Please find my review attached.

The manuscript of Casado et al. addresses the urging and relevant topic of high-precision isotope ratio measurements of atmospheric water vapour with special target on extremely low amount fractions (~1 ppm). The authors present the development of a highly complex spectroscopic setup based on a high-finesse cavity ring down (CRDS) technique in combination with frequency stabilized laser source using the optical feedback from a V-shaped cavity. This approach allows for outstanding precision and long-term stability, compared to the performance of commercial instruments, in particular to Picarro L2140i. Thus, the proposed analytical tool can be highly relevant e.g. for the polar atmospheric research.

Despite the very impressive instrumental developments, the manuscript has a few caveats in presenting this otherwise very promising topic (see general comments for details). The authors should consider consolidating the wording, language, and the structure of their manuscript. A more precise formulation of the thoughts/facts would substantially improve the readability and the scientific value of the paper. Therefore, I can recommend this manuscript for publication after some clarifications and corrections are made.

General comments

The measurement technique is heavily advertised as a new spectroscopic instrument. A total of 17 instances for "new instrument/generation spectrometers" can be found in the manuscript, which is clearly an exaggeration considering the fact that all of the key elements were already published by the authors almost a decade ago, i.e. in 2014/15. These include the OFFS-CRDS, the VCOF, the MZM, and even the Lamb dip approach for a highly accurate reference frequency measurements. As such, the "new" attribute should be avoided and used only in the context of the molecular species, i.e. H₂O, measurement. Therefore, I suggest to use an abbreviated name for the spectroscopic setup, e.g. VCOF-CRDS and use this throughout the manuscript.

The key aspects relevant to the reader should be placed more prominently, e.g. in the abstract or conclusion:  1) the current instrument cannot be deployed in the field, 2) the lowest measured H₂O amount fractions are 25 ppm, while the performance at 1 ppm is merely an extrapolation, and 3) the recorded absorption spectrum is limited to 7 data-points only, which requires tedious frequency drift correction approaches.

Especially, the validity of the extrapolation is questionable due to the spectral interferences from the CH₄ absorptions, the most prominent being at 7199.9547 cm^-1 (S = 3.715E-24) and at 7200.0287 cm^-1 (S = 1.742E-23), the former overlapping with H₂¹⁸O absorption line, while the latter biasing the baseline at the 3^rd data-point according to the Fig.2a. Without the discussion of these aspects and their impact on the atmospheric H₂O isotope measurements, the statements regarding the precision and accuracy at low amount fractions of H₂O remain highly speculative.

There is a great deal presented related to drifts and their suppression. The authors successfully demonstrate the benefit of the Lamb dip method for frequency self-referencing. While the comparison with the standard optical frequency comb is convincing, the authors should describe the details of the measurements: i.e. that the instrument is operated in slow-pace measurement mode specifying the resolution and scanning time, quantifying the impact on the stability of the optical cavity when switching the gas pressure from 35 to 0.1 mbar, and the overall duty-cycle of the measurement. Also the sudden transition shown on Fig.3 around 90 h requires some discussion to better understand the reason and its effect on the retrieved precision. Furthermore, it would be interesting to see the effectiveness of this approach in case of larger temperature variations, mimicking real field conditions.

Many statements about limitations or issues are generalized; however, these are mainly specific to CRDS, or more strictly speaking to Picarro instruments. Benchmarking of the setup to a L2130i/ L2140i is fine, but then the discussion should be confined accordingly.

Specific comments

The title needs some revisions. The "low humidity water vapour" is not appropriate. I suggest the "low amount fraction" instead. In general, the humidity should be replaced by amount fraction or mixing ratio throughout the paper, especially when is expressed in ppm. The term "new infrared instrument" is very general (see general comments) and thus, I recommend to be more specific here, e.g. "frequency stabilized CRDS". In my opinion, the "reliable" is somewhat far-fetched and misleading, mainly because the low (< 20 ppm) H₂O amount fractions were not measured directly (and extrapolation is not straightforward) and there was no demonstration of any calibration with V-SMOW standard materials or reference gas mixtures with different isotopic composition. The authors should either include such data or refrain from creating false expectations.

