Zeng et al. present NH3 retrieval results from measurements of the Geostationary Interferometric Infrared Sounder onboard FengYun-4B. Compared to previous work, their data product now allows for studies of the diurnal cycle of NH3, due to the geostationary orbit of the satellite. 

The topic of the manuscript is certainly within the scope of AMT and of scientific importance. However, before publication in AMT, major revisions are necessary, as described below.

General points:

- Labels and ticks are for most of the figures very small and hard to see without zooming the PDF to 200%. I would suggest to increase font size and line width considerably.

- Until the beginning of Section 4 it was not clear to me, if one retrieval using both microwindows is presented here, or if two retrievals using one of the microwindows each are compared against each other. This should be stated much more explicitly in the beginning of the manuscript. It should be also motivated, why two retrieval approaches are compared against each other here. What should the reader learn from this exercise? Or are there also retrieval results shown using both microwindows (as indicated by the caption of Fig. 10)? I am very confused and I am not sure if I know which kind of retrieval was shown in which part of the manuscript.

- In Section 3.2, I miss general information about the retrieval. What quantities are derived from the retrieval? Is the NH3 concentration derived as a profile or only as a total column? What about temperature and the other trace gases? Until now, this section mainly collects references to theory. This is also important, but does not help to understand what kind of retrieval is performed in this work. Further, specific advantages and drawbacks of the selected retrieval approach should be at least mentioned here. Also the importance of the a priori profile in case of optimal estimation.

- In Section 4, the retrievals using different microwindows are compared. It is certainly of interest, which spectral region is suited best for the retrieval (or maybe both regions together?), but one important point is missing: There are certainly (temporally and spatially) co-located measurements by IASI, CrIS, ..., which could be used to judge, which of the retrieval approaches better matches the established NH3 columns. I really think this kind of validation approach is missing here. Just saying that microwindow #2 has a stronger signal was already clear by looking at the spectral signature in Fig.2. So what is the point of the comparison of the different retrievals, if not comparing them to independent data?

- Section 5 seems to be out of place and the title "Discussion" is misleading. In my opinion, this section shows attempts to quantify the errors of the measurements. This is an important task, but it would be interesting to know the uncertainties of the measurements, before diurnal cycles are discussed in Section 4. I suggest to add both subsections 5.1 and 5.2 to section 3, since they also describe the retrieval.

- Some sources of errors are mentioned in the sections 5.1 and 5.2. Are there any other random or systematic error sources, that are expected to influence the retrieval? I guess the calibration of the spectra should also come with uncertainties, which should be considered for the error of the retrieval results.

Specific points: (line numbers are given in the beginning of each point)

- 14: "our retrievals are implemented using two different absorption micro-windows": For me this sounds like both microwindows are used in the same retrieval. The following sentences sound more like two retrievals based on the two microwindows are performed and compared. Please clarify. (see also general comment)

- 17: Typo: no -> No

- 28-36: In this introduction to NH3, it would be also helpful to mention typical atmospheric loss processes of NH3. 

- 37-44: For completeness, it would be also worthwhile to mention that NH3 has been measured in the upper troposphere of the Asian Monsoon by infrared limb sounders.  

- 70: Suggestion: "developed by the group (Zeng et al., 2022) based" -> "developed by Zeng et al., (2022), based"

- 70: Typo: optical -> optimal

- 71: absorption feature -> absorption features

- 72: The sentence "The primary goals are ..." feels out of place there. It is between two sentences explaining the spectral properties of the retrieval. The sentence could be moved to the beginning of the paragraph (before the sentence "The retrieval algorithm uses ...").

- 74: "our retrievals are ...": It is still not clear to me, if the authors are explaining a single retrieval setup using both microwindows (in this case, this phrase should be used in singular "our reatrieval is ..."), or if two different retrieval approaches with different microwindows are compared (in this case, this should be written much more specific). 

- 75: "their difference in retrievals": Same comment as for line 74.

- 90: The introduction of Fig. 1a is misleading. It should be clearly stated in the text that this panel shows NH3 inventory data and no measurements!

