Check of CO₂ Calibration Performance

In order to check the performance of the individual calibration procedures, every group measured between one and four NOAA/CMDL CO₂ standards in the nominal concentration range of 250-500 ppmv as "unknown samples." Depending on the individual calibration procedure, different CO₂ standards were measured. Figure 4 shows the results of this exercise. It should be pointed out that this check was carried out on the last day of the exercise (June 17). Therefore no data are available for system "A," which had to prematurely quit the exercise on June 13 because of major technical problems. For system "C" only one standard could be measured because the measurement range had been fixed to an upper limit of 400 ppmv, which was slightly exceeded by the relevant NOAA/CMDL standard (403.85 ppmv). System "F" required all six standards for initial calibration, which could therefore not be measured as "unknown samples."

The results show that essentially all checked NDIR instruments were calibrated to an accuracy on the order of 1 ppmv or better over the whole concentration range of 250 to 500 ppmv. Only system "G" shows deviations of 3 to 6 ppmv. This is indicative of a systematic problem associated with the calibration of the CO₂ analyzer or with the measurement itself. Such deviations are clearly not tolerable and need to be addressed thoroughly. System "B" makes use of the factory calibration of the LI-COR LI-6262 instrument, which only requires the adjustment of "zero" (with a CO₂-free gas) and "span" (with a single CO₂ standard). It appears that some accuracy is lost by this somewhat crude calibration technique¹, and the "classical" approach using at least two CO₂ standards spanning the range of anticipated CO₂ mixing ratios is preferred.

¹This problem has been shown to be very reproducible. The calibration routine of system "B" therefore includes a small correction for this error which routinely removes it from the final xCO₂ values of system "B". Figure 4 shows uncorrected xCO₂ values only to demonstrate the magnitude of this error if left unaccounted for.