This paper describes a new portable infrared spectrometer that has is able to provide spectral radiance observations from 100 to 1000 cm-1.  This spectrometer (FIRMOS) is the successor of the REFIR-PAD system, and is serving as a technology demonstrator for the future FORUM instrument.  It was deployed to a dry, high-altitude site at Mount Zugspitze in Germany, where there are other instruments which were used to help evaluate this instrument.

In general, the paper was well written, especially the first half of the paper that discuss the technical details of the FIRMOS instrument.  The latter part of the paper compares temperature and humidity retrievals from the FIRMOS against radiosondes, Raman lidar, E-AERI, and ERA5 reanalysis.  Due to the relatively few observations that were available, this latter section is somewhat weak.  However, there are always data limitations, so this isn’t necessarily crippling to this paper.

This paper would greatly benefit from having a comparison with both the E-AERI in radiance space.  Due to the differences in the spectral resolutions, I would recommend the authors use the “double difference technique” outlined in Tobin et al. JGR. 2006.   As the two instruments are essentially collocated (although vertically offset by 4 m), the spectral differences between 405 and 600 cm-1 should be within the instrument noise (if both systems are well calibrated). 

Line 253: did you assume any cross-level covariance in your a-priori?  Were there any cross-correlations between temperature and humidity?  There should certainly be cross-level correlations in temperature due to the atmospheric lapse rate, and a long analysis of radiosonde data from the region (or ERA5 data) should indicate if there should be other correlations in the a-priori.  If you assume the a-priori is a diagonal matrix, that will essentially increase the information content (DFS) of the retrievals.

Line 260: It is important to note that the gradient in a cloud-free measurement is zero only because it is so dry at the Zugspitze location.  If you were in a tropical location, there would be a negative slope.  This needs to be stated.

Line 279: In the selection of the 625 cases, did the Raman lidar (or the backscatter lidar, which was briefly mentioned later in the paper) confirm that these were cloud-free?

Line 288: it was not clear if the uncertainty used in the retrievals was the NESR or the sum of the NESR and the CalErr.  Please clarify this in the text.  If the latter, then the chi-squared term being less than 1 could be due to the very conservative estimate of the thermistor error in the blackbodies (stated on line 230).

Line 293: the mean residual also will contain any bias error in the forward model (not only instrument calibration error).

For the two comparisons in Fig 15: it would be nice to include the integrated water vapor (IWV) amount for the two cases.  Also, for line 309, the authors suggest that the DOF depends on the surface water vapor content, but it is really dependent on IWV?  Turner and Löhnert (JAMC, 2014) showed that the DOF in the water vapor retrieval using AERI observations in the 538-588 cm-1 region depends on IWV.

Figure 20: please replot using a skew-T approach, so that differences of a few degrees can be more easily identified and quantified. 

Fig 23 and resulting analysis: this is pretty unsatisfying.  I realize the purpose is to show that the FIRMOS is capturing the evolution of the event well, but the very coarse resolution of the ERA5 data in a mountainous region is totally inadequate to the task.  I highly recommend that the comparison be made against higher-resolution NWP output, such as the (order)2-km resolution ICON data from the DWD.  And that the figure include a subpanel showing the bias and RMS difference between the NWP model and the FIRMOS.

Question: Why did the authors not perform KLIMA retrievals using the E-AERI spectra, and then compare the retrievals from the E-AERI with the FIRMOS?  This seems like it would be a relatively simple comparison and include a lot more data (e.g., there seems to be hundreds of points in Fig 22), and open up an interesting discussion because their spectral differences between the two instruments.

This paper describes the FIRMOS instrument and presents data obtained during a measurement campaign at Zugspitze in 2018/2019. Details on the instrument design are given, and the level 1 and level 2 processing is described. The spectra are characterised in terms of NESR and calibration error, and vertical profiles of temperature and water vapour are presented along with their error bars and averaging kernels. Finally, the retrieved data is compared to co-located radiosonde, lidar and FTS measurements, and small time series are compared to ERA5 data.

The paper is well and clearly written and it fits the scope of AMT. I have, however, several comments, which should be considered in order to improve the manuscript.

Section 2.1.2 Radiometric Calibration Unit

This section is not clear to me. I think I can follow the description of the blackbody setup. However, a sketch showing the positions of the 4 sensors and indicating which sensor is T1, T2, and T3, respectively, would be helpful.

I would also be grateful for a few words on how the temperature stabilisation is realised. How are the BBs heated/cooled? How is the thermal homogeneity assured?

The accuracy of 30 mK given in Table 3 is not traceable for me. You state that you use a high-accuracy (30 mK) PT100 sensor as T1, but you do not give an accuracy for the temperature reading by the FIRMOS controller. The accuracy of the Lakeshore temperature monitor is given as 0.6%, which would correspond to 1.8 K at 300 K, which is not very likely and would make this monitor useless to correct for a possible offset of 200 mK in the FIRMOS controller. Please give more details concerning the accuracy of the different sensors and their readout electronics, also for T2 and T3.

I am also missing a comment on the emissivity of the blackbodies. How is a deviation from 1 handled?

line 146f:

"the temperature of the PT100 sensor was recorded after the switching of the stabilisation."

What do you mean with "switching of the stabilisation"? What is switched?

line 147f:

"The difference between the PT100 reading and the

stabilisation temperature of each BB is reported in Figure 3 (a)-(d)."

