Thank you very much for taking the time to carefully review the manuscript. We reply to the individual points below regarding the revision.

1. Impact of Model Biases and the Validity of Proxy Comparison

As we described these biases in the text, we believe they are not negligible; thus, we agree that care must be exercised when interpreting and drawing implications from our results. We will provide further discussion on this issue. We note that sea ice concentration, notably in the Barents and Kara Seas, improved with the cloud parameter C, rather than A, as illustrated in the comparison with the HadISST2 data. The simulated present-day sea ice thickness has not yet received sufficient attention in the interpretation of paleoclimate modeling results, perhaps due to the lack of a long-term thickness observation dataset. Our results suggest that it may affect the model-data comparison. Our intention is not to declare that our simulation is successful, but rather to present cases in which the representation of present-day sea ice thickness and the temperature dependency of cloud phase, both controlled by the cloud parameterization here, affect the LIG Arctic simulations and can also be a contributing factor, or at least a factor to be concerned about, to the model spread.

There are a couple of other reasons why we did not claim which model version is more realistic (more appropriate wording may have been “more correct”): 1) While the temperature dependency of cloud phase in the parameter set C is designed to be closer to modern satellite observations in the previous study (Sherriff-Tadano, 2020), there is no guarantee that the same relation holds at the LIG environment, when the aerosol distribution is likely different from that of the present-day. Thus, we intended to emphasize that the aim of the study has a nature of sensitivity experiments; 2) the agreement of surface temperature difference of LIG and PI between the model and proxies is still insufficient (Fig. 8), as commonly seen in other models; and 3) there is some disagreement in the reconstruction of summer sea ice cover between the two datasets we used. We will clarify these points in the revision.

2. Initial Condition Sensitivity

The model was integrated for at least 2,000 years, a sufficiently long period to reach a quasi-equilibrium. Therefore, the results are not sensitive to the choice of imposed initial conditions, as our study does not focus on the transient response. The experiment protocol follows the PMIP4 LIG experiment. In the real world, the LIG does not strictly represent the equilibrium climate, and the complications arising from its transient nature require investigation in separate studies, including the effect of freshwater fluxes. We consider that the initial conditions are not a critical element of the interpretation of our conclusion.

3. Overemphasis on Annual Mean Temperature

The seasonal summary of temperature, as well as surface radiative and heat fluxes in the Arctic region, is presented in the form of bar graphs throughout the paper. Our intention is not to emphasize the annual mean temperature; we will consider adding polarized seasonal maps of Fig. 2 in the supplementary materials in response to the referee’s suggestion.

4. Oversimplified Explanation for Polar Asymmetry

We agree that the complete response of Southern Ocean cooling near Antarctica is subject to feedback through changes in atmospheric circulation, moisture transport, and other factors. However, we argue that these factors are secondary for two reasons: 1) the difference between the two model versions in PI simulations is the cloud parameters, and thus the cloud parameterization is the cause of the difference (though not necessarily the dominant contributor to the complete response); and more importantly 2) the energy balance analysis at the surface clearly shows year-round cloud longwave warming and the dominant cloud shortwave cooling effect during summer near the Antarctica, compared to other terms. We will discuss the suggested potential feedback factors and clarify these points.

5. The attribution of autumn/winter warming

We agree with the referee’s comment. Indeed, our interpretation, as stated in Section 5.4, is: “As discussed above, the LWP and low-level cloud amount increases in LIG simulations are likely associated with decreased sea ice cover from PI simulations. As the reduction in sea ice cover is much larger in ΔLIGvC than in ΔLIGvA, a larger cloud response occurs in ΔLIGvC than in ΔLIGvA.” We state that both a larger reduction in sea ice cover and stronger cloud feedback in the model with cloud parameter set C are constructively responsible for the different response from the model with cloud parameter set A. We will reconsider the title of Sect. 5.4, which may have been misleading, and clarify the writing.

