected in the same way. As shown in this report, the crossover data for DIC are statistically improved when the correction is applied. We also recommend the TAlk data be reported to the database manager in a similar way, appending a "meta" file containing a description of the CRM results. In addition, it is useful for both CRM corrected and uncorrected TAlk data to be submitted.

4. CONCLUSIONS
The comparison of the carbon system parameters during the WOCE and OACES cruises in the North and South Pacific has provided unique information on data quality at the crossover locations. For DIC, fCO2, and pH, the agreement at most crossover locations is well within the design specifications for the global CO2 survey, despite the lack of CRMs for both fCO2 and pH. In a statistical analysis performed on DIC data that were corrected to CRM values vs noncorrected values, results indicate there is a significant difference between the two. On the other hand, although normalized to CRM values for TAlk, the comparisons made in this report for that parameter were not as good. The outcome of this comparison stresses the importance of CRMs, as well as the value of building some redundant measurements into the program to provide an independent check on data quality.

Since the inception of this document, we have made every attempt to include the most up-to-date information available; however, large data sets are constantly evolving. Some of the data presented in this report are expected to change as the data are further evaluated. To access the latest data sets, please check the web sites listed in Section 5.