In the abstract, the key findings should be mentioned: precision in d-values, integration time, H₂O amount fraction, frequency stability, laser emission frequency, optical path length, SNR value, drift value, etc. The last sentence about the new constrains on the fitting technique requires more details in the manuscript and a clear statement in the abstract about the key factor that makes the biases less disturbing.

The statement on Pg2, L45: "Applications of infrared spectroscopy techniques to trace detection and isotopic monitoring is dominated by two techniques: OF-CEAS and CRDS" is simply overlooking the scientific fact that the most accurate and field-deployed measurements are performed with the classical direct absorption spectroscopy (e.g. Aerodyne TILDAS and similar techniques) extending the capabilities even toward clumped isotopes.  

The authors should consider adding more details regarding their setup: the supplier of the DFB laser, optical power, operating parameters (current, voltage, and temperature), the optical power reaching the detector, detector type and vendor, the supplier of the CRD mirrors and the coating, piezo actuator, etc.

Please motivate the selection of the CRD cavity length: why 48 cm instead of the more optimal 23 cm?

Without a detailed comparison between slow- and high-pace mode, it is rather difficult to understand why the high-pace mode is favoured in this work. Does the water vapour isotopic composition change faster than the time required for a full spectral scan and if even so, would this high temporal resolution be useful in any context? Contrary, there are many beneficial aspects for using the slow full spectrum mode, such as higher spectral point density, less prone to frequency drifts, accurate and robust fitting, etc. I do not really understand what is meant by changes in the gas composition during the slow scan and the additional noise with their auto-correlated features. These statements require more clarification about the underlying effect and processes.

Fig.2 caption: by adding experimental parameters such as gas pressure, H2O amount fraction and optical path length would help the reader to better interpret the illustrated spectra.

As the speed-dependent Nelkin-Ghatak profile is a function of seven parameters, it is not clear how this profile can be applied to a spectrum containing 7 data-points. Furthermore, it is not mentioned how the non-Voigt parameters were determined for the selected transitions.

It is shown that at low gas pressure the Lamb dip can be generated. While this is a very elegant and robust approach to determine absolute frequency, it leads to another question that was not mentioned in the paper, namely, the saturation effect also at higher gas pressure. It is well known that the field energy build-up in the cavity can be sufficiently high to induce non-linear effects and saturate the absorbing gas. How does this affect the accuracy of the isotope ratio measurements?

Fig.4 shows the Allan-deviation (ADEV) plot over 2 days. This extended period is somewhat obsolete considering the optimal stability range of the spectrometer of 150 s, and also the fact that the authors do not disclose their gap-filling method to estimate the ADEV beyond the 10 h limit. On the other hand, the observed behaviour can fully be dominated by the water vapour generator. Thus, it remains an open question, why the authors did not use e.g. a pressurized air cylinder for the low H₂O amount fraction measurements.

The term d-excess needs to be defined and it would be beneficial to also include the δD data. Eventually, move Appendix B into the main text by replacing Fig.4 by Fig.B1.

Pg9, L201: The biases mentioned here are in the context of using the Picarro instruments and therefore generalization to any infrared spectrometer should be avoided.

More information about the new fit parameters are required. How are they obtained, what exactly is optimized to reduce the biases and how much improvement is obtained compared to the case with improved frequency stabilization scheme alone?

I recommend a revision of the Discussion section. Instead of recalling the limitations described in earlier publications (this can easily be moved to the introduction), the authors should keep the focus on their own results and present a concise evaluation of the measurements, perhaps by answering all the above questions and addressing the raised issues.   

For sake of scientific rigour, Fig.8 should be slightly modified: the grey bar corresponding to the OFFS-CRDS performance should be drawn to the 20 ppm level, while the extrapolation toward 1 ppm should be indicated by a lighter gray or a dashed contour.

