Azevedo et al. aim to characterize two Senseair K30 NDIR CO₂ sensors for use in UAS-based atmospheric measurements. They try to isolate the impact of individual environmental parameters on CO₂ readings and investigate the sensors’ time response to pressure. Overall, this topic is of great interest to readers of AMT, as the low cost and easy handling of NDIR sensors has made them widespread in the monitoring of atmospheric CO₂. While analyses of this nature are not novel, high-quality data and analyses - especially with regards to improving sensor accuracy beyond design specifications are of great interest.

The presentation quality of this paper is good, especially with regards to language. However, this paper falls short on the quality of its data and analysis. Regarding the influence of environmental parameters, the presented data is subject to external interferences which are left untreated and the analyses of said data lack the scientific rigor required to provide a significant contribution. The investigation of the time response to pressure is of interest, since these kinds of phenomena are difficult to treat in post-processing, however the quality of analysis is insufficient to be of practical use to readers. I therefore propose to reconsider the publication of this manuscript after major revisions.

General comments:

• It is difficult to review the scientific validity of the author’s comments regarding the different sensors’ dependences on environmental parameters, as the analyses are mostly of a qualitative nature. Increasing the level of quantification would also enable a comparison of their results with other analyses and would be of chief interest for readers of this paper, as they will potentially be looking to apply these corrections to their sensors.
  • Some examples (not complete):
    • L 148 “This hypothesis is supported by a stronger correlation between the test and independent sensors.”
    • L 156/157 “In this experiment, the absence of humidity dependence is evident.”
    • L 174/175 “This experiment showed an extreme dependence between the CO₂ concentration values […] and pressure.”
  • For reference on possible metrics and desired level of quantification and rigor, please consult publications in this journal like Arzoumanian (2019; doi:10.5194/amt-12-2665-2019)
• For all fits in this study, only r-values are provided and slope as well as intercept including respective errors are missing. Additionally, the factors found for the corrections (incl. time response correction) are not presented for any of the analyses, so neither the absolute dependence on environmental variables (or time response) nor the difference in dependence between the two sensors can be evaluated or compared to the scientific literature.
• Since not all environmental parameters are controlled in Benchtop experiments, showing the other environmental factors in the time-series and analysing them is important to ensure the isolation of relevant parameters which is a central goal of this paper.
• Some of the experiments were conducted months apart. Since NDIR sensors’ optical properties (and thus dependences on environmental parameters) are known to change with time, I recommend against putting these experiments simply next to each other without some further justification or analysis.

Scientific comments

• L 60 “We also isolated the effects of pressure, temperature, and relative humidity on an NDIR sensor and analyzed their impact separately.”
  • Did you consider also performing a linearity analysis?
• L 79 “[…] creating appropriate conditions to simulate UAS flights.”
  • Is this the case for a pressure range from 1050 to 600 hPa (~4km height) as used in the Mesonet Pressure Dependence Experiment?
• L 87 “[…] this article considered only the CO₂ concentration values reported by each sensor unit.”
  • How were the Senseair K30-FR units configured? Which kind of onboard compensations do they already have and which ones were active?
• Li-840A temperature independence
  • Please check your assumptions here, as the manual provides a total drift of the Licor of 0.4 ppm/°C. For the temperature range of 10 to 40C, this would equate to a drift of 12ppm, for 20 to 40C, a drift of 8ppm.
• Reference sensor deployment
  • Why did you choose to setup the 840A inside the chamber and the 820 outside? Was this an arbitrary choice?

• Mesonet Temperature Dependence Experiment
  • Why is the data range for CO₂ 160 to 320 ppm? These measurements don’t seem to be realistic.
  • If you consider the independent sensors to be actually independent from environmental factors, why don’t you use those to correct for external disturbances? Analysing the difference between the test sensors and the independent sensor for temperature dependencies would, then, yield more robust results.
  • As you performed the pressure experiment already 9 days before the temperature experiments, the high pressure dependence was already known. Did you perform any pressure measurements during the temperature experiment to disentangle the influence of the ambient pressure changes from the temperature changes? Or did you assess the possible effect the pressure dependence might have on the measurements?
  • Did you measure the temperature at the external independent sensor? Did you preclude any temperature interference here?
  • Quantification:
    • As one of the major sources of temperature dependence in NDIR spectrometers is a change in optics properties, did you consider time-lags in increasing sensor reading?
    • For which time periods did you calculate the correlation coefficients between sensor reading and temperature? If these were the actual transitions, this might be problematic because of time-lag, non-linearities or material effects.
    • After identifying an external interference in your measurements, it seems to me like you should not be able to draw conclusions about a sensor temperature dependence without EITHER: discarding the data and repeating the experiment without this source of interference (even Gaynullin, 2016 says that an “[a]bsolute elimination of contaminating leakages from ambient air” must be provided “to provide a reliable test environment”) OR: treating the interference by e.g. analysing the difference in readings between the Licor and K30 sensors (under the assumption that the Licor readings are independent of temperature, which has to be proven given its design specifications – cf. above)

