General comments:

This manuscript is a novel attempt to assimilate C-band polarimetric microphysical retrievals into a convection-resolving model in Germany compared to the legacy assimilation approach using Z. The authors find generally positive benefits, particularly for LWC, with IWC neutral/beneficial in stratiform cases but with harmful effects in a convective case. This is not particularly surprising, as the retrievals used are not formulated for rimed ice/hail, but it is nonetheless a good demonstration of what can happen if the retrieval equations are not applied appropriately/selectively to how they were formulated. The results overall indicate that the addition of LWC and IWC retrievals, with Z, results in the best forecasts, and encourages further exploration. The manuscript is very well written, with a thorough introduction section that clearly establishes the state-of-the-art of polarimetric radar DA and its challenges. The basis and results of the study are novel and timely as polarimetric radars are increasingly implemented and their information content explored. There are a number of improvements I’d like to see to the manuscript primarily regarding clarity or requests for additional context/information. While I don’t think it is necessary for publication, I did feel the manuscript could benefit from one or two plots actually depicting the results of the different assimilation experiments (e.g., actual post-assimilation hydrometeor and/or moisture fields, or QPF fields, vs. the observed QPE; the distribution/statistics of the IWC/LWC fields, etc.), so that readers can get a visual sense of the effects these are having rather than solely relying on only statistics. Despite this, I believe the manuscript will be ready for publication once the following minor comments are addressed.

Specific comments:

1. L26: Please define FSS and FBI for the abstract.
2. L103: It may be worth mentioning that beyond the rudimentary treatment of PSDs, hydrometeor shapes and orientations are rarely (if ever?) taken into account at all within model microphysics schemes.
3. L160: Which microphysics scheme is used in the ICON-D2? A citation is needed here.
4. L180: In this study, when Z is assimilated (either in CNV+Z, CNV+[LWC + IWC]/Z, or CNV + LWC + IWC + Z) does this mean it is both directly assimilated and assimilated through LHN? It is not explicitly clear to me how Z is being used as the reference assimilation study. (Edit: I see now on L338 it is stated that LHN is excluded in this case, but this wasn’t clear at first when mentioned in the context of the operational Z assimilation scheme).
5. L204 and elsewhere: Is there a specific reason the variables are included as ZDR, PHIDP, KDP, and RHOHV rather than Z_DR, Φ_DP, K_dp, and ρ_hv? It isn’t critical, but I think use of the variables are more standard notation, would clean up the equations/discussion, and not suggest that ZDR, RHOHV, PHIDP, and KDP might be acronyms to unfamiliar readers.
6. L206: Strictly speaking, isotropic scatterers will have an intrinsic ZDR of exactly 0 dB rather than one close to 0 dB.
7. L217: This value of ρ_hv seems much lower than what I would have expected. At S band, ρ_hv in dry snow/ice (rather than melting particles, which are neglected in these retrievals) is usually nowhere near 0.85 unless it is very large hail experiencing resonance scattering, in which case the use of the IWC retrieval equations would be inappropriate anyway (both due to it being hail rather than snow and due to the non-Rayleigh scattering). With all the tests done to find the optimal DAP, were any tests done to examine the impact of these thresholds and find optimal values? What did the typical distribution of ρ_hv values actually look like above the ML? I am concerned that using such a low ρ_hv threshold aloft will necessarily retain data that is going to have very noisy Φ_DP/K_dp and thus noisy/erroneous IWC retrievals that should not be assimilated. I would be curious if a much more stringent ρ_h threshold (perhaps something like 0.92 or greater) would result in better retrieval/assimilation results.
8. L221: How is the bottom of the melting layer determined, as shown in Fig. 2? I assume some sort of ρ_hv threshold when using a QVP, but for the convective case (for which I presume model data was relied upon) it is less clear.
9. L225: I appreciate why such a large window is needed for consistency, but depending on how small and isolated the convection is, could 9 km be too large a window and end up heavily incorporating data boundary regions into the Kdp calculation for precipitation regions?
10. L246 and elsewhere: I assume log here is log₁₀ and not ln? It may be helpful just to clarify.
11. L256: It may also be worth mentioning in addition to just hail that R(A) at C band struggles from the resonance scattering of medium-large-sized raindrops, which causes R(Kdp) to outperform R(A) for moderate to heavy rainrates.
12. L207: I am a bit confused by this. The Figure 2 caption says LS = 10 km, which would make it equivalent to LC, but here it makes it sound like LS != LC.
13. L305: While the results of these tests are not shown here, previous studies such as Liu et al. (2020) have demonstrated the same thing for hydrometeor mixing ratios and could be cited here, if desired.

Liu, C., M. Xue, and R. Kong, 2020: Direct variational assimilation of radar reflectivity and radial velocity data: Issues with nonlinear reflectivity operator and solutions. Mon. Wea. Rev. 148.

1. L331: I may be misunderstanding something simple, but if the assimilation process during this period is what is done operationally (CONV + Z + LHN), what is the reason the data was obtained 24-35 hours beforehand and further spun up rather than just using the operational data from the model initial time (e.g, 00 UTC 13 July 2021 instead of 00 UTC 12 July 2021)?
2. L336: I am curious about the assimilation of Z data only within the melting layer, where the relationship between Z and the microphysical state variables becomes most complicated and obfuscated. If anything the data in the melting layer is often neglected because it can result in some large errors.
3. L512: I apologize if I’m misunderstanding, but these two sentences seem contradictory to me. It is stated that the best forecasts are achieved when there is limited influence from the radar-based retrievals (I assume in terms of impact to the analysis by way of a larger lower threshold, rather than spatial localization?). But subsequently it is stated that a smaller observation error standard deviation, which I believe would enhance the weighting toward observations, is also beneficial.
4. L548: This, to me, is an absolutely crucial piece of information for interpreting the results of assimilating LWC/IWC in this study and needs to be discussed earlier in the paper in the assimilation section. While I think the adjustment of the moisture variables (rather than precipitation variables) is in general an important aspect of radar data assimilation, I also feel the impacts of assimilating a “bad” retrieval (say above the ML in what is actually graupel/hail) on moisture may be even more deleterious than if it were acting on qs, which could precipitate out relatively quickly and lead to little harm to the forecast. This seems like an important follow-up study since it is likely not what readers would have expected at first.
5. L565: Could additional studies be done that only assimilate the IWC retrievals in areas identified as snow in convective cases, thus potentially limiting the presumed harmful impacts of assimilating poor/inappropriate retrievals of IWC? With benefits coming from Z alone, I am curious if it could still be wise to assimilate IWC only selectively, in addition to LWC and Z (and V and CNV).

Technical corrections:

1. L31: “adequat” should be “adequate”
2. L330: “and including LHN” should be “including LHN”
3. L538: “adjusted” should be “developed”

see attached file