GENERAL COMMENTS

SGLI onboard GCOM-C is a powerful instrument to cover wide spectral range of both solar reflected light and thermal emission. By adding O₂A information, understanding vertical distribution of clouds will be much improved. Authors referred their former studies. However, the description of why several cloud parameters such as base height and thickness can be retrieved from space is essential. How many parameters can be retrieved by assuming how many parameters from how many spectral channels should be described. In addition, the degree of freedom and uncertainties for each retrieved parameter using the optimal estimation method should be presented. I recommend major revision before its AMT publication.

I have the following general questions.

(1) SGLI covers the O₂A band with one spectral channel, of which spectral radiance depends on observation geometry, cloud height and fraction, surface albedo etc. The algorithm cannot use individual lines within the O₂A band. Large airmass causes saturation in strong absorption lines. Do viewing geometry and solar zenith angle affect the quality of retrieval? If uncertainties from retrieved cloud parameters varies with latitude etc., it should be presented.

(2) Forward calculation: retrieved parameters must be defined in the forward model. Definition of vertical layers, cloud top and bottom height, optical thickness of high-altitude cirrus cloud and aerosol help readers’ understanding.  Which cloud-related parameters are retrieved, and which are assumed? Do authors assume a single pixel is fully covered? Do they consider popcorn like clouds?

(3) By assuming the signal to noise ratio of SGLI and other uncertainties such as none-linearity of electronics, radiometric calibration error, what is the expected detection limit or uncertainties of these parameters from theoretical optimal estimation method? The values of a prior distribution and ranges are well summarized in Table 1. How about posterior? What are the results using real SGLI data versus posterior? These descriptions will improve readers’ understanding of the validation part of this paper.

(4) For the last 10 years, line parameters of the O₂ A band have been much improved by innovative laboratory spectroscopy. Which database the authors used? Do authors use line by line calculation for the O₂A band or look up tables in their forward model?

(5) A priori information. How many A priori information such as aerosol type, surface pressure, wind speed over the ocean are included? How much uncertainties are assumed?

SPECIFIC COMMENTS

(1) Page 1, lines 21-22

What is the difference between “systematic” bias in line 20 and bias in line 21?

(2) Page 11, Line 267

What do authors mean by “negatively affect cloud retrieval”? Generally speaking, by properly considering uncertainties, adding spectral channel for retrieval provide information.

(3) Page 19, Line 433,

What are the definitions of mid- and high-level clouds? What is the difference from “lower-level” in line 363?

TECHNICAL CORRECTIONS

(1)　Line 544, “CTT”

It appears first in this paper. It looks typo.

A new algorithm was proposed to retrieve the CBH using the SGLI 763 nm channel in combination with several other SGLI channels in the visible, shortwave infrared, and thermal infrared regions. However, there are some critical aspects that require more detailed elaboration and clarification to enhance the clarity of your findings.

1. The optimal estimation algorithm is undoubtedly the heart of your study. Please expand on the methodology section to provide a comprehensive and step-by-step description of the algorithm. This should include the mathematical formulations, assumptions made, and any pre-processing or post-processing steps involved. This will enable readers to fully understand  your work.
2. Clarify how you address the non-Gaussian distributions of the observations. Discuss the limitations and potential biases introduced by these assumptions.
3. Provide details on how you estimate and incorporate the covariance matrix of the observations, particularly addressing the correlations between different channels. Discuss any challenges in estimating these correlations and the strategies employed to mitigate their impacts on the estimation accuracy.
4. Explain how you account for angular biases or other systematic errors in the observations, particularly as they relate to the state variables.
5. Elaborate on the methodology used to determine the background error covariance for the state variables. Specifically, discuss how you handle correlations between different state variables and how you arrived at the values presented in Table 1. Consider discussing the sensitivity of your results to these assumptions and any validation performed to support the chosen values.
6. Detail how you estimate the uncertainty in cloud-base height (CBH) from your optimal estimation algorithm. This should include a discussion of the error propagation and any assumptions made in the uncertainty analysis.
7. Consider performing a sub-analysis by classifying clouds into different types and reporting the results separately. This would help isolate the impacts of cloud type on your findings and provide valuable insights into the variability in estimation performance across cloud types.
8. Some new cloud base height retrieving method should be cited in the Introduction.  such as:   Retrieving cloud base height from passive radiometer observations via a systematic effective cloud water content table, Remote Sensing of Environment, 294 (2023), 113633.