• Benchtop Temperature Dependence Experiment
  • CO₂ readings seem to decrease at the beginning of the experiment. Did you ensure the experiment was in a steady state with respect to CO₂?
  • L130/131 “Even though there is a slight 10 ppm increase […], it occurs a full minute after the temperature is brought back near its original state.”
    • Again, might this be due to the temperature response time of optics/electronics, a large mixing volume or even some drift in timestamps of the recording PC (especially since this experiment was conducted 18 months after the other ones)? If this were the case, the increase in signal would fit very well with the expected one of the Licor due to temperature drift, which is 8ppm (cf. above).
    • Where was the temperature measured? Was the possibility of a time-lag between temperature measurement and sensor measurement of an air parcel excluded?
  • Since this experiment does not control RH, it is difficult to use this for a impact analysis isolating the environmental parameters, which is the stated aim of the paper. This would need some further justification leveraging measured RH values during the experiment.

• Mesonet Relative Humidity Dependence Experiment
  • The same comment regarding the external interference applies here (cf. Mesonet Temperature Dependence Experiment).
  • The same comment regarding the CO₂ data range (120-300 ppm) applies here (cf. Mesonet Temperature Dependence Experiment).
• Benchtop Relative Humidity Dependence Experiment
  • Details of the humid air source are missing – is it guaranteed that temperature stays constant and only RH increases?

• L 188/189 “the correction needs to be based on the variation magnitude from the initial state”
  • Can you provide some physical justification for this assertion? This is not discussed for temperature or humidity and to my knowledge, it is not common practice.
• L 202 “The results demonstrate four instances […]”
  • Where there further instances where this method was applied unsuccessfully?
  • Is this a robust method with respect to e.g. time and temperature?
• Fig 12: This dependence seems very linear, almost as if it were following the ideal gas law.
  • Again, please add some comments regarding pressure correction being enabled or disabled in the Senseair K30 units.
• Fig 13: There seems to be some kind of change of pressure dependence with time (similar to hysteresis effect)
  • Was this analysed further?
  • Could this be caused by the large pressure range?
• L 211 “Because the pressure chamber is completely isolated from the external environment […]”:
  • In the experiment before, this chamber was not isolated which is why you had the outside reference sensor. In L 207/208 you write “the pressure correction experiment setup […] was used again” Is this not the same, non-isolated setup?

• L 222/223
  • Why did you use a different data source for estimating the exponential correction time constant and the time shift? Did you consider correcting the time shift first to then estimate the exponential correction time constant from this corrected data?
• L 225 “[…] an idealized signal was created […]“
  • How did you create this idealized signal? The experiments with this data are not shown in this paper. It would be important to at least add them to the supplement for the sake of documentation and repeatability.
  • Why did you opt for creating an idealized signal rather than applying the pressure correction to the exponential and time shift corrected data? Is the purpose of the exponential- and time shift correction not to improve the pressure correction?
• L 231 “Unfortunately, the attempted correction was not as effective on the gradual pressure changes”
  • Where do you see a difference in expected and real outcome? Did you quantify this?
• L 234 “we recommend repeating these experiments on a better quality chamber”
  • Why do you think the quality of the chamber is deficient? Why would smaller pressure changes help?

Technical comments

• L 48 None of the sensors available in the market was designed for UAS-based deployment.
  • None of the sensors available on the market were designed for UAS-based deployment.
• Fig 1:
  • Naming the sensor “CO2 Independent Sensor” is misleading as one might understand the sensor to be independent of CO₂ instead of temperature. I suggest amending the names.
• Fig 2:
  • Line colors:
    • Please consider choosing colors which have more contrast. Without zooming in, it is difficult to differentiate the lines.
    • Why do test sensor 1 and 2 have different levels of saturation? Visually, it seems like sensor 2 is less important.
  • Line types:
    • It is difficult to see the dashed and dash-dotted lines and doubly so to differentiate the two. Again, maybe rethink the figure layout.
• Figure 3:
  • For scatter plots between two CO₂ readings, always plot the 1:1 line, as the linear fit will be misinterpreted visually to be this 1:1 correlation. Alternatively, one can also restrict the line of the linear fit to the data and use the same x- as y-limits albeit for a larger span, ensuring the 1:1 line is at a 45 degree angle.
• Fig 3, 6, 8, 10:
  • Please consider using shared x- as well as shared y-axes. This makes such plots easier to read.
• Fig 16:
  • RMSEs by definition cannot be negative
  • Please add Pressure readings as one aim of this figure is to show the quality of the pressure correction.
• L 196 “All cases emulate a typical UAS-based CO₂ vertical profile”
  • Did you mean flight profile?
• L 196/197 “[…] there is a dwell period (in this case, 1.5 minutes) to ensure samples from the previous altitude are discarded from the system after a change in altitude.”
  • change altitude to pressure or use “simulated altitude”
• L 207 “No mention of such affect was found […]”:
  • No mention of such an effect was found […]
• L 233 “For those who”
  • For those, for whom
• L 257
  • Please consider adding the statements with regards to long-term stability also to section 1.1 or 2.

