Review of "Diurnal cycles of cloud cover and its vertical distribution over the Tibetan plateau revealed by satellite observations, reanalysis datasets and CMIP6 outputs" by Y. Zhao et al., proposed for publication in ACP

Major point

Minor points

• l. 20-21: "we find that cirrus clouds... show significant... spatial and temporal distribution characteristics" -- what this means is unclear to me. Please rephrase
• l. 22: "Be different from tropic" -- do you mean "Unlike in the tropics"?
• l. 23-24: "The cloud cover... are influenced"
• l. 32 and many others : "convection activities" this should be described as "convective activity" (singular).
• l. 60: "ERA-20C" this is not defined.
• l. 69: please check the order of references here and throughout the paper. Here Zou et al. 2020 should be cited last.
• l. 80: "detect": "document" would be better
• l. 83-87: this is a very long sentence. Please consider ways to split it up.
• l. 86: as written, it looks like Yorks et al. 2016 talks about CALIOP data, which is not the case.
• l. 90: "this makes it possible for the CATS to analyze..." CATS does not analyze, CATS is an instrument. A scientist can analyze CATS data.
• l. 90 and throughout: instead of "the CATS" and "the CALIOP", please use "CATS" and "CALIOP" instead.
• l. 108: "to clarify the cloud/aerosol layer and retrieve its properties": this is confusingly written, please revise.
• l. 116: "rotation" do you mean "operation"?
• l. 122: please see main comment #1.
• l. 146: "the detection range of the AHI moves daily": unclear, does this mean the detection range changes from one day to the next? How is this "range" defined? The range of what? It is unclear to me what was your intention when including Himawari-8 imagery into the comparison -- the results show quite clearly that its retrieved cloud covers are less robust than the other datasets you've considered. The strength of the AHI imagery appears to be its very high horizontal resolution compared to the other datasets, is that the reason for its inclusion? Do you think the insights it provides justify its inclusion?
• l. 153: "unlike satellite observations... cloud characteristics from reanalysis data largely depend on atmospheric numerical models and data assimilation schemes" true, but data assimilation is the process by which observations (including satellite) are taken into account in reanalysis datasets.
• l. 155 you specify the number of vertical levels for MERRA-2, please do the same for ERA5.
• l. 166: if the dimensions include latitude, longitude, height and time, then it is a 4-dimensional dataset.
• l. 169: I understand that in what follows you've averaged the cloud covers from these 12 models. CMIP6 includes more than 12 models. How were these specific 12 models selected for your particular study? How does this selection affect your results? How did you reconcile model outputs that were on different spatial grids? What did you select as the main grid?
• l. 172: you mean the CMIP6 historical runs stop at 2014, right? Couldn't you use RCP8.5 runs? (their emission scenarios follow quite closely the actual emissions). They might cover the 2015-2017 period. Otherwise you are comparing satellite observations over 2015-2017 (ie two rather recent years) with an average over 1979-2014 (ie 33 more years). The observations will likely be much more affected by climate change than the model output. If you can't use RCP8.5 output, please address this somehow in the text.
• l. 174: Unclear. The paragraph opens by saying that you use the 3-hourly cloud area fraction from 12 CMIP6 models, and now it says the CMIP6 simulations are unavailable for the 3-hourly cloud area fraction? How can you use that data is it is unavailable?
• l. 176: "Table 1..." please make this a new paragraph.
• l. 181: Are results shown in Figure 1 aggregated over the entire time periods? Please make this explicit.
• l. 187: "especially the total cloud cover": Figure 1 shows only the total cloud cover. Especially compared to what?
• l. 199: "CERE"
• l.213: "The cloud cover from MERRA-2 is lowest": Himawari cloud cover is lower overall
• l. 215: "except in the ITCZ": this is not relevant here
• l. 221: "cannot be overlapped": cannot overlap
• l. 222: this sentence is not useful.
• l. 229: why do you start by describing the southwestern TP (Fig. 2c)? You don't say a lot about this figure and very quickly switch to northwest TP. It seems to me most of the discussion of northwest TP applies equally to southwest TP. What logic drives the order in which you discuss the figures?
• l. 231-233: the same can be said from cloud covers in the southwestern TP.
• l. 237: this is very interesting and quite surprising. Can you reference other works in which the diurnal cycle of ERA5 cloud cover has also been found so weak?
• l. 240: do you mean "half as large"?  As I see it the Himawari cloud cover is much smaller than the CATS cloud cover.
• l. 241: the sentence mentions western TP (figures 2a and 2c), and references figure 2b. ???
• l. 244: why "partly"? What other explanations are there? If all instruments had the same detection sensitivity and coverage, wouldn't their cloud covers match exactly?
