Revised methodology for CO<sub>2</sub> and CH<sub>4</sub> measurements at remote sites using a working standard-gas-saving system

Sasakawa, Motoki; Tsuda, Noritsugu; Machida, Toshinobu; Arshinov, Mikhail; Davydov, Denis; Fofonov, Aleksandr; Belan, Boris

doi:https://doi.org/10.5194/amt-18-1717-2025

Articles | Volume 18, issue 8

https://doi.org/10.5194/amt-18-1717-2025

© Author(s) 2025. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/amt-18-1717-2025

© Author(s) 2025. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 18, issue 8

Research article

|

22 Apr 2025

Research article |

| 22 Apr 2025

Revised methodology for CO₂ and CH₄ measurements at remote sites using a working standard-gas-saving system

Motoki Sasakawa, Noritsugu Tsuda, Toshinobu Machida, Mikhail Arshinov, Denis Davydov, Aleksandr Fofonov, and Boris Belan

Download

Final revised paper (published on 22 Apr 2025)
Supplement to the final revised paper
Preprint (discussion started on 07 Feb 2024)
Supplement to the preprint

Interactive discussion

Status: closed

RC1:
'Comment on amt-2023-246', Anonymous Referee #1, 26 Mar 2024
The manuscript by Sasakawa et al. describes the analytical set-up, the standardization and uncertainty evaluation of CO2 and CH4 measurements at several tall tower observational stations in Siberia over several years. Those data have been made available recently and such measurements are of immense scientific value as they are extremely scarce in that part of the world. They provide constraints for global inversion studies needed to improve the assessment of the carbon cycle. Therefore, it is very appropriate to publish all details that allow an assessment of the accuracy of this data set and AMT certainly is a magazine where such an article is well placed.
While acknowledging the relevance of the subject the manuscript there are several topics where I would ask the authors for revisions.
General comments:
The manuscript title suggests that the gas saving system is being published in this manuscript. However, the setup has already been described earlier (as cited in the manuscript) so a rewording of the title might be considered.

A detailed report of the uncertainty should consist of a complete list of the uncertainty components. While the manuscript addresses instrument repeatability, the calibration procedure and the reproducibility within about one week, there are some topics (listed below) not being included. For a subset of sites results of a comparison between two different analyzers calibrated by the same set of standard gases are presented. The assessment of the uncertainty is found to be consistent with the comparison results for CO2 and at two of the three test sites for CH4 which seems to be taken by the authors as a confirmation for the completeness of the uncertainty evaluation. Yet, I see the following aspects either missing or not adequately accounted for in this uncertainty assessment:

The accuracy of the assignments of the standard gases is not discussed. An estimate for the uncertainty introduced by the calibration transfer procedure from the respective NIES calibration scales to each individual standard is not provided nor the analytical method for the calibration transfer. It is described that the standard gases were prepared from pure CO2 diluted with purified air. This could mean that the isotopic composition of the CO2 in these synthetic standard gases is different to the atmospheric CO2 which would cause different sensitivities of the NDIR analyzer to the standard gas and the sample CO2, respectively. A respective clarification should be provided and if needed an uncertainty term for the isotopic discrimination be added.

The stability of the CO2 mole fraction of a standard gas during the entire period of usage cannot be assumed to be certain. At least at the KRS site calibration standards apparently have been replaced. Have those standards been recalibrated after usage at the NIES laboratory for confirmation of their stability?

For the scale relationship between the NIES scale and the WMO scale the results of the WMO Round Robin 6 Experiment are quoted. However, the period under review extends from 2005 to 2019 and three Round Robin Exercises (4-6) have taken place during that time. It would appear reasonable to consider all these results and deduce an uncertainty of the scale relationship. If other evidence for the relationship between the WMO scale and the NIES scale existed that might further help to get a better constrained uncertainty for this.

The NDIR detection with the specified instrument is known to be not perfectly linear. In the manuscript no test results for this are being presented / discussed. A different presentation of the NDIR-CRDS differences (residuals, i.e. difference NDIR-CRDS vs CO2 plots instead of CO2(CRDS) vs CO2(NDIR) plots) could allow to judge how significant this non-linearity is (taking the CRDS as a linear reference).

The authors state that “From the point of this procedure, the SWS-gas equates with the 'target tank' or 'surveillance tank' mentioned in the GAW report (WMO, 2018)”. However, the scatter of the SWS target with a usage time of ca 1 week mostly reflects the output drifts within 12 hours that are not accounted for with the employed calibration approach. This SWS target gases can certainly not assess the long-term reproducibility throughout many years or even just one year (where different seasonal temperatures within the measurement container could cause an effect). For this a real long-term surveillance tank would be needed. A stronger independent check is the parallel measurement by the CRDS instruments. It is very convincing that the two measurements agree within the expectation and the annual mean differences remain pretty constant. I propose to add another figure for the Supplement displaying the NDIR-CRDS differences and TOS-CRDS differences vs time. This would allow to identify or exclude any season dependent inter-instrument offsets.

Reference is being made to the GAW report with the WMO CO2 Expert’s meeting recommendations. However, the compatibility goals that are specified in this report for such background observations are not quoted (0.1 µmol/mol for CO₂ and 2 nmol/mol for CH₄). It is fully appreciated that the immense logistical challenge to operate these measurements at such remote places set boundary conditions that take their toll. I encourage the authors to actively discuss the discrepancy between the WMO compatibility goals and the results they achieved. The measurement uncertainties achieved should also been put into the context of the signals observed in the records and into the context of the gradients within this JR-STATION network (e.g. as presented in the Sasakawa et al 2010 paper) or relative to global reference stations (e.g. Mauna Loa, Alert, Mace Head). As stated these data are extremely scarce and will certainly be valuable for many applications also with larger uncertainty as targeted with the WMO compatibility goals. This could be an additional summarizing section 3. Currently if feels that after the individual descriptions of uncertainties in sectin 2.4.2, 2.4.3 and 2.5 a summarizing part is missing.

Comparison measurements are available only for some stations and the last few years. The case of the failing catalyst for the TOS CH₄ sensor at NOY demonstrates that an independent long-term check is needed to realize instrumental failures that can happen. The authors should comment if this failure could have been detected also by other measurement parameters not related to the CRDS measurement.

