This manuscript presents a comprehensive and detailed explanation of methods and apparatus that keenly solve some of the formidable challenges that have perplexed field scientists attempting to make and calibrate high quality measurements of water isotopes and water vapor concentration for some time.  For those who can successfully implement the technical designs and concepts presented here and apply it to their research, I am convinced it will represent a step change in their ability to make more useful and reliable measurements in a variety of environments.  It is clear that this represents a culmination and evolution of concerted efforts, both on the part of the authors, and many whom they cite, in a long term quest to address the challenge of stability in analytical systems of this type.   I whole heartedly recommend its acceptance for publication, with some minor edits, mostly technical clarity, and slight improvements to 2 of the figures.  If possible, a few photos would really help illuminate the not only the complexity of the system, but it’s likely compact nature and portability.

One of the more important innovations is the advanced PID control of the headspace pressure in the vials by metering between pressurized air and vacuum, leading to remarkable steady metrics in both isotopes and water vapor concentration.  An equally head-line worthy finding is the speculation on the effects that cavity temperature control have on the precision of the Picarro instrument. This may not be surprising to some but seeing it here along with all the other metrics is compelling.  Not only have the authors achieved in creating a remarkable instrument/inlet system, but they have tested the system thoroughly and produced metrics that quantify and demonstrate its stability in a convincing manner.

General comment

This paper presents a new water vapor generation module meant to improve the accuracy and precision of the analysis of water isotopes in the liquid or vapour form, which is a critical issue for the study of certain second-order parameters like d-excess or 17O-excess in atmospheric water vapour or ice cores. The presented calibration module is an efficient technical solution to solve a number of problems encountered by the authors and listed in the introduction, namely reduce the memory effect, increase the robustness and reliability for field calibration and adapt the system for multiple standard analysis while offering a large humidity range and a stability over several days. These technical issues are indeed encountered by many teams working on this specific subject and the first part of the articles gives a very detailed description to duplicate the proposed solution, including technical references.

In a second part, the performance of a Picarro analyser and the vapor generator are presented and discussed, using several tools such as the Allan deviation or the wavelet coherence analysis. By comparing two Picarros, or comparing the new vapor generator module and a commercial Picarro autosampler and vaporizer, the authors are able to determine whether the performance originate from the analyser or the vapor generator module in a very convincing way.

Scientific questions

I have a few questions and comments that might shed some additional light on your discussions:

• You demonstrated in this paper a reduction of the memory effect using the new vapor generator module compared to the Picarro vaporizer, especially visible on dD. Can you precise whether the residual memory effect is dominated by the Picarro response or the humidity generator?
• The wavelet coherence analysis gives a good indication on the correlation between the cavity temperature and the delta measurement. Did you plot the Allan deviation of some of the studied Picarro parameters (cavity temperature, pressure, etc) to check for the presence of the same bump? It could be interesting to compare it on the two Picarro analysers who do not show the same bump. Also, other parameters such as the cavity temperature or warm box PWM can be interesting to check.

Technical corrections

• The legend of fig. 7 should be removed or made smaller to avoid covering the Allan deviation curves
• Table 2 could be easier to read in a graphic way, by putting for example the humidity as x-axis and the other metrics as y-axis
• For an easier reading of fig. 8, I suggest a centering of the delta values around zero (by subtracting the mean value to the raw dataset) and share the same y axis for d17O (a and b), d18O (c and d), dD (e and f), dexcess (g and h) and D17O (i and j).
• Fig. 3: If possible, I would be interested in seeing the temporal signal of d18O and dD (below the H2O curve for example). Maybe with a rolling average the memory effect can be directly observed?
• Fig. 2: I would be interested in having a global 3D drawing of the water vapor generator to understand how the two pieces are connected. Otherwise, a photo of the module would be appreciated.