The submission by Nagao et al. combines a 4-channel 4-property cloud retrieval with a cloud phase differentiation to get liquid and ice cloud properties output in a single product. This is a neat idea and I found the paper pretty well structured and easy to read. I appreciate the authors found whatever validation data they could and I liked the comparison of cloud base heights (CBH) with reanalysis lifting condensation level.

I have a few comments I would request the authors address but I would be happy to support publication following improvements. My comment areas can be summarised as: (1) fixing some issues in the literature review and process description, (2) clarifying some method details, (3) expanding slightly on the verification step, (4) doing some quick theoretical error quantification calculations.

I don’t believe my requests would greatly change the results or conclusions, but would improve the quality of the manuscript.

1. LITERATURE REVIEW AND BACKGROUND

Beginning L53 I read it as talking about studies that retrieved BOTH cloud top heights (CTH) and other geometric information (e.g. thickness), but it appears Desmons et al. (2017) gets a single pressure and Davis et al. (2018) talks specifically about how DISCOVR can only get one piece of vertical information. The latter is a two-part analysis and I think both are worth citing, including doi: 10.1016/j.jqsrt.2018.09.006 . Either rephrase, remove these citations, or see my comments a couple of paragraphs below here to see where they could fit.

On L204 you state “Therefore, for a given CTH, the amount of oxygen above the clouds is uniquely determined”, but isn’t it for cloud-top pressure?

Next up, I would suggest using the path-length framing for parts of your description as you judge appropriate. Lower or thicker clouds = photons travel further and there is more absorption by O2. Then explain how it is hard to work out how much absorption is above versus within cloud, which is why your retrieval relies heavily on another source for CTH. I personally find this framing much simpler and clearer and I think some readers who are not A-band experts will appreciate it.

Finally, on L131—135 you mention you would need a couple of channels to separate cloud-top and cloud-thickness, but isn’t that assuming you already know other stuff like optical depth? Actually, I think you’re underselling the difficulty of getting geometric thickness here and readers should know that you’re attempting something that is very challenging. You can cite O’Brien and Mitchell (1992, doi: 10.1175/1520-0450(1992)031<1179:EEFROC>2.0.CO;2) and/or Richardson et al. (2018, doi: 10.5194/amt-11-1515-2018) as those papers show the challenging spectral resolution requirements for pure A-band approaches. The Desmons and Davis references could fit here.

1. METHODOLOGY DETAIL

I think you did a nice job here, being efficient in your explanations and I believe I could reproduce a lot of your retrieval thanks to details like Table 1. However, I haven’t used RSTAR. Can you specify the gas absorption spectra source, including if/how line broadening is handled. Is there anything else that’s relevant: e.g. effective spectral resolution or the angular resolution of your scattering calculations?

Could you also comment briefly on the cloud flag – how does it handle optically thin or broken clouds? I understand your retrieval assumes plane parallel, but a general statement on whether we should expect lots of broken clouds to be retrieved would help me to interpret things.

1. VALIDATION

When you compare against the ground-based ceilometers you use the local standard deviation and “empirically determined” some data selection. I don’t think we can be sure that the result is appropriate to compare with your values estimated from the previous section, because you might have filtered for “best behaving” scenes. Is there a way to identify these “better” scenes from your satellite product? If not then you should explicitly state that this as a potential issue for the comparison.

Can you do the CloudSat comparison split by phase? It would be cool to see, but it’s not worth much extra effort on your part so I am not pushing hard for this. If you have to reprocess all the granules to get your fields then don't bother - it's only a minor request.