The main factors hindering the instrument to be deployed in the field are not fully clear. In particular, the issue caused by the thick ice layer is hard to understand. Furthermore, there is no attempt by the authors to discuss the potential improvements or alternative solutions to overcome this serious limitation. What are the constrains that can eventually be removed by further engineering and is there any fundamental issue that is difficult to address? As a side note, a similar technology (see e.g. the ProCeas from AP2E) has been proven to be market ready.  

Without addressing these aspects, the final statement in the Conclusion remains highly elusive. Nevertheless, I'm firmly convinced that the scientific value of the manuscript is appropriate and with an adequate revision the objections can be completely overcome.

Technical corrections

Abstract: Pg.1,L12: add "water" before vapour

Abstract: Pg.1,L16: replace "suffers" with "exhibits" or "demonstrates"

Pg.2, L56: define VCOF-CRDS

Pg.4, L75: replace "Distributed FeedBack diode (DFB)"  with "distributed feedback (DFB) diode laser"

Pg.4, L91: check grammar "cavity can be ever so slightly adjusted"

Pg.4, L93: change "Bronkhorst pressure and flow controllers" to "pressure and flow controllers (Type/Model, Bronkhorst)"

Pg.4, L94: use SI units, i.e. Pa instead of mbar. Check all instances.

Pg.4, L96: check grammar: "The instrument is set so the laser source produces …"

Pg.4, L100-104: consolidate the phrase

Pg.7, L148-150: revise the phrase, especially the "the accuracy of the instrument of humidity to isotope relationship" is difficult to interpret.

Pg.7, L158: check wording "water level stable levels lasting"

Pg.7, L166: change "a normal law" to "white-noise"

Pg.8, L190: replace "no" with "not"

Pg.9, L210: add space between value and unit. Check for all other instances.

Pg.10, Fig.6 caption: check "local tap distilled tap water"

Pg.13, L268: what is "fractionation coefficient"?

Pg.14, L301: replace "phrase" with "phase"

Pg.15, L310: rephrase "outside of the confine of a fully equipped spectroscopy lab", e.g. outside of temperature controlled laboratory environment.

This manuscript reports on the development and characterization of a laser-based spectrometer for measurements of isotopic composition of water at low humidity levels, relevant for future field measurements in central Antarctica. This work fits very well within the scope of the AMT journal. The cavity ringdown spectrometer is based on a DFB laser whose frequency is locked by optical feedback to a temperature-stabilized reference cavity. The long-term frequency stability of the laser is ensured by monitoring a Lamb dip signal in a selected transition at regular intervals. The authors test and discuss the performance of the spectrometer for measurements of δ¹⁸O and d-excess at various levels of humidity between 20-1500 ppm.

I have several major comments on this manuscript that I believe the authors should address before the paper is published.

The paper claims that the spectrometer is able to measure isotopic composition down to (or even below) 1 ppmv water mixing ratios, conditions found in central Antarctica. I see two problems with this statement. First of all, the lowest water mixing ratio used in this work is 25 ppmv, and the result at 1 ppmv is an extrapolation. What is the rationale for extrapolating to 1 ppmv? One could extrapolate even further and discuss the expected performance at around 0.1 ppmv. Second, the measurements are performed at room temperature in the laboratory, and the discussion about how the instrument will be prepared for operation at -80 C is missing. On the contrary, it is stated that the instrument in its current form is not ready for field deployment, e.g., because of the fragile design of the reference cavity (line 293). Will it be possible to construct a stable reference cavity, which is the heart of the system, that will operate at -80 C? In fact, the temperature stabilization of the presented spectrometer is not discussed in the manuscript. The authors should add this information, and comment on future developments that will enable field deployment.