Specific comments
l. 27: “A CRDS (Cavity Ring-Down Spectroscopy) analyzer is more stable and does not need frequent checks of the drift (ICOS RI, 2020).” I feel this might be misinterpreted in a way that you do not take care of response drift. The ICOS document recommends as starting point bi-weekly calibration and 7 hourly target gas measurements so perhaps that could be presented as an example.
l. 51: “The measurement system was housed in a freight container insulated to reduce temperature variation.” There is no further detail given on whether any active temperature control was performed within the container, whether the temperature was recorded and what temperature changes occurred within the container over the years and within the 12 h calibration intervals.
l. 79: “To protect the cavity of the CRDS from an inflow of the dissolved chemical desiccant (Mg(ClO4)2 or P2O5) in the accidental case of a broken pump etc., we equipped the CRDS with two poppet check valves (“PCV” in the analyzer unit)”. As I understand Figure 1 the CRDS is not positioned downstream of P₂O₅ or Mg(ClO₄)₂ so perhaps either the figure or the text need an update.
l. 82: pure CO₂: If the CO2 in the working standard gases is entirely from pure CO2 the isotopic composition matters. Is this available? If purified air means that the matrix is ambient air already containing natural CO2 and only moderate amounts of pure CO2 are added that might matter less.
l. 84: it would be good to rather refer to WMO X2019 than to X2007 scale (or provide in addition the information of the WMO X2007 to X2009 conversion term).
l. 85: As stated before it is not clear why only two data points from one WMO Round Robin are being taken as basis for the NIES – WMO scale relationship whereas more Round Robin data is available.
l. 170 “The analysis precision for this system was uniformly estimated as 0.3 ppm for CO2 (Watai et al., 2010).” I suggest to add: The reported precision for this system under laboratory conditions was estimated as 0.3 ppm (Watai et al., 2010)
l. 180 To make it clear, the language of this sentence needs a bit of rephrasing like: “We estimated the outputs in voltage of three standard gases at [..] difference of the outputs in voltage of the SWS-gas only when both standard gas measurements satisfied the criteria described in section 2.3”
l. 249: The GAW report 255 (WMO, 2020) explicitely states: “Standard deviations are distinct from measurement uncertainty and should be reported in addition to rather than in lieu of measurement uncertainty. Standard deviations provide a measure of atmospheric variability (AV) plus instrument noise.”. So a better metric for the repeatability term might refer to the repeatability of the adjacent SWS measurements rather than the standard deviation of 3 min of atmospheric measurement.
l. 255 “The fourth term expresses the contribution from the variation in x. In most cases, the uncertainty for the concentration of standard gases is not given, or it can be negligible compared with other uncertainties. Thus, we neglected this term.” That it is negligible appears like a speculative statement that I do not feel can be kept as is. To justify the neglect of this term there needs to be a quantified small uncertainty for the assignment and also the stability of the mole fractions for the period of use.
l. 265: We calculated a calibration line only when the SWS-gas measurements closest to both sides of the sample measurements were normal.
l. 267: The following text should be reformulated: “However, a gap would appear if the system temporarily became unstable. To detect any gap, ..” The gap only appears when the criteria are applied that are explained below. If I got it right I guess a sentence like the following describes it better: “When the system temporarily became unstable the corresponding large changes of the analyzer output could not be corrected by the SWS to a sufficient degree. To identify such occasions…”
l. 280: “We determined a threshold for the standard deviation (σsws; 1 ppm for CO2, 10 ppb for CH4) and the fluctuation range (5 ppm for CO2 and 20 ppb for CH4) obtained from the estimated independent data set. All the data that exceeded the threshold were deleted.” Does one outlier invalidate the entire week? From this text I would assume YES, from Figure 5 I would guess that NO.
l. 284: “From the point of this procedure, the SWS-gas equates with the 'target tank' or 'surveillance tank' mentioned in the GAW report (WMO, 2018).” I do not fully agree, see general comment above.
l. 285: GAW report 242 (WMO, 2018) has a more up-to-date succeeding GAW report 255 (WMO, 2020).
l. 303f: “Only data from the period when the system was normal was extracted.” I propose to rephrase: Only data from periods was extracted when the SWS results fulfilled the reproducibility criteria outlined in the previous section.
l. 310: Please add “No significant difference in the instrument offset between the different intake heights was found for CH4 concentration either.” Else the sentence contradicts the following sentence.
l. 311-316: Reading “it is expected that the temperature regulation of the catalytic unit did not work well and the catalyst did not function” it remains unclear to me if there were any other indications other than the increasing and instable bias from the TOS to the CRDS that allowed to identify the problem with the TOS catalyst (e.g. more variable calibration fit residuals or clearly implausible atmospheric CH4 mole fractions). Fig. S5 and S6 do not reveal any problem. If there is no such additional indicator that raises the question if the same problem might also have happened unrecognized at one of the sites without CRDS?
l. 320 / 328: Figures 6 and 7: to judge the agreement between the two instruments the way of display is not ideal: presenting the residuals (NDIR – CRDS) would allow to zoom in to the differences which is more informative
Figure S1-6: black dots are named to represent the “uncertainty for the CO2 measurement”. As stated above (specific comment to l. 290) uncertainty is not appropriate term if the standard deviation of 60 three second averages of atmospheric measurements is referred to. Please change the terminology in the caption here.
If it does not need too detailed descriptions some short general explanations could be considered what general problems have caused the large data gaps in the beginning.
The supplement figures S1-S6 only display data from the NOY site, so there is no information on the relative performance at the other sites. It might help if not only NOY as an example was presented but also the same information of the other stations.
Figure S2: The sample uncertainty (black dots) of NOY in the last months of 2012 has a remarkable two level pattern. Is this a 12 hour shift day-night pattern of atmospheric variability? Something else? Why only in that year?
Figures S7 and S9: The legend description "WBT unstable" is not further explained in the caption and the abbreviation WBT not used anywhere in the manuscript.
Figure S10: The data of NOY 2018 appear to group along two lines. Is there any explanation that the authors could offer?
Citation: https://doi.org/10.5194/amt-2023-246-RC1
- AC2: 'Reply on RC1', Motoki Sasakawa, 30 Jul 2024
  
  Response to reviewer #1.
  Thank you very much for reviewing our manuscript in detail. You looked at it deeply, asked essential questions, and made us aware of many things. We have responded to all comments as best we could and believe we have made it more comprehensive. Sorry for the delay. Below are the responses.
  
  Best regards.
  Comment:
  The manuscript title suggests that the gas saving system is being published in this manuscript. However, the setup has already been described earlier (as cited in the manuscript) so a rewording of the title might be considered.
  Reply:
  I appreciate your suggestion. We have reworded the title as follows.
  Revised calculation method of concentrations and uncertainties for in-situ CO₂ and CH₄measurements with a working standard gas saving system.
  
  Comment:
  The accuracy of the assignments of the standard gases is not discussed. An estimate for the uncertainty introduced by the calibration transfer procedure from the respective NIES calibration scales to each individual standard is not provided nor the analytical method for the calibration transfer. It is described that the standard gases were prepared from pure CO2 diluted with purified air. This could mean that the isotopic composition of the CO2 in these synthetic standard gases is different to the atmospheric CO2 which would cause different sensitivities of the NDIR analyzer to the standard gas and the sample CO2, respectively. A respective clarification should be provided and if needed an uncertainty term for the isotopic discrimination be added.
  255 “The fourth term expresses the contribution from the variation in x. In most cases, the uncertainty for the concentration of standard gases is not given, or it can be negligible compared with other uncertainties. Thus, we neglected this term.” That it is negligible appears like a speculative statement that I do not feel can be kept as is. To justify the neglect of this term there needs to be a quantified small uncertainty for the assignment and also the stability of the mole fractions for the period of use.
  Reply:
  We have added the following description in Section 2.4.2.
  “The NIES09 scale is based on the gravimetric primary standard gases using a one-step dilution (Tohjima et al., 2006), and its overall uncertainty, including transfer, is estimated to be 0.043 ppm (Machida et al., 2011). This value is equal to , but this term is not included in the calculation because cannot be determined precisely; as for the value of , it distributed between 0.025 ppm to 0.043 ppm, since there are three WSs.”
  As for the isotope effect, Tohjima et al. (2009) reported that the errors for isotopic influence for the three NDIR analyzers range from -0.04 to -0.08 ppm. However, the apparent NDIR CO₂ mole fraction error depends on its individual sensitivity to the optical filter property. The installation of this observation system started in the early 2000s, but the characteristics of each NDIR are not known, as this was not yet taken into account at that time. We have described the possibility that the results could be low to the extent of 0.08 ppm or less in Section 2.1.
  Tohjima, Y., Machida, T., Mukai, H., Maruyama, M., Nishino, T., Akama, I., Amari, T., and Watai, T.: Preparation of gravimetric CO₂ standards by one-step dilution method, in Report of the 13th WMO/IAEA Meeting of Experts on Carbon Dioxide Concentration and Related Tracers Measurement Techniques, Boulder, September 19 – 22, 2005, WMO/GAW, 2006.
  Tohjima, Y., Katsumata, K., Morino, I., Mukai, H., Machida, T., Akama, I., Amari, T., and Tsunogai, U.: Theoretical and experimental evaluation of the isotope effect of NDIR analyzer on atmospheric CO₂ measurement, J. Geophys. Res., 114, 10.1029/2009JD011734, 2009.
  
  Comment:
  The stability of the CO2 mole fraction of a standard gas during the entire period of usage cannot be assumed to be certain. At least at the KRS site calibration standards apparently have been replaced. Have those standards been recalibrated after usage at the NIES laboratory for confirmation of their stability?
  Reply:
  Unfortunately, for various reasons, it was impossible to recalibrate any of the standard gases used for the observations in Siberia, as it is difficult to return equipment and high-pressure gas cylinders brought into Russia from Japan. As far as possible, the standard gas used was replaced by a new standard gas before the residual pressure of the standard gas used fell below 0.2 MPa.
  
