|Review of revised manuscript: Mechanisms influencing seasonal-to-interannual prediction skill of sea ice extent in the Arctic Ocean in MIROC” by Ono et al.|
This is a review of the revised manuscript. The authors have taken some effort to address my and the other reviewer’s comments. However, I still have two major concerns with the manuscript in its present form: (1) the choice of Arctic domain and its influence on the results and conclusions of the study; and (2) the proposed advective ocean mechanism for winter prediction skill. These major concerns and some additional minor comments are outlined below.
1) Choice of Arctic domain
In my first review (RC1), I outlined a number of concerns directly related to the author’s choice of using an Arctic domain defined as all gridpoints north of 65N. This choice creates confusion throughout the manuscript, making the results difficult to interpret (see RC1, major comment 1). In my opinion, the authors’ response to RC1 has not provided a compelling justification for this choice. If the authors insist on retaining this domain choice, the results throughout the manuscript need to be carefully caveated, so that readers do not misinterpret the findings. Below are some changes that would need to be included if the choice of domain is retained (note that this list may not be exhaustive).
3.33: A line needs to added here explicitly stating that these results are not directly comparable with other hindcast studies. For example, “It should be noted that the results of this study are not directly comparable with other hindcast studies that focus on pan-Arctic SIE (e.g., Chevallier et al., 2013; Sigmond et al., 2013; Wang et al., 2013; Msadek et al., 2014; Peterson et al., 2015; Guemas et al., 2016; Sigmond et al., 2016), due to this choice of Arctic Ocean domain.”
4.11-12: I recommend explicitly noting the reason for this discrepancy: Winter SIE_AO variability is dominated by changes in the Barents and GIN Seas, which have long persistence timescales relative to other regions of winter ice variability.
4.21-22: This sentence should be removed, as a comparison with Tietsche et al. (2014) is not appropriate given the different domains used in these studies.
4.13-30: To alleviate issues with the choice of Arctic Ocean domain, the authors should consider using a normalized RMSE (NRMSE) metric, where the RMSE values are normalized by the standard deviation of each month. As it currently stands, these RMSE values are difficult to interpret giving the large seasonal cycle in RMSE. Put differently, it is hard to know from Fig. 2b what is a “good” or “bad” RMSE value.
5.10-19: The authors have chosen to remove Fig. S3 of the previous supplementary material. This Figure should be added back into the supplement, as it provides important context for the results of the study. Also, a comment should be added to the text that summer-to-winter differences in SIV-SIE correlations are much less pronounced when using a northern hemisphere domain for SIE.
2) Advective ocean mechanism for winter prediction skill
The authors claim that Figs. 4c-f provide evidence for subsurface OHC anomalies advected from the North Atlantic. While plausible, this mechanism is not convincingly shown by Fig. 4. Aside from a small patch of positive correlation west of Novaya Zemlya in panels 4c and 4d, there is no clear evidence of OHC anomaly advection. Rather, the main signal appears to be a large stationary patch of OHC anomalies throughout the Barents and GIN Seas, which is present at lags of 9, 6, 3, and 0. The origin of these anomalies is not clear from Fig. 4.
The authors claim that OHC anomalies “flow into the BS through advection.” What is the evidence for this? Either more evidence needs to be provided or the statements about regarding the advection mechanism need to be appropriately qualified. For example, the advection hypothesis could be referred to as a “plausible mechanism.”
5.33-6.1: As mentioned in RC1 and RC2, this apparent advection from the North Atlantic is likely an artefact of only defining OHC below the mixed layer. When North Atlantic mixed layers deepen, this quantity becomes undefined, creating the undefined regions seen in Fig. S4. Therefore, this sentence should be removed.
3.26-27: This statement remains unclear. How are the drifts removed? An additive correction? A regression based approach? Is the drift removal lead-time dependent? More detail is needed here.
4.4: The autocorrelation coefficients are the skill of a persistence forecast. I recommend changing this line to: “We first examine the potential predictability of SIE_AO (Fig. 1), based on lagged auto-correlation coefficients, which is the skill of a persistence forecast.”
4.9: Change to “…similar to those…”
4.10: Change to “…due to differences in the observational time period.”
4.25: Change to “…mid-1980s and mid-1990s.”
5.14: Change to “…similar features…”
5.17: I believe Fig. S3 here is referring to the old Fig. S3 (which is no longer in the supplement)
5.21: Change to “…model-predicted December SIE_AO”
6.20: Rather than sampling uncertainty, I would suggest that a more likely explanation is errors in sea ice thickness initial conditions and model drift.
7.1-3: This sentence needs to be removed (the author’s claimed to have removed it in their response, but seem to have forgot).