Sayer et al. presented the CHROMA algorithm and provided simulated retrievals for OCI on the forthcoming PACE mission and for OLCI on Sentinel-3. The CHROMA algorithm retrieves cloud optical thickness and cloud top pressure/cloud height, and surface albedo simultaneously and provides uncertainty estimates for each pixel. Based on the simulated retrievals, the authors found that the CHROMA algorithm should meet the PACE mission requirement for CTP.

There are new features in the CHROMA algorithm, although the cloud and aerosol retrievals from O₂ A band is well-understood. This paper provides important insight about the sensitivity of CTP to the cloud phase and ice cloud habits (asymmetry parameter). The CHROMA algorithm uses cloud top effective radius as a function of cloud top pressure for ice and water clouds separately instead of a homogeneous cloud layer. The cloud layer has a fractional geometric thickness, 0.5 for water clouds and 0.25 for ice clouds. The estimation of the covariance matrix S_y is another relevant feature of this paper.

I think CHROMA algorithm has taken into account lots of details in the cloud setup and retrieval. The authors explained the theoretical basis and performance of the algorithm clearly. It is a pleasure to read this paper. I think the paper can be published as is.

Specific comments:

1) Page 9, line 233 ‘The cloud layer is divided into 5 contiguous sublayers, running from the CTH to the cloud base height (CBH).’

It is not clear how the sublayers are defined. Are the top and base heights of the sublayers equally distributed in the cloud layer? Is 5 sublayers accurate enough for the libRadtran simulations for geometric thick clouds?

2) Page 18, close to the last line ‘…, which is a consequence of CER -dependent aspect ratio in other Yang et al. (2013a) ‘

Please check if ‘other’ is correct in the sentence.

3) Figures 8 and 9. Caption. Please provide the central wavelengths from OA12 to OA15 in the caption. I think only people work on OLCI would know the band numbers.

4) Figure 9 might need more explanations. In the paragraph started from line 470 in page 20, the first two sentences related to OCI would be better to move to the previous paragraph. This paragraph explained Figure 9 but it was not so straight forward.

General comments :

The manuscript presents the theoretical basis, performance, and usage of an algorithm, called CHROMA, whose main purpose is the retrieval of cloud top pressure or altitude from future PACE OCI measurements. Radiative transfer simulations are extensively used to understand sensitivities and expected performance of CHROMA applied to OCI, also to OLCI measurements. While OLCI measurements are already effective from Sentinel 3A and 3B satellites since 2016, CHROMA applied to OLCI does not actually use real measurements, but only RT of it, which can be surprizing at the first sight. The authors announce it for a follow-up study. The results of this study is that the performance of CHROMA applied to OCI or OLCI should meet the PACE mission goal for CTP error.

The abstract of the paper is clear. The paper is well written and easily readable. The authors are generous in providing informations that help to follow the information content of the measurements, the performances and weaknesses of the algorithm. A significative section is dedicated to the estimate of the covariance matrix Sy. Authors don’t neglect the fact that cloudy atmosphere are not rarely multilayered, that challenges all cloud algorithm, and in particular the CHROMA algorithm. The given perspectives of the paper are particularly interesting.

I think that this manuscript needs only minor revisions before being accepted in AMT. The minor revisions consist in the following: shorten the paper by making the choice to move some figures and tables to an appendix. Indeed, 17 figures and 5 tables are a lot, and a global feeling reading the manuscript is that the figures are not fully exploited and not entirely necessary in the main text. For example Figure 10, Table 4 could be moved to an appendix.

Another point, maybe not a necessary revision, is the idea that a few lines could be added in the conclusion about the overall differences and similarities obtained between OCI and OLCI, as a certain number of results are not shown. It would reinforce the clarity of exploiting OLCI as a proxy to future OCI measurements.

Minor comments (not sorted by importance) and points that need clarity :

* about the ratio of reflectance inside/outside the A band

Information about Cloud Top Height comes mainly from the ratio of reflectance inside/outside the A band. As the authors are pedagogical in this paper, I wonder if it wouldn’t have been possible to add to Figure 2 the dependence of such a ratio to CTH, COT, albedo and fractional depth (FD), that would have shown that the signal is sensitive to CTH, and FD, but less strongly to COT ? A naive question arises : what if the measurement vector consists in ratio of reflectances  instead of reflectances? Would have it make a difference in terms of algorithm performance ?

* line 243 to 246 :  

“unfortunately” is maybe not necessary,

“sometimes with minimal vertical extent”: can you be more precise ?

* line 267: error on the writing of “crystals”

* line 279 to 282:

the choice you make for ice clouds is a linear profile: but is it a lower triangular profile, or a trapezoidal shape? Also, the ice cloud’s vertical profiles in Feofilov (2015) are not all linear and exhibits often a maximum of IWC somewhere between the cloud base and top heights. Actually, the normalized vertical profile of ice clouds depends on the Ice Water Path, see for example Carbajal et al (2013, http://dx.doi.org/10.1063/1.4804794). As the authors point out later, the results are sensitive to the vertical extent of clouds, and also with a magnitude that could be studied, the normalized vertical profile, so that a perspective in follow-up work could be the definition of correct correlation between vertical extent, profile and cloud density (from more statistics that would to go beyond the use of three ARM site measurements). It is important as authors are interested by the forward model uncertainty.

* line 403: the authors perturb forward model parameters ‘within reasonable ranges’. A question arises about the probability density function they define for each cloud parameters: for example in Table 2 : are the uniform distributions for CTH and for FGD realistic? Do they come from some climatologies? Could have they be better chosen? One can think that, if not optimized, they could eventually degrade the performance of the retrievals.

* line 453: why a systematic error source? Not clear. Please elaborate.

* line 520: you can be maybe more informative in terms of SWIR use.

* line 545: ‘COT clouds are shallower’ : what does it mean?

* line 546-550: the statements seem contradictory or the conclusion are not clear: what would be the strategy to account for within-cloud scattering?

* line 567: precise here how is evaluated the ‘retrieval’s estimate of retrieval uncertainty’.

* line 571-573: comments about Figure 15a and b are not sufficient: why this underconfidence? What have to be revisited?

* line 583: ‘will be useful to data users’: please elaborate.

* line 588: the reference of Desmons et al (2017, JAMC, DOI:10.1175/JAMC-D-16-0159.1) could be added.

* line 610: or in the legend of Figure 17, for clarity, give the definition:

(CTPlow – CTP) / (CTPlow-CTPhigh)