The manuscript by Wang et al. presents a sensitivity study to define the best spectral domains to measure middle and upper atmospheric wind with a THz limb sounder. Such analysis complete the few existing works on that matter by considering the most relevant lines over a large frequency domain (100 GHz-5 THz) and the effect of the spectrometer frequency resolution, bandwidth and noise. I believe the study is well suited for AMT, but I have a lot of comments listed below. I believe they will help to clarify issues without major modifications.

General comments:

• The method to select the spectral bands is the key point of this analysis and it is not clear for me. Do the authors utilize Fig. 3 to recognize the individual lines relevant for wind retrieval and adjust the radiometer band for each line to center it? If yes, how do the authors deal with a group of lines with moderate intensity (e.g., 359 GHz)? How did they identify the cluster at 655 GHz?
• The authors chose to conduct their study by pre-selecting relevant molecules. Doing so they ignore lines that could potentially improve wind retrieval, such as strong lines from isotopes of H2O and O2 as well as from minor molecules such as NO. The authors should mention this limitation.
• The authors focus mostly on retrieval errors and do not discuss the vertical and horizontal resolutions. There are trade-off between all these parameters that cannot be overlooked. To account for these issues, the authors should clarify the default observation strategy they assume for these simulations (antenna size, satellite altitude, scan velocity, integration time, …) and could propose possible improvements for future studies. If the default setting is that of TALIS, they could discuss how this study could help to improve its design.
• Why the authors have chosen to limit the altitude range to 120 km? The atomic oxygen lines (2 and 4 THz) provide a strong signal above 120 km.
• The diurnal changes of O3 and O are not discussed though they strongly impact the measurement performance (especially O3 between 60-80 km). This should be discussed.

Specific comments:

Line 2: The statement “A new method to derive line-of-sight wind is derived” is not correct since the method described in this study is the same as that used for MLS and SMILES wind measurements.

Line 8: “resolution” should be replaced with “spectral resolution” to not confuse with the spatial one.

Line 25: I find the sentence “…have achieved to observe” a bit strange  (but I am a non-native English speaker). I would rephrase it as  “… have been able to observe wind profiles between 35 and 75 km altitude.”

L27: Same for “ in the previously research”. It should be “in a previous research.”

L37: SMILES-2 is designed to measure winds up to 160 km and not 110 km by using the OI line at 2 THz (Ochiai et al. 2017)

Lines 48-50: These sentences are not clear for me. Do they mean that the simulations are performed with TALIS instrumental and observational characteristics? If yes, they should be provided in this manuscript (scan velocity, 1 or 2 LOS ?, antenna size, integration time, …)

L52: “The sensitivity of atmospheric molecules to wind …“ should  be “the sensitivity of wind retrieval to atmospheric molecules…” ?

Eq 2: y and F(x) are not defined. It should be stated that F includes instrumental effects, and it is not only Eq 1.

L89: The sentence could be rephrased as “The retrieval error of the following simulations …”. I would use “retrieval error” instead of “precision” since the null-space error component may include systematic errors.

L64: The statement is not correct “the variation of …induced by wind is amplified by the spectral line broadening”. Broader lines make difficult the detection of changes induced by wind. For instance, the lower limit of wind retrieval using THz lines is about 25-30 km because of the large line broadening induced by pressure (as it is explained in L133).

L69: I think that the following fact is worth mentioning in this paragraph:  The anti-symmetric signature of the wind  makes it possible to retrieve wind measurements simultaneously with other parameters that have symmetric signatures (temperature, pressure, VMR).

L94-96: The sentence ”Since the brightness temperature….” is unclear. A simpler statement such as “The following molecules are expected to provide a useful wind signal based on their spectroscopic line-strengths and  typical Earth’s VMR: …”. Note that because of the saturation of optically thick lines, the statement “the brightness temperature is proportional to … (VMR)” is not always true.

L98: Are the tropical and mid-latitudes profiles an average of day/night conditions?

 L104: What  the “on-the-fly” mode means?

L112: There is no comment for the altitude of 30 km though it is shown in the plot. It could be mentioned that no signature of winds are noticeable at 30 km because of pressure broadening of the lines.

Tab 1: The lines O2@773 GHz and H2O@752 GHz considered for SMILES-2 (Baron et al., 2020) are not selected. It would be interesting to explain why.

L115: How the retrieval altitudes are defined?

L116: What is the size of the antenna. Is-the effect of the scan velocity included in the antenna pattern?

What is the integration time (or scan velocity)?

L126: This sentence is not clear for me. Do you mean that you do not retrieve other parameters than wind velocity?

