To the Authors,

Thank you for your careful consideration of Dr. Shultz's comments in particular. Upon examination of the revised manuscript, along with your letter and Dr. Schultz's initial comments, I am largely satisfied that you have mostly separated the topic of clustering as a useful tool for assessing model differences, from using clustering as a way to find MMM subsets that better match observations. However, I suggest a final objective alteration of the text to make this still clearer to the reader, so that clustering is not used naively by readers for downselecting models for (for example) choosing future model projections to use in ensemble analyses. As one arbitrary and non-exhaustive example, you could highlight in the Abstract the SH example of the primary cluster 'unselecting' ensemble members that are in fact closer to observations than the most populous cluster as a case where clustering in fact selects poorer models. This would provide an important caveat.

In net, it would be great if you could perform a final sweep to more carefully elucidate clustering as a useful ensemble analysis tool *in general*, without *particularly* emphasizing choice of a particular cluster as a 'best' MMM over other clustering applications (though this is still a finding worth presenting). Then I believe this manuscript would stand as a novel demonstration of cluster-based ensemble analysis for the general purpose of understanding inter-model similarities/differences. I believe this would address Dr. Schultz's concerns.

Please don't hesitate to provide a rebuttal to my suggestion for final revisions, as you see fit.

Jeremy