This paper reports on an extensive study of mid-to-high latitude marine clouds, over ocean and sea ice, on both hemispheres, focusing on cloud phase and potential dependencies on different aspects. The manuscript is fairly well written, the methods is generally well conceived and some of the results are interesting. The paper could be publishable with some – maybe quite a bit – more work, so I’ll settle for major revision on this one.

Main concerns

My first concern is that the authors are using several standardized datasets based on remote sensing and modeling without showing much insight into the uncertainties in these and how the accumulate uncertainty to the final result. There is a sub-chapter on uncertainty – that comes after the results! This should obviously have come up front at the beginning and be more insightful.

Starting with the primary remote sensing data, it is well understood – and documented – that CloudSat has so-called ground clutter problems, making the lowest several hundred meters unusable – or at least very uncertain. This takes a rather large chunk out of the height interval for the “Low” class clouds here; at least in the Arctic low clouds with a cloud base lower than a few hundred meters tend to dominate. It is also well known that a cloud radar cannot distinguish between cloud layers (or a cloud layer and the surface) if precipitation is falling out of the cloud(s), especially if that is frozen as the radar is completely overwhelmed by the size of ice crystals.

Hence, I wouldn’t trust the CBH in these datasets very far and then one can’t trust cloud thickness either, except when the lidar detects both top and bottom of the cloud. Therefore, the only way to be sure single-layer clouds are sampled is to only use profiles where the lidar can see the surface; else there is no way to know for sure if the lidar is extinguished or not and if so where. The consequence of this would be a quite strongly biased dataset, with only thin clouds; the subset used here, excluding obvious multi-layer clouds, have already roughly halved the cases (Figure 14; “Other”). So, some – or even many – of the single layer clouds in this study may in fact be two-or-more layer clouds with precipitation in between. And even if the DARDAR data set seems more advanced than the previously used, one must also still remember that it is all a retrieval; it is all dependent on a lot of somewhat ad hoc choices combined with a priori model data, which also has it limitations. This doesn’t make this type of data useless; one just has to be ultra-careful and multi-suspicious.

On top of this, all the CRE data are based on calculations, not measurements, and all of the aerosol data is also modeling; probably the best modeling one can get, but it is still surrounded both by uncertainty, errors and other problems because of the way the model is designed and the modeling is set up. Moreover, the modeling is using a limited set of aerosol parameters because of the complexity of the problem, with a lot of uncertain parameterizations, and the availability of computational power. This – again – need not be a show stopper, but it has to be acknowledged and discussed. While I agreethat reference to data sources should be as good as references to the models used, with descriptions of the uncertainty in methods, there has to be a discussion here of what this means for this study, and these results.

Another problem I is find is the inherent “apples and pears” comparisons between the Southern Ocean, which to a great extent is mid-latitude, and the Arctic which is not. The study of both is important, but I would almost have wished these were two separate papers so the authors didn’t fall into the trap of doing this comparison. To start, there is nothing north of 82° N in the Arctic, which is a fairly sizeable fraction of the Arctic Ocean, while there is nothing similar for the Southern Ocean. For example, comparing cases with and without sea north of 60°N with those south of 40 °S is not a fair comparison since there are vast areas of the latter that never see any sea ice at all. Why not limit the latter to south of 60 °S to make a better comparison? Starting at 60°N includes the northern north Atlantic (roughly down to Iceland) with rather particular climate conditions due to the AMOC.  And this data was collected over two years, which is not a lot to begin with, but nothing is said about seasonality; for example the mid-winter darkness leaves a much larger foot print when sampling 60-82 °N than when sampling 40-82 °S.

Finally, several multi-panel figures are on the very small side and I have a hard time reading the labeling without magnification. This include Figures 5, 6, 11 & 13, while Figure 10 could have more separation between markers; there is certainly room for it.

Detailed concerns:

Lines 63-75: While this is a necessary discussion, this problem is inherent to all remote sensing datasets; as the technology and techniques develops the criteria are shifted. This means progress, but is of course also a problem.