1. UNCERTAINTY TESTS

We know that cloud properties correlate, but I’m fine with you sticking with a diagonal prior given the data we have available. Also, we know that the forward model also has error, which it seems you don’t include in your optimisation. Therefore your retrieval might be too “tight” to the observations, right? I think these are legitimate concerns that would take a full other project to appropriately quantify. However, you could address them by re-running a subset of the retrieval footprints and evaluating how the retrievals change. I request 3 tests:

• Try with different priors to see whether your results are meaningfully sensitive to them. Shift means term-by-term, and/or scale the standard deviations. The most interesting would be to introduce a cross correlation e.g. between tau and CGT/CBH. Even if it’s only a small correlation.
• A subset of simulations with y elements perturbed in turn? The perturbation magnitudes should probably be proportional to the standard deviations implied by S_e for consistency. This would tell us something about how errors in each particular channel would feed through into the overall retrieval.
• Some tests with a relaxed S_e, just scale it to be larger. This would represent the effect of model error.

Taken together these tests would go a long way to assuaging my concerns about potential retrieval issues. They might be best presented as supplementary info.

MINOR COMMENTS:

L280 – why the change in resolution? Why specified dates?

L319—327: “it is not surprising” and associated text seems like a long winded way of saying that a cloud with 3 km cloud top cannot be 6 km thick… you could cut this down for brevity.

L407: “SGLI, VN3, SW3, and SW4 channels” I think the first comma should be removed to read “SGLI VN3, SW3, and SW4 channels”. The current writing makes it seem like SGLI is part of the list.

L481—484: this bit just seems obvious. You could lose the last two sentences.

L519—520: “One is the presence of optically thin cirrus clouds overlying opaque clouds, which can only be detected by CALIOP. In such cloud vertical structures, the CTH and CBH retrieved by our algorithm are expected to correspond to those of the opaque clouds” – won’t this be a bit more complicated and you might end up with something in between? How far into the clouds does the TIR channel typically see?

"Retrieving cloud base height and geometric thickness using the oxygen A-band channel of GCOM-C/SGLI" (Takashi M. Nagao, Kentaroh Suzuki, and Makoto Kuji)

This is a well-written comprehensive paper, presenting the use of O2-A band information for CBH/CGT and multiple cloud retrievals, which would be applicable to a few upcoming sensors. Below are just minor corrections and additions for clarification which would help improve the manuscript.

Suggestions/comments:

Evaluations using ground, ship, and space-based observations are very impressive. Just for more completeness of the paper, especially for the regional or global analysis cases, intercomparisons with other satellite sensor cloud data like MODIS or Himawari (even a visual inspection on the general patterns) would be wonderful here, rather than just addressing the consistency or well-known cloud regimes with the sole SGLI products.

Section 5.2: I thank the authors for including this section which contains many answers about my questions and thoughts already. Looking forward to seeing your next research outcomes based on these perspectives.

Minor comments:

Line 50: It would be good to add “narrow” before nadir

Line 77-78: Remove CBH and CGT here from other fundamental properties

Line 125: It is not very clear. Please rework on this sentence.

Figure 2 caption: TH0, TH1, FAI in Fig. 2 caption have ever defined somewhere?

Table 1: How to come up with Table 1 values?

Line 206-207: This sentence lead to ask why for "rather than CBH". Just a slight revision could be done for clarification.

Line 208: Does it mean that this TIR region was missing in the original RTM?

Line 218: "TOA radiance product" means SGLI operational L1 data? Reference or data link?

Line 218: “SGLI-measured refelctances and radiances -> Information specifically for which channels will be helpful.

Line 226-227: could you add a little bit more details about how to correct it, not just added flags.

Line 232: Add the MERRA-2 data source here or to the data availability section at the end.

Figure 3: Add the time to Fig. 3 caption

Line 244: add 'RGB' color composite

Line 385: I understand the difficulties to obtain matchup data, but still 30 min average seems like quite a relaxed threshold.

Line 507: remove "," after SGLI

Line 543: Maybe "Moreover" would be better, if the authors intended to address it is good to be sensitive to other cloud properties.

Line 566-567: “This underestimation of the CTH also suggests a systematic

underestimation of the CGT by the SGLI. “ Any suggestions or further thoughts for this?

Line 596: the “current” CTH retrieval -> just for clarification