- 93: "GIIRS is in principle capable of measuring trace gases": Which other trace gases than NH3 have been retrieved or are planned to be retrieved in the near future?

- 99: "comparable to existing infrared sounders": I suggest to give some examples of NedT for infrared sounders, which also measure NH3 columns.

- 100: I do not understand the formulation "provide strong constrain" in this context, but I am no native speaker.

- Figure 1c: It would be helpful to zoom into the used microwindows (in an additional panel?) and highlight the spectral features of NH3, which are used for the retrieval.

- 124: It would be good to also mention the possible interfering species and discuss a bit the content of Figure 2.

- 128: Please define the acronyms ECMWF and ERA5

- 130: Pleas define the acronym CAMS. Are the concentrations or the fluxes used from this data set? Why did the authors decide for this specific CAMS data set? Why are other trace gas data taken from ERA5 instead?

- Figure 2: N2O is mentioned to be considered for the radiative transfer model, but there are no spectral features shown in this figure. Why?

- 134: "Since the NH3 is more ...": It seems like the retrieval grid is introduced and motivated here. I do not understand, why this part starts with a comparison to CO. Maybe, the retrieval is compared to previous work of the authors? I would suggest to start a new paragraph here and then explain, how the retrieval grid is chosen for the NH3 retrieval, motivate the choice and then compare it to similar retrieval setups with the same instrument. I am also missing the information, if the retrieval is on pressure or altitude grid, since some numbers are given in pressure, others in kilometers.

- 136: "The number of layers for retrieval ranges from 7 (at high altitude) to 11 layers (at low altitude).": I do not understand this sentence. Maybe it becomes clearer, if the altitude grid itself is introduced properly before.

- 147: "The auxiliary parameters include ...": I do not understand the meaning of "auxiliary parameters" in this context. Does that mean that all of these quantities are used by the forward model? Or are some of these parameters also fitted? What exactly are "scale factors"? Are these factors adjusted by the fit?

- 149: "...which maximize the a posteriori probability given the FY-4B/GIIRS spectra." I do not understand this formulation. Please rephrase.

- 157: A_ij has not been introduced until here. Suggestion: "where each element A_ij of A represents ..."

- 160: Suggestion: "by 2.0 times" -> "by a factor of 2.0"

- 175: "(2) it is not ...": I am not sure, if I understand this correctly. Are the authors trying to say that they would need to rely on model profiles (which may have an incorrect diurnal cycle), if they want to use time dependent a priori profiles? 

- 178 and following: I am not sure, if I understood this correctly: Only one profile is used for all of the retrievals. This profile is an average over all land regions in the given lat/lon boxes for 2022. Please try to formulate more precisely here!

- 180: Since the sentence before was discussing a model, it should be "One year of simulation" instead of "One year of measurements".

- Figure 3: The error bars of the average NH3 profile reach negative mixing ratios. How is that possible for model simulation results? Is the profile averaged using the mean of all profiles? Or the median? Or something different? Further, which kind of mixing ratio is shown here? Volume mixing ratio? Mass mixing ratio? Please add!

- Section 3.4: I would expect to perform cloud filtering before doing the retrieval to avoid unnecessary fits of cloud contaminated spectra. I think the subsection headline "post-filtering" is not adequate for the cloud filtering part then. Further, I would expect this part earlier in section 3, since it is performed before the retrieval itself.

- Figure S1: It would be more interesting to see the histograms before applying the filters and with an enlarged x-axis.

- 199: Suggestion: "suggesting satisfying goodness of fit." -> "suggesting good fit quality."

- 200: I do not understand the formulation "do not have enough constrain from the observed spectra" in this context. Please rephrase.

- 215: "Fortunately, no large systematic bias is observed ...": I think, there is a banana-shape feature of the correlation (but I may be wrong, see my comment on Figure 4). If there is this kind of banana-shape, what does this mean for the retrieval performance? How do you know, which microwindow selection is better? As already mentioned in the general points, I think such correlations are more helpful against independent measurements, e.g. from IASI, CrIS, ...