Line 65: Why drag geostationary satellites into this? They have limited or no cover in these regions, and currently there is nothing geostationary for the higher latitudes. Even though it could be done...

Line 94: What ECMWF product? Operational IFS, ERA5 or something else?

Lines 101-120: Have you seen Arctic sea ice in summer? I would argue it is equally or maybe even more horizontally heterogeneous than the clouds above it! But I agree on the temporal side…

Section 2.3: And has it ever been evaluated how good this is in the polar regions?

Lines 119-123: All of which are uncertain! None of the satellites actually measure any of this first hand.

Lines 127-130: A figure with maps of these two areas in polar representation would be useful, to see where different land areas are etc., possibly with lines for maximum and minimum ice extent.

Lines 132-137: And do you trust the provided cloud mask blindly?

Line 142: Curious why you selected 2 km as the upper limit, and also what the lower limit was set to.

Equation (2): Is this really necessary?

Equation (3) & (4): So, you just assume that half of the mixed-phase pixels are liquid? Is there any basis for this, and why not 0.4 or 0.6? Or is it a “matter of convenience” lacking a better solution?

Lines 168-172: What are the intervals?

Line 185: “… satellite track.”?

Line 187: Heterogenous variability in a single cloud layer on scales > 1km could still be the same cloud, especially if it is less than solid. I have no other suggestion; you’re probably erring on the safe side here and the scales could be larger.

Lines 198-207 and elswhere: Please be a bit more imaginative when describing a figure; I can see myself what it shows, so I want the interpretation; not a repetition of what I can see.

213-216: Do not compare apples and pears!

Line 226-237: When comparing to another study, it is essentiall to know where and for what time period that study was done. Maybe differences are to be expected, if the seasons or regions are very different.

Line 243-245: While it seems intuitive that liquid fraction would increase if you use only profiles that are not fully extinguished, and also that mixed-phase fraction must decrease if also ice fraction is increasing, the latter is not obvious; could you elaborate.

Figure 4 and related text: To a first order, this only shows that temperature decreases upwards in the atmosphere; rather trivial don’t you think?

Figure 5: Why is HML lacking liquid when M has plenty? Is that because the lidar can’t see through more than the top of juicy deep clouds, and there is really some liquid farther down that is missed?

Lines 268-269: Isn’t this rather trivial? The vertical structure is such that there is simple much more vertical distance with sub-freezing temperatures than with temperatures above zero. Hence the likelihood of finding a really deep ice cloud must much larger than to find a similarly deep liquid cloud.

Lines 281-293: Could it be that the lidar is just extinguished and then you can’t know how thick the liquid layer really is? While for mixed-phase clouds you have use of the radar even when the lidar is extinguished?

Figure 6: Way to small; make two figures – and please recalculate everything to precentages instead of “# of clouds”.

Figure 7 and text on the “dips”: I find this intriguing and at the same time I do not find the suggested explanation very credible. I can buy that there are regime shifts as temperature changes, but that there would be a band of temperatures where something particular would happen while both higher and lower temperatures would be similar. At the same time, the results for the north and the south are sufficiently similar to make this either common (global) processes or – and this would be my guess – a consequence of something in the retrieval process. The latter would at least have to be excluded before trying to “fit a round object into a square hole”.

Line 340-341: The models do what they are told to do; this is all parameterized in the models and if we can’t explain these dips, why would you expect the models to be able to?

Lines 350-351: I think this would be contrary to many surface-based remote sensing studies, as well as a few aircraft profiles. Maybe you have included ground clutter here?

Section 4.2.4.: Like mentioned earlier I would like this to be more comparable, by restricting the are to the ocean where there is some ice possible.

Lines 380-381: I’m no expert here, but sea-salt I get is an excellent CCN. But why do you think it is also a good INP? I’d like to see that argument.