  Comment:
  For the scale relationship between the NIES scale and the WMO scale the results of the WMO Round Robin 6 Experiment are quoted. However, the period under review extends from 2005 to 2019 and three Round Robin Exercises (4-6) have taken place during that time. It would appear reasonable to consider all these results and deduce an uncertainty of the scale relationship. If other evidence for the relationship between the WMO scale and the NIES scale existed that might further help to get a better constrained uncertainty for this.
  Reply:
  The NIES CO₂ scale agreed with the WMO scale to within 0.1 ppm, even with the results of Round Robin 5, which was measured in 2009 at NIES. This point was added in addition to CH₄ scale. The results for Round Robin 4 are not shown as they were with a previous scale (NIES 95 scale). No other published data are available to directly compare the scales.
  
  Comment:
  The NDIR detection with the specified instrument is known to be not perfectly linear. In the manuscript no test results for this are being presented / discussed. A different presentation of the NDIR-CRDS differences (residuals, i.e. difference NDIR-CRDS vs CO2 plots instead of CO2(CRDS) vs CO2(NDIR) plots) could allow to judge how significant this non-linearity is (taking the CRDS as a linear reference).
  Reply:
  NDIR calibrations were performed by linear regression, as shown in Figure 5, where the coefficient of determination is greater than or equal to 0.999; if it is less than this, the data is rejected. We have replaced the correlation plots with residual plots (Figure 6), which show linearity on the NDIR, taking the CRDS as a linear reference.
  
  Comment:
  The authors state that “From the point of this procedure, the SWS-gas equates with the 'target tank' or 'surveillance tank' mentioned in the GAW report (WMO, 2018)”. However, the scatter of the SWS target with a usage time of ca 1 week mostly reflects the output drifts within 12 hours that are not accounted for with the employed calibration approach. This SWS target gases can certainly not assess the long-term reproducibility throughout many years or even just one year (where different seasonal temperatures within the measurement container could cause an effect). For this a real long-term surveillance tank would be needed. A stronger independent check is the parallel measurement by the CRDS instruments. It is very convincing that the two measurements agree within the expectation and the annual mean differences remain pretty constant. I propose to add another figure for the Supplement displaying the NDIR-CRDS differences and TOS-CRDS differences vs time. This would allow to identify or exclude any season dependent inter-instrument offsets.
  Reply:
  It is certainly not a gas to verify long-term reproducibility, but it is still a good guide for repeat measurements of about one week, so we changed the notation to that. We have added the figures for differences vs time in the Supplement (Figures S20 and S21), which show a good agreement.
  
  Comment:
  Reference is being made to the GAW report with the WMO CO2 Expert’s meeting recommendations. However, the compatibility goals that are specified in this report for such background observations are not quoted (0.1 µmol/mol for CO₂ and 2 nmol/mol for CH₄). It is fully appreciated that the immense logistical challenge to operate these measurements at such remote places set boundary conditions that take their toll. I encourage the authors to actively discuss the discrepancy between the WMO compatibility goals and the results they achieved. The measurement uncertainties achieved should also been put into the context of the signals observed in the records and into the context of the gradients within this JR-STATION network (e.g. as presented in the Sasakawa et al 2010 paper) or relative to global reference stations (e.g. Mauna Loa, Alert, Mace Head). As stated these data are extremely scarce and will certainly be valuable for many applications also with larger uncertainty as targeted with the WMO compatibility goals. This could be an additional summarizing section 3. Currently if feels that after the individual descriptions of uncertainties in sectin 2.4.2, 2.4.3 and 2.5 a summarizing part is missing.
  Reply:
  We have added the following sentences in Section 2.4.3.
  “On the other hand, the CO₂ (CH₄) concentrations in these observations can fluctuate on the order of ppm (10 ppb), even during a few hours of daytime when the atmosphere is well mixed (Sasakawa et al., 2010, 2013). It is, therefore, considered adequate for observations carried out in the vicinity of strong emission and absorption sources, such as those in the Siberian interior. The GAW report states that the compatibility goal for CO₂ (CH₄) in the Northern Hemisphere is 0.1 ppm (2 ppb), but this is a target for background sites such as coastal areas and does not need to be strictly adapted to an observation area such as this study.”
  
  Comment:
  Comparison measurements are available only for some stations and the last few years. The case of the failing catalyst for the TOS CH₄ sensor at NOY demonstrates that an independent long-term check is needed to realize instrumental failures that can happen. The authors should comment if this failure could have been detected also by other measurement parameters not related to the CRDS measurement.
  311-316: Reading “it is expected that the temperature regulation of the catalytic unit did not work well and the catalyst did not function” it remains unclear to me if there were any other indications other than the increasing and instable bias from the TOS to the CRDS that allowed to identify the problem with the TOS catalyst (e.g. more variable calibration fit residuals or clearly implausible atmospheric CH4 mole fractions). Fig. S5 and S6 do not reveal any problem. If there is no such additional indicator that raises the question if the same problem might also have happened unrecognized at one of the sites without CRDS?
  Reply:
  Previous notations were not good. Significant differences in CRDS and TOS values were observed at NOY and DEM, which are due to issues with the catalyst's temperature adjustment. The differences were small at KRS, where no defects in temperature control were observed (Figure 7). At the other sites, the temperature controller operated normally, so it is expected that the TOS produced nonbiased values, as is the case with KRS. This explanation has been added to the text.
  
  Comment:
  27: “A CRDS (Cavity Ring-Down Spectroscopy) analyzer is more stable and does not need frequent checks of the drift (ICOS RI, 2020).” I feel this might be misinterpreted in a way that you do not take care of response drift. The ICOS document recommends as starting point bi-weekly calibration and 7 hourly target gas measurements so perhaps that could be presented as an example.
  Reply:
  We have revised it as follows:
  “The CRDS analyzer is a more stable device, and a calibration frequency of every two weeks to every month is recommended (ICOS RI, 2020).”
  
  Comment:
  51: “The measurement system was housed in a freight container insulated to reduce temperature variation.” There is no further detail given on whether any active temperature control was performed within the container, whether the temperature was recorded and what temperature changes occurred within the container over the years and within the 12 h calibration intervals.
  Reply:
  Two thermometers are mounted inside the container, one near the ceiling and the other near the floor. According to the upper thermometer, the room temperature in the container during the year was kept above 15ºC and the temperature difference in the 12-hour calibration interval was kept below 3ºC on average during the year. Since the introduction of the CRDS, a simple cooler was installed to prevent the temperature inside the container from rising too high during the summer months due to the heat generated by the CRDS. We have added this explanation.
  
  Comment:
  79: “To protect the cavity of the CRDS from an inflow of the dissolved chemical desiccant (Mg(ClO4)2 or P2O5) in the accidental case of a broken pump etc., we equipped the CRDS with two poppet check valves (“PCV” in the analyzer unit)”. As I understand Figure 1 the CRDS is not positioned downstream of P₂O₅ or Mg(ClO₄)₂ so perhaps either the figure or the text need an update.
  Reply:
  If the pump in the pump unit stops and the CRDS pump is running without the PCV on, it is possible that the desiccant line is pumped backwards instead of from the inlet direction. To prevent this, if the above situation occurs, the flow stops at the PCV upstream of the P₂O_5, and the PCV before the Nafion is designed to take in container air when the upstream flow stops and suction pressure increases (data from this case has been deleted). This explanation has been added.
  
  Comment:
  82: pure CO₂: If the CO2 in the working standard gases is entirely from pure CO2 the isotopic composition matters. Is this available? If purified air means that the matrix is ambient air already containing natural CO2 and only moderate amounts of pure CO2 are added that might matter less.
  Reply:
  The working gas is made by diluting pure CO₂. The details of this are also explained in Tohjima et al. (2009), but its effect is different for each NDIR depending on the characteristics of the optical filter. Although the characteristics of the individual NDIR optical filter installed at the sites are not known, the bias due to this case is estimated to be less than 0.1 ppm in the negative direction. We have added to the text.
  