Figure 4: I think, a 1:1-line would be very helpful here to guide the eye. For me it looks like there is a banana-shape in both correlations, but this is not discussed at all in the text. Maybe my eye is wrong here, but a 1:1-line would help!

- 229: I suggest to replace "non-source" by "background"

- 239: "A typical “butterfly” shape can be seen in almost all cases." What about the other cases, e.g. North-India in July? A short comment would be helpful. 

- 244-251: Fig. S2 shows very large TC for the Tibetan Plateau for BJT 14-15 h. However, the DOFS are very low here (almost zero). How does this agree with the mentioned "strong correlations between DOFS and TC"?

- 256: "We can see the diurnal AK values ...": First, the quantity "averaged averaging kernel row" should be introduced and what we can learn from it. Further, in Fig. 7 the x-axis-label should be corrected to "averaged averaging kernel row". I also miss a altitude-resolved averaging kernel plotted for different altitudes without averaging it, at least for an example in the supplement.

- 293: "The data gaps ...": I would have expected that cloud filtering would also considerably impact data availability, in particular in North India during the Monsoon season. In fact, in Fig. 10, no time series can be shown for North India in July-August. I miss a comment for this at all in Section 4.3.

- 297: "Interestingly, ...": Comparing the steps of the diurnal cycle of different months in Figures 8 and 9 with the half-year average inventory in Fig. 1a seems not like a fair comparison. On the one side, diurnally- and seasonally-resolved satellite measurements with data gaps in North India (a region specifically mentioned in the comparison) are compared to a half-year average. It's apples and oranges. Further, in July-August, there are times, in which North China Plain shows higher NH3 columns than North India.

- 307: Typo: capture -> captured

- 308: "In addition, ...": To which months does this sentence refer to? Further, would more precipitation also wash-out the highly water soluble NH3 from the atmosphere?

- Section 4.3: This section should be restructured. In the first paragraph, the diurnal cycles shown in Fig. 10 are already briefly introduced, then the next paragraph starts with some explanations of the observed diurnal cycles, while in the end of the section, the diurnal cycles are again introduced again, but in more detail. I suggest to start a new paragraph for the description and discussion of Fig. 10. In this paragraph, please first describe, what can be seen in the figure, and then explain, why it makes sense, what we see. 

- Figure 8/9: I do not understand, why the order of the panels is different compared to Fig. 5. It would be easier, if this would be consistent throughout the manuscript. E.g. keep it in a way that the diurnal variation is visible within a row and the seasonal changes are visible within a column. Maybe it would be best to change Fig. 5 (and S2).

-339: "The error bar represents one standard deviation.": So the error bars rather show variability within the region than an error/uncertainty of the measurement? Since this variability is quite large compared to the observed diurnal cycle, this variability should be discussed in Section 4.3 somewhere.

- Section 5.1: In the end of the section, a noise error is estimated for the retrieval. It would be interesting, how this error is for the diurnal cycles shown in Fig. 10 compared to the signal of the diurnal cycle itself. Is the error larger or smaller than the variability (which is shown in Fig.10 with error bars)? Is the amplitude of the diurnal cycle larger than the scaled random error from this exercise?

- Figure 11: For me, it looks like there is a low bias for the NH3 retrieval: There are considerably more dots far away from the 1:1 line in the lower-right half of the plot than in the upper-left part. And a non-negligible number of the extreme outlier have a DOFS > 0.5. This should be discussed in the text. Do the authors have any idea about the source of this systematic error? 

- 385: I think, the reference should be to Section 5.1

- 386: In Section 5.1, also absolute uncertainties are given. That would be also helpful here.

Figure 12 b/c: A 1:1-line would be helpful here. Further, it seems like the correlation splits into "branches", in particular for Fig. 12b. So, the correlation does not evenly scatter around the 1:1 line. This should be discussed in the text. Are there any explanation for this behavior?

- 429: The given link only leading to the FY-4B/GIIRS CO data, but no NH3 data is available there. Please add the missing data to the website, or give the correct link. For this and all other mentioned data resources, it would be further better to have the data versioned and tagged with a DOI. 

