Conclusions

We have provided a common infrastructure to all participating groups in this exercise. We also have carried out several checks to exclude possible sources of error. Furthermore all raw data were run through the same calculation procedure. All these measures were taken in order to reduce as much as possible controllable sources of error. In this respect the exercise was technically a full success, and as summarized in the following discussion we also think that the exercise was a success scientifically.

We have demonstrated that the results of three out of seven underway systems agree to within about ±2 µatm throughout the cruise. This is not only the case for underway seawater fCO₂ measurements but also for measurements of atmospheric xCO₂. Interestingly, these three systems represent differences in such aspects as the design principle of the equilibrator, the volumes and flow rates of water and air involved, and the choice of wet or dry NDIR measurements. Thus this perfect agreement shows that-at least for NDIR instruments-the variety of designs used in the scientific community does not necessarily give rise to comparability problems or, to put it the other way round, systems of different design can produce reliable and consistent results.

We have also demonstrated that significant offsets of up to 10 µatm can be found in underway fCO₂ measurements under typical and identical field work conditions. Although in at least one case this may be a consequence of a technical failure, it is an indication of significant systematic differences in other cases. We certainly cannot claim that the observed differences are representative for these fCO₂ systems in general. They may also be typical only for the specific conditions of this particular cruise. There is, however, no indication that this cruise provided in any way untypical circumstances that could be made responsible for some of the observed deviations.

Finally we were able to demonstrate that discrete fCO₂ measurements agree with the results of the three most consistent underway fCO₂ systems. Therefore, measurements with these quite different approaches can be made with sufficient consistency, and the horizontal and vertical fCO₂ profiles generated from these different techniques can be expected to match in surface waters. In conclusion, therefore, three main messages can be derived from this exercise:

• Underway measurements of the CO₂ fugacity in surface seawater and overlying air can be done to a high degree of agreement (±1 µatm) with a variety of possible equilibrator and system designs.
• Even well-designed systems, which are operated without any obvious sign of malfunction, can show significant differences of up to 10 µatm.
• The discrete fCO₂ measurements are in good agreement with the three most consistent underway fCO₂ data sets, at least to within its nominal accuracy of 1%.

These results pose the important question of how fCO₂ data sets acquired from different groups can be combined into a common database in light of possible incompatibilities of up to 10 µatm. Although the results of this exercise do not solve this problem, they underline the importance of this aspect which must be taken into account in the construction of a consistent global fCO₂ database. Contributing to this dilemma is the fact that, in contrast to this exercise, other sources of error (temperature and pressure measurements, calibration gases etc.) further contribute to this uncertainty in field data.

In addition to this more general outcome, some of the results in more concrete terms follow. These may also serve as recommendations for future fCO₂ work in the ocean.

• The exercise shows no "best choice" for the type of the equilibrator (i.e., "showerhead," "bubbler," or "thin film") nor specifics on its dimensions and flow rates of seawater and air in regard to the achievable accuracy of the fCO₂ system.
• In contrast, the equilibrator type and its flow rates of seawater and air are important aspects with respect to the time constant of the equilibration process.
• Wet measurements can be done on the basis of the LI-6262 CO₂/H₂O gas analyzer (LI-COR Inc., U.S.A.) without necessary loss of accuracy when compared with traditional dry measurements.
• The factory calibration of the LI-COR LI-6262 CO₂/H₂O gas analyzer, which only requires the user to adjust "zero" and "span" of the instrument, seems to result in a loss of accuracy, which can easily be avoided by establishing an individual calibration curve on the basis of measurements of standard gases.
• The importance of rather accurate measurements of in situ and equilibrator temperature does not seem to be addressed adequately in the community. The observed differences between temperature measurements are clearly above a tolerable level and contribute-if representative and usually left unaccounted for-inconsistencies of several µatm (up to about 7 µatm in the present exercise).
• Calibration gases are an important issue. Even with the provided suite of consistent calibration gases, the NDIR analyzers could only be calibrated to an accuracy of 0.5 to 1.0 ppmv. We feel that this is about the tolerable limit. So any further error contribution from the calibrated standard concentrations worsens the situation. Use of calibration gases that are traceable to the same primary standards, such as the WMO primary standards maintained at SIO, would be desirable.