  Comment:
  84: it would be good to rather refer to WMO X2019 than to X2007 scale (or provide in addition the information of the WMO X2007 to X2009 conversion term).
  Reply:
  There have yet to be published results with WMO X2019. However, it matches WMO X2007 to within 0.1 ppm, so if you convert to X2019, you can currently do so as if you were converting from X2007.
  
  Comment:
  85: As stated before it is not clear why only two data points from one WMO Round Robin are being taken as basis for the NIES – WMO scale relationship whereas more Round Robin data is available.
  Reply:
  The results from Round Robin 5 have also been listed.
  
  Comment:
  170 “The analysis precision for this system was uniformly estimated as 0.3 ppm for CO2 (Watai et al., 2010).” I suggest to add: The reported precision for this system under laboratory conditions was estimated as 0.3 ppm (Watai et al., 2010)
  Reply:
  Thank you for your comment; we have added: “under laboratory conditions.”
  
  Comment:
  180 To make it clear, the language of this sentence needs a bit of rephrasing like: “We estimated the outputs in voltage of three standard gases at [..] difference of the outputs in voltage of the SWS-gas only when both standard gas measurements satisfied the criteria described in section 2.3”
  Reply:
  Thank you for your comment. We have corrected it as such.
  
  Comment:
  249: The GAW report 255 (WMO, 2020) explicitely states: “Standard deviations are distinct from measurement uncertainty and should be reported in addition to rather than in lieu of measurement uncertainty. Standard deviations provide a measure of atmospheric variability (AV) plus instrument noise.”. So a better metric for the repeatability term might refer to the repeatability of the adjacent SWS measurements rather than the standard deviation of 3 min of atmospheric measurement.
  Reply:
  The uncertainty of the analysis system is expressed in s_sws using SWS-gas, as denoted in Section 2.4.3, not here. As required by each element of equation (22), equation l.249 is part of the uncertainty in the estimated concentration (x), including atmospheric variability over a 3-minute period. We consider this equation to be appropriate.
  
  Comment:
  267: The following text should be reformulated: “However, a gap would appear if the system temporarily became unstable. To detect any gap, ..” The gap only appears when the criteria are applied that are explained below. If I got it right I guess a sentence like the following describes it better: “When the system temporarily became unstable the corresponding large changes of the analyzer output could not be corrected by the SWS to a sufficient degree. To identify such occasions…”
  Reply:
  Thank you for your comment. We have corrected it as such.
  
  Comment:
  280: “We determined a threshold for the standard deviation (σsws; 1 ppm for CO2, 10 ppb for CH4) and the fluctuation range (5 ppm for CO2 and 20 ppb for CH4) obtained from the estimated independent data set. All the data that exceeded the threshold were deleted.” Does one outlier invalidate the entire week? From this text I would assume YES, from Figure 5 I would guess that NO.
  Reply:
  It is YES. We have revised Figure 5.
  
  Comment:
  284: “From the point of this procedure, the SWS-gas equates with the 'target tank' or 'surveillance tank' mentioned in the GAW report (WMO, 2018).” I do not fully agree, see general comment above.
  Reply:
  We have modified it as follows.
  “From the point of this procedure, the SWS-gas is similar to the short-term 'target tank' or 'surveillance tank' mentioned in the GAW report (WMO, 2020).”
  
  Comment:
  285: GAW report 242 (WMO, 2018) has a more up-to-date succeeding GAW report 255 (WMO, 2020).
  Reply:
  Updated.
  
  Comment:
  303f: “Only data from the period when the system was normal was extracted.” I propose to rephrase: Only data from periods was extracted when the SWS results fulfilled the reproducibility criteria outlined in the previous section.
  Reply:
  Thank you for your comment. We have corrected it as such.
  
  Comment:
  310: Please add “No significant difference in the instrument offset between the different intake heights was found for CH4 concentration either.” Else the sentence contradicts the following sentence.
  Figure S10: The data of NOY 2018 appear to group along two lines. Is there any explanation that the authors could offer?
  Reply:
  Previous notations were not good. The following text has been rewritten.
  “At NOY and DEM, it was discovered that the temperature controller of the catalytic unit was not functioning correctly. Since the TOS is sensitive to CO and H₂ in the air, it could produce unusually high values without a proper catalytic unit. For the period, only the data from the CRDS should be published. Since no catalytic unit errors were identified at the other sites, the ambient atmospheric values were detected, as is the case with KRS.”
  
  Comment:
  320 / 328: Figures 6 and 7: to judge the agreement between the two instruments the way of display is not ideal: presenting the residuals (NDIR – CRDS) would allow to zoom in to the differences which is more informative
  Reply:
  We have replaced the correlation plots with residual plots (Figures 6 and 7) and discussed the agreement.
  
  Comment:
  Figure S1-6: black dots are named to represent the “uncertainty for the CO2 measurement”. As stated above (specific comment to l. 290) uncertainty is not appropriate term if the standard deviation of 60 three second averages of atmospheric measurements is referred to. Please change the terminology in the caption here.
  Reply:
  As explained above, we believe “uncertainty for the ambient air CO₂ concentration” is proper. It's rewritten as such. It is not “uncertainty for the CO₂ (CH₄) measurement”.
  
  Comment:
  If it does not need too detailed descriptions some short general explanations could be considered what general problems have caused the large data gaps in the beginning.
  Reply:
  The main causes of missing measurements are power failure, insufficient flow due to line blockages or pump deterioration. The TOS, in particular, can require several hours for normal measurement once they stop measuring. Observation sites are all located in remote areas, and maintenance can be carried out once a month at best but sometimes once every two or three months, especially in the northern sites. There is no internet access, so problems may be discovered on site and not be repaired immediately on the spot, so they are dealt with at the next maintenance visit. This situation makes continuous observation in Siberia extremely difficult. However, this explanation does not know the appropriate place to describe it within this paper.
  
  Comment:
  The supplement figures S1-S6 only display data from the NOY site, so there is no information on the relative performance at the other sites. It might help if not only NOY as an example was presented but also the same information of the other stations.
  Reply:
  We have added the figures for KRS and DEM.
  
  Comment:
  Figure S2: The sample uncertainty (black dots) of NOY in the last months of 2012 has a remarkable two level pattern. Is this a 12 hour shift day-night pattern of atmospheric variability? Something else? Why only in that year?
  Reply:
  During this period, the system was not at its best; the low intake side was clogged (no data available), and the high intake side was also somewhat clogged. Probably due to this, fluctuations in the output voltage are visible. However, the fluctuation is not so large that it hangs over the set flag, and the average CO₂ concentration over a three-minute period shows a stable value, so it has been adopted. One side of the line was clogged for a number of months in other years, but no data was ever obtained as in 2012, the cause of which is unknown. The previous original data was outdated due to a mistake, so it was re-plotted with the data that is now publicly available.
  
  Comment:
  Figures S7 and S9: The legend description "WBT unstable" is not further explained in the caption and the abbreviation WBT not used anywhere in the manuscript.
  Reply:
  The WBT-based data selection has been indicated in Section 2.5. It is now not shown in the Figures.
  
  Citation: https://doi.org/10.5194/amt-2023-246-AC2
RC2:
'Comment on amt-2023-246', Anonymous Referee #2, 13 Apr 2024
Review of “Working standard gas saving system for in-situ CO2 and CH4 measurements and calculation method for concentrations and their uncertainty”
Sasakawa et al.
General Comments
This is a useful paper, describing GHG measurements in a poorly measured part of the world. Furthermore, the description and evaluation of this technique for using locally-prepared artifacts for tracking instrument drift is valuable for the community.
I have a few general comments for improvement:
It seems to me that an opportunity has been lost to explore the performance of the sub-working standards. Twice a day for the week or so duration of a given SWS (A or B), you have assessments of the SWS concentration, using both the NDIR / TOS system and the CRDS. Do you observe a clear dependence of the reported concentration from the SWS over time after filling? Any dependence on the cylinder itself? Dependence on ambient temperature? Dependence on the concentration of the gas? Dependence on pressure in the cylinder? Similiarly, there are 100-200 measurements of each SWS using the CRDS, which would show the behavior of the SWS with much higher temporal resolution. Correlations to ambient temperature, pressure, etc could be performed with much higher fidelity. These would be interesting statistics to collect and present. It would allow those who want to use this approach to have some idea of ways in which the recipe could be modified without danger of exceeding the boundaries of the validation that you have done. I would like to see some of these data explored, summarized in the main text with additional data in the supplement.