I suggest the authors modify the phrasing at a couple of places to better reflect the focus of this work. For example, in the Abstract the second sentence states: ’During polar winter, the temperatures are so low that current commercial techniques are not able to measure the vapour isotopic composition with enough precision’ which puts focus on measurements at low temperatures, which are not shown in this work. I suggest focusing on low humidity levels, and stating clearly that the instrument was tested down to 25 ppmv. Similarly, in the Discussion, it is stated that ‘the new instrument can circumvent all the hurdles that limit the monitoring of water vapour isotopic composition monitoring in the coldest conditions, such as found in Antarctica in winter’ (line 288), and that ‘The performances of the instrument in the field should be roughly the same, provided that the temperature stabilisation of the instrument is as good as in the AC rooms of the lab (line 296)’, which are far-fetched claims, since it is not even discussed how the temperature was stabilized in the laboratory.

The authors state that one of the keys to the good performance of the instrument are ‘new constraints on the fitting technique’ (line 17) and ‘the new fit parameters’ (line 207), but the details of the fitting technique are missing. It is also not clear what the authors are comparing to when saying ‘new’. On line 117, the authors state that the absorption profile is fitted using a speed-dependent Nelkin-Ghatak profile, but it is not clear how and why this is done, since only one point is recorded per absorption line. The fits and residuals are not shown, it is not stated what the fitting parameters are, and if any parameters are fixed – what are they fixed to? Where are the line intensities, frequencies and broadening parameters taken from?

The authors refer to the 2 x 10^-12 cm^-1 noise level in the residuals, but they do not show these residuals. Does this number refer to the noise in the fit to the 7 points shown in Fig. 2a? On line 281 it is stated that the performance of the previous systems was limited by drifting fringes. How were the fringes avoided in this system? Is there no baseline in these measurements? The authors should show and describe the fits, including the residuals.

Related to the fitting process, please explain how the measurement points for the high pace mode were selected. The laser frequency is tuned in steps of cavity FSR, how is it ensured that some points coincide with the line centers? Is this a lucky coincidence, or was the FSR chosen carefully? Why would a cavity length of 23 cm be better than 48 cm (line 263)? That would make the FSR larger, which would make hitting specific frequencies harder. What is the ‘parking method’ mentioned in the figure caption on line 113?

Minor comments

• Line 53: ‘we present a new generation of infrared spectrometers’ – I suggest toning this down to ‘we present a new spectrometer’.
• Line 100: ‘with performances of 10^-12 cm^-1’ – ‘performances’ should be ‘sensitivity’, and the time over which this sensitivity is obtained should be stated.
• I suggest unifying the name of the technique, it is called OFFS-CRDS or VCOF-CRDS at different places.
• Line 105: ‘spectral resolution is of only multiple of the FSR’ – this is sample point spacing, not resolution.
• Lind 135: I suggest removing references in the caption of figure 3, because they suggest that the data are taken from these papers, while they are not.
• Line 164: ‘at the resolution of the instrument (3 Hz)’ – It is not clear where this resolution comes from. Is this time or frequency resolution? I guess it is the latter, because 0.7 s does not correspond to 3 Hz. Please clarify.
• Line 160: State how was the Allan deviation modified.
• Line 189: ‘roughly 10 times larger’ – larger than what? Further on – I do not understand where it is shown in Fig. 4c what happens after 800 s (‘dropping down to … after 800 s’).
• How was the isotope-humidity response measured? What does anomaly mean here? Is it the same as in Fig. 2b?
• Line 224: ‘The main source of uncertainty of the measurement comes from the drift of the laser source’ – Is it frequency or intensity drift?
• The language needs to be carefully revised at many places, for example:
  • Line 16: the instrument exhibits (not suffers)
  • Line 47 and 106: the interest of – the advantage?
  • Line 96: stable light – what is stable about the light?
  • Line 101: monitor water vapour isotopic (what?)
  • Lines 148-150 are hard to follow
  • Line 190: while we were not able to …. generated as described
  • Line 267-269 are incomprehensible
  • Line 300 is incomprehensible