L136: “The 474 … show a little better … “ should be rephrased (e.g., “The 474 GHz are a little better …”)

L137 Should be rephrased (e.g., “while the 620 GHz band errors are poorer …”)

L138: Add “s” at “show” in “and the 655 GHz band shows…”

It should be indicated that this band was selected for SIW and SMILES-2.

L145: It should be stated that the OI@4 THz is much stronger than OI@2 THz: The former is better to measure the altitude above about 140 km while the latter is a better choice for lower altitudes.

L146: “from the stratospheric” -> “on the stratospheric”

L147: Another way is to increase the number of lines with moderate intensity by increasing the bandwidth or using DSB (Baron et al. [2015]).

L153: I am not convinced by this selection of lines. I don’t mean it is not correct, but one needs more explanations: Do you sort automatically the lines by considering all molecules together or molecules by molecules. Which altitude range is considered, ...

L160: The sentence “It is clear …” sounds strange for me. It could be rephrased as: “It is clear that higher resolution leads to better retrieval precision under the same noise conditions."

L166: “0.5, 1, and 2 MHz resolutions has” -> “resolutions of 0.5, 1, and 2 MHz have”

L169: “655 and 840 GHz band” -> “… bands”

Sect. 4.1: The results suggest that the need for high resolution (< 1 MHz) for upper-atmosphere wind retrieval decreases with increasing frequency. This could be stated as a general result of this study. It could be explained by the fact that the line width increases with frequency (Doppler broadening).

L176: "Is the targeted spectral line set at the center of the bandwidth?"

L186: “Baron etalBaron et al. (2013)” -> “Baron et al. (2013)”

L193: From Baron et al. [2015] it looks like the combination (555,655,2000) is better than (487,655,2000). It would be interesting to explain why you prefer the latter one.

Tab 2: I think that it is enough to show  a single NEDT  (factor 1) since the other values are simply obtained by multiplying or dividing the value by 2 as explained in the text (L198).

 L218: I think that the issue of the antenna size and the vertical resolution should be discussed a bit more.

L220: “Whether a new band is needed to improve a little precision still needs to be considered.”

This sentence is not clear for me. Do you mean: “The benefits of a  small gain in  retrieval precision by adding a new band should be considered regarding the instrument design complexity and cost”?

L226: What is the meaning of “total” in “total precision”? I would rather use retrieval error instead of precision (the calculated error includes null-space error which may have systematic errors).

L227: “The 118, 448 … are the final selected…” -> “…have been selected….”

L228: The statement “for different advantages and disavantages” is too vague.

L229: “of receiver” -> “of the receiver”

L232: The sentence “The results show …” is not clear. It could be rephrased as: “Stratospheric winds can be retrieved with good precision by using O3 lines within a relatively large bandwidth of 4-8 GHz. For MLT retrieval, a small bandwidth of 1-2 GHz is enough.”

L233 “can derive” -> “can provide” and  “very significant” -> “significant” and “at 0.5 MHz” -> “at a 0.5”

L236 The 4K cooled radiometer is not shown in the analysis. I don’t think it could be mentioned without additional information.

L238: Why is a THz receiver a superior method? To what is it compared? Measuring middle-atmosphere winds with an IR sounder has been proposed in the past.  

References:

Baron et al. [2015], “Measurement of stratospheric and mesospheric winds with a submillimeter wave limb sounder: results from JEM/SMILES and simulation study for SMILES-2”, SPIE proceeding, https://doi.org/10.1117/12.2194741

General comments

This manuscript presents a sensitivity simulation study to select the optimal frequency bands for middle and upper atmospheric wind measurement using a THz limb sounder. The sensitivity of 0.1-5 THz to wind speed was comprehensively analyzed in a typical profile scenario, and the effects of spectral resolution, bandwidth and system noise were quantitatively analyzed. From my point of view, this sensitivity analysis is meaningful for the specification of future limb sounders and can be suitable for publication in AMT. However, a minor revision is required to clarify the issues that are described below.

(1) The first section introduced the past and planned radiometers for measuring wind, but the relevant frequency bands were not described explicitly later in the manuscript. More discussion or analysis should be provided.

(2) As stated in referee report #1, the vertical resolution that is a critical parameter for satellite observations, has not been discussed throughout the manuscript, which should be included in the revised manuscript.

Specific comments

Sect. 2, Line 64: Why “the variation of brightness temperature (BT) induced by wind is amplified by the spectral line broadening and can be detected”? How the doppler shift can be obtained using a spectral resolution that is larger than the doppler frequency shift?

Sect. 3.1, Line 105: What are the differences in simulation results between these typical profiles? Do these differences affect the final conclusion?

Sect. 3.1, Line 103: Where is the BT at tangent heights of 20 km in Fig. 3?

