Dear authors,

Based on your extensive response to the reviewers, I would like to invite you to submit a revised version. Your response has helped to clarify the purpose and potential impact of the paper. I would certainly encourage extra-care by making sure that a model is not simply said to be "better" because it is closer to the ensemble mean. You seem to be aware of the pitfall, but it has to be clear in the text.

Sincerely,

Michel Crucifix.

Dear Dr. Schwarber,

First I would like to thank you for your patience, and the carefully drafted response to the reviewers comments. Unfortunately, I am not coming with good news. One reviewer accepts the clarification that you are testing SMC, and hence retracts his/her original criticism, but still is not fully convinced in all respects, without, alas, giving more information. The author reviewer fully appreciates that you are testing SMC, but still formulates a major objection: the long-term evolution of the temperature average of a multi-model ensemble an unreasonable target for validating an SMC because we have no reason to consider that it is close to the truth than individual members. I understand that they oppose your working hypothesis that "[...] posit that the responses closer to the multi-model mean are likely to more accurately represent that particular response pattern", and this may indeed cause concern about the way your conclusions could be used or interpreted, namely those expressed in lines 285-290. I share this concern and, accordingly, regret not to be able to proceed with acceptance of your manuscript.

Most sincerely,

Michel Crucifix.

Dear Dr. Schwarber,

Further to your appeal I am pleased to confirm that your article is now considered to be publishable with the provision that you address the revisions listed in the correspondence thread below. Most sincerely,

Michel Crucifix:

Dear Dr Schwarber,
(cc: Chief Editors, Steve Smith, Michel Crucifix, Anna Wenzel/Copernicus)

We have now discussed your appeal among the chief editors and with the editor, Michel Crucifix, and I apologize for the time it took (due to summer travel and vactions).

Based on this interaction, we will revert the decision to “minor revision”. Please find the correspondence from Michel Crucifix pasted in below about one aspect that needs to be addressed in the minor revision. Also, Valerio Lucarini provided an informal review (pasted
below) that should be helpful as well.

I will contact Copernicus next to change the status of the manuscript in the system.

Thank you for your patience and I am looking forward to receiving the revised manuscript.

Best regards,
Axel Kleidon


==== Re-evaluation by Michel Crucifix ===

Dear Axel,

I have read again the whole correspondence associated with the manuscript of Adria K. Schwarber et al. , and I am happy to revise my decision. The manuscript status should be upgraded from 'rejection' to 'minor revision'.

The reviewer #1 develops a valid point to criticise an element in the response of the author --- namely, the ensemble-mean of low-dimensional quantities (like the global mean temperature) cannot be expected to be closer to reality than that of individual models. Reviewer
#1 appears to have based his advice of rejection mainly on that argument. Fortunately, this argument is not critical to the overall validity and relevance of the article.

It is therefore justified to revise the recommendation from 'rejection' to 'minor revision'. Nevertheless, I would request the authors to address this comment, and revise the text accordingly (mainly the section called "Conclusions", which in fact contains discussion
material). I share the concern of the reviewer that the Flato argument may be incorrectly applied and interpreted, especially by scholars not so familiar with the physics and statistics of the climate system. Reviewer #2 seems to still have some concerns but is not
explicit about them, except for a note about the AR5-IR model that should be addressed as well.

I trust that this outcome will satisfy the authors. This said, this issue could have been resolved more quickly with a simple and open dialog. In five years serving as editor of Earth System Dynamics, one would fail to find any dismissive or hostile sentence in my
communication with authors, and this particular manuscript has been no exception. A formal appeal to the editor-in-chief in this case has appeared to me as an unnecessary formality.

All the very best,

Michel.

==== Informal review from Valerio Lucarini ====

I see no huge problems in this paper. Indeed, they do a reasonable job. With some (important) caveats.

What is entirely missing is the he theoretical framework for what they do. Indeed, the recent work of M. Ghil and myself goes in that direction and provides a pretty solid background for what they do. Testing response to forcings makes perfect sense if one wants to compare
models; response can be (approximately) linear even if models are fully nonlinear; this is the essence of Ruelle’s response theory in statistical mechanics (and variants thereof).

Traditionally this has been done quite empirically, but the use of a sum of exponential can be similarly justified on very solid grounds, taking into consideration the transfer operator method (also from statistical mechanics).

Obviously, they need to provide full details on they model setups because results have to be reproducible by other people.

Finally, I believe that the interpretation of the results should be expanded.

If they provide a solid theoretical background and give full details on their methodology, I think they get a good paper out of it.

Thank you for the thorough response to reviews, and impressive supplementary information material. I am pleased to recommend publication of this manuscript.