|The authors have responded to by comments and criticism, and the manuscript has been improved. The new analysis of ocean heat transport convergence is useful, and the new figure 9 summarizing prediction skill is most helpful. However, the new figure 9 also highlights what is still my major concern. Throughout the text the authors highlight the skill differences consistent with their proposed mechanism, but these differences are sometimes marginal and (based on the new figure 9) do not seem to be significant. Considering that this is the key result of the paper, this apparent lack of significant differences should be better discussed. |
To conclude, although the manuscript has been improved, I still believe that it is in need of some substantial revisions before it is ready for publication. Detailed comments are found below.
1. Differences in prediction skill: As mentioned above, it is not clear from the text and figures to what extent there is a significant increase in prediction skill by considering the state of the AMOC. On l.310 it is stated that “significant skill difference (defined by non-overlapping confidence intervals) between strong and weak AMOC phases”, but in my opinion the differences in skill shown in figure 9 do not reflect how the authors present the results and conclude. Comparing with the similar figure 11 in Borchert et al. (2018), it looks like the skill differences in the present study are much smaller (less significant). Hence, I think the presentation and discussion of skill differences could still be improved.
2. Ocean heat transport convergence: The authors assess OHT convergence as the difference between two latitude bands (l.109-110, l.200-201). Why not calculate the actual convergence? From the location of Box 1 and 2 it is hard to believe that zonal transports are negligible. This should at least be checked and mentioned in the text.
3. Surface heat fluxes: The authors find negative correlations between SST and surface heat fluxes, i.e., the ocean forcing the atmosphere (l.243). However, as stated in the manuscript (l.228), it is known from observations that seasonal SST anomalies are strongly linked to atmospheric forcing. Why would the model be different, and would this in any way influence your interpretation of your results?
l.3: It is not clear what “a seesaw-like mechanism” means.
l.24: You should consider explaining the SST anomalies associated with the tripole.
l.37-41: Suggest to move these sentences up to l.34 (before “[Here] We evaluate…”)
l.44: delete “as”
l.45-46: “by sub-selecting ensemble members that meet certain physical criteria, thus filtering atmospheric noise in the ensemble” -> I think this argument needs to better explained.
l.47-48: I don’t see the logic between the first and second part of this sentence (“…which is why it focuses”).
l.133-135: The use of smoothing is still confusing. In their reply, the authors state that “applying the low-pass filter only for plotting time series (e.g. Fig. 1, 4) , but not for any analysis of seasonal means”. This is not easy to understand from the text. Are the grey lines in Fig.1 used/necessary? You could rather show some form of error bars.
l.153: “up to 7 months” – ahead?
l.156-157: I don’t think you need a new paragraph here.
l.160-161: I’m still not convinced that any pronounced displacement of correlation is seen along the northern boundary of the STG. Could you help the reader somehow by e.g., adding the mean barotropic streamfunction to one of the panels?
l.177: “Hence,…” Not sure this sentence is consistent with the previous two, but rather agrees with what is stated on l. 173.
l.186: No need for a new paragraph here
l.203-205: The authors state that in the tropical lobe (Box 1) the correlation is significant (negative), whereas in the subtropical lobe (box 2) the correlation is weakly positive. However, in Fig.5 it is stated the mean correlation for Box 1 is -0.33 and +0.46 for Box 2. So I’m not sure I understand the authors interpretation here. Also, as the relation between AMOC and SST vis OHT convergence/divergence is central in the mechanism of D16, I think these correlations (or lack thereof) should be discussed in more detail (l.209-210 just states “other factors”, could be elaborated).
l.216: “an assessment of an attribution” – unclear
l.217 (and elsewhere): the authors refer to other drivers than AMOC as “non-oceanic”. However, this is not justified and there are also other sources of oceanic variability.
l.238-239: Some repetition, consider rewriting.
l.273-275: Some repetition in the description of skill. Consider rewriting.
l.288-289: How does this statement resonate with Fig.9? Except for MAM in Box 1, it seems like the confidence intervals are overlapping for “strong AMOC” and “all years”.
l.299: “skill differences we find agree to some extent with D16’s physical mechanism” – the authors should better explain what skill differences they refer to and more specifically how this relates to D16’s mechanism.
l.308: Again, how do relate this to what is shown in Fig.9? In Box 1 there is a skill decrease for “strong AMOC” versus “all years” (although overlapping).
l.310: “only during summer a significant skill difference” – Confidence intervals seem to be overlapping for JJA. Ans what about MAM in Box 1?
l.353: Better skill for MAM in box 1?
l.399-400: Again, from Fig.9 skill improvement is only seen in JJA (but not significant?).
l.402-403: This I assume refers to only the small patch also referred to earlier in the text. The area-averaged skill is still only 0.4 (Fig. 9).
l.407-408: I know I’m repeating myself, but the skill improvement the authors refer to is marginal for DJF and not significant for JJA.
Figure 4: It’s very hard to see the significance lines.
Figure 9: As in Borchert et al (2018), you should also add the persistence skill here as a comparison.