In a similar vein, you have long time series measurements of the WS tanks, both with the CRDS and the other sensors. The CRDS could give you some idea of WS variability and WS to WS accuracy as new WS tanks are swapped in, and would help you tie your measurements down to the absolute scale. And the WS measurements on the NDIR and TOF would give an idea of the nature of the drift of these sensors. In particular, it would give insight into whether drifts appear as an offset in voltage or span change in voltage (see comment #3, below), and the nature in the errors in the sensor with ambient temperature, ambient pressure, time of year, etc. These analyses would be helpful to the readers of this article.

The approach to using the SWS to extend the life of the WS is essentially to perform linear interpolation in voltage from the twice daily calibration of the SWS using the three WS tanks. That assumes that the correction to the sample data from the SWS measurements is applied as an offset to the sample voltage, regardless of the SWS measured. But could the correction also be applied as a slope correction? That is, a correction that is proportional to the measured voltage? How do you know which is the right approach for each sensor? Similarly, NDIR is known to have a nonlinear response. Does the TOS also have a non-ideal (i.e., not logarithmic) response? Are these nonlinearities significant relative to your target uncertainties? Do your data shed any light on these concerns? Please add some discussion of these issues in the text.

Specific Comments
In general, there are too few references. It would be good to add a few more references pointing to other calibration strategies used in the GHG monitoring community for CO2 and CH4, references to the individual measurement technologies, references to similar data sets in the regions (e.g. ZOTTO), etc. It would help to put this manuscript in better context relative to similar work.

L105-108: It’s not clear the sequence of preparing and using the two SWS cylinders A and B. For example, when A is being used, is B being filled? Does a cylinder ever get used once it drops below 0.1 MPa? Is the cylinder pumped out prior to refilling? Do you have a time delay to allow the gas in the cylinder to equilibrate? Do you fill the cylinders at a certain time in the day?
The plumbing schematic of the SWS and associated manifold for filling and dispensing gas, with the double arrows and the 3-way valve bridge, is confusing. Showing the NC, NO, and COM on the valves, and then a table with the states (e.g., FILL A, USE B, etc..) would be helpful here in reproducing the setup.

L126-127: “had a more significant standard deviation than a determined threshold (2 mV for CO2, 5 mV for CH4).” Reporting fluctuations in mV without describing the mapping of voltage to concentration, it difficult to understand the context of what this noise, and the selected thresholds, mean. A short section describing the response of the detectors prior to this would be good.

L135: It’s often considered a good idea not to perform calibrations at exactly the same day and time. Why was 12 hours chosen as the interval? Do you see persistent differences between the day and night calibrations?

L319 and L327: Correlation plots are useful, but residual plots (TOS – CRDS and NDIR – CRDS) would be instructive in understanding any concentration dependence to the results.

Similarly, residual plots over time in the supplement would help understand the potential for biases in the measurements.
A particularly useful plot would be to show σ_sws vs the std. deviation of the difference between the sensor data and the CRDS over time and/or as a scatter plot. A plot like these would illuminate whether your uncertainty estimates properly capture the actual uncertainties in the sensor measurements using the SWS.

L319: There is a striking set of outliers just below the 1:1 line in the 2018 panel (at about 415 ppm CO2). There are two similar though less obvious outlier groupings in the 2020 data panel as well. What is the cause of these discrepancies? Shouldn’t they have been trapped by the data quality checks? Some explanation in the text would be helpful.

Technical Corrections
L12: “to a level less than one order of magnitude smaller” – unclear whether change is greater than or less than 1 order of magnitude. Reword for clarity.
L14: “…consisted of this system installed at nine sites…” – awkward. Reword, please.
L22: “respectively” – unnecessary.
L50: “sucked” – colloquial language. Try “pulled” or another word.
Citation: https://doi.org/10.5194/amt-2023-246-RC2
- AC1: 'Reply on RC2', Motoki Sasakawa, 16 Jul 2024
  
  Thank you very much for reviewing this detailed paper. You looked at it in great depth, asked essential questions, and made us aware of many things. We have tried to respond to all comments as best we could, and we believe we have made it more comprehensive. Below are the responses. Best regards.
  Comment:
  
  It seems to me that an opportunity has been lost to explore the performance of the sub-working standards. Twice a day for the week or so duration of a given SWS (A or B), you have assessments of the SWS concentration, using both the NDIR / TOS system and the CRDS. Do you observe a clear dependence of the reported concentration from the SWS over time after filling? Any dependence on the cylinder itself? Dependence on ambient temperature? Dependence on the concentration of the gas? Dependence on pressure in the cylinder? Similiarly, there are 100-200 measurements of each SWS using the CRDS, which would show the behavior of the SWS with much higher temporal resolution. Correlations to ambient temperature, pressure, etc could be performed with much higher fidelity. These would be interesting statistics to collect and present. It would allow those who want to use this approach to have some idea of ways in which the recipe could be modified without danger of exceeding the boundaries of the validation that you have done. I would like to see some of these data explored, summarized in the main text with additional data in the supplement.
  Reply:
  
  As described in Section 2.1, the cylinder for SWS (A) was filled immediately after the switchover (approximately 5 hours), and when the residual pressure in the other cylinder used as SWS (B) dropped below 0.1 MPa (after approximately one week, or three days with the CRDS), the flow path was switched and used as SWS. Therefore, there was no information on changes in concentration due to differences in the time between filling and the start of use. After introducing the CRDS, the working standard gas measurement interval was changed from 12 hours to 48 hours, which we modified in Section 2.1. At the same time, since the consumption flow rate has doubled, the SWS-gas changeover time has shortened. We have added the time series of SWS using the CRDS (Figs. S1, S2). No cylinder-dependent characteristics were observed; there were no changes dependent on the SWS concentration; there was also no change dependent on the residual pressure of the cylinder. Schibig et al. (2018) reported that the CO2 concentration in a 29.5 L aluminum cylinder increases by 0.090 ± 0.009 µmol mol-1 when dropping from 150 bar to 1 bar, but also note that this change is smaller if larger cylinders are used. As a 48 L cylinder was used in this system and the maximum pressure was 3.5 bar, it is considered that there was no increase in CO2 concentration due to the decrease in residual pressure. The method of this system is to use the SWS output to compensate for variations in sensor output that depend on overall environmental changes, including temperature and air pressure, so it is not intended to look at variations in SWS output in response to individual factors. Based on the flags described in the text, when the output showed abnormal fluctuations, the data were rejected as a failure of stable measurement. We quoted the result by Schibig et al. (2018) in the text and added an explanation.
  Schibig, M. F., Kitzis, D., and Tans, P. P.: Experiments with CO2-in-air reference gases in high-pressure aluminum cylinders, Atmospheric Measurement Techniques, 11, 5565-5586, 10.5194/amt-11-5565-2018, 2018.
  
