The authors have clarified some of their reasoning in the paper and provided more arguments for the changes in ice retreat and advance timing. However, there is no new analysis in the study. This is disappointing. I think that there are many interesting elements to the work presented here but feel that the authors have not dug in deeply enough to understand these elements. Because of this, I believe that the important new material in the study is modest - basically that solar heating occurring over the summer is slow to be released during the fall freeze-up and delays the ice advance. This is based on simulations with a simple single column model and only limited results on this mechanism (just the SST annual cycle) are shown from the coupled simulations. There are a number of questions that remain regarding the changes in ice retreat and advance timing as outlined in the major comments below. I would encourage the authors to address some aspects of these questions and thereby provide more insights on why the trend in ice advance timing exceeds the ice retreat trends in the 21st century (as raised in my original review). I recommend another round of major revisions and would encourage some new analysis to address these comments.
1. The discrepancies between the long term and short term variations in ice retreat and advance timing are interesting but not at all explored in the study. The authors offer a number of possibilities for why there might be this difference, with references to previous work, but nothing is conclusive or backed up by analysis. More analysis on what drives the ice retreat timing, ice advance timing and how this varies across timescales would be useful.
2. The authors provide results from the 0-layer thermodynamic model but with very little comparison to the coupled climate simulations. The 0-layer model excludes many factors (not just dynamics). For example, in the coupled model solar energy can penetrate through leads and ice so will not have such an abrupt transition as in the 0-layer model and the surface fluxes will affect the atmospheric state and feed back onto the surface fluxes, among others. I’d suggest that more analysis is done with the coupled model to assess whether the 0-layer model results are indeed the dominate process occurring in the coupled runs, how things change with time scale, etc.. For example, on Figure 6b, why are the solar and net heat fluxes for IPSL not shown in addition to the SST? Do these look similar to the 0-layer model in Fig 6a? Do they change on interannual versus longer timescales (see 1 above)? Making more comparisons between the 0-layer results and the coupled model would help to strengthen the paper. It could also help to answer why the simple mechanism at work in the 0-layer model does not act on short timescales in the coupled modeling systems (again, see 1 above).
3. The 0-layer model results provide a reason why the ice advance timing is delayed but do not clearly indicate why (at least to me) it would be delayed more relative to retreat timing in the 21st century. Some physical arguments for why this occurs are suggested on lines 309-321. However, there is no analysis to back them up. It would be useful to assess some of these factors in the coupled model simulations to determine their influence, relative magnitude, etc. A clearer reason with corroborating material as to why Rlong increases to greater than one (which is one of the main conclusions of the manuscript) would be useful. I think that this could be done without drastically inflating the manuscript, especially if it is just to show support for the mechanisms that are already speculated in the manuscript.
Line 54: Wording/typo – need to include “be” in the sentence
Line 72-73: Wording needs to be revised (“in accord the observed in situ increase …”)
Line 109-112: It might also be good to say what the caveats are from a forced model. For example, the surface state doesn’t feedback to the atmosphere which can affect the processes that drive changes in ice advance and retreat timing.
Section 3.1. You should mention here that there is an inherent discrepancy in comparing observations, which are influenced by internal variability, to a multi-model mean.
Section 3.1. Last sentence: It would be useful to mention that the Barents Sea is subject to internally-generated decadal scale variations driven by ocean heat transport anomalies (for example see Yeager et al., 2015) which could explain some of the discrepancies between observations and models in that region.
Lines 244-245. It should be mentioned that internal variability may also play some role here in the discrepancy between the climate model and satellite data.
Lines 361-362. “This points to dynamical processes as most likely drivers …” I believe that you are making this argument because of the results from the 0-layer model. However, as mentioned above, the 0-layer model is missing many things in addition to dynamics. For example, the coupling to the atmosphere, which will change atmospheric conditions and fluxes could play an importan