This work by El Kattar et al. describes a portable NIR LHR and investigates the expected performance of such kind of spectrometer by performing an information content analysis. LHRs offer the prospect of replacing or at least complementing traditional FTIR approaches, so this is a very relevant study, which fits well into the scope of AMT. However, the current manuscript requires substantial revision before final publication can be considered.

Comments:

Section 3.1, line 90: I can hardly believe a Gaussian line shape is assumed for modelling the absorption lines, I guess you refer to the Voigt profile?

Section 3.1, line 102: please rephrase: “ …. while TCCON a-priori information is used for CO2 and H2O atmospheric profiles.”

Figure 2: The calculated H2O signatures seem undetectable in the measured spectrum - is the slant column used for the calculation not matched properly? In the figure description, please specify SZA of the observation and total integration time of the measurement shown.

Section 3.2: in the opening section of this treatment, information needs to be given concerning which quantities are fitted in the retrieval (please add a table listing all components of the state vector): I guess in addition to the gas mixing ratios of CO2 and H2O, spectral shift (or scale) is fitted? Further fit variables are needed for describing the continuum background level. Is the solar spectral abscissa scale fitted (usually required in spectrally high-res measurements to compensate for residual LOS errors)?

Section 3.2, line 126: “the ith measurement” -> “the ith spectral channel in the measured spectrum”

Section 3.3: unfortunately, the construction of the a-priori (see table 1) is so oversimplified that I doubt any useful conclusions can be drawn from the current information content analysis. If one wants to compare the expected performance of the presented LHR with existing FTIR setups in a sensible manner, the a-priori covariance matrix needs to be far more realistic. Moreover, the information content analysis needs to discuss explicitly the expected errors on column-averaged abundances, which are the products of current networks. Note that the current networks do not provide gas columns, but XGAS values, which are constructed with the help of co-observed O2 columns. Your claim “The LHR exhibits unique advantages … in retrieving gas columns with better vertical discretization [should be: resolution]. It is therefore a promising alternative instrument for local scale measurements or for satellite validation”. This might be correct, but needs to be supported by the results of information content analysis. In the application context you refer to in the manuscript (especially satellite validation), high vertical resolution is mainly useful via improving the reconstruction of XGAS amounts over current techniques. (It needs to be kept in mind that the satellite also measures XGAS.)

For achieving a meaningful comparison with current state-of-the-art, I would suggest to proceed in the following manner:

(1) construct sensible S_a matrices for CO2, H2O, and T. Note that the relevant variability here to be reported in S_a is the variability between the actual profile and the TCCON a-priori. This S_a matrix for CO2 can be constructed from aircore launches (the French community is quite active with this technique, so a sufficient amount of data should be available for constructing an S_a matrix). The equivalent matrices for H2O and T can be derived from meteorological soundings, ideally launches which were not used in the model assimilation underlying TCCON a-prioris (perhaps H2O and T are by-products of aircore launches anyway?). When constructing the S_a matrices, it is crucially important to maintain the covariances, which inform about characteristic lengths of variability along the vertical. Only by maintaining the diagonal elements a meaningful S_a matrix is constructed.

(2)  For the performance comparison with TCCON and COCCON for the XGAS values, the propagation of T errors into a O2 retrieval from the 1.26 um band needs to be included. This will alter (expectedly improve) the uncertainty budget for the target quantity XCO2, as this is calculated using the ratio of CO2 and O2 columns. This error compensation is lacking in the LHR approach. Moreover, note that SZA errors cancel out in this rationing approach, so in the discussion of model errors, the resulting error contribution for the LHR needs to be estimated (from the assumed SZA errors).

(3) A further important model parameter is the ground pressure. Ideally, it should be included in the error analysis, as the sensitivity of the LHR very likely differs from that of TCCON and COCCON due to the high spectral resolution and due to the fact, that there is no rationing over the O2 column. But if you clearly state in the text that you assume the availability of an ideal sensor, one might skip this item.

Using (1) and (2) and your error propagation equations, you can realistically establish the desired performance comparison between the LHR and current techniques. I would expect that the LHR is superior wrt the smoothing error, while the current networks might be more robust wrt the impact of model parameter errors (T and SZA). With respect to the smoothing error, it might be interesting for TCCON + COCCON to work out the smoothing error both for the operational setup (scaling retrieval) and a possible future data processing which performs a profile retrieval fit of CO2. The latter result would specifically reveal the improvement introduced by the high spectral resolution achieved by the LHR. If, however, you feel this is beyond the scope of your work, restrict the investigation to the operational setup.  

In table 3, please add integration times, otherwise the comparison of SNR figures is not meaningful.

In table 4, the column errors seem unrealistically large to me for all instruments. It therefore would be good to split the reported error into different contributions (spectral noise, model parameters, smoothing). This would make transparent which error source drives the total budget and would allow to explicitly verify at least the calculated noise error contribution, as this can be easily deduced from data retrieved from actual measurements.

The treatment provided in section 4 needs to be refined. The authors claim that from this investigation, the preferred CO2 channels to be measured can be deduced, saving observation and data analysis time. This is in principle correct, but we need to realize that the presented LHR is operated as a ground-based solar absorption spectrometer. In this configuration, a model for the continuum background level needs to be included in the fit (because a solar reference measurement outside of the atmosphere is not doable). This in turn requires a sufficient number of background channels (spectral positions largely free of absorption) to be included both in the measurement and in the fit. The analysis, however, suggests using only channels with strong CO2 absorption, which seems unrealistic.