This paper develops a Newtonian sea ice relaxation approach for models with multi-category sea-ice, and then examines how the approach improves the solution in two regional high latitude model configurations. The authors consider their scheme both at lateral boundaries as well as within the model domain. Both concentration and thickness can be nudged. The authors show that the approach significantly improves sea-ice and upper ocean fields in their experiments.

This is an interesting paper and the topic of sea-ice assimilation is relevant to a number of groups modelling the pan-Arctic region. Thus the topic is appropriate for GMD. The paper is well written and the figures are clear and easy to follow. That said, there are ways I believe the manuscript could be improved and thus recommend major revisions. Details and specifics are provided below.

Big Picture Comments:

The authors relax sea ice concentration and thickness to PIOMAS, which is a reanalysis product. That is understandable given the lack of observational thickness data. Yet, I’d still be interested in seeing the authors consider examining the sensitivity of their results to other products. Maybe another reanalysis like GLORYS? But more importantly, as well as assimilating to PIOMAS, the authors also evaluate their results in comparison with PIOMAS. I would definitely like to see some evaluation using independent products, both reanalysis, but also observational. There are some satellite sea ice thickness fields available, as well as lots for concentration.

The authors state there is an initialization step, where the grid points where nudging will occur are identified? Does that mean these grid points are fixed in time? They don’t change between seasons? Or years, given there can be significant inter-annual variability in the Arctic sea-ice.

If I am following section 4.2.2 correctly, with the multiple categories and figure 2, it seems like most of the ice goes in within one category, with small changes in the thinner ice categories. Is this optimal? Might a smoother distribution work? Or addition of some ice in all categories? I think some more discussion and potentially some sensitivity analysis would help convince readers the author’s approach is optimal. And for the thickness assimilation, does it go in at the maximum thickness in a category, the mean, or just whatever value you take from PIOMAS? Given this is a key aspect of the work, I think more detail and evaluation is justified.

Is the relaxation approach the same in the sea ice pack interior as within the marginal ice zone?

For ARC10K and looking at figure 3c shows the 60N nudging domain going through the middle of the Hudson Bay Complex, across the Labrador Current and into the Baltic, all areas of active sea-ice. Is this latitude the best? Wouldn’t a slightly lower latitude work better?

No multi-year hindcasts with ARC10K? In any case, for the single-year hindcasts, why those specific years? Might the relaxation work differently in high vs low sea-ice years?

In terms of evaluation, would like to see how the relaxation impacts the sea-ice velocity. And for the upper ocean, how the stratification and mixed layer depths respond.

For the discussion, would the authors expect this to work for other models? What about for sea-ice in the southern hemisphere? In a changing/warming climate?

Specific Comments:

Figure 1 caption – For ARC10K, since the caption mentions relaxation north of 60N, it would be good to indicate that latitude with a line on the figure.

Table 1 – Is the background kinematic viscosity really zero?

Section 3.2 title – Model validation metrics or Model evaluation metrics? I’m assuming the model code is valid, so this should more likely be the latter option.

Line 449 – “This supports the assumption that the Bering Sea ice state is weakly sensitive to Arctic LBCs.” Is this surprising given the flow through Bering Strait is generally to the north. What about during a northerly wind event when ice is exported to the south (which has been shown to occur in the literature).

Line 571 – Why might the relaxation being having no effect during the freeze-up?

Figure 13 – Looks like the different experiments have different run length. I.e. one goes to 2025. Unless there is something relevant to be discussed in that extended period, just truncate in in the plots and show all results for the same time frame.