Comments on the manuscript “High-precision δ13C-CO2 analysis from 1 mL of ambient atmospheric air via continuous flow IRMS: from sampling to storage to analysis.” from Joana Sauze et al., 2025.

Sauze et al., (2025) developed a method for the preparation and handling of small air sampling vessels with a volume of 1 mL STP for the analysis of stable carbon isotopes in CO2. The authors describe a series of experiments that lead to a proposed protocol for the preparation and handling of the air sample containers. They find the preparation steps are critical to achieve their best precision. The authors suggest this method opens a door to new analyses of stable carbon isotopes in CO2 from environments where the amount of available sample volume is critically limited, such as rhizosphere and chamber studies. Those applications hold the potential for important improvements of our understanding of carbon cycle processes, which is of great importance to atmospheric science and biogeochemistry, and the future of life on Earth as we know it in general.

I share the view that sampling volumes that small for high precision analyses is technically challenging. The topic of the manuscript is well suited for publication in AMT. However, I’d suggest significant rewriting, including further analysis existing data, and potentially making additional measurements. Some of the results do not seem to be of sufficient quality. The manuscript lacks fundamental basics on conventions and guidelines in isotope research. I will list a few examples and refrain from commenting on details.

Accuracy

The elephant in the room seems to be the lack of accuracy in the presented experiments. The authors focus entirely on “precision” as a quality objective (I can’t remember seeing a definition of precision, but assume standard deviation of a set of experiment results?) Almost all figures, including all of the results underpinning the finally suggested protocol (Figure 7) show significant offsets between target values and the achieved average values of a series of experiments. To me, the lack of accuracy suggests that something is not quite right with the method, and a critical experimental process not sufficiently controlled. The manuscript does not seek to explore the causes for the inaccuracy. I may be convinced otherwise, but the lack of technical details leaves a lot of room for speculation on the lack of accuracy. I suggest the accuracy problem to be fully explored, ideally with additional measurements, and the details to be provided.

Protocol and gases used to prepare sample vials

The authors report a difference when flushing the vials for 8 s with CO2-free air. Unfortunately, the flow rate of that flushing is not stated, which should be reported to understand the protocol. Afterwards, the protocol includes four cycles of evacuation to 0.1 bar, followed by filling with pure N2. The authors find a substantial reduction in variability of d13C-CO2 values when the initial flush with CO2-free air is included in the protocol (Figure 2). This suggests to me that the residual CO2 is not sufficiently removed by the four evacuation-flush cycles alone. CO2-free air and pure N2 should lead to the same result, if both have sufficiently low CO2 blank, and if both gases were used to quantitatively remove residual CO2. Have the authors tested the effect when pure N2 or CO2-free air are used interchangeably for flushing or evacuation/fill cycles under the same conditions? Demonstrating the effectiveness of the sample vessel preparation with increasing initial flush flow rate and/or flush time, as well as numbers of evacuation-fill cycles would be useful to determine a protocol that leads to accurate and precise values.

Standard practice in atmospheric science

It should be noted that preparing vessels for atmospheric sampling and isotope analysis by evacuating, flushing and even thermal treatment is absolute standard practice. A comprehensive protocol for glass flasks was previously published (Steur et al., 2023, DOI:10.1080/10256016.2023.2234594). Successful protocols work based on a large number of gas exchanges within the vessel to eliminate remains from previous samples (memory) as well as dealing with the small but significant amount of surface water on the internal surfaces.

Use of d18O-CO2 as indicator of analytical performance

The protocol described by Steur et al., 2023 is especially relevant for the analysis of oxygen isotopes in CO2, which are not considered in this manuscript. d18O-CO2 can be a very useful indicator for analytical problems. Therefore, I wonder if the d18O-CO2 data could help to identify the cause of the inaccuracy? For the purpose of method refinement, the precision (standard deviation) of d18O-CO2 from different experiments might be useful.