The manuscript "A versatile water vapor generation module for vapor isotope calibration and liquid isotope measurements" by HC Steen-Larsen and D Zannoni describes an excellent addition for those in the laser spectroscopy community interested in pushing the limits of their water isotope instrumentation. It clearly meets the scope of Atmospheric Measurement Techniques and addresses a frustration by many atmospheric water vapor researchers by providing a method for reliable and automatic calibration of their vapor measurements across a wide range of concentrations. The manuscript should be accepted for publication in AMT after attending to my comments and suggestions described below.

My main criticism is that the manuscript frequently reads like a rough draft that was hastily put together and submitted. To be clear, the content of the study is sound, the science is sound, all of the pieces are present, and it is encouraging that my only criticism is related to the conveyance of the material. Still, the authors need to read through it very carefully, ask a colleague to proofread it, or preferably, both. While the team at AMT will help provide a final polished version, the authors need to take it upon themselves to do much of the work. 

The measured vs expected results presented in Table 3 are clouded by a confusing description of how these data were calibrated. I would not be able to reproduce your calibration strategies based on lines 384 to 389. Are the mixtures calibrated to the pure versions of SW and WW or to SP and BER? While within error, the bias is interesting and points to a calibration issue. Do the authors have calibration weights for their balance used to create the mixtures? The authors provide no evidence to support the claims in the sentence spanning line 400.

What is the maximum consumption rate of liquid water from a 2 mL vial? Given the variety of concentrations one could choose from for a deployment, what volume of reservoir do the authors recommend, or perhaps more to the point of this paper, what is the maximum reservoir volume given the dry-air-pump enrichment noted on line 465.

Regarding the data presented in Figure 6 and the different patterns observed in dD compared with d18O, I wonder if the authors could speak to the relevance of memory. What order were the standards analyzed in and were the vapor concentrations always stepped through from high to low or low to high? Or did the authors rule out memory and were left with suggesting spectral fitting. If the pattern is spectral, I wonder if the 18O of laser 2 shows a different pattern compared with the 18O of laser 1. What about 17O? Does it show a unique pattern?

Thank you, authors, for this contribution. I am looking forward to employing these techniques and citing your study.

1) Elucidate that what are the advantages or improvements of the UAV-based EC system developed by authors over other existing UAV-based flux measurement systems. 

2) Show that what are the differences or improvements in the calculation method of wind or turbulent flux for the current UAV-based EC system compared with manned airborne EC systems. 

3) Measurement precision or reliability is an important metric for the successful application of the UAV EC methods, the current manuscript only gave the mathematical precision (or instrumental error) in measurement of wind and turbulent flux. I recommend that the authors could make a direct comparison between the measurement from UAV- and ground-based EC systems. 

The paper reports general improvements in the author's previous device. I found it challenging to understand the manuscript in detail simply because it is not written for broader readers. Too many self-citations (14/43=32%) imply a narrow target readership. It would be preferable to clarify and discuss the position of this study from a broad perspective, as related studies exist. For example, Graaf et al. (Isotope ratio infrared spectroscopy analysis of water samples without memory effects, Rapid Com. Mass, 2021) reported no memory effect for liquid measurement. Although these test results would be helpful for certain people, this manuscript should be thoroughly revised before acceptance.

(1) There are many typos. I cannot point out all. Following are the typos found only on page 1.

L17: show > shows

L22: Remove "as" at the end of the sentence.

L23: have > has

L23: document > documenting (or documented?)

L30: falls > fall

(2) An example of the limited readability only on Figure 1.

L.132: "...module is an improved and revised version of an original prototype in 2014…"

> Please briefly explain the previous system. A few sentences will be fine.

L. 134: "A four-ovens version was used in this study."

> What is a four-oven version? Why does Figure 1 illustrate only one oven?

In Fig.1, caption: "… in open-split mode…"

>Please explain the open-split mode.]

Fig. 1: There are no explanations about abbreviations and lines. Please define or explain "P1", "T1", "Tn", "Vacuum line", "red dotted line", "black dotted line", and "black solid line ".

Please add several photos of this system in supplementary information.