• l. 257-259: Since H8 features the lowest cloud cover of all datasets here, and that ERA5 and ISCCP cloud covers match quite well CATS cloud cover, wouldn't it be more appropriate to say that H8 underestimates the cloud cover by 10% compared to ERA5 and 20% compared to compared to ISCCP?
• l. 261-286: it's not clear to me what is gained by this analysis. It is not referred to at all in the rest of the paper. Please consider what would be effectively lost by moving this paragraph to an appendix?
• Section 3.2: this part is quite long and could benefit from being split up
• l. 305: the differences appear smaller in Figure 4a, and bigger in the subregions. This suggests that the subregions are perhaps too small for the sampling of CATS observations to be representative of what is going on with the clouds in that region. Also, you don't specify over which period you've used the CALIPSO/CloudSat and CALIPSO dataset. If you've used anything longer than 2015-2017, differences with CATS could come from that too.
• l. 310-314: As you say, this bias is probably due to optically thick clouds masking the bottom of the atmosphere in CALIPSO data, but its impact should be limited by taking it into account when documenting the number of available profiles in every height level, as suggested in my main comment
• l. 317: "status"
• l. 320: why don't you discuss the vertical distribution predicted by CMIP6 models? At least acknowledge why you think it is not a good idea
• l. 336-337: this is an important result I think
• l. 353, l. 363: we know that in your results CALIPSO understimates low-level clouds due to optical extinction from higher clouds, the same is probably happening for CATS data here. This effect might get less important if data analysis is revised (see main comment #1)
• l. 366: Do you have confidence in these results? Could you check its robustness by eg plotting out the number of profiles that are sampled by CATS over that region in that time period? Does it appear in all seasons? If you find it is robust, could you propose a mechanism responsible for producing this weird-looking sudden +7km increase in cloud altitude at 6PM LT (and its subsequent rapid decrease)? 
• l. 374: How did you obtain the diurnal cycle of tropopause height that is described here? What is the original data source? How was it processed?
• l. 388: "TAU" might look better as a greek letter
• l. 433: do you mean that overshooting over the TP can increase polar ozone consumption? Could you expand on that by explaining the mechanism?
• l. 435: why should your results be considered preliminary? What is it that you don't trust here?
• l. 437: As I'm sure you know, this kind of analysis is strongly dependent on the dataset considered for the tropopause altitude. It makes it even more problematic that you do not explain how this tropopause altitude was obtained and how comparisons with cloud altitudes were performed. Do you compare cloud covers and tropopause altitudes as local-hour averages over the entire period of CATS operation? Or do you perform overshooting detection on individual profiles (as Dauhut et al. 2020 did)? 
• Section 3.4: I find it problematic that the diurnal cycle of cloud cover of cirrus clouds (at 10km above the surface) is compared to the diurnal cycle of surface properties (T2m and 10m wind speed) as if the latter were driving the former. Please clarify the description as to explain that cirrus cloud cover and surface properties might all be driven by the same underlying mechanism.
• l. 452: "standardized": What does this mean? How did you get the standardized cloud column?
• l. 454: "statistical results": which statistical results? If you're referring to the correlation coefficients you are about to describe, please move that statement after their description
• l. 470: "the correlation provides only limited insights": so, what are they good for?
• l. 483: "radiational" radiative
• l. 484: 250hPa and 2m are quite different altitude levels. Please see my comment for Section 3.4 above
• l. 488: midnight and 03:00LT are different things
• l. 503-504: I think there is a misunderstanding here. You write that changes in the temperature at 2m somehow drives the diurnal evolution of cirrus cloud cover. I find this hard to believe. How do you propose that would work? Would surface infrared emission somehow lead to changes in cirrus cloud cover? What i could believe is, that the diurnal evolution of the temperature at 2m and of cirrus cloud cover are both driven by the same mechanism, which is heating from solar illumination. If you had access to the temperature at 250hPa, that might be easier to show. This is unfortunately harder to get. Please check your explanations.
• l. 514 this "air mass uplift" is what happens in deep convection. Please clarify your text here.
• l. 515 "positive correlation" is technically true but could be deceptive. A 0.01 coefficient correlation is a positive correlation, but it is not high enough to be meaningful. Please revise.
• l. 532: Compared to CATS they underestimate cloud cover almost as much as H8. Please mention that H8 underestimates the cloud cover as well.
• l. 540: peaks