The paper by Zeng et al., presents a NH3 retrieval product from the GIIRS infrared sounder on board FY-4B. Geostationary NH3 measurements are novel and of high interest to the scientific community.  The paper is not bad, but there is certainly a lot of room for improvement, both in content, form and depth, which is why I would recommend a major revision, after which the paper needs to re-reviewed. I would also encourage the authors not to see the revision as a hurdle that needs to be overcome to publish, but rather as an opportunity to improve the paper, and to increase its impact. 

A. Major Comments

A.1 General comments on the figures

 - As a general rule, the font size of the figures should match that of the text, or slightly smaller.  A lot of text is currently simply unreadable on print-out. 

 - Again, as a general rule, try to decide whether a figure should be a one-column figure or a two column-figure in the final manuscript, and then utilise exactly the full width or exactly half of the full width of the text. This applies at least to Fig 1C (bottom row could be full page width), Fig 2, Fig 4, Fig 6, Fig 7 and Fig 10.

 - For large multipanel figures with the same x and y axis, the axis tick labels and axis labels can be removed from the middle panels. This in effect increases drastically the space available for the actual figures. For instance in Figure 8, the degrees East can removed on top, between the first and second row, and the second and third row (in this way only keep it at the bottom). Likewise, the degrees North can be removed between the first and second column and between the second and third column, and only kept on the left side. After this, all panels can be enlarged. This applies to Figure 5, Figure 6, Figure 8, Figure 9 and Figure S2.  For Figure 5, one colourbar suffices, which can go to the bottom, again allowing for larger and more visible panels.

 - Wherever possible, please export figures in a vector format (https://en.wikipedia.org/wiki/Vector_graphics, e.g. pdf), to avoid pixelation of lines and text, especially visible when zooming in on screen. For instance figure 10 or 12 (but ideally for all figures).

A.2: Other major comments

 - 1 km layering at the bottom of the atmosphere is not sufficient (at least 500 m is required below 2 km). 1 km is not sufficient for modelling fast varying temperature (e.g. thermal inversions at night), nor for representing the fast NH3 (especially at night) and H2O variations (always). I can confidently say that the fit residuals will decrease once a finer layering is adopted (even just for the H2O fit). The fact also that there was a need to multiply the spectral noise by two, gives a hint that the fits can and should be improved. I realize that redoing all the fits is a major undertaking, but worth it, if the product is going to be further extended in time and used by the community. If a coarse spectral resolution emissivity database was used, I would also encourage to look for better alternatives.

 - Section 3.1 "forward model" already introduces elements of the retrieval. It makes it confusing and not well structured. E.g. the layering used for the atmosphere doesn't necessarily have to match the layering used for the retrieval.

 - Section 3.2 I think a comprehensive table with all the forward/retrieval parameters would go a long way in making this section more understandable, and the paper more reproducible. For an AMT paper it is essential that these important technical aspects are fully transparent. E.g. a table with 1st column: Parameter name, 2nd column, levelling used in forward model, 3rd column: levelling used for the retrieval, 4th column: a priori, 5th column covariance matrix (i.e. % variability if diagonal, and description of off-diagonal elements, e.g. correlation length). This table should at least include all parameters that are retrieved, but also the important atmospheric and surface terms (temperature/pressure)

 - Figure 3: please provide an additional panel with the a priori covariance matrix. Showing the a prior NH3 profile is only showing half of the story. How was it calculated? Also from model output? Please discuss this matrix at some length, as the choice of covariance matrix is absolutely key (both diagonal and off-diagonal elements).

 - Section 4: comparing two retrievals is very interesting, but obviously also opens the door to a lot of questions. Currently the reader remains rather unsatisfied.

    * For the column retrievals, would it not be better to apply averaging kernels in the comparison, to remove the impact of the a priori all together (and to what extend it is used by both retrievals). Perhaps both could be shown (with and without AVKs applied). I am not sure what is the best approach (which averaging kernel to apply to what), but there is certainly literature out there to guide you. 