Lines 407-408: Because you can’t see the liquid in thick deep clouds doesn’t mean it’s not there!

Line 412: What is the rationale for this assumption? Is it some insights into the modeling of sea-salt aerosols, or is it just that it would fit well with something you like to believe? I suggest you look into what the model really does and what the source term(s) for sea-salt aerosol is.

Line 426-435: Drop this whole paragraph; it is irrelevant in the context of this work. The satellite data has no way of distinguishing between coupled and uncoupled systems and CAMS in fundamentally incapable of simulating it, and hence any effects it would or could have on the aerosol contribution. Youre grasping for straws here.

Figure 13: How are these results averaged? What about sea ice or not, and what about the seasonal winter darkness, when shortwave CRE does not exist by definition.

Line 563: As for vertical thickness and cloud phase, the only thing you have shown really is that temperature decreases with height in the atmosphere. More low temperatures, more ice. And in the horizontal it is not thickness; it is distance.

Lines 567-570: I would be careful here; it might also be an artefact of the method. The fact that the Southern Ocean and the Arctic are so similar despite having very different aerosol climates suggests that it is not a microphysical process; rather something else artificial.

Characterisation of low-base and mid-base clouds and their thermodynamic phase over the Southern and Arctic Ocean

This study presents a thorough characterization of low-base and mid-base clouds in the Polar regions, including their cloud phase and cloud radiative effect (CRE) decomposed by different cloud types. The authors use 2 years of satellite data (CloudSat and CALIPSO) to investigate the influence of aerosols, surface type (ocean/sea ice) and cloud type on the cloud phase and CRE, respectively. This characterization based on observational data provides an important tool to validate model simulations in the future, as models currently struggle to simulate cloud phase in the high latitudes correctly. The study uses a consistent definition of cloud types, which is very important based on the presented results and needs to be considered when comparing to other studies.

The manuscript is extremely well structured and is well written, so it is easy for the reader to follow the storyline. The methods are clear and described in a concise way. The figures are generally of very good quality. Some figures display rather complex content to decompose the relationships with respect to the different cloud types. However, the authors made a good job in choosing different ways to visualize this. I would like to mention that I find the introduction and the discussion of the results with respect to other studies outstanding. This makes it much easier to put the findings into context. I do not have any major comments, but please find some specific comments and technical corrections below.

Specific comments

Line 24: Could you specify ‘complex microphysics’ a bit more?

L44: specify where this underestimation occurs – at the surface?

L124: It would be helpful to mention that you are investigating the CRE at the TOA already earlier in the paragraph (closer to the equation).

L144: I think you could build a bit upon how you include Fig. 2 in the manuscript. This seems to me like a very nice justification of the introduction of Zmax for cloud classification based on atmospheric temperatures, however, so far you only introduce Fig. 2 with regards to the dashed lines. Please elaborate a bit here.

L148: You are giving estimates about how often different cloud types occur over the different surface types, and also that you reduce the number of profiles by 50% by excluding multi-layer cloud scenes. Another helpful estimate would be how many clouds actually have CBH between the lower CloudSat detection limit of about 1 km and your lower threshold of 2 km? What is the exact reason that you are limiting yourself to clouds with CBH above 2 km instead of 1 km?

L167: What are the ranges in occurrences referring to? It seems like the calculation of fraction of cloudy profiles over e.g. open ocean would yield one value only.

Methodology: I suggest clarifying/summarizing at the beginning the two different approaches: (1) cloud object analysis of horizontally connected clouds, (2) individual cloud profile analysis. Please also specify that you are calculating the statistics over all profiles that are e.g. over open ocean, and that no spatial analysis is done. The use of the word ‘pixel’ in the vertical dimension for the liquid fraction calculation is a bit misleading, at least for me. I relate ‘pixel’ to something in the horizontal dimension. Maybe you could instead use vertical bin to clearly distinguish the two dimensions?

