Summary

The authors perform a systematic assessment of in-atmosphere cloud radiative effects across CMIP models, and compare these with two observationally-based estimates. They also examine the climate change response of the cloud radiative heating in models and show that the upper tropospheric response is well modeled by the assumption that clouds shift upwards following the isotherms, consistent with theoretical constraints in the literature. The results are interesting and important, and the paper is well written. I have only minor comments and recommend acceptance after they are considered.

General Comments

1.    The paper occasionally uses language that seems a bit self-promoting, which kind of grates on me for a paper (it might be appropriate in a proposal). For example, I note the following phrases describing the work: “the first systematic assessment”; “the most comprehensive assessment ever generated”; “the most comprehensive assessment …to date”; “is a major step forward”; “we generate the most comprehensive assessment of cloud radiative heating in global climate models to date”. Suggest using this language sparingly.

2.    Can the mean state cloud properties and the upward shift of clouds be separated as is done here or are they inextricably linked?  By this I mean, if a biased model had the “right” mean-state cloud radiative heating profile, would it still shift upwards by the same amount as it currently does, or would that corrected mean-state cloud radiative heating then lead to a different response of either the temperature profile or the circulation influenced by the diabatic heating? If not, does the fact that the warming-induced high cloud response can be modeled as a simple upward shift somewhat undercut the message that cloud radiative heating is super important for the atmospheric circulation and its response to warming (used as motivation for the analysis).

3.    The “pattern effect” results: While the amip-future4K experiment does impose a pattern that is not uniform, it is not a strongly heterogeneous pattern and typically most pattern effect studies contrast the uniform or 4xCO2 warming pattern with something more distinct, like the observed warming over the last few decades. Hence while I don’t doubt that the high cloud response is going to be pretty similar even with a more heterogenous warming pattern, it is too strong a statement to state unequivocally that the difference between these two simulations “quantifies the extent to which the response of cloud-radiative heating to surface warming depends on the pattern of surface warming” and to conclude that  “the pattern of surface warming has little impact”, that “cloud-radiative heating is essentially independent of the pattern of surface warming”, or that “the response of upper tropospheric cloud-radiative heating is essentially insensitive to the details of the surface warming.” All that has been tested here is whether the high cloud response is different between amip-p4K and amip-future4K, which is a weak litmus test given how similar these patterns are. I suggest either weakening / appropriately caveating these statements; contrasting the cloud heating response in experiments with more distinct warming patterns; or just dropping this part of the analysis, which seems a bit tangential anyway.

Specific Comments

•    L83: Unequivocally is misspelled.

•    Figure 1 description: I suggest dedicating some more text to explaining the basic features of this figure (or Figure 2) for those readers that are not used to looking at in-atmosphere cloud radiative heating rates.

•    I am surprised that the same Beta parameter works for every model in Eq. 5. Is this because they are all subject to the same uniform+4K of SST warming and have roughly the same upward shift of isotherms?  If one were to estimate the “best fit” Beta for each model, how much would it vary, and would that lead to even better predictions?  This connects back to General Comment #2 where it seems that the basic atmospheric response is not very dependent on cloud radiative heating such that one can model each response using a single model-invariant Beta value.

•    Figure 4: suggest showing the multi-model mean or median in the last open panel.  Rather than overlaying the isotherms, I wonder if it might be helpful to instead overlay the amip control climate cloud radiative heating contours.

•    Figure 7: Suggest noting in the caption rather than in the text that the colormap is centered on 4K. 

•    L345: In this discussion, I suggest citing Yoshimori et al (2020) [DOI: 10.1175/JCLI-D-19-0108.1], who make many of these points.

This paper examines the cloud radiative effect from an angle different from many other studies. Instead of measuring the cloud effect by the radiative fluxes, the analysis here is focused on the radiative heating of the atmosphere. This relatively new angle presents many interesting results and mainly for this reason, I think the paper potentially makes a valuable contribution to our understanding of the climate effects of clouds. I would recommend acceptance if the following comments were addressed.

1. Method

It is well recognized that the cloud radiative effect, if simply measured by clear- and all-sky difference (eq. 1), would be subject to large biases due to the non-cloud changes causing different radiative changes in clear- and all-sky (a “masking” effect). The kernel-based adjustment method of Shell et al. (2008, https://doi.org/10.1175/2007JCLI2044.1), for example, is widely adopted to correct this bias in the cloud feedback analysis. It can be expected that similar issues will occur for the heating rate analysis. What measure can/should be applied here? Tests, discussions, and/or recommendations should be made, in the context of continued kernel developments. For example, one latest kernel dataset published by Huang and Huang 2023, https://doi.org/10.5194/essd-15-3001-2023 extended the kernels from TOA to surface already. Would you recommend making available layerwise kernels for heating rates or (equivalently) for flux profiles?

Line 149. Given that CRH profile is discrete, I am concerned about the impacts of the interpolation on the resulted profiles. For example, are the CRH features at the cloud boundaries, e.g., at the top of high clouds and around boundary layer clouds, dislocated or blurred by this processing? Some discussions, preferably with supporting plots, would be appreciated to clear this concern. This could for example affect the replotting of the CRH results in different coordinates later on.

1. Results

The paper is claimed to be “the most comprehensive assessment of atmospheric cloud radiative heating” (line 48, 324). I found the lack of such crucial results as the longwave vs. shortwave decomposition of the heating rates, at odds with the claim. These may be necessary for interpreting some results, e.g., the cancellation noted in Line 239.

Line 170: warming near surface seems not as noticeable from Li et al.

(https://doi.org/10.1175/JCLI-D-14-00825.1)?

Line 177: Is this difference due to cloud difference or other variables? Perhaps good to overlay these plots with cloud amount climatology.

Line 188: what’s causing their difference, given they’re based on the same active sensors? Any systematic bias in obs, e.g., due to their limited time/space sampling, when compared to GCMs?

Line 227: beta seems a crucial parameter in this analysis. More explanation and discussion on how this value is set would be helpful.

Line 247: “very similar” sounds subjective to me.

Line 292/298: there seems latitudinal difference which may be a different aspect of Ts control?

1. Literature review

The introduction and comparison of results to previous works would benefit from a more complete inclusion of relevant papers, such as:

Zhang et al. (2017), https://doi.org/10.1007/s00382-016-3501-0

Kato et al. (2019), https://doi.org/10.1029/2018JD028878