• l. 543: "Over 7% of the subvisible cirrus clouds exist at night": Does this mean that 93% of subvisible cirrus are found during daytime? Or do you actually mean that the cloud cover of subvisible cirrus reaches 7% at night? Is that a lot or a little compared to the daily average? Please add some details to help the reader who will only read the conclusion
• l. 543: what is difficult to detect?
• l. 555-556: this is an important result
• l. 556-557: this supposes that (global-scale) climate change is strongly dependent on the cloud diurnal cycle in the TP region. I'm not sure this has been conclusively demonstrated
• l. 558-560: in a general sense, it is unrealistic to assume the cloud cover can be defined outside of actual instruments with their own detection sensitivities. The cloud cover does not exist without an instrument to measure it. It will always be impossible to completely reconcile cloud covers observed with instruments based on different observation methods.
• l. 565: detection based on solar backscatter will still be daytime-only, and subvisible cirrus are more frequent over nighttime, as you showed in your results, so it doesn't solve the problem 
• l. 569: what is GRAPES-GFS and how is it relevant to the results you present here? Please avoid introducing unrelated elements right before the conclusion
• l. 584: how come Gasparini 2019, Zou 2021 and Zhang 2021 were able to propose mechanisms responsible for the formation of cirrus clouds, and you're not? I don't think they were equipped with more data than you are. Here you are using non-sun-synchronous spaceborne lidar data from CATS and CALIOP, output from climate models, ISCCP and geostationary imagery, and not one but two reanalyses datasets. I'd say your dataset is pretty comprehensive. You have all the elements to propose an interpretation of the processes responsible for cirrus creation.
• l. 585: "Further comprehensive investigations...": again, I don't think investigations can get much more comprehensive than yours.
• l. 587: are you suggesting that aerosol loading could be one of the major influences driving the diurnal cycle of cirrus clouds when averaged over many years?
• l. 591-596: all the instances of "is" here should be replaced by "are" (data is plural)
• l. 602: "and carried them out". Carried what out?
• l. 603: "maintain"?
• Check the order of references. For instance, the many Li et al. references are not in chronological order. They are not alone with this problem.
• Figure 1: when first looking at Figure 1, I would have liked to see a figure showing maps of correlation coefficients between each pair of datasets -- CATS vs ISSCP, CATS vs H8, CATS vs ERA5, etc. As a grid. It would help quickly visualise in which regions the diurnal cycle of which datasets are well/not well correlated. Please consider building this figure and including it if it brings anything of value to the discussion.
• Figure 4 : It could be useful for the reader if you could locate for each region the topmost surface height altitude. I expect the TP surface altitude above the sea level to be quite high, but its variation across the TP is unknown to me. I'm assuming here that all the altitudes shown in the paper are above the mean sea level and not in reference to the surface, please make that explicit somewhere in the text.
• Figure 5: Where do the tropopause altitudes come from? How were they processed?
• Figure 6: please find a way to show the optical depths of cirrus clouds in each subplot. Please explore ways to make the y-axis limits of the 4 figures as consistent as possible.
• Figure 7: I am particularly concerned by the fact that apart from strong peaks (cover > 0.01), the small overshooting fractions appear to follow a pattern that makes them maximum at 7:00, 9:00, 11:00, 13:00, 15:00, 17:00, 19:00, 21:00... and minimum at 6:00, 8:00, 10:00... etc. Could you please check that this is not an artifact, for instance related to the variation of the number of available profiles at each time step? In a more general way, could you somehow discuss the uncertainty of these results? For instance, if at a given local time only one profile features overshooting, I'm not sure if the result could be called representative. Could you quantify the cloud cover that would be reached if only one profile was found as overshooting?

This manuscript reports a study for the diurnal variations of total cloud cover, cloud vertical distribution, and 

cirrus clouds and their relationship to meteorological factors over the Tibetan Plateau (TP) based on active and passive satellite observations, 

reanalysis data, and CMIP6 outputs. The diurnal variation and vertical distribution of clouds affect the radiation budget very much but very few studies have been conducted in this field. In this work, the authors studied the clouds variations, espectially the diurnal cover change, with different datasets, most importantly with lidar data, which can detect super-thin clouds that other passive instruments cannot find. The topic is of significance. The data and method they use show no apparent problem. English is good and may only needs a little improvement. Only major suggestion is that the authors need qualitively/quntitatively address in the Conclusion and Abstract what effect this research can have on climate modeling/local climate/weather.

This reviewer recommends it be accepted for publication after minor corrections.