Summary of the main contribution of the paper:

Bertossa et al. study the evolution of Arctic radiative states (Transmissive Vs Opaque) using a high-resolution climate model designed for polar regions (Polar-WRF). They combine observations from satellites (CALIPSO-CloudSat) and ground station (ARM-NSA) to evaluate the representation of Downwelling Longwave Radiation (DLR) from climate models and reanalyses. They found that the bimodality of the DLR distribution is misrepresented for almost all seasons (which is critical as we advance in simulating future Arctic trajectories). Using the PWRF (with some specific microphysical schemes) allows us to get a DLR distribution close to the observed one, with two distinct radiative states: Transmissive and Opaque, although some biases remain (overproducing opaque clouds or too frequent intermediate conditions). A sensitivity study showed that the biases come mainly from over sea ice and need further improvements. However, PWRF does a good job in allowing us to see the broader picture of DLR and CRE-LW evolution as the climate warms. Specifically, the authors show that going from past-present to future climate, the DLR will shift toward larger values with a higher occurrence. Decomposing into contributions, the DLR-clear will shift toward larger values, and the CRE-LW will increase the occurrence of large values (higher than 70 W/m2), mostly due to the low cloud cover increase.

General comments:

The study is well presented and nicely constructed. The findings are interesting. I suggest some minor/specific comments.

Specific comments:

• L-63 : not in DJF, why?
• Figure 2: How is the CRE obtained from the ground station? Is it the same definition as the satellite? (F_dwn – F_up)_all - (F_dwn – F_up)_clr ? The maximum of occurrence ARM-NSA Vs CSC does not occur at the same value of CRE-Lw (75W/m2 Vs ~65W/m2). It looks like the satellite retrieval occurs at higher values for all seasons and agrees the most in SON. Can the author explain why?
• L-175. Why is P3 overproducing opaque clouds while the Morrison one the intermediate mode? I mean, what is the microphysical configuration that causes these discrepancies? Figure 6 shows that almost all the DLR clr have the same distribution across the different datasets/configurations and the differences are in the CRE-LW.
• L-190: interesting because climate models usually underestimate opaque clouds. Wonder what in P3 scheme creates the overproduction of opaque clouds.
• L-193. CALIPSO can get fully attenuated in opaque clouds, and CloudSat suffers from the strong echo from the surface that can alter low cloud detection over rough surfaces, but shouldn’t be an issue over the Arctic. Is the issue the coarse vertical resolution of the satellites?
• Figure 7: Did the author do some sensitivity studies on the vertical resolution of the radiative computation? Maybe it can explain some of the discrepancies. Or authors can add uncertainty bars for the different datasets and check whether they overlap.
• Figure 8: The mean cloud fraction profile in the Morrison scheme (over ice) shows a bimodal distribution with a large population of low clouds (~500m) and mid-level clouds (5-6 km). I don’t understand why we don’t observe a bimodality in the CRE-LW distribution that follows the cloud fraction one. Could the authors clarify this point?

Figure comments:

• Using green and red in the same plot is not color blind friendly. I suggested changing the line style when these two colors are used.
• In fig. 6, the legend says ‘Obs’ for the black line. I suggest kipping CSC for satellite retrieval and ARM-NSA for the ground station (or similar) throughout the paper. Same for the simulations, keeping the same notation can be helpful.

This paper investigates Arctic radiative states—specifically the transmissive and opaque regimes—using case studies, climatological analyses, and future projections based on regional climate modeling and observations. Several advanced microphysical schemes are evaluated. The results demonstrate that PWRF is a valuable tool for simulating Arctic radiative conditions; however, it still exhibits a tendency to overproduce the opaque state (P3 & Morrison), likely linked to excessive low-level, optically thick clouds over sea ice. Future projections driven by CESM1 also suggest an increased frequency of opaque conditions.

I agree with the authors’ conclusion that PWRF simulations are highly sensitive to the choice of microphysics scheme (e.g., P3, Morrison, and likely Thompson). As the authors note, individual schemes can outperform others under certain climate conditions—for instance, P3 may better represent the opaque state, despite its known tendency to overestimate cloud cover. That said, I would encourage the authors to expand the discussion on why P3 and Morrison behave differently from a microphysical-process standpoint (model configuration), particularly in their treatment of opaque clouds. Such clarification would provide deeper insight into the scheme-specific sensitivities highlighted in the results.

Overall, the paper is well written. I personally would suggest merging some of the shorter paragraphs to improve flow and emphasize the main points more directly. Most of my remaining comments concern the PWRF configuration and interpretation. Thus, I would recommend a major revision.

Some PWRF setting questions:

1. 52-vertical level is used in this paper, is this number enough, what is the lowest model level? Like ASR v2 is 71 levels, with lowest level at 4m. I don’t think re-run any model is necessary, but I would suggest authors provide more info in the table 1. PWRF default (at PMG) is 71 levels, with more levels at lower altitude. That is designed to better capture near surface conditions, and cloud formation.
2. I might miss it. Is unified Noah or Noah MP is being used here?
3. Is there any nudging applied? If so, what variables for what levels?
4. This is not a suggestion, but more like a discussion. P3 (option 50) in general provides pretty good results, especially related to super-cooled liquid water in clouds. However, we also notice sometimes it can produce large value of cloud LWP (e.g., like unrealistic large for a short period of time at WAIS compared to obs. While Thompson & Morrison did a better job). I believe Hines et al. 2019 also kinda mentioned that. I am wondering if authors encounter similar issue in Arctic.
5. I am not familiar with ERA5 default surface type for Arctic regions (aka whether it is good enough). For Antarctic region, ERA5 use a quite old land surface cover (which still have Larsen A & B ice shelf included…). Thus, a more up-to-date land-ocean description is usually needed (Like REMA). For high-resolution simulations, more accurate SST and SI observations are usually recommended to be included as initial fields. As authors mentioned in Ln 220, difference surface type matters. I am wondering if more information can be provided here regarding the simulations has been done in this study.
6. There is no need to add any simulations. I am wondering authors have ever tested Thompson (aerosol aware) by any chance?

Minor:

Ln 15-20: Several research suggested a poleward shift of atmospheric river activities for Arctic region, which will enhance the transport of moisture, energy and warmth. I think this is worth mentioning as background change for Arctic region.

Zhe Li, Qinghua Ding ,A global poleward shift of atmospheric rivers.Sci. Adv.10,eadq0604(2024).DOI:10.1126/sciadv.adq0604

Introduction structure:

The Intro is in a good shape in general. This is just a suggestion. Introduction section has a decent number of short paragraphs, I am wondering authors ever think about merging some of those, to deliver the key messages more clearly.