Review of “Large ensemble assessment of the Arctic stratospheric polar vortex” by A. Kuchar et al.

The authors present a mostly statistical assessment of CMIP model performance in relation to the geometry of the perturbed polar vortex, plus