One of the primary responsibilities of the CCL is to disseminate the WMO scale with small enough uncertainty to prevent significant biases from influencing the interpretation of atmospheric data. This scale transfer uncertainty is separate from the total uncertainty of the scale and relates solely to how well a tertiary standard measurement is related to the scale. This is the relevant measure of uncertainty to use when considering the significance of differences of replicate calibrations of a cylinder when evaluating drift.

In Table 1 we estimate the scale transfer uncertainty (at 68% CL) for each instrument used for CO₂ calibrations. We use the reproducibility of target (or surveillance) cylinders measured over long time periods at the tertiary level on the calibration systems (see Hall et al., 2020). Currently we run "short term" target tanks (TT) approximately weekly to monthly and "long term" target tanks approximately yearly to provide information on the stability of the scale over decadal time scales. The current calibration system based on laser spectroscopy (https://www.esrl.noaa.gov/ccl/co2_calsystem.html) has been in service since 2016. TTs show very good reproducibility across the range of the scale. Prior to 2016, a series of NDIR analyzers from Siemens and Licor were used on the calibration system. We are limited in the number of TTs and in the mole fraction range covered for the earlier NDIR calibration system. These estimates are generally applicable to most of the range of the scale, but they may be underestimated for cylinders on the extremes of the scale and are not applicable to any measurement made beyond the scale range. Measurements made on an extension of the WMO scale are provided but we do not assess the consistency of these measurements.

We note that the performance of the calibration system is continuously evaluated and these estimates will be updated when warranted. We also note that there are potential issues with cylinder regulators that may not be apparent with TTs. NOAA uses standardized procedures for flushing and conditioning regulators and moderately high flow rates so we do not expect large issues with regulators. However, these procedures have been in place for many years and as we move to try to lower the scale transfer into the sub 0.01 ppm realm this needs to be reevaluated.

Table 1: Scale transfer uncertainty (at 68% CL) of analyzers used for CO₂ calibrations

Notes for table entries:

1) The laser spectroscopic system (co2cal-2) has a comprehensive set of TTs. Both short term and long-term TT sets are run on fixed schedules (weekly/monthly or yearly) and cover the nominal range of 300 to 600 ppm, nearly the entire range of the scale. As reported in Hall et al. 2020, assuming no drift is occurring in the TTs the average standard deviation of the measurements is 0.010 (± 0.003) ppm for all 23 TTs when analyzed on co2cal-2. Allowing for linear drift in the TTs reduces this only slightly to 0.008 (± 0.003) ppm. There is no strong mole fraction dependence to this; however, above 500 ppm this estimate may under-estimate the scale transfer uncertainty. It should be noted that the same set of secondary standards have been used on co2cal-2 since 2016. Changing secondary standards may impact the long term reproducibility. The reproducibility of the δ¹³C and δ¹⁸O measurements on the laser spectroscopic system is estimated at 0.2‰.

2) The scale transfer of the NDIR calibration system (co2cal-1) was previously estimated at 0.03 ppm on the X2007 scale. Several historical TTs measured since near the beginning of the development of the NOAA scale are still available and are measured on a yearly to bi-yearly schedule. Hall et al. 2020 used these tanks to verify that the reproducibility of the NDIR system was the same on X2019 and we continue to recommend this. It should be noted that the agreement of replicate measurements over shorter time scales often appeared better than the long term reproducibility of the system. This is likely due to changing the secondary standards in use. No distinction is made for the particular Siemens and Licor analyzers used except for S4 (see note 3 below).

3) Instrument S4, a Siemens Ultramat 5, was in use from December 1988 through Sept 1989. This was during the transition from using CO₂ in artificial air to CO₂ in natural air and the transition from using steel to aluminum tanks. This particular analyzer may have been sensitive to matrix effects due to variable Argon content in the artificial air. Its results were often clearly biased. We estimated the scale transfer uncertainty of S4 by looking at 33 cylinders measured on S4 that had enough of a calibration history on other analyzers before and after the S4 measurements to determine the offsets. The average offset was -0.01 ± 0.10 ppm. We assume the scale transfer uncertainty for this particular analyzer is 0.10 ppm.

4) The early calibrations performed by NOAA (mostly from 1979 to 1986, with a few calibrations through 1988) were done on Maihak UNOR and Hartmann-Braun URAS NDIR analyzers. During this time, NOAA used CO₂ in artificial air standards in steel tanks made by a commercial gas supplier. The steel tanks often displayed rapid growth of CO₂ leading to high drift rates. In addition, the composition of the artificial air was not as tightly controlled. We think Argon may have varied and the UNOR and URAS instruments may have been sensitive to this. We do not have much TT data from this period. Instead we used the calibration histories of secondary standards to get an estimate of the scale transfer uncertainty for these analyzers. NOAA was not the CCL during this time period and did not calibrate many cylinders for outside organizations on these analyzers. We use this estimate for our internal standards when weighting measurements on the UNOR and URAS analyzers compared to later measurements on the Siemens and Licors to assess drift.