Dear Authors:
Thank you for your patience with the first round of reviews as an unusual number of nominated reviewers declined our request. Thank you also for your thorough revision in which you went beyond what was asked for, and which further improved your manuscript. The Reviewers are content with your responses and I am happy to accept your manuscript for publication in NPG, subject to technical corrections. While looking at your responses, I spotted two locations where minor changes may improve the presentation:

1) In your response to Referee 2, point #4, I find the first sentence of the added text unnecessarily winding and long:

"To summarize, in the geopotential height case where extreme events are defined not just by large anomalies but persistent structures, methods such as k-means, or the more heavily regularized convex coding applied here, that better identify such structures provide bases that are more amenable to interpretation in terms of physical extremes and so may provide a more direct starting point for analyses of such events."

2) In the 2nd sentence at the same place, could you rephrase the little awkward expression "less well suited"?

"Unregularized convex coding and archetypal analysis, on the other hand, are less well suited"

Please submit an updated version of the manuscript at your convenience, and thank you for choosing NPG for publishing your research results. With regards
Zoltan Toth