Control of residual CO2 and possible impact on d13C

The authors evacuate their sample vessel to 0.1 bar. In other words, around 10 % of the previous gas would still be present in every following preparation cycle. This seems to include the internal volume of the manifold (Figure 1), which is substantial in comparison to the volume of individual sample vials. This could potentially result in different gas compositions across the vials, especially as the manifold is filled with N2 from one side, pushing the gas from the previous filling towards the other side (pump side) of the manifold, where the vials at the pump end may potentially receive larger fractions of the previous gas filling. This contains some degree of speculation on my side, but I am not convinced that four cycles of evacuating to 0.1 bar and filling with N2 are a guarantee for quantitative replacement of the previous gas. It only takes 1 % of atmospheric CO2 with a d13C of around –8 ‰ and 99 % of a working standard CO2 with a d13C of around –36 ‰ to cause an offset of 0.3 ‰. Even a 1 % blank of a small sample peak might appear relatively small in the blank test the authors performed (line 93). Because of the small sample volume the presented method is targeting, system banks are very important, which the authors are well aware of. Vessels with septa can easily be evacuated to fractions of a mbar. Why did the authors choose not to evacuate to much lower pressure levels?

Blank test data

The authors performed blank tests (i.e., line 93), but do not present blank data, instead stating they didn’t find a detectable blank signal. I haven’t yet seen a system that has virtually no blank. There has always been some blank, and that blank is ideally smaller than the defined limits, above which Isodat/Qtegra automatically identifies a peak. Just because the software does not report the peak, this doesn’t necessarily mean there is no blank. Especially when measuring isotopes in small sample quantities, a small blank can have significant impact. I am not totally convinced that the variability in the “no flush” scenario shown in Figure 2 is not resulting from incomplete removal of the previous gas in the vial (memory). The d13C range in the “no flush” experiment is almost 2 ‰. A quick back of the envelope calculation suggests that around 6 % of ambient air CO2 with d13C of –8 ‰ would be needed to shift a CO2 with d13C of –38.7 ‰ by 2 ‰. This amount might well be detectable in the peak sizes of the measurements. It doesn’t seem that insufficient memory and blank control are fully explored in the manuscript and the underlying experiments. These information or experiments should be delivered in a manuscript that seeks to establish a new sample vial preparation protocol as the primary objective.

Undisclosed modifications to the IRMS instrument

The authors state their IRMS method includes modifications from Fiebig et al, (2005) to work for ambient CO2 mixing ratios (line 57), and “adapted here for high precision at trace levels of CO2” (line 112). However, there is no reference, no description or proof of what is done differently from Fiebig et al., (2005) and how that improved the analysis. I regard this as essential information. What are “trace-level CO2” in an atmospheric research journal? A fraction of lower tropospheric mole fraction averages?

Insufficient specification of used components

A large part of the success of the suggested sample vessel preparation protocol seems associated with the use of Terostat. Terostat seems to be a brand name for a range of sealing products and not a unique product. At no place do the authors explain what specific product they use and what it is made of, so it is impossible for a reader to gauge, or even to follow and adopt. Also, the description of how this is applied to top, and bottom could be more detailed, as the authors suggest this is a significant part to achieving high data quality.

Accurate description of system performance in context of literature

The authors state that the method is of “high” precision in title and throughout the text (i.e., line 113). However, 0.1 ‰ precision in an air sample of 1 mL STP is not particularly high or novel, i.e., Brand et al, (2016), DOI:10.1002/rcm.7587, achieve 0.04 ‰ for d13C in CO2 on 1 mL or air with GC-IRMS. Schmidt et al, (2011), https://doi.org/10.5194/amt-4-1445-2011, achieve 0.05 ‰ on around 3 mL pre-industrial air sublimated from an ice core sample. These methods have demonstrated accuracy to within their much smaller measurement uncertainty or better. The additional challenge and potentially the source of the additional uncertainty/inaccuracy of the method described by Sauze et al., (2025) may thus be associated with the sample vials, not really with the sample amount.

Data quality objectives and indicators for instrument performance

The authors seem to develop their quality control criterion around the precision value of 0.1 ‰. At no point do they provide an explanation why this value is needed for any analytical purpose. The smaller the precision, the better the protocol seems to be the paradigm. Given the precision criterion is the only data quality objective the authors seem to apply, the rationale for the choice of this value should be presented. A good example to question that approach is shown in Figure 6: The values from storage temperatures of –20C show a precision of 0.24 ‰ and are distributed around the target value with good accuracy. In contrast a precision of 0.1 ‰ is found at –80C, but then the values appear inaccurate. Yet, the storage at –80C is preferred because of the better precision, accepting inaccurate results. I’d suggest being very cautious of that rationale. Accuracy and reproducibility are at least as important as precision.

