Dear authors,

Thank you for submitting your revised manuscript. I have received one further review from one of the original referees, and have a couple of small comments of my own on this version. We both agree that this version of the manuscript has addressed the concerns from the previous version, and I appreciate the efforts you put in to updating the neural network (to, e.g. use reflectance, and assess dust nonsphericity) in the revision.

I feel the paper needs a few more very minor revisions, after which I believe it will be ready for publication in AMT. I do not expect it will need another round of review. Further comments from the prior reviewer are:

- P3, L11. Also mention in the introduction that you compared your typed AOT retrievals to CAMS.

- P8, L8. After "multilayer perceptrons", cite: Werbos, P. J.: Beyond Regression: New Tools for Prediction and Analysis in the Behavioral Sciences, PhD thesis, Harvard University, Cambridge, MA, United States, 1974

- P8, L17. I would replace "operations" with "functional elements".

- P10, L10. Aren't you actually multiplying by the square ratio between the actual Sun-Earth distance d and the mean Sun-Earth distance d0? Otherwise your modified radiance would be expressed in a strange unit (a radiance times the square of a length). Note that, even if d0=1 AU and thus numerically d/d0 equals d in AU, dimensionally what you are taking is still the ratio between the two quantities. [Editor’s note: I think the reviewer is correct here, please check.]

- P11, L5.Maybe you can emphasize that, if you look at Fig. 2, the spectral signature of sulphate and brown carbon look very similar (except for a scaling factor - which means that it may be possible to confuse sulfate with a certain AOT with brown carbon with a higher AOT) whereas dust has a markedly different spectrum thanks to its absorption at shorter wavelengths. Therefore, one may expect some degree of confusion between sulfate and carbon and a better capability of identifying dust.

- Section 5.5. The sulfate AOT plot looks quite encouraging. It would be also interesting to see in how many of the 21 considered scenes CAMS and your NN agree on which is the aerosol type with the largest AOT.

Further comments from me are:

Page 2 line 8: the reference should appear as “Pope et al.”, not “Pope III et al”. The surname is “Pope” and the author is the third of that name in the family. I am not sure if you are using BibTeX for references but if so a guide for typesetting names containing constructs like that is here: https://tex.stackexchange.com/questions/557/how-should-i-type-author-names-in-a-bib-file I think the correct syntax for the author name in the bibliography file is “Pope, III, C. A.”

Page 2 line 18: it would be better to state that MODIS/MISR etc have “near-global” rather than “global” coverage, as there remain several systematic sampling gaps.

Equation 4: from this, it would be good to note somewhere in the manuscript that the surface is assumed to be Lambertian. (A familiar reader can infer this from the Equation, but it would be good to state it directly in the text too.)

Equation 10: this quantity is commonly referred to as “top of atmosphere reflectance” in the aerosol remote sensing community (typically after also scaling by pi). I realise you may not wish to use the term reflectance due to the potential confusion with surface reflectance. I leave it up to you whether you wish to note this identity in the manuscript.

Page 11 line 5: the comment about SSA is wavelength-dependent, as dust is fairly strongly absorbing in the blue and UV. So I would add a qualifier about which wavelength range you are talking about here. One might instead right that dust and sulfate “are less strongly absorbing across most of the relevant spectral range” than carbon. Also, I think you mean “uncertainty” rather than “accuracy” here (since we’re discussing total error, not bias).

Figure 13: what is the background image here? I didn’t see that stated; it should also have a colour bar if the colours are to be interpreted quantitatively.

Please let me know if you have any questions.

Best wishes,

Andrew

Dear authors,

Thank you for making these additional minor revisions. I am pleased to accept this version of the manuscript for publication in AMT.

Best wishes,

Andrew