Dear editors,



This paper is scientifically sound and prepared by a team of scientist known to be experts in the field.

I reviewed this paper together with my colleague Dr. Viju John. Both of us concluded that the paper needs some major, but mostly minor revisions. Our impression was that the paper is written hastily. This can only be an impression, but figures, tables, and use of terminology we found inconsistent. With little effort it should be possible though to make the paper more focused, concise, and coherent.

Kind regards, Rob Roebeling

General Comments:

This study aims to simulate AVHRR radiances, with subsequently produced cloud products, from VIIRS radiances with similar satellite orbital configuration. The stated goal is extension of the CLARA CDR, and I believe the potential benefits and uses would extend beyond that. The authors’ arguments are well constructed with conscious use of independent validation data, and the writing is strong. The clear presentation of results with added exploration of cloud parameter validation is appreciated.

I think the one major area of improvement falls in the area of the linear SBAF methodology explanation. It is stated that the linear SBAF is determined by looking at simultaneous observations, but it isn’t clear to me how the authors would be able to separate spectral bias from radiometric bias with this approach. That is, the inter-calibration sequence has three components to account for, 1) radiometric bias, 2) spectral bias, 3) retrieval biases. The authors mitigate possible retrieval biases with careful angle-matched colocation, and then seem to claim that any remaining bias is explained by spectral differences. If that’s true, then the authors would be suggesting that AVHRR and VIIRS have absolute radiometric consistency that is stable across the record. If this assumption is the case, then the authors should make that clear and share their justification.

Regarding discussion of the NASA SBAF tool: It should be noted that the NASA-derived SBAFs are limited by the fact that IASI covers the continuous spectral range of 3.60 - 15.50 µm, and thus computations that consider spectral response below 3.60 µm must rely on assumptions, which are likely imperfect. This should not be framed as a “problem” (Line 146) with the NASA-derived SBAFs, but rather a knowledge limitation due to an observation gap. The AVHRRs and VIIRS have significant response below 3.6 µm, thus computing SBAFs for these would be questionable to begin with. That is, if the use of such would not be recommended, then their impact should not be framed as “negative” (Line 146).

I recommend acceptance after the above minor issues are addressed, as well as the minor specific comments below.

Specific Comments:

Line 87: “satCORPS" should be identified as “satellite cloud and radiation property retrieval system (SatCORPS)”

Line 139-141: Given the limited spectral range of IASI data combined with the effects of solar contribution in the 3B channel range, “serious deviations” should not necessarily be unexpected (see discussion above).

Line 146: See discussion above regarding “problems” and “negatively.”  

Table 1: What’s the reason for using Training 1, Training 2, and Validation as a naming convention rather than the more typical Training, Validation, and Testing?

Line 233: Please define “channel quota.”

Table 5: Were other activation functions tested, e.g., ReLU?

Lines 278-281: The authors should explain why < 0.20 and < 0.4 were specifically used. Was a sensitivity analysis done?

Line 301: Table 2 shows the spectral relationship of VIIRS and AVHRR channels. It does not explicitly show how radiances compare unless such is broadly inferred from the spectral information (which would be scene-dependent). The authors should clarify how the table shows the radiance comparison, or is this a mistake?

Lines 543-544: This concluding statement regarding “implicit object type identification” is excellent.

Lines 545-547: I agree that using more VIIRS channels in the NNet would lead to better-simulated AVHRR channels with improved consideration of specific scene types. Is there a reason the authors did not use additional training channels to being with? There is seemingly no need to limit selection to comparable channels when it comes to training purposes.

Lines 572-573: It should be mentioned that the basic radiances also would need to maintain stability.