  Comment:
  
  In a similar vein, you have long time series measurements of the WS tanks, both with the CRDS and the other sensors. The CRDS could give you some idea of WS variability and WS to WS accuracy as new WS tanks are swapped in, and would help you tie your measurements down to the absolute scale. And the WS measurements on the NDIR and TOF would give an idea of the nature of the drift of these sensors. In particular, it would give insight into whether drifts appear as an offset in voltage or span change in voltage (see comment #3, below), and the nature in the errors in the sensor with ambient temperature, ambient pressure, time of year, etc. These analyses would be helpful to the readers of this article.
  Reply:
  
  According to comments from reviewer #1, the need to calibrate the CRDS at 2-week to 1-month intervals, as recommended by ICOS RI (2020), has been added. The CRDS was calibrated by the WS. Since the introduction of the CRDS, there has only been one WS replacement at each site. Unfortunately, however, the time to stay at the site was limited due to access difficulties, and measuring the old WS with the new WS was impossible. We would like to conduct such an analysis next time we get that opportunity. The span of each sensor is calibrated by measuring the WS every 12 hours. It is difficult to assess the dependence of each sensor output on temperature and pressure individually from this observation system itself, and this system is characterized by its ability to correct for such dependence with SWS while reducing WS consumption.
  
  Comment:
  
  The approach to using the SWS to extend the life of the WS is essentially to perform linear interpolation in voltage from the twice daily calibration of the SWS using the three WS tanks. That assumes that the correction to the sample data from the SWS measurements is applied as an offset to the sample voltage, regardless of the SWS measured. But could the correction also be applied as a slope correction? That is, a correction that is proportional to the measured voltage? How do you know which is the right approach for each sensor? Similarly, NDIR is known to have a nonlinear response. Does the TOS also have a non-ideal (i.e., not logarithmic) response? Are these nonlinearities significant relative to your target uncertainties? Do your data shed any light on these concerns? Please add some discussion of these issues in the text.
  Reply:
  
  As explained in Chapter 2.4, the outputs of the three WSs at the time of the sample measurement of the target were interpolated by weighting the two independent values ΔB and ΔE by the time from the sample measurement time, so both slope and intercept were corrected. Linearity was checked over the concentration range of the standard gas used; the output of the TOS varies logarithm to the concentration, so the linearity of the output was checked against the log value. The uncertainty was described at the end of Section 2.4.2. Also shown in Figure 5, the data were rejected when the linearity was poor. We consider that this corrected method has no issues regarding the linearity of the calibration line.
  
  Comment:
  
  In general, there are too few references. It would be good to add a few more references pointing to other calibration strategies used in the GHG monitoring community for CO2 and CH4, references to the individual measurement technologies, references to similar data sets in the regions (e.g. ZOTTO), etc. It would help to put this manuscript in better context relative to similar work.
  Reply:
  
  The following sentence has been added to the introduction.
  
  “After this, in Central Siberia, Winderlich et al. (2010) developed a measuring system without dehumidification using a CRDS (Cavity Ring-Down Spectroscopy) analyzer, an ingenious way to reduce the frequency of cylinder replacement.”
  Winderlich, J., Chen, H., Gerbig, C., Seifert, T., Kolle, O., Lavric, J. V., Kaiser, C., Hoefer, A., and Heimann, M.: Continuous low-maintenance CO2/CH4/H2O measurements at the Zotino Tall Tower Observatory (ZOTTO) in Central Siberia, Atmospheric Measurement Techniques, 3, 1113-1128, 10.5194/amt-3-1113-2010, 2010.
  
  Comment:
  
  L105-108: It’s not clear the sequence of preparing and using the two SWS cylinders A and B. For example, when A is being used, is B being filled? Does a cylinder ever get used once it drops below 0.1 MPa? Is the cylinder pumped out prior to refilling? Do you have a time delay to allow the gas in the cylinder to equilibrate? Do you fill the cylinders at a certain time in the day?
  Reply:
  
  We have rewritten as follows.
  
  “An aluminum cylinder (0.048 m3) for SWS-gas was automatically exchanged when the inner pressure decreased below 0.1 MPa, then soon air from the highest inlet was compressed by a pump (LOA-P103-NO, GAST, USA) into the cylinder for about 5 hours to approximately 0.35 MPa, after having been passed through a similar triple dehumidification path as the sampled air (a stainless steel water trap, a semipermeable membrane dryer (SWF- M06–400, AGC, Japan), and magnesium perchlorate). It was preserved for about one week (three days with the CRDS) for usage until the inner pressure in one used for measurements decreased below 0.1 MPa.”
  
  Comment:
  
  The plumbing schematic of the SWS and associated manifold for filling and dispensing gas, with the double arrows and the 3-way valve bridge, is confusing. Showing the NC, NO, and COM on the valves, and then a table with the states (e.g., FILL A, USE B, etc..) would be helpful here in reproducing the setup.
  Reply:
  
  Corrected with reference to the comments.
  
  Comment:
  
  L126-127: “had a more significant standard deviation than a determined threshold (2 mV for CO2, 5 mV for CH4).” Reporting fluctuations in mV without describing the mapping of voltage to concentration, it difficult to understand the context of what this noise, and the selected thresholds, mean. A short section describing the response of the detectors prior to this would be good.
  Reply:
  
  This is a step at the beginning of the data processing to remove obviously incorrect data, and the voltage set is empirical. It is plausible that if voltage values are shown at this stage, it is not clear what they mean, so the relevant sentences have been deleted.
  
  Comment:
  
  L135: It’s often considered a good idea not to perform calibrations at exactly the same day and time. Why was 12 hours chosen as the interval? Do you see persistent differences between the day and night calibrations?
  Reply:
  
  Thanks for the comment. It was set up like this when it was designed by Watai et al. (2010), so it was left as it was, but we will consider setting it up for 13 hours in the future. The day/night calibration makes no difference.
  
  Comment:
  
  L319 and L327: Correlation plots are useful, but residual plots (TOS – CRDS and NDIR – CRDS) would be instructive in understanding any concentration dependence to the results.
  
  Similarly, residual plots over time in the supplement would help understand the potential for biases in the measurements.
  Reply:
  
  We have replaced the correlation plots with residual plots.
  
  We have also added residual plots over time in the supplement.
  
  Comment:
  
  A particularly useful plot would be to show σsws vs the std. deviation of the difference between the sensor data and the CRDS over time and/or as a scatter plot. A plot like these would illuminate whether your uncertainty estimates properly capture the actual uncertainties in the sensor measurements using the SWS.
  Reply:
  
  After introducing the CRDS, the working standard gas measurement interval was changed to 48 hours. At the same time, since the consumption flow rate has doubled, the SWS-gas changeover time has decreased to approximately three days, which makes the number of SWS measurements with NDIR/TOS only three times at most. As shown in the revised Figures S1 and S2, the concentrations of SWS measured with CRDS were stable between WS measurements, which indicates that the concentrations in SWS didn’t change for the periods. Thus, σsws is considered to capture the actual uncertainty. Of course, the values obtained from a few data are for reference only (red dots in Figures S3-S17).
  
  Comment:
  
  L319: There is a striking set of outliers just below the 1:1 line in the 2018 panel (at about 415 ppm CO2). There are two similar though less obvious outlier groupings in the 2020 data panel as well. What is the cause of these discrepancies? Shouldn’t they have been trapped by the data quality checks? Some explanation in the text would be helpful.
  
  Reply:
  
  Thank you for finding these data. The 2018 data were observed in January, and the 2020 data were observed in July and November, which had been dropped from the quality check. We have removed them and will also remove them from the published dataset in the next update.
  
  Comment:
  
  L12: “to a level less than one order of magnitude smaller” – unclear whether change is greater than or less than 1 order of magnitude. Reword for clarity.
  Reply:
  
  In response to Reviewer #1's comments, the relevant sentence has been corrected as follows.
  
  “We have revised a calculation method of concentrations and uncertainties for in-situ CO2 and CH4 measurements with a working standard gas saving system.”
  
  Comment:
  
  L14: “…consisted of this system installed at nine sites…” – awkward. Reword, please.
  Reply:
  
  We have rewritten as follows.
  
  “JR-STATION (Japan-Russia Siberian Tall Tower Inland Observation Network) was made up of this system, which was installed across nine different sites in Siberia”
  
  Comment:
  
  L22: “respectively” – unnecessary.
  Reply:
  
  We have removed it.
  
  Comment:
  
  L50: “sucked” – colloquial language. Try “pulled” or another word.
  Reply:
  
  We have replaced it with “pulled”.
  
