Han et al. examines the role of ice nucleating particles (INPs) and thermodynamic conditions on mixed-phase microphysics, with particular focus on the proportion of liquid to ice in deep convective clouds. They use cloud-resolving simulations to show that increasing the INP concentration leads to glaciation at warmer temperatures and a lower fraction of liquid cloud pixels, and they examine the microphysical mechanisms driving this effect. Furthermore, they apply multiple satellite forward operators to their model output and compare the results to remotely observed cloud properties in order to explore how well these changes driven by INP concentrations might be detected from satellites.

Overall, I think this study tackles valuable questions given the uncertainty surrounding mixed-phase clouds and their interactions with INPs. The use of satellite forward operators to make an apples-to-apples comparisons of remote sensing observations and model data is novel and interesting. That being said, the paper is not always clear about the causes for discrepancies between the models and observations and how these comparisons contribute to the overall goals and science questions. I have some questions/suggestions on these and a few other points that may improve the analysis and focus the discussion. I therefore recommend this article be accepted with major revisions. Please see more detailed comments below.

Major Comments:

1. Role of comparison with satellite data: The satellite forward operators allow the authors to directly compare model simulated properties to remote sensing retrievals, but it is not clear whether the goal of these comparisons is to evaluate the performance of the model microphysics or of the forward operators/retrieval algorithms. In some places, it seems that the aim is to evaluate how well the model replicates the remotely sensed cloud phase distributions (i.e., the satellite obs are treated as the “ground truth”). In the conclusions, the authors state that the goal is to “evaluate cloud microphysical processes of numerical models using satellite observations directly”. However, the discussion does not focus on which processes drive the discrepancies between the modelled and actual cloud phase distributions. If this is the overall goal, then there needs to be more analysis of the microphysical differences between simulations which do match the observed data well (e.g., the DEC03/05 simulations) and those which do not, to actually evaluate which processes are not being treated correctly in the models. In other places, it seems that the aim is to evaluate the retrieval algorithms and whether they are able to capture the differences that we expect to see based on the simulations (i.e., the model data fed through the forward operators is the “ground truth”). Most of the discussion fits this framing, which also makes sense given the comparisons of the CLAAS-2 and ML-based retrievals. The discussion about why the CLAAS-2 retrieval performs worse than the ML retrieval should be expanded on beyond saying it was due to the “loss of information through the postprocessing” (e.g., were the pixels with high uncertainty in the ML retrievals on cloud edges?). Also, why does the CLAAS-2 retrieval not capture the differences in cloud phase distribution as a function of INP when it does capture differences as a function of the thermodynamic environment, despite the two perturbations having a similar order of magnitude effect on the modelled phase distribution at cloud top? Either of the two options could make for a valuable contribution to the literature—but the authors should clarify what they are aiming for in comparing the satellite observations to the models.
2. Model representativeness: Section 3.1 compares the retrieved cloud properties between the ICON model using a satellite forward operator and actual satellite observations. There is some discussion of the differences between the model and observations which is generally attributed to “model physics”, but it would be useful to see at least a bit more discussion of the factors potentially causing this (e.g., 1.2km grid spacing isn’t resolving entrainment fully). Though I do appreciate that the simulation is not going to perfectly match the observations, the article would be improved if the authors consider how any under-resolved model physics might affect the validity of their findings (or at least argue why they think this isn’t the case). For example, a good amount of the discussion focuses on differences in ice microphysics between cloud top and in-cloud—if entrainment is under-resolved, is it possible that the simulated difference between the two regions is magnified compared to reality? The results here may certainly still be applicable especially towards cloud cores that more closely resemble the very homogenous clouds simulated here, but a caveat about realistic cloud edges might improve the discussion.
3. Figure 6 and Figure 7 are the same figure. The authors have probably inserted the wrong figure for one of these unless I’m missing something here.

Minor Comments:

1. Lines 86-87: Suggest citing some work on the transition to deep convection around mixed phase regions (e.g., Li et al., 2013, Sheffield et al., 2015, Mecikalski et al., 2016).
2. Lines 112-123: A lot of this paragraph focuses on INP impacts on precipitation, but this doesn’t seem to be a focus of the rest of the article, so this can be abbreviated to a short statement that the impact of INPs on precipitation from deep convective clouds is still uncertain and may depend on precipitation/cloud type.
3. Lines 141-146: These sentences/references would fit better in the methods.
4. Lines 183-193: Are the fitting constants described uniform in space and time, or does the INP concentration only depend on the ambient temperature?
5. Lines 204-205: How frequently is the model output/sampled for the analyses?
6. Lines 222-225: The description of the temperature perturbations was confusing. Suggest rephrasing to “The temperature increment is linearly increased/decreased with height from 0K at 3km to +/-3 or +/-5K at 12km, […]”.
7. Lines 347-351: Is there a portion of the domain near the lateral boundaries that is excluded from the analysis?
8. Section 3.2: Does changing the time period considered impact the results (here and in the rest of the sections) at all? Are the trends as a function of INP concentration relatively consistent in time?
9. Fig 4, 5, 8: The color scheme used makes it difficult to see trends as a function of INP. Maybe use a color scheme with warmer colors for increased INP and cooler colors for decreased INP (or something similar).
10. Fig 4-11: Figures would all be improved by indicating important altitudes/levels with a dashed line or similar, or adding a shaded region to indicate the mixed phase region.
11. Fig 5: It would be helpful to add a panel to show what percentage of the overall ice formation is heterogeneous vs. homogeneous.
12. Fig 6-7: This figure format could use a dashed line at 0.5 liquid fraction since the glaciation temperature is referenced multiple times in the discussion. Also, it is hard to see the differences in the panels e-h, a difference plot relative to the control instead (or in addition) would be clearer.
13. Section 3.3: Are the trends in liquid mass fraction presented here monotonic/is there a consistent trend among all the INP concentrations tested? The authors don’t need to show these in the article itself but would be good to make a statement about that and include those figures as a supplement or as a reply to this comment.
14. Section 3.4: “Liquid cloud pixel number fraction” is a confusing term. Especially since Section 3.3 is about the “cloud liquid mass fraction”, I initially thought “number fraction” was referring to the number of liquid drops versus ice crystals, rather than the number of pixels which are mostly liquid. Maybe the term could be changed to something like “liquid cloud pixel fraction”?
15. Lines 605-606: The authors say that the magnified impact of INPs at cloud top compared to in-cloud has “implications for analyzing cloud products […]”. What are the implications?
16. Lines 614-615: Found this sentence confusing: “Total cloud ice mass concentrations do not increase but decrease with increasing concentrations in the simulated deep convective clouds.” Does this just mean that the total cloud ice mass decreases with increasing INP concentrations?
17. Lines 627-632: Add a sentence here about the relative impacts of INP and thermodynamic perturbations on the cloud phase distribution.

Within the manuscript, the authors retrieve cloud properties from ICON simulations of a deep convective cloud day over central Europe using a satellite forward operator and remote sensing retrieval algorithms. They test the influence of lower and higher concentrations of ice-nucleating particles (INPs) on the liquid-to-ice partitioning in-cloud and at cloud top as well as how different initial thermodynamic states influence the cloud microphysics. The authors convincingly show that microphysical and thermodynamic adjustments to the model setup can be equally important for the simulated cloud microphysics. Overall, the high INP and high convective available potential energy (CAPE) scenarios show the best match with satellite observations from SEVIRI.

The manuscript reads exceptionally well with an easy-to-follow structure. From a methodological side, particular emphasize is given to the use of a satellite forward operator for a more ‘apples-to-apples’ comparison between the ICON model output and the satellite observations. In remapping the ICON output to the SEVIRI resolution before applying the satellite operator, the authors can disentangle the effects of remapping and of the satellite operator on the simulated cloud microphysics.

I want to emphasize that it has been a great pleasure to read the manuscript, and I recommend it for publication after only very minor revisions. I structured my review into general and specific comments below, before I will give some hints for technical corrections.

General Comments

(1) It looks to me that Figures 6 and 7 are identical and the authors may have accidentally placed the same figure twice into the manuscript? I cannot identify any difference within the two figures. However, based on the text, it sounds like this is an important figure with a very interesting interpretation, so I suggest to double-check Fig. 7 if it indeed shows cloud top.

(2) In Figure 3 you compare the spatial distributions of liquid water path, ice water path, and cloud optical thickness for the CTRL case and the CLAAS-2 satellite product. It shows clear discrepancies between the ICON output and the satellite observations in terms of intensity and spatial coverage. The authors state that, as Geiss et al. (2021) reported, the primary source of these deviations stem mainly from model assumptions on subgrid scale clouds. However, as you find a much better match between the model and satellite observations for the high INP case (with SEVIRI_ML), it would be interesting to reproduce Figure 3 for the high INP case and compare it to the satellite observations and see if you find a better match with the LWP/IWP/COT maps.

Specific Comments

Line 102: I cannot completely follow the line of argumentation here. With respect to which parameter does the frequency distribution of ice water fraction have a U-shape? Please clarify this.

Figure 6: As the color bars are the same for Figs. 6a-d and 6e-h, respectively, I suggest to just show one color bar each and rather have a y-axis title at each individual panel, as right now it is a bit confusing with w (m/s) corresponding to the color bar but not to the y-axis title for the panels on the right column. In addition, could you explain what you mean with normalized counts in panels 6e to 6h? The normalization is a bit unclear to me.

L489: Could you discuss at this point how much higher the cloud is extending above the mixed-phase temperature range and can you somehow diagnose the sedimentation rate in the model to investigate this statement?

L547 and Fig. 8: This is only a suggestion, but as you talk about a temperature shift as compared to Fig. 8c (on the SEVIRI grid), maybe you could move panel e below panel c and move the legend to where panel 8e has been before? Thus, the temperature shift would be immediately clear, and it is a bit easier to compare the shape of the curves.

Fig. 9: maybe adding a rough estimate where the cloud top height is in these simulations would help to interpret the vertical profiles of vertical velocities.

Technical Corrections

L349: cloud water plus cloud ice

L434: that is Sect. 3.3, is this the correct reference?

L547: noisier

L551: of approximately 1 above -10°C and 0 below approximately -30 °C,

L563: as the CAPE increases