Thank you very much for taking the time to review the manuscript carefully and for your many valuable comments, including suggestions for additional literature.

Cloud parameterisations

We will add the relevant variables for additional model performance checks. Historical simulations are not available for all model versions, as is sometimes the case for the designated paleoclimate simulations. We plan to discuss some of the differences between the historical and preindustrial simulations in the revised manuscript.

Fixed-angle calendar

There is no perfect or universal way to adjust the calendar for comparing past and present, and we share the view that the paleoclimate modeling community should establish several different robust methodologies for specific purposes. Our feedback analysis aims to explain the differences in surface temperature for each month, independently of the other months, using the calendar definition employed in many previous studies. This is useful for comparing with summer proxies, as the referee suggested, and for interpreting the response of the same season. As the referee correctly points out, the time integration over a year under the fixed-angle calendar does not equal the annual average; thus, the interpretation of the analysis for annual mean values is not apparent. The annual mean surface temperature difference can be interpreted using the annual mean of the additional fixed-date calendar feedback analysis. We note that no calendar adjustments are applied before computing the presented annual mean values, and thus, they are not affected by the calendar definition. Sea ice minima are simulated in September, and the comparison with corresponding proxies is not expected to be impacted significantly. The main conclusion remains unchanged when calendars are defined at the data analysis stage. Nevertheless, the point raised by the referee requires attention that the community has overlooked, particularly in the context of ocean heat or ice mass budget analysis to explain their tendencies. We plan to discuss these points with the available data in the revised manuscript.

Observations

Thank you very much for suggesting the specific summer reconstruction dataset. This is very helpful for further comparison.

Marine core data

Thank you very much for clarifying the reliability and reinterpretation of published data. We also appreciate your guidelines for reviewing the current understanding of the LIG Arctic sea ice state. The descriptions provided in the comment regarding the latest interpretation are invaluable.

Thank you very much for taking the time to carefully review the manuscript. We reply to the individual points below regarding the revision.

1.

Although we have not examined the sensitivity of the mid-Holocene (MH) Arctic simulation to the cloud phase representation of the model, we expect that the representation is particularly important for the LIG. The summer insolation anomaly from today is larger at the LIG than the MH because perihelion was close to the summer solstice at the LIG (rather than the autumnal equinox at the MH), and the reconstruction suggests a warmer Arctic at the LIG. As shown in the manuscript, sea ice cover and cloud cover changes are closely related. Thus, we expect that the warmer Arctic makes the cloud phase feedback more critical. We will add this point to the revised manuscript.

2.

We will merge the two tables so that readers can easily grasp an overview of the experiments.

3.

To solely demonstrate the effect of cloud phase representation differences, it may be sufficient to use only the model with dynamic vegetation, as we believe that dynamic vegetation feedback is vital for realistic simulations. We also examined the sensitivity of cloud phase representation using the model without dynamic vegetation because most PMIP models that conducted the LIG simulation in previous studies did not include dynamic vegetation feedback. There was a concern that the difference in simulated LIG climate state (with and without vegetation feedback) may result in differing sensitivity to the cloud parameterization. Thus, we aim to examine whether the effect of cloud representation on the simulated LIG climate is robust and point out the potential of explaining a portion of the spread in the simulated LIG climate among many models, even without the dynamic vegetation feedback. We will clarify this motivation in the experimental design of the revised manuscript.

4.

Thank you for highlighting this important process. This was stated briefly and may be implicit in the text. We will make the statement more explicit in the revised manuscript.

5.

Thank you for pointing this out. We will improve the corresponding captions.

6.

This is an important point, which the first referee also raised. We will add the detailed discussion on this uncertainty. Please see the reply to comment #1 of the first referee.

7.

We will change the phrase as suggested.

Thank you very much for taking the time to carefully review the manuscript. Thank you also for the encouraging and helpful comments. They surely help improve the presentation and readability of the manuscript, and we will do our best to incorporate your suggestions in the revised manuscript.