  Citation: https://doi.org/10.5194/amt-2023-246-AC1

Peer review completion

AR: Author's response | RR: Referee report | ED: Editor decision | EF: Editorial file upload

AR by Motoki Sasakawa on behalf of the Authors (02 Aug 2024) Author's response Author's tracked changes Manuscript

ED: Reconsider after major revisions (08 Sep 2024) by Tim Arnold

Thank you for your replies to the reviewers’ comments. The reviewers point out the importance of these measurements and therefore publication of the methods behind them is of significant value. I have looked at the authors’ replies and concluded that more work is needed to adequately address the reviewers’ concerns. Please see more specific information below.

In reply to reviewer 1’s comment starting ‘The accuracy of the assignments...’ the authors suggest, as uncertainty can’t be determined precisely, it should not be included. Alone, this is not a convincing reason to leave it out of the total uncertainty budget, however, it is a sensible idea to report an estimate for the scale (and scale transfer) uncertainty separate from your in situ measurement uncertainty. This is a commonly understood approach – to report scale/transfer uncertainty separately. I would therefore suggest rewording and reformulating this section appropriately. It is also not explained why the U(x) is distributed between 0.025 and 0.043 ppm.

The working standards are prepared from a CO2 and methane source (with isotopic distinctions). Is this isotopic issue (scale transfer from WS to air) treated as an error or is it a consistent bias? If it is a consistent bias why is it not corrected for? Is it the isotopic composition difference that goes some way to explain the offset with NOAA/WMO scale? The authors need to make more effort to explain this problem and how it should be treated. This issue comes up again later by reviewer 1.

Regarding reviewer 1’s comment ‘The authors state that “From the point of this procedure, the SWS-gas equates with the 'target tank' or 'surveillance tank' mentioned in the GAW report (WMO, 2018)”. However, the authors’ response needs slightly more explanation – how exactly are consecutive measurements of the SWS used to calculate short term repeatability?

Regarding reviewer 1’s comment ‘Reference is being made to the GAW report with the WMO CO2 Expertʼs meeting’, the authors have not made sufficient attempt to reply to the reviewer. Users of the data should have an understanding of how ‘compatible’ the measurements are with other global measurments. The authors allude to this when they say ‘but this is a target for background sites such as coastal areas and does not need to be strictly adapted to an observation area such as this study’. This might suggest to potential users of the data (e.g. global modellers) that these data cannot be used as a harmonised dataset with the global measurements. Some insight from the authors here would be useful. What about mention of the extended WMO compatibility goals? Do the measurements fall within those goals? This will help the ultimate end users on consideration of how they can use these valuable measurements.

Re X2019 and X2007 scale comparison, please add more information in the text on how to do this, as suggested by the reviewer.

Reviewer 1 brings up the issue of using atmospheric variability in the uncertainty calculation. As the reviewer states, the measurement uncertainty is distinct from measured variability. It should be up to the end user to decide how to incorporate the air variability, as this may depend on how they use and how the data is ultimately consolidated e.g. averaged over minutes vs hours depending on the model for interpretation. Why not calculate measurement uncertainty and report separately the air variability within that time interval? Later in the reviewer’s comments this comes up again and the authors argue that ‘uncertainty in the ambient air CO2 concentration’ is appropriate. I agree with the reviewer that this represents a metric for real variability within a time period, not uncertainty. The equation may well not be appropriate as it stands and the reviewer’s concern needs to be technically addressed thoroughly.

Many of reviewer 2’s comments are adequately answered in the reply, however, please consider adding more explanation to the manuscript as readers are likely to have the same questions and will not refer to the discussion of the mauscript.

Reviewer 2 highlights the need for more references and more context. The Introduction section is significantly lacking in detail and scene setting. For a reader closely involved in this work the introduction might be adequate but the authors are missing an opportunity to explain the problem and motivation for the work thoroughly.

Reviewer 2 asks several questions in their comment ‘L105-108: Itʼs not clear the sequence of preparing and using the two SWS cylinders A and B…’ but many of these questions have not been answered or it is not already clear of any answer.

Reviewer 2 comment: ‘The plumbing schematic..’ Please can the authors be more thorough in their reply. Has the suggested table been created? If not, why not? Etc.

Reviewer comment about timing of calibrations: Again, please also reply to the comment within the manuscript itself. Why is 12 hrs the calibration interval and was any day/night difference observable?

Reviewer 2 found data that should have been flagged originally (Jan 2018, July/Nov 2020). This is quite concerning and an explanation isn’t given as to why these were not flagged in the routine quality procedures. I think the reviewer is correct to request information on the quality checks and procedures that go in to delivering the published datasets. This could be a devoted subsection.

Some further comments:

- Strictly the measurement is of an ‘amount fraction’ or ‘mole fraction’, not a concentration. Although concentration is often used for general communication, for a specialised measurement journal it is advised to use the correct measurement terms. I encourage changing the text appropriately throughout.
- Title suggestion? ‘Revised methodology for CO2 and CH4 measurements at remote sites using a working standard gas saving system’. It can be assumed that uncertainties are part of this.
- In the abstract ‘after gas replacement time’ – not clear what this means.
- Please replace wording around ‘ingenius’ in abstract with ‘..analyzer as a way..’
- Line 88 ‘in the selector’ written twice
- Line 144: delete ‘the spane’, ‘calibrate the sensors’ is sufficient.
- Line 423 remove ‘and timed’

Please prepare a revised manuscript which addresses all the reviewers concerns more thoroughly and answers the above highlighted issues.

Hide

AR by Motoki Sasakawa on behalf of the Authors (20 Sep 2024) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (23 Sep 2024) by Tim Arnold

RR by Anonymous Referee #1 (07 Oct 2024)

Suggestions for revision or reasons for rejection

Thank you to the authors for their revisions. What I still feel is not yet complete is a combined uncertainty covering all aspects that would provide the information on the data compatibility. This involves a few items that are remaining.

1. Authors reply:
We have added the following description in Section 2.4.2.
“The NIES09 scale is based on the gravimetric primary standard gases using a one-step dilution (Tohjima et al., 2006), and its overall uncertainty, including transfer, is estimated to be 0.043 ppm (Machida et al., 2011). This value is equal to , but this term is not included in the calculation because cannot be determined precisely; as for the value of , it distributed between 0.025 ppm to 0.043 ppm, since there are three WSs.”

a) This only addresses CO2. It does not give information on the CH4 propagation uncertainty.
b) In Tohjima et al 2006 on the one hand side an uncertainty estimate is made for the methodology that the NIES 09 scale is based on that results in an uncertainty of 0.032 ppm (based on uncertainties from gas purities, mass measurements etc.). The associated uncertainty terms would likely bias all primary standards prepared by the method in the same way and would not introduce scatter within a set of standards prepared in the same way. This very scatter of the deviations for individual primary standards relative to a common reference (NIES95 scale) is also provided in Tohjima et al. 2006 as 0.042 ppm for 15 individual standards produced in three batches each a few months apart. This number then has been quoted as “preparing uncertainty of one-step dilution cylinder” by Machida et al, 2011. Machida et al. 2011 quote an additional uncertainty for the scale transfer to the set of scale secondaries with the “uncertainty in standard transfer process by our NDIR calibration system was estimated to be 0.010 ppm”. This scale propagation was essential for maintaining the NIES 09 scale (CO2 in the primary standard cylinders was suspected to drift) so it is clear that utmost attention has been spent for this transfer. I wonder if the same effort has been spent each time in assigning the Siberian WS. Machida et al, 2011 also explain that comparison between the scale secondaries and another set resulted in variable results that were within ± 0.06 ppm. Therefore, perhaps this might be a more realistic uncertainty estimate to quote for the assignment of the WS (in addition to the scale uncertainty)? The uncertainty that the WMO-CCL has estimated for assignments of standards using the NDIR technology is 0.03 ppm that they came up with based on very thorough quality control documentation. Achieving a measurement uncertainty of 0.01 ppm for assignment of standards over a long period using NDIR appears like an extremely challenging task. Therefore, I feel that such an estimate requires to present more evidence than the reference of the 2009 meeting report.
The following sentences (starting l. 265) are confusing to me “The production of standard gases is typically conducted concurrently using the same equipment (such as dilution vessels and pressure gauges). This process may introduce common biases, albeit to varying degrees, potentially resulting in correlated (non-independent) mole fractions.” My thinking has been that the JP-Siberian WS were assigned by NDIR comparison measurements vs. the NIES09 scale standards and I do not understand what the referred hardware (as dilution vessels and pressure gauges) should add as bias.
In my understanding the scale uncertainty of 0.043 ppm is specifying the uncertainty estimate made for the scale and thus cannot be reduced by using multiple standards at stations. As stated before I assume there needs to be an additional scale propagation uncertainty term be accounted for each standard assignment and the number of calibration standards employed then will reduce the uncertainty.