Summary:

Azevedo et al. 2022 present a laboratory study on lower-cost commercial CO2 sensors from SenseAir AB (type: K30). Building on previous work they extend the range of environmental variables to find a calibration strategy that allows the use of such sensor on small unmanned aerial vehicle (UAV) systems in quantitative studies. The paper is clearly structured, easy to follow and well-written. The topic itself is of great interest as UAV monitoring systems for greenhouse gases continue to increase in their importance for local and facility scale monitoring.

However, there are serious concerns about the reliability of the results. Should the authors be able to address these major comments the paper could suitable for publication, but the amount of additional work necessary might be too challenging to be achievable in a few weeks.

General comments:

The first major concern is the lack of a calibrated and reliable reference dataset for ambient air CO2 dry air mole fractions (or as referred to in manuscript: the CO2 concentration). The authors LiCOR LI-840 and LI-820 systems as a reference. However, it is obvious that neither of the two instruments was appropriately calibrated. An indoor air concentration of 200-300ppm CO2 as reported in Figure 3 is completely unrealistic. Ambient clean air data from NOAA can be found here: Global Monitoring Laboratory - Carbon Cycle Greenhouse Gases (noaa.gov) demonstrating that 200-300 ppm is not possible unless in artificial gas mixtures or environments. Given that the LiCORs are not calibrated it is also unlikely that they were properly tested for cross-sensitivities, offsets or non-linearity. 

The LI-820 and LI-40 have known temperature dependent drifts. According to the LI-840 manual this the calibration drift is <0.4ppm per degree C 840A_Manual_10690.pdf | Powered by Box (boxenterprise.net)

How can the reader be sure that there isn’t a residual drift in the reference data?

2.) The range for the temperature calibration is too small and only a single test (at only one RH level) was conducted. Atmospheric temperatures outside the tropics frequently reach values below 10^oC, which seems to have been the lowest temperature setting tested in section 3. Also, the chamber experiment holds the temperature stable for multiple hours. Is this really a realistic temperature profile for a drone flight?

3.) It is unclear how/if the lab bench setup described in figure 4 was able to provide a homogeneously heated air-stream to all instruments. It would be necessary have many more temperature sensors placed around the 2xK30 and the Li-COR to be sure they measured the same (temperature) air. Furthermore, the lab bench tests measured a response to a short-term temperature change within a few minutes, while the chamber test duration was over 6 hours with 2 hours time for instruments to equilibrate. How can those to experiments be compared? The low correlation seen in Figure 6 could well be related to the change in time scale of the experiment.

Specific comments:

L1: Suggestion to mention that this study focusses on (lower-cost) NDIR sensors

L14: Please clarify: what does “mentioned measurement systems” refers to. Also please add a citation of studies that demonstrated the claim that no suitable measurement systems existed for local and regional scale work. Since the 2010s cavity ring down spectroscopy (CRDS) and integrated cavity output spectroscopy (ICOS) systems have been in regular use for atmospheric CO2 measurements and have allowed high-resolution and accurate measurements, even on mobile platforms (e.g. Chen at al. 2010, AMT - High-accuracy continuous airborne measurements of greenhouse gases (CO2 and CH4) using the cavity ring-down spectroscopy (CRDS) technique (copernicus.org)).

L85: You mention the need for instrument specific correction coefficients, but only decided to measure 2 instruments. How representative are two units? Martin et al. 2017 (AMT - Evaluation and environmental correction of ambient CO2 measurements from a low-cost NDIR sensor (copernicus.org)) tested at least 6 units of the K30 series.

Figure 2. It is very difficult to distinguish the time series of the different instruments.

L144, Figure 8: A linear fit does not seem appropriate for the left panels. Did you consider a non-linear instrument response?

L170: The temperature experiment, especially Figure4 clearly show that the concentration inside and outside the chamber can (and do) differ. Why do you consider the LI-840 on the outside as a reliable reference here, especially after the potential ‘unknown external interference’?

L214: This is a major concern: Can the results reported here be useful to other researchers, If the time delayed response to pressure changes is specific to the inlet and housing design?

L240: The accuracy of the instruments has not been investigated at all. No gas standards from NOAA or NIST was used here, neither were calibrated reference instruments.

The authors did demonstrate that they can reproduce measurements within 2.5ppm for same air sampling of another optical instruments under certain conditions.