Sect. 3.1, Line 113: How did you select the spectral lines from Fig. 3? It appears that some of the frequencies in Table 1 are not apparent in Fig. 3.

Sect. 3.2: For O3 lines at 1028 GHz the retrieval error seems acceptable, why is this frequency not selected?

Sect. 4.1: Why does the 0.5 MHz spectral resolution play such a big role in 118 GHz retrieval, while other bands not?

Sect. 5: As we know, interferometers are mainly used for middle and upper atmospheric wind field measurements. , What would be the advantages and disadvantages of using THz limb soundersthen? Please expland the relevant discussion.

This manuscript uses the radiative transfer model and the optimal estimation algorithm to simulate the inversion of atmospheric wind speed for terahertz limb sounding, screen out the target molecules in the THz band and the corresponding optimal frequencies, and analyze the influence of spectral bandwidth, resolution and NEDT on the inversion accuracy. But I have the following questions, which need to be revised and answered:

• In “As of now, direct measurements of middle and upper atmospheric wind are still scarce” and “However, measurements of wind speeds are still lacking in these altitude regions, particularly between 30 and 60km, which is also known as the “radar gap””, could some more recent references be added here?
• For “and the observation method is limited”, compared with existing equipment, what are the advantages of space THz equipment?
• In “THz limb sounder (TLS) is a concept instrument for lower thermospheric neutral wind/density/temperature and can provide wind profiles of 100–180km”, can a brief description of the pros and cons of this THz instrument be made?
• In “However, investigations on the retrieval performance using a combination of different bands and the impact of instrument parameters were insufficient in the previous studies.” 。please briefly talk about where the previous research is insufficient and why it is important.
• For “Figure 1 shows the wind measurement principle of a radiometer”, it is necessary to briefly describe the detection principle and advantages, and further, explain the diagram in Fig.1.
• For “Therefore, the spectral resolution of measurement does not need to reach as high as 100kHz and the retrieval problem can be solved by the linear least-squares method”, maybe there is a mistake here, the optimal estimation algorithm is used, not the least square method.
• For the “2.1 Retrieval method”, suggestions are that 1. Introduce the variables in OEM in combination with the parameters to be inverted; 2. The parameters in the formula are not introduced one by one, such as b, A, I, and Gy.
• In “Since the brightness temperature is proportional to the molecule line intensity and volume mixing ratio (VMR)”, the “proportional” is true?
• In “the Doppler shift calculation is performed in the "on-the-fly" mode”, can you explain the characteristics of this model?
• For “The results in Fig. 3 demonstrated that BT induced by wind can be as large as 10 K and different molecules are sensitive to different altitudes”, “Although the BT variation (i.e. wind signal) becomes larger with increasing frequency, the system noise temperature also increases correspondingly. From the perspective of SNR, most of the high frequencies do not have better SNR than the low frequencies. The BT of wind at 50km comes mainly from the numerous O3 spectral lines, and the BT induced by wind at 70km is significant at the O2 and H2O spectral lines. Furthermore, wind can be obtained up to 90km by strong spectral lines of O2 and H2O. Finally, BT of wind at 110km exists only at two spectral lines of OI. Therefore, based on the above simulation results”, this paragraph needs to be re-expressed, because it is difficult to clearly conclude Table 1 from the above simulation results (Fig. 3), at least to explain the specific basis for the selection of these bands
• “The spectral bandwidth and resolution of the radiometer in this simulation is 4GHz and 1MHz. The system noise temperature for a double sideband (DSB) radiometer at ambient temperature can be simply calculated as a function of frequency (Hubers, 2008)”, can you briefly explain the connection between SSB and DSB?
• In “Due to the purpose of channel selection, the a priori molecule profiles” which will affect retrieval precision is not considered in this study.”, dose the "priori molecule profiles" refer to the accuracy of prior profiles? But the accuracy of the prior profile is included in Sn, how to explain?
• In “Two cases of the same noise (i.e. noise from 0.5MHz resolution) and normal noise” are compared.”, how to understand the normal noise?
• Figures 16, 17, 18, etc. do not mark a,b, but a,b is mentioned in the article.
• In “The measurement noise induced by receivers is the main error source in OEM retrieval. System noise temperature and NEDT can be calculated according to the (9) and (10)”, since the accuracy of the wavelength has a great influence on the inversion accuracy, does the measurement noise include the uncertainty of the wavelength? The accuracy of possible wavelengths also needs to be considered.
• For the whole article, does the time resolution impact inversion accuracy? Maybe you can briefly explain the principle of the radiometer in the THz band, and then analyze which factors influence the inversion results.