2. Authors reply:
The NIES CO2 scale agreed with the WMO scale to within 0.1 ppm, even with the results of Round Robin 5, which was measured in 2009 at NIES. This point was added in addition to CH4 scale. The results for Round Robin 4 are not shown as they were with a previous scale (NIES 95 scale). No other published data are available to directly compare the scales.

Machida et al. 2011 provide a NIES 95 – NIES09 scale conversion function for CO2 with the following statement “The uncertainty of CO2 calculation using equation (2) is 0.01 ppm”.
CO2[09] (ppm)= 1.0519x10-4x(CO2[95])2 + 0.922433xCO2[95] + 14.313
So it is not clear why NIES95 data are assumed to be not available for a comparison.

3. Authors reply:
NDIR calibrations were performed by linear regression, as shown in Figure 5, where the coefficient of determination is greater than or equal to 0.999; if it is less than this, the data is rejected. We have replaced the correlation plots with residual plots (Figure 6), which show linearity on the NDIR, taking the CRDS as a linear reference.

While the rejection criterion of 0.999 ensures that bad calibrations are filtered out, it does not mean that there is no remaining uncertainty from a linear regression fit that is caused by a non-linearity of the response. The Figure 6 and the corresponding figures S18 and S19 in the supplement material are useful. Yet, it is not possible to confirm or invalidate a systematic mole fraction dependent bias resulting from the assumption of a linear NDIR response based on the single data 3-minute average comparison plots. Those plots are too busy and the y-axis scale too large. That might perhaps become visible if a quadratic fit was made through the data points presented in Fig. 6, S18, S19 and could be quantified from that. Else an estimate for the uncertainty caused by application of a linear response curve based on laboratory or literature results would also be appropriate. Machida et al 2011 applied a cubic fit curve to a multipoint NDIR calibration.

As the authors state most of the scatter of the difference between the two analysers measurement data will be due to the atmospheric variability. That should cancel out, though, and there is a systematic -0.1 to -0.15 ppm CO2 discrepancy (NDIR-CRDS) at all sites and all years that are displayed. I do not follow the authors conclusion: “There was no significant difference in CO2 mole fraction regardless of the inlet at both altitudes (Fig. 6, Supplement Figs. S18-S20).” just because the mean absolute offset is less than the SD in those figures because the SD will mostly be resulting from the atmospheric variability and several thousand data points this will be significant.

4. Authors reply:
We have added the figures for differences vs time in the Supplement (Figures S20 and S21), which show a good agreement.

These figures are a good way to provide that information. To display the longer term reproducibility it would be helpful to aggregate them (e.g. have daily means of the inter-instrument differences and the std.dev. of those daily data points) and have a smaller y-axis range instead (e.g. by factor of ten). The large y-axis scale range currently limits the information content considerably.

5. Authors reply:
We have added the following sentences in Section 2.4.3.
“On the other hand, the CO2 (CH4) concentrations in these observations can fluctuate on the order of ppm (10 ppb), even during a few hours of daytime when the atmosphere is well mixed (Sasakawa et al., 2010, 2013). It is, therefore, considered adequate for observations carried out in the vicinity of strong emission and absorption sources, such as those in the Siberian interior. The GAW report states that the compatibility goal for CO2 (CH4) in the Northern Hemisphere is 0.1 ppm (2 ppb), but this is a target for background sites such as coastal areas and does not need to be strictly adapted to an observation area such as this study.”

The GAW compatibility goals shall incorporate all uncertainties of the atmospheric measurements (with atmospheric variability not being part of this) and that is something that the manuscript has not yet fully provided. So I still see the usefulness of a summary of all uncertainties that combines assignment uncertainties, method repeatability (as approximated by sigma SWS), a non-linearity uncertainty estimate, isotopologue sensitivity uncertainty, …

6. Authors reply:
“As for the isotope effect, Tohjima et al. (2009) reported that the errors for isotopic influence for the three NDIR analyzers range from -0.04 to -0.08 ppm. However, the apparent NDIR CO2 mole fraction error depends on its individual sensitivity to the optical filter property. The installation of this observation system
started in the early 2000s, but the characteristics of each NDIR are not known, as this was not
yet taken into account at that time.”
In order to check each optical filter property, we need to get them back to our laboratory in Japan, which is impossible. So, we have described the possible bias: “The CO2 values measured by the system may appear low in the range of up to 0.08 ppm.”

While more depleted reference standards compared to the atmospheric samples will cause a bias, the referenced publication (Tohjima et al., 2009) also shows that different individual analysers have slightly different isotopologue discriminations. So there will also be this different behaviour between the calibration NDIR instrument in the NIES lab and the analyzer in the field. The apparent differences between three such NDIR analyzers presented in that 2009 publication suggests an additional 0.07 ppm uncertainty resulting from this.

7. Authors reply:
The uncertainty of the analysis system is expressed in sigmasws using SWS-gas, as denoted in Section 2.4.3, not here. As required by each element of equation (22), equation l.249 is part of the uncertainty in the estimated concentration (x), including atmospheric variability over a 3-minute period. We consider this equation to be appropriate.

I guess I do not fully understand this but the previous suggestion might also have been misunderstood. My point was to rather not use the 3-min-atmospheric-variability for the first term of equation (23-1) but rather approximate the detector’s 3 min measurement scatter by the respective 3-min-SD of the weighted SWS SD in order to not include the atmospheric variability (following the advice of GAW report 255).

8. Authors reply:
The WBT-based data selection has been indicated in Section 2.5. It is now not shown in the Figures.

Please introduce the acronym WBT (WarmBoxTemperature) somewhere in the manuscript. This is missing.

One small detail:
All Figures S4-S17 refer in their caption to “The same Figure as Figure S1…” this should rather be S3.

Hide

ED: Reconsider after major revisions (11 Oct 2024) by Tim Arnold

AR by Motoki Sasakawa on behalf of the Authors (05 Jan 2025) Author's response Author's tracked changes Manuscript

ED: Publish subject to minor revisions (review by editor) (04 Feb 2025) by Tim Arnold

AR by Motoki Sasakawa on behalf of the Authors (05 Feb 2025) Author's response Author's tracked changes Manuscript

ED: Publish subject to minor revisions (review by editor) (06 Feb 2025) by Tim Arnold

AR by Motoki Sasakawa on behalf of the Authors (11 Feb 2025) Author's response Author's tracked changes Manuscript

ED: Publish as is (12 Feb 2025) by Tim Arnold

AR by Motoki Sasakawa on behalf of the Authors (13 Feb 2025)

Short summary

Standard gases are essential for accurate greenhouse gas measurements. However, exchanging cylinders at remote sites presents logistical challenges, requiring systems that minimize gas consumption. We developed methods for calculating greenhouse gas mole fractions and uncertainties using our original system designed to reduce standard gas use. We validated its long-term stability through instrument comparisons. The system has proven effective for maintaining observations at remote sites.

Revised methodology for CO2 and CH4 measurements at remote sites using a working standard-gas-saving system

Download

Interactive discussion

Peer review completion

Suggestions for revision or reasons for rejection

Revised methodology for CO₂ and CH₄ measurements at remote sites using a working standard-gas-saving system