The authors have addressed all the remaining concerns of the reviewers. There correction to the ERA-I has substantially changed the results, however, the change further strengthens the main results that SnowModel performance is superior, and so addresses the comment that, based on the previous assessment, ERA-I did as well as SnowModel in matching the in situ snow depth. Therefore, I judge that the manuscript should be accepted.

There are two points that the change in ERA-I has made more clear to me, and I would recommend that some short additional discussion be added to clarify these.

1. In the prior revision, I made the following comment:

“One thing you did not point out is that snowmodel takes as its input precipitation from Polar-WRF, which will be different from ERA-I. So the comparison between ERA-I and snowmodel and in situ (at east for the CS-2 comparison) mostly just shows that the retrieval is sensitive to errors in snow depth, and not which method is necessarily better (snowmodel would presumably be better where, as you note, snow redistribution matters, but you have not shown that this is a factor here).”

To which you responded:
It is stated in the manuscript (L244-248 second version, L251-255 latest version) that hourly atmospheric forcing were generated by version 3.5 of the polar-optimized version of the Advanced Research Weather Research and Forecasting Model. The revised ERA-I dataset also now show that ERA-I is not suitable for retrieving sea ice thickness in this region at least.

Due to poor wording on my part, this misses my point. The point is that the precipitation fields from WRF-ARW that are used to drive SnowModel are not that same as ERA-I. In fact, I believe Polar WRF initialization and boundary conditions are provided by NCEP GFS (this may have changed) so that it also cannot be viewed as a downscaled model of ERA-I. This means that ERA-I performing poorly does not suggest that SnowModel is responsible for the much better fits. It could be that the WRF-ARW precipitation fields forcing SnowModel are much better and responsible for most of the improvement. The results do suggest that snow redistribution in SnowModel may be important, as the snow depth distribution from SnowModel matches in situ distribution much better than ERA-I (figure 3), but it is not clear if the high-resolution model WRF-ARW uses in this area actually produces similar spatial variation in precipitation that ERA-I cannot, nor whether the match to mean snow depth is due to SnowModel or primarily due to better precipitation in WRF-ARW.

This was not necessarily a goal of the study, and not raised by the reviewers, but I would suggest at least that some comment in the discussion about this point and its significance for the results would be helpful.

2 In Figure 6, the cause of the range of the thicknesses is not clear, and not really discussed much in the manuscript. I can understand the range in ERA-I, since there is a large discrepancy here. But in the manuscript it is stated that SnowModel has a mean snow depth that is only 2 cm more (SWE, so ~5cm in snow depth) than in situ. Now, since equations 1-3 are all linear, I would have thought this offset (dTs ~ 5 cm) would cause an offset in the bounds shown by some modest amount (e.g. dTs*pw/(pw-pi), and dTs*(pw-ps)/(pw-pi) from the in situ bounds, or 30-50 cm). But in Figure 6 these changes are much greater (~ 1m). Also, since the Pd range shown would cover this ~5 cm bias, I do not understand why the range of the dots do not span the in situ depth indicated by the plus sign

Either Figure 6 is showing something different, or I am not understanding how the bounds are being determined (is it showing the range of CS-2 derived thicknesses across the domain, and not the mean? Or possibly the 2 cm SWE bias is not scaled spatially, so the difference is mostly due to the larger bias in region 2?). It would be helpful to the reader for there to be a few sentences more explanation of what contributes to the differences in the results in Figure 6.

Other minor comments:

Line 532-543, page 17. To clarify, ASPECT is from shipboard underway observations. These do not necessarily avoid ridged areas, but the estimated snow depths are probably somewhat biased as they are often estimated based on the snow depth on the level ice portion of the ice, and perhaps less from ridged areas, and the ship likely does avoid heavily ridged, thick ice.

Line 605, page 19 – confidently, not confidentially