    * Figure 4b is misleading, as it is highly saturated with the colourbar as-is. I would propose to use a colourbar with log scale (e.g. 1 to 10000 or higher). Also, the vast majority of observations have a column below 5 10^16,  and for these points, there seems to be a clear bias between the two retrievals, where the micro-window 1 is low-biased.  I disagree strongly that "no large systematic bias is observed" (perhaps non-surprisingly because of the lower information content - but in that case applying averaging kernels would show this, and the bias should disappear).  I would  not show observations above 1e17, to allow higher level of detail for the low columns.

    * For the comparison, it would be good to add a figure showing the mean residual (calculated-observed as a spectrum) for one or two selected regions, for both retrievals. For a well-behaved retrieval, there should be no systematic features, and noise should average out.

    * Did you also try to retrieve using the entire NH3 band? What were the results? Can both retrievals be combined to provide a combined column? If columns are simply averaged, what to do with the corresponding retrieval uncertainty and averaging kernel? A discussion is missing on this. In the NH3 dataset that is shared online, what are the variables that are included?

    * Are there differences in retrieved profiles? Could you again for selected region(s) show the average retrieved NH3 profile (+ a priori?) with both windows? Alternatively, can you show the averaged scaling factor profile (retrieved NH3 profiles divided by their a priori). 

    * Figure 10: Can you add thermal contrast to this figure? (compare e.g. Fig. 4 of Clarisse et al., 2021). This would allow discussing better the variable sensitivity as a function of time. You could also add the mean or median DOFS (unfiltered, total average).  

 - Section 5

    * I would call this rather "Retrieval experiments".

    * Did you add synthetic noise? (ideally generated randomly from the noise covariance matrix)

    * I found section 5.2 not that interesting. Much rather it would be nice (see also comment above) to show and discuss the retrieved NH3 profiles along side temperature profiles and averaging kernels. This could be done with averages, but also on individual observations, showing cases e.g. of extreme temperature inversion (which forces the profile to a narrow band), very large thermal contrast, etc.. The paper does not nearly go deep enough on this aspect.    

B. Minor comments

- Please detail exactly how thermal contrast was calculated (because "lower atmosphere") can mean a lot of different things. Also, is surface temperature used or the brightness temperature of the surface (the difference being the surface emissivity)

- Figure 1c: x-label should be wavenumber, not frequency

- Figure 2:  y-axes labels inconsistent (diff vs difference)

- Figure 2: I would not show a priori x1 or x2, it doesn't contribute or provide any new information. In this way, the second row can be removed altogether, since this difference is shown in the bottom panel. Also please refer to the appropriate section for the definition of a priori (which is not introduced yet at this stage of the paper).

- Line 159: How was the measurement error covariance matrix determined?

- Line 170: this is technically incorrect. It all depends on the magnitude of Sa. If the retrieval is poorly constrained (very large Sa), then the DOFS will naturally be high, as all info comes from the measurement rather than the a priori. Conversely, if the retrieval is very (too) tightly constrained, the DOFS will always be small. 

- Line 191: I guess this should read satellite zenith angles instead of solar zenith angles?

- Line 234: this is factually incorrect, one can have a lower thermal contrast with increasing surface temperature. TC is a temperature difference, and for instance (Tsurf=290, Tair= 280 K) has a higher thermal contrast then (Tsurf = 300 K, Tair = 299 K). In fact the entire passage lines 233 - 237 need to be rewritten. I suggest the authors consult again the papers they reference just above, for proper terminology and physical explanations. 

- What is the spectral resolution of the emissivity atlas that was used. Please mention this in the manuscript.

C. English. Although in general not bad, there are several typos/grammar mistakes. I definitely did not try to be exhaustive, so here are just some that I noted: 

- Line 15 and 75: to imply > to quantify / study / analyse (?)

- Line 310 availabel

- Line 77: remaining > remainder of (?)

- Line 83: inconsistency east Asia vs East Asia

- Line 100: constraints (?)  +  remove "measuring" + remove "column"

- Line 113: "can be referred to" > can be found in

- Line 134: the NH3 > NH3

- Line 135: remove therefore

- Line 233: "detectivity" please rephrase

- Line 241: Please rephrase

- Line 244: This strong correlations > This strong correlation

- Line 294: "are resulted" > result from