No, it's not. Figs. 3(a) and 3(c) show the absolute temperature over time. Please provide a difference plot, either with respect to T1 or with respect to the stabilisation temperature (as suggested in the text). Please choose the ordinate such that the thermal inhomogeneity becomes more visible. From the plot it seems that the thermal gradient in the hot BB is rather 1 K than the 0.3 K given in Table 3, but it is hard to see from this kind of figure. Temperature variations (over time) in Figs. 3(b) and 3(d) are only given for T1 and not for each sensor as stated in the figure caption. 

line 140 and caption of Fig. 3:

In line 140 you state that the sensors T2 and T3 are of the type Dallas DS18B20, in the figure caption they are named DS60B18. Please clarify.

line 227f:

"The corresponding calibration error CalErr is spectrally correlated but independent from one measurement to another"

I doubt that the blackbody temperature error is independent from one measurement to another. In contrast, I would assume that the error is dominated by systematic effects constant in time (e.g., a resistance offset or a temperature gradient). Please justify, why you can assume the calibration error as independent from one measurement to another.

line 229f:

"which is conservatively assumed to be equal to 0.3 K."

In line with my comments on Section 2.1.2, I am not convinced that this is a conservative assumption. Please review this number after re-assessing the accuracy of the blackbody temperatures (and emissivities).

lines 275 / 278:

In line 275 (and Fig. 9) you give a range of (-1,1) for the slope, while in line 278, the range of the slope is [−5·10−5,5·10−5]. Is this the same slope? Please clarify.

Section 4.1

In this section, I find a thorough analysis of the data in terms of fit quality and vertical resolution, but I am missing the data itself, except for two exemplary profiles. Having 625 clear sky profiles in total, it should be possible to provide a meaningful 2D plot as time series (like in Fig. 23) for both water vapour and tmperature, when cutting out the times without measurements. This would give an overview over the actual measurements and would allow the reader to comprehend the statement that the variable number of DOFs in water vapour is related to the form of the vertical profiles.

line 284:

A short sentence explaining the meaning of "reduced chi-square" would be helpful.

Fig. 14: Also the temperature shows a certain variability in the number of DOFs, although less pronounced than for water vapour. Do you have an explanation for this variation? Is it also related to the actual profile?

line 314f:

"Instead, when the effect of water vapour content

on temperature retrieval is less significant, both the Averaging Kernel profiles and the vertical resolution show little variation."

It is not clear to me what you want to express with this sentence. Maybe the words "instead" and "when" are misleading here. Do you want to say something like:

The effect of water vapour content on temperature retrieval is less significant, both the Averaging Kernel profiles and the vertical resolution show little variation (Fig. 17)."?

Or do you really want to say: "*when* the effect of water vapour content

on temperature retrieval is less significant ...". Then my question would be in which cases the effect of water vapour content

on temperature retrieval *is* significant?

Fig. 20: Difference plots would be helpful here.

line 355:

"The agreement of the lidar measurement with the CFH data was outstanding below 5 km, in the

upper troposphere and lower stratosphere in the case of the best time overlap"

This is confusing, because the upper troposphere and lower stratoshere is not below 5 km. Do you mean: "... was outstanding below 5 km *and* in the UTLS"?

line 413:

This section reads more like a summary than a conclusion.

line 456:

I would not call this an overestimation, if it is due to spatial variability.

line 463ff:

"The advantage of the FIRMOS observations is the higher time resolution of 1 minute compared to ERA5 (1 hour), allowing to catch faster

atmospheric cycles."

This is a strange argument. Obviously, a local measurement is something completely different to a global set of assimilated data. I would simply omit this sentence.

Technical corrections:

line 21:

(for a detailed discussion, see Harries et al., 2008; Palchetti et al., 2020).

line 23: 

contributions to

line 131:

Montecarlo -> Monte Carlo

geometry of BBs -> geometry of the BBs

line 132:

Palchetti et al. (2008) -> (Palchetti et al., 2008).

line 180

signal -> signals

line 193:

As noted in Bianchini et al. (2008) all the quantities used in the calibration procedure, are complex

Please add a comma afer "(2008)" and remove the comma after "procedure".

line 202:

Remove the comma after "NESR".

Fig. 5, caption:

NESR on -> NESR of

line 225:

very optically thin clouds -> optically very thin clouds

line 236:

(see Eq. 1) is larger, in this case. -> (see Eq. 1), is larger in this case.

line 237:

weighed -> weighted

line 309:

are shown respectively, -> are shown, respectively,

line 311:

I'd suggest "In contrast" rather than "instead".

Table 4, caption:

launches for -> launches on

line 336:

acquired on the 5th of February -> acquired on 5 February

line 337:

Figure -> Figures

line 338:

"The retrieved products are the red curves ..."

Please put a comma after "red curves"

Fig. 18, caption:

between 3 Km to 10 Km -> between 3 and 10 km.

line 347:

were -> was

line 393:

Infact -> In fact

line 424:

W/m2-sr-cm−1 -> W m-2 sr-1 cm

line 429:

atmospheric retrieved parameters -> retrieved atmospheric parameters

line 430:

a set of radiosondes were -> a set of radiosondes was

line 439:

were -> was

line 449:

during the day 6 cirrus clouds passed -> on 6 February cirrus clouds passed (otherwise it sounds like there were 6 cirrus clouds passing ...)