Composition of applied gases and gas equipment

Unfortunately, the authors do not disclose the compositions of the applied gases. Besides “ambient” there is no statement on the CO2 mole fractions in any of the applied gases. In a technical manuscript on CO2 isotope analysis, with particular focus on small sample sizes, knowing CO2 mole fractions of the applied gases is essential to understand experimental processes and results. Including data on the composition of used gases is obligatory for such a manuscript. For all experiments, the compositions of the gases and especially the CO2 mole fractions must be stated. The manuscript should be clear on mole fraction scales, impurities, calibration uncertainties, etc., as well as manufacturers and models of pressure regulators on those cylinders.

Isotope conventions

The manuscript ignores basic isotope conventions. All isotope reference gases used need to be stated with uncertainty and traceability chain (Camin et al., 2025, https://doi.org/10.1002/rcm.10018). This is important best practice, even though this manuscript does not show atmospheric data that a reader can compare to other measurements. The authors may have used a cylinder containing liquid CO2 as one of the reference gases (Figure 3), and possibly for some of the gas mixing etc. It should be stated whether this contains liquid CO2 as well as gaseous CO2, as this may affect the isotopic composition over time or with different use, i.e., when used for mixing. When referring to an isotope ratio, the isotope (d13C) is combined with the molecule (CO2) as d13C-CO2 or d13C(CO2) when referred to isotope values, where the “delta” is italicised. Negative isotope values are expressed with a long dash, rather than simple dash (Coplen 2011, https://doi.org/10.1002/rcm.5129)

References

The manuscript includes a lot of old references. There is nothing wrong with old references and credit should be given to original ideas, but in many cases, things have moved on and improved over several decades.

Review of Sauze et al. “High-precision δ¹³C-CO₂ analysis from 1 mL of ambient atmospheric air via continuous flow IRMS from sampling to storage analysis”

General Comment

This manuscript presents methodologies of vial treatment and continuous-flow IRMS measurement for δ¹³C-CO₂. They are fairly developed for ecosystem measurement applications and worth being shared in the isotope measurement community; therefore, this study is well within the scope of the journal. However, I do not recommend publication of the current manuscript until the following issues are addressed. In addition, Referee#1 already provided a long list of constructive comments, which I totally support and express sincere respect to the thorough evaluation.

Given that not a few papers on continuous-flow δ¹³C-CO₂ measurement in similar principles are already available over the last 20 years (those both cited and not cited in this study), I do not find the method novel, in contrast to what the authors write. They improved sampling and laboratory treatment methodologies for a reduced sample amount, and these careful descriptions are of value. However, from my point of view, many places in the manuscript look exaggerated in particular for novelty and outlook. I strongly recommend toning down with focus on actual application fields only.

One simple way to minimize storage effects is to employ a glass flask with a stopcock valve with a rubber O-ring, which has been long used in the atmospheric monitoring community (e.g., Worthy et al. 2023). The flasks are in many cases much larger (up to liters), but stability of sample air has been ensured for months to a year as the flaks are transported worldwide including Antarctica. In this regard, one might wonder why the authors take riskier septum, despite of availability of the safer method. I surmise there are several reasons such as practice in field deployment, familiarity and cost, but I did not find them well explained in the manuscript.

Throughout the manuscript, precision, which the authors consider important, is not consistently evaluated and described. In this type of experiments, I suppose that efforts are made to minimize measurement uncertainty, which includes uncertainties associated with every measurement process such as pre-treatment, storage, and isotope analysis. These processes affect both variability (precision) and systematic offset (accuracy). I believe that the authors would like to ensure minimized systematic offsets with acceptable measurement variability. Description and discussion on the former are incomplete in the current manuscript, thereby it is difficult to consider that the present method was evaluated sufficiently. Regarding the latter, the authors should define “precision” explicitly throughout the manuscript. I would use “repeatability” or “reproducibility” depending on the context.

The authors presumably consider that “precision” of 0.1 per mil is a criterion to evaluate quality of the measurement, but it is not clear why this value is appropriate. If the authors assume ecosystem measurement applications, one would see relatively large variability through a series of samples (e.g., some ‰), therefore relaxed “precision” might be acceptable.

I do not think that Table 1 is useful for readers, because the laboratory settings could change and some part of information is clear from Figure 3. Alternatively, a schematic of the whole measurement system including a sample vial connection (GasBench), ConFlo, and IRMS might be presented. The schematic along with the current Figure 3 should be presented in section 2.3.

Below my specific concerns are also detailed, but I think the authors might need corrections which span the entire manuscript, as several comments apply to several relevant places.

Specific Comment

P1 L10: The “stable” carbon isotopic “ratio”, if only δ¹³C of CO₂ is referred. One might include ¹⁴C if the term “composition” is preferred.

P1 L11: …, yet its extended application remains limited due to analytical and sampling restrictions.

P1 L14: “Terostat” here and other first place where it appears, the authors should describe what it is, not the trade name, e.g., sealant/adhesive tape?

P1 L19: “carbon stable” at another place, “stable carbon” was used.

P1 L23: “composition” to “ratio”. The bracket “(δ¹³C)” should come just after “ratio”, i.e., “The stable carbon isotopic ratio (δ¹³C) of atmospheric CO₂…”

P2 L37: It is weird that the authors mention to the analyzer of Picarro as emerging instrument, but no reference paper is cited. The authors later cite Sperlich et al. (2022), but their instrument was not Picarro. As Picarro is not the only availability, these sentences might be reformulated in a more balance way.

P2 L40: Picarro could measure sample air with 1/10 atmospheric CO₂ concentration (e.g., ~40 ppm), but it would make the precision worse. According to their data sheet, detection limit is not clearly defined, and also in principle it would not make sense. I would think that it is matter of whether the worse precision at low CO2 concentration is acceptable. Moreover, dilution measurement with laser-based instrument would cause different problem; change in matrix could cause measurement offset.

P2 L45: …during “mixture between atmospheric and ecosystem reservoirs.”

P2 L58: At other places, the term “concentration” is used. Is “mixing ratio” used with a different meaning?

P3 L86: “0.5 bar” is this absolute pressure or above ambient?

P4 Figure 1: I wondered that there may be a close valve at the vial side of the 3-way valve, otherwise the vials cannot be evacuated with N2 or exhaust connected.

P4 L107: Explain about “Terostat” here, so that readers without prior knowledge about the product can follow. And how was it used at where of the vial? Describe explicitly (“apply” it, as in the manuscript, says almost nothing).

P4 L109: As in my earlier comment, a schematic figure including the whole measurement system might help readers. The figure in the reference (Fiebig et al. 2005) represent only part of the system and the authors should present the system more in detail as they improved from the earlier one.

P4 L117: What is the cryofocus unit? Is it a tubing or a capillary of which material and size?

P5 L122: “the cryofocus was raised” to “the cryofocus column (or tubing as appropriate) was lifted out from liquid nitrogen”

P5 L129: This sentence does not describe traceability clearly. The authors should mention to the international reference material (RM) to which the reported values are traceable eventually. It is unclear if the authors determined δ¹³C value of the working standard against an RM or if they have a suit of different gases whose δ¹³C values were determined against an RM.

P5 L133: “…, and analytical precision consistently reached ±0.1 ‰ for δ¹³C” here and other places, it is unclear if “reached” means whether the value is larger than 0.1 ‰ or not. I would avoid ambiguous verbs or adjectives when large or small matters. For instance, here the sentence could be rephrased like: …analytical precision of δ¹³C was consistently <0.1 ‰. Note my comment on “precision” and consider it consistently throughout the manuscript.

P5 Table 1: See my earlier comment.

P6 L147: In section 2.1, the authors described (O₂-free) N₂ was used to prevent possible N₂O production at ion source, but here they use air containing both N₂ and O₂.

P6 L155: It is unclear if the δ¹³C value is a nominal one from the gas company (Air Liquid) or that determined by the authors’ laboratory so as to be traceable to an RM. This is important because it would be no wonder if the values have independent traceabilities.

P6 L158: See my earlier comment. If the term “precision” refers to standard deviation, define so at early place of the manuscript.

P7 L169: This section and Figure 3 could be merged into section 2.3.

P8 L186: “50 μL” this amount depends on the CO₂ concentration of sample air. How much concentration corresponds to the sample amount 50 μL?

P8 L194: Same comment as P5 L133. Does “exceed” mean smaller or larger than 0.1 ‰?

P9 L210: “reduce” delete “s”; the risk of gas leakage, diffusion “and associated” isotopic drift

P10 L233: Same comment as P6 L155.

P11 L11: “…improved measurement precision” it seems to me more important that the low temperature storage resulted in the values in agreement to the nominal value, than the magnitudes of the error bars.

P11 L257: Same comment as P6 L155.

P12 L274: “composition” to “ratio”

P13 L297–L301: For integrity, the authors should discuss total uncertainty, not only “precision.”

P13 L302–305: Appropriate reference should be mentioned. Otherwise, readers cannot understand whether the description is an established knowledge or the authors’ speculation. At least I do not find this discussion convincing with the current style.

P14 L320: As the authors explain at P14 L323, I agree that diffusion is plausibly the dominant process that caused the result with δ¹³C offset and larger variability. The authors might come up with adsorption or desorption, but I could not find any signal that support these processes occurring. This paragraph could be reformulated.

P14 Section 4.4: I think discussion of this section is too general and thereby reads exaggerated. For instance, isotope equilibrium with water vapor during storage largely matters in δ¹⁸O of CO₂ measurement, thereby it would go much less smoothly. The authors additionally explain prior treatment of vials, but storage over days, weeks and months are totally different, and I would not be optimistic as written currently. If the authors have a specific application plan deemed feasible only with the result presented in this study, they could focus on it in this section, but otherwise I think it is difficult to keep this section in a style convincing to readers.

P15 Conclusion: The conclusion (as well as abstract) section should be reformulated after all issues are addressed.

P15 L360: I do not think that the exact methods in this study can help reduce cost of long-term monitoring network, because they analyze δ¹³C of CO₂ using sub aliquot of original flask samples (use of extra vials would increase cost).

P15 L362: A week is too short for international collaborations. Extra efforts to cool samples would also complicate logistics. The following sentence also reads like overstatement.

Sauze et al. (2025) developed an analytical workflow to measure the carbon isotopic composition of CO₂ in small atmospheric air samples (1 mL). Their work demonstrates that all the successive steps described in the paper are necessary to achieve a good precision (0.1‰). The ability to measure isotopic composition in such small quantities opens new perspectives in ecological and paleoenvironmental research. For this reason, the study fits well within the scope of the journal.

I find this study interesting, particularly in the way it tests different parameters. However, as already noted by the other two reviewers, several points need to be addressed before publication. I also believe that the authors should explain in more detail why achieving a precision of 0.1‰ is novel. This could be related to Berryman’s 2011 study, where a precision of 0.3‰ was reported.

Quality of the figures

Although this study focuses on optimizing precision in δ¹³C measurements, it is difficult—based solely on the figures—to assess the difference or similarity between the measured precision and the accepted values reported in the literature.

For example, in all figures the mean value is shown in red, but what are the corresponding precisions and accepted standard deviations? How far are the measured values from the reference ones?

Need for more quantitative information

Since this study is based on the analysis of very small amounts of CO₂, more quantitative detail and precision are needed. For example:

• L93: “A control set of vials (i.e., blanks) was analyzed and showed no detectable CO₂ signal.”
• L249: “These blank vials showed detectable CO₂ signals, confirming the integrity of the sealing protocol and the absence of background contamination during storage under ultra-low-temperature conditions.”

The authors mention possible contamination, but without quantifying the blanks and background, how can one assess the effect on the results? If the δ¹³C of atmospheric CO₂ is around –8‰, could this explain the higher δ¹³C values measured compared to the reference value of the working standard? A possible mixing process could be discussed.

Accuracy

Although this study focuses on improving δ¹³C precision, it does not address accuracy. There are often systematic offsets between the mean measured values and the reference values, but these are not discussed. For example, why does the single-septum test (l. 189, 3.3) show an offset from the reference value that is not observed in the other tests?

Even if measurements were performed using a continuous-flow setup, it would be useful to test different IRMS settings. What were the instrumental parameters and operating conditions used during the study? Providing this information could help other researchers reproduce the method.

Additional aspects that could be discussed include:

1. the impact of instrumental drift during analysis,
2. the memory effect, and
3. the background signal.

Data accuracy is strongly linked to background levels, with increasing background often leading to less accurate and less precise results. What are the quantitative values associated with these three parameters, and how were the data corrected for them?

Finally, the authors mention diffusion as a plausible cause of some effects. It would be useful to provide references or a brief explanation to support this hypothesis