General Comments

The manuscript details a field-deployable custom vapor generation system that will facilitate field measurements of water isotopes; the analysis demonstrates the analytical advantages of the method relative to vaporization units that rely on discrete injection. The manuscript will benefit readers of AMT who utilize water isotope analysis systems in the field and laboratory. The authors claim that the custom unit will benefit field measurement campaigns because the vaporization unit is portable, easy to service, and can measure reference waters more quickly than commercially available units and with less isotope memory. The authors tested the system and show that it meets analytical precision targets. They also conduct a stability test of the vapor generation module by measuring the vapor stream with two Picarro instruments simultaneously to show that the noise from the vaporizer system is small relative the noise of the instruments. More information about operational protocols and maintenance of the vaporizer unit will help convince a reader of its field-worthiness. The apparent relationship between cavity temperature fluctuations and d18O on timescales of ~tens of minutes is an important finding of this analysis and should be highlighted; as the authors suggest, if the instrument temperature could be stabilized, this would improve precision for d18O, but this also has likely significant implications for the deuterium- and 17O- excess measurements as well. 

Reorganization and reprioritization of some discussion points would improve this article. As it is written, the abstract does not summarize or highlight the most important outcomes of the study. Instead of focusing the abstract on theoretical future applications (i.e. instead of saying what it “could in principle” do), please revise the abstract to document the novel advantages of the vapor generator and highlight the new results that are achieved by the tests described in the manuscript. The vaporizer module is of course important for its potential to field-calibrate, but it is also useful for the analysis of instrumental noise, examining isotope-humidity dependence, and other performance metrics that are shown in the paper and deserve emphasis. For example, the abstract could perhaps highlight the vaporizer’s ability to generate continuous water vapor for a wide range of humidity levels (which is an advantage for vapor measurements over CFA vaporizer configurations like Gkinis, Jones, or Davidge that target a single water vapor concentration) with the good stability at each level but increasing memory issues at low humidity. The comparisons with discrete injections (such as in Figure 4) are important documentation of the advantages of continuous over discrete vaporization, but do not uniquely reflect the contributions of this vaporization unit over other continuous vaporizers designed for CFA – the abstract (and manuscript) could better contextualize the new contributions made by the authors with this custom vaporizer through direct comparisons to other published calibration methods and continuous vaporizer units.

Finally, the authors aim to demonstrate that the custom vaporizer has sufficient signal stability to measure 17O-excess, but the authors need to be more cautious about the treatment of raw data to assess performance for 17O-excess. It is misleading to assign units of per meg to the raw data since the raw signal variability is not equivalent to the calibrated signal variability. This treatment potentially affects data shown in figures 3, 5 and certainly in figures 8(i,j) and S2. Previous work has established accuracy (not precision) as the dominant error in CRDS 17O-excess data, so while the precision seems great for these data, more work is needed to show that system operating conditions like automated, variable flow rates or secondary air dilution do not create calibration bias for 17O-excess. Though small, the systematic offset in the calibrated 17O-excess data further suggest that there likely is a bias in the calibration, which should either be examined further to better characterize the limitations of this method or should be qualified appropriately in the text when claims are made about the quality of 17O-excess data.

Specific Comments

Some sections of the text need to be revised for clarity and completeness. In some sections, the claims that are made are not directly supported by the evidence provided in the tables/figures so it can be difficult to evaluate some statements. The whole manuscript would benefit from a careful reread and review by the authors. 

The abstract claims that the vapor stream is constant for 90+ hours and that it therefore could be used in the field for more than three months, but it is unclear what “constant” means in this context, especially since the water vapor data that is shown in figure 3 exhibits some notable variations. It is also unclear (in the abstract) how operating for 90 hours in the lab translates to three months in the field – this is explained at the very end of the manuscript (in that it will theoretically be measuring standards for 1h/day) but is confusing in the abstract since it is not explained. Further, the 90hr to 3 month relationship is speculative at best, since it has not been demonstrated that the unit will, for example, not clog with precipitates or encounter other operational setbacks over the three month window. More information about typical or intended operations of this system will help the reader better understand its advantages and limitations.

There is a lot of confusing or vague language throughout the document and also many acronyms that have not been defined – please try to define acronyms before using them in the text or figure captions and make sure details of system components are explained the first time they are mentioned.