L221: Could you give numbers for that? Are these clouds more frequent in the Arctic with regard to the relative frequency, or in absolute numbers? (as the overall relative frequency of low-level clouds in the Southern Ocean is higher).

L227: The paragraph where you compare the cloud type occurrences to Sassen and Wang (2008) seems very relevant, however it seems like you are comparing it to your frequency values based on Fig. 3. Moving this paragraph before you start discussing the dependence on CTT (Fig. 4) would make a bit more sense to me.

L254: I find it a bit counter-intuitive that you get more low-level liquid clouds if you only consider profiles where the lidar is not fully attenuated as compared to all cloudy profiles. Your ‘phase flag’ approach basically assigns a liquid/mixed/ice flag based on your calculated liquid fraction. This should not lead to a higher frequency of liquid clouds because you can detect more liquid (as you say on L243), as your phase flag is not necessarily sensitive to the total amount of liquid water. The lidar gets mainly attenuated by the liquid part of the profile, so most often you should have the case where you miss ice in the lower parts of the cloud as the lidar is already attenuated by liquid above (e.g., in the very abundant case of liquid-topped mixed-phase clouds in the high-latitudes). Is this due to the fact that these are only relative frequencies here, and due to the larger vertical extent of mixed-phase clouds you actually filter out more mixed-phase clouds than liquid-only clouds when accounting for only not fully attenuated profiles?

L295: Could you state the extensions (horizontal/vertical) from your analysis here as well to compare the values? As the horizontal extent is given in number of profiles and not in km.

L335: Very nice job in putting the results into perspective to other studies. Regarding one point you are mentioning briefly I was wondering whether you have looked into Fig. 7 on a seasonal basis? I am wondering whether sea ice leads to the fact that supercooled liquid clouds can get maintained at much colder temperatures without the open ocean as a potential source of INPs? A seasonal look into this very interesting data set could potentially disentangle some of the reasons the authors mention.

L404: In addition, as Papakonstantinou-Presvelou et al. (2022) investigate ice-only clouds, they might look into former mixed-phase clouds at a different stage in their lifecycle. These might already be completely glaciated clouds, so a comparison of the ice crystal number with the liquid fraction presented in this study is not easy.

Fig. 11: Have you performed a similar significance test as in Fig. 9? Maybe the plot is getting quite busy with all the symbols then, but mentioning in the text whether these differences are significant or not would be helpful.

L492: do you mean larger optical thickness due to a larger vertical extent?

Technical corrections

L20: I suggest citing a specific chapter (e.g. Chapter 7, Forster and Storelvmo et al. 2021) instead of the entire WGI report.

L22: showed

L36: delete ‘differ’

L46: shows

L49: underestimated

L50: showed as well

L51: single-column model simulations

L60: delete ‘a’

L79: split into

L108: provides

L111: IFS has been introduced earlier

L180: comma after percentile

L188: separate clouds

L209: delete ‘further’

L219: single-layer, as a function

L236: also showed

Figure 5: Please increase the size of the figure a bit.

L243+245+250+Fig. 5 caption+…: I would use attenuated instead of extinguished

L263: such as the cloud phase as a function of…

Figure 6 caption: with one vertical profile having a horizontal… (instead of one vertical profiles)

L284: result

L286: ground-based

L303: CTT has been introduced before

L328: temperatures

Fig. 7 caption: the liquid fraction of each profile (not profiles)

Fig. 8: I suggest labelling the x axes ‘Fraction of liquid/mixed-phase pixels’ to clearly distinguish it from the liquid fraction that has been used up to now.

L357: further research is needed

L361: ground-based

Fig. 9 caption: clouds over sea ice

Fig. 12 caption: left panel shows

Fig. 12: could you increase the size of the figure a bit?

L420: seem to play

L465: further analysis shows

L539: compared to considering all cloud profiles

L540: and not tropical

L553: and 2B-FLXHR-LIDAR for CRE?

L588: in lower parts of the cloud