This study presents model-data comparisons on the scattering phase function of frozen-droplet and frozen-droplet aggregates in deep convective clouds. The data are obtained from the CIRCLE−2 field campaign. Gaussian random spheres and droxtals were proposed as possible candidates for representing the forms of observed frozen droplets and frozen droplet aggregates. The authors generate a total of 120 individual models of Gaussian random spheres and 129 models of droxtals and by attaching the individual models in both a homogeneous and heterogeneous manner, a total of 315 models of three different types of droplet aggregates model were generated. This study has some interesting findings, such as the differences between faceted particle and quasi-spherical particles on the phase function comparison. Given that the study provides quite many comparisons on different types of models, this study might have some reference value on this topic.

Specific comments:

1. What are relative measurement uncertainties in each region of scattering angle?
2. Could you provide some in-situ images for the support of these idealized shape models?
3. The model only considers surface scattering effects and aggregate configurations, what about internal scattering?
4. Why is the distortion parameter only set up to 0.3? Since high distortion provides the best fit, presumably higher distortion could give even better fit. It is suggested to increase the distortion up to at least 0.6 and perform the comparison again.
5. It is suggested to have more discussion about the differences among the three different regions, that is, forward, lateral, backward scattering. How these angular regions related to the particle shapes.
6. It would be better reducing the number of significant findings to highlight the most important ones in the conclusion section.
7. It may be more useful to not just state the relative differences, but also state the signs of relative differences in three different scattering regions.

General comments:

The authors provide a conclusive and comprehensible assessment of the scattering behavior of small quasi-spherical ice particles in the cirrus anvil region of high convective clouds. They vary the shapes of individual particles and their aggregates over a large range, calculate their scattering properties on the basis of geometrical optics and find an optimal constellation by minimizing the difference between modelled and observed scattering behaviour. Interestingly, the authors find very good agreement with previous work (Baran et al. 2012), which argues for a universal scattering behavior of this particle type. The introduction to the topic and the description of the methodology are very good. I expressly endorse the publication. However, large parts are rather a kind of painstaking work due to the very repetitive procedure for the different crystal types and their aggregates. I suggest shortening it substantially and not going through every variation including illustrations. Perhaps one could only occasionally refer to the results.

The use of aggregates consisting of identical particles could be omitted, as the scattering properties of both do not differ that much, as the authors themselves show, and which is also known from the literature.

I would be very interested to know whether the results at the end of the mixed particle aggregates could be obtained by a simple averaging of random Gaussian spheres and doxtrals, i.e. by something like

pf_best(c, t1, t2) = c*pf_rgs(t1) + (1-c)pf_dox(t2)

with a fraction c and two optimal distortions t1 and t2 for the Gaussian random spheres and the doxtrals, respectively. This could also be shown in Fig. 15.

In fact, you could shorten section 4.1 and 4.2 substantially and focus more on 4.3.

Another concern of mine is that the PN measurements do not cover the full range of forward scattering. And because of the very strong forward scattering behaviors, the measurements miss let' s say 99% of the scattered energy. So, does it make sense to tune to observations that only cover 1% of the scattered energy?

Specific comments:

line 24: "cloud radiative forcing" -> "cloud radiative effect"

l 51 - 52: One could argue that the radiative effects of deep clouds are to a certain extent "saturated" due to the asymptotic behavior of the radiative fluxes with increasing optical thickness. Therefore, subtle changes in scattering properties may not play an important role for this cloud type.

l 72 - 74: As projected area is radiatively more relevant than number concentration, this means that more than half of the scattering is not by FDs and FDAs! Do you know the shape of those particles?

l 112 - 114: Sentence duplications from above

l 206: Is this justified by observations? And if so, is the scattering at the aggregate significantly different from that of their individual components?

l 252: are you comparing at discrete angles or for angular intervals?

l 272: "compared with the PN measurements": The PN show several scattering maxima at about 25, 45 and 55 degrees. Could these be halo features or other indications of hexagonal structures of the ice crystals?

Btw, is the variability of the PN measurements also caused by specific orientations of the particles? Probably not, as they are quasi-spherical. Just curious. And how do you know that the PN does not contain measurements of other particles than FDs and FDAs?

l 285: No, the tilt angle do not mimic surface roughness or inclusions.

l 287: azimuth is tilted between 0 and 2pi, zenith between 0 and pi

l 291 - 292: Fig. 6). "A single Gaussian random sphere with t= 0.3 was the best-fit model...": I have no doubt that this is the case. However, it would be nice to see results for t = 0.4 and 0.5, just to see that the modeled phase function again deviates more from the observations.

l 377: doubling: ". The aggregates of Gaussian random

spheres showed a" -> "showing"

l 380: "...discrepancies in the lateral scattering angles remain.": Yes, because the scattering of aggregates is close to that of their individual (and identical) components. It is therefore possible that this exercise can be omitted as long as the individual FDs in the aggregate are all identical. See also my general comment above.

Figs 10 and 13 could be merged into one figure by using different colors for the different particle shapes.

Should a good scattering model not only fit to the mean observations but also to their variability?

Summary and Conclusions: I suggest to not repeat all the numbers (percentage differences) here but rather to provide a qualitative statement on the (dis)agreements between results from models and observations.

This study aims to find optical models for small ice crystals identified as frozen drops or frozen drops aggregates. Several optical models are compared to data obtained during the CIRCLE-2 campaign. The manuscript is well written and the topic is of interest for ACP. However, I have several questions that should be addressed before I would recommend publication.

• The observations are an important part of this paper but they are hardly described at all. More details need to added. Specific questions are:
  1. What does the PN measure and how?
  2. Are these single crystal or bulk observations?
  3. What selection of data is made? Are there only FD or FDA’s in this sample?
  4. What do the grey ranges in figure 5 and others mean? Are the spikes seen in these ranges real features or more like noise?
  5. How can we interpret the mean of the observations that is used as a target?

• Related to the last question about the data: Why is the mean of the observations an appropriate target of an optical model? Should the variation not be represented by a set of models?  On line 598 it is stated that “the assumption that FDAs consist of homogeneous components was found to be inadequate for interpreting the in situ measured single-scattering properties.” What is the criterion for calling these other models inadequate? They are close to the mean of the observations and within the grey range. So how well should any model fit the mean of the observations to be deemed an adequate model?
• The effect of distortion on the models is investigated. The best match is found for the highest distortion applied. Therefore, I suggest to also apply higher distortion values to show that the optimum is indeed at 0.3, or whether it is at a higher value. Also please indicate for which specific case of Gaussian random sphere and which specific droxtal type the distortion is applied. I also suggest adding an extra panel just as in Fig 6, 8, 11 and 14 showing the change in g as a function of t.
• On line 603 you state that “the findings of this study have significant implications for improving the accuracy of simulations regarding the radiative impacts of deep convective clouds and associated anvils on the Earth's climate system”. You did not show this. What is the basis for this statement? Often optical models with smooth phase functions and asymmetry parameters close to those you are finding are used in such calculations, so I would not expect a large impact. Please discuss a firm basis for this statement or remove or weaken this statement, also in the abstract.
• In the conclusions, the findings numbered 2, 3, 5, 6, 7, 8 are too detailed in my opinion and I suggest removing them.