Ln 53-55 suggests that excess values require a “relatively high output of individual number of samples measured” – can you clarify what you mean here?

Ln 72-75: define “low uncertainty” and “large quantities”.

Ln 91: Davidge et al 2022 utilizes a unique vapor generation system so “this system” should instead say something like “a similar system”.

ln 151: instead of noting the differences between this vaporizer and an earlier version of this vaporizer, this might be a good place to describe the differences and advantages of this system over other types of vaporization systems (e.g. it adopts the multi-channel selector valve of Jones et al. 2017, similar continuous vaporization setup to Gkinis but with vacuum pump, PID control for humidity, additional mixing tee, etc.).

Many of the details noted on page 6 would be better left to the later sections of the paper since a general reader might not understand the specifics about salt deposits, number of ovens, valve circuitry, etc. at this point in the paper.

Ln 190: if the pressure regulator resolution is 0.01psi, why is the regulated range of pressures that is listed so large (0.5-3.5psi)? How is this system typically operated for each humidity level, and how much does the pressure fluctuate for each humidity level during a typical measurement? More operational details are needed to help a reader duplicate this work.

All tables and figures should have sufficiently detailed captions so that the reader can easily understand what data is shown – please reread all figure and table captions and try to add more information.

Ln ~265: perhaps it would be advantageous to highlight the regions of the plot that the author is describing in the text when talking about the slope of the allan deviation with time.

Ln 271-272: figure 3 is important both because it documents the ability of this vaporizer unit to generate consistent isotope data, but also because the authors have highlighted the difference in the allan variance analysis when truncating the data to account for memory effects within the system. The vaporizer unit exhibits excellent performance and this should be highlighted, but it is incorrect to say that the performance is better than that of other systems, since other systems show similar precision at these averaging times (e.g. Gkinis et al. 2010 for dD and d18O, Steig et al. 2021, or Davidge et al. 2022 for 17O-excess). It is therefore also incorrect to suggest that no previous work has  managed system memory effectively.

The authors might consider showing the data from Table 2 as a figure, or perhaps showing the full sequence of measurements made over time to help the reader understand the tests that were conducted.

The 12500 ppmv threshold finding is very interesting, especially since the allan variance test shows the worst performance in this region. Have you tested whether this is a physical effect of the capillary diameter or some other design choice? Were all data generated for this study using the 127um capillary? 

Ln 397: this replication looks promising, but more information about the calibration is needed here. This section should also include that this performance is comparable to other continuous vaporizers that have been developed for 17O-excess measurements (i.e. Steig 2021, Davidge 2022).

Ln 401: The offset is likely due to calibration, so additional details about how these data were calibrated would be useful. It seems unlikely that the age of the reference water has modified the 17Oexcess values if they have been in sealed containers and cold storage, but it is possible that the systematic bias is due to an error in the calibration standard assignments or that it is generated by the vaporizer unit itself, which must be carefully examined. Which standards were used for calibration? How frequently were they measured? How stable were the raw values in d18O/d17O between reference water measurements? Without this information it is impossible to speculate what might be the cause of these offsets.

Throughout the paper and abstract there are speculations about what could be done in principle, but it is important to properly document what can be done in practice, especially since these systems do require maintenance and cannot run indefinitely or measure standards frequently enough for perfect calibrations. How often is it necessary to clean the capillary? What maintenance was required for this system during the study period and with what frequency? How can one best operate a system like this within those maintenance limitations to maximize the quantity and the quality of the data? Details about operational controls, conditions, and maintenance would help the reader better understand the performance of this vaporizer system.

Ln 465 should also acknowledge the increase in the delta values.

Ln 469 seems like a major limitation of this method – perhaps the authors should repeat this test with larger vials to eliminate this enrichment issue.

Ln 480: please define “relatively clean”

Ln 483: is 1h per day sufficient for the calibration of all water isotopes? Why has this duration been chosen?

