Data integration attempts to improve the spatial coverage of observations used in modeling studies by combining measurement records maintained by independent laboratories into a cooperative globally-consistent data set. The term "globally-consistent" implies that when extending and integrating measurements records from independent programs, inconsistencies related to maintenance and propagation of calibration scale, sampling, analysis, and data processing and selection are minimal and do not introduce significant uncertainties when interpreting model output. At the moment, assessing consistency among measurements from independent programs relies on results from round-robin exercises (e.g., WMO/IAEA sponsored experiments) and comparisons with other sites and reference records. A major limitation of the round-robin experiments is that they take 2-3 years to complete. By the time the experiment is completed, the results give only a snapshot at one point in time; we do not know from this experiment how discrepancies may evolve in time. More frequent and ongoing comparison experiments such as the flask-air intercomparison between NOAA and CSIRO [Masarie et al., 2001] and the CarboEurope Sausage Comparison [Levin et al., 2005] are critical to this effort.

It is not an easy task to assess the level of comparability between two independent sets of observations due to the differences in scales, methodology and the inherent complexity of making high-precision atmospheric trace gas measurements.

The WMO/IAEA carbon cycle atmospheric measurement community recommends target levels of comparability required to address current carbon cycle scientific issues. Table 1 summarizes their recommendations.