The discussion in ln ~510 and data shown in Figure SM2 suggest that the performance of this vaporizer is changing over these 48h of analysis due to the automation of air/water ratios and necessary reduction in flow rate to accommodate the formation of salts in the capillary. Have you tested this system with milli-Q or other treated water to minimize this effect, and have you seen any improvement in this performance? A change of ~3 per mil dD over this short timescale should probably be investigated further. The way this is accounted for in CFA systems is by keeping the liquid/air injection rates as constant as possible, because otherwise it is impossible to know what the effect of the memory is at any point during the analysis when the flow rates are changing during analysis. What range of flow rates does this study utilize and can you attribute any of the changes in system performance to these variables?

Ln 521: define “relatively high measurement uncertainty” and other vague quantities throughout the paper.

Ln 523: change “error, which is” to “error that is” for clarity/correctness.

Ln 524: without additional analysis of calibrated data it is a stretch to say that the module is “optimal for 17O-excess” but it is certainly promising to see such nice signal stability in the 17O-excess record. Previous work has established calibration as the major limitation on laser spectroscopy measurements for 17O-excess, so without additional analysis of calibrated data it is hard to accept these claims.

Ln 525: similarly, while the unit can operate over long periods, unless it is possible to measure sufficient durations of the calibration standards for 17O-excess in between vapor samples, the resulting data could have large errors – this issue is discussed in both Steig et al. 2021 and Davidge et al. 2022. How long can the vaporizer operate before the capillary clogs or the flow rates change? This data would be important for understanding limitations around 17O-excess calibration.

Though I look forward to following your updates in the future, Section 4.5 is not analysis of the data that is presented in this paper and could be better suited for a proposal. 

Table and Figure Comments

Figure 1 – this will be difficult for many readers to follow. I suggest additional labels and defining acronyms and process control symbols.

Figure 2 – recommend additional labels to help the reader understand this figure

Figure 3 – clear figure. Can you comment on the variability in the water vapor concentration? Especially since earlier studies have linked water vaporization inconsistencies to isotope fractionation it seems important to better characterize these vapor fluctuations.

Figure 4 – great visualization of this relationship that shows a major advantage of continuous injection modes for laser spectroscopy. The legend is perhaps a little confusing because it is unclear what the legend means without also reading the caption.

Figure 5 – Why not show all water isotopes here? Also please consider revising the label on the y axis if the data used for this analysis has not been calibrated.

Figure 6 – great figure with important implications for low-humidity measurements. Maybe instead of defining d18O_diff in the caption you could just label it d18O_3500ppmv – d18O or similar?

Figure 7 – if possible, moving the legend away from the data would make it easier to read panel A. This is a very compelling and disturbing result!

Figure 8 – the authors should either calibrate the 17O-excess data or find another way to describe the spread of the raw spectroscopy response – showing values of 300 per meg is very misleading! Large variability in the 17O-excess raw data suggests that the errors in d17O and d18O are not perfectly correlated, which is likely to cause accuracy issues in the calibrated 17O-excess values. Because the data is not calibrated it is difficult to evaluate this data – please provide a record of calibrated 17O-excess over time in the revision of this manuscript.

Figure 9 – this figure would be more useful if calibrated 17O-excess values were shown.

Table 1 – define BER and SP. What is the uncertainty of the values of the excess measurements and the SP measurements? Where were the data measured? Please specify whether these values are measured relative the VSMOW-SLAP scale. It would be helpful to combine Tables 1 and SM1 and show all waters here since the reference waters from the supplement are referred to in the text.

Table 2 – as noted above, a plot of this data could be helpful for the reader to understand the different tests that were conducted. Is the standard deviation calculated for the raw data over the full duration of each test?

Table 3 – please include details about how these measurements were calibrated.

Table S2 – Please check for rounding error in the 17O-excess reference water assignments; from the isotope values and mixing ratios given I calculate 20, 11, and 17 per meg (not 21, 12, and 18), though this does not substantially change the result or interpretation.

Figure S1 -  the memory effect for dD appears to be different during these two analysis windows; it could be interesting to examine the operating conditions during these runs and consider whether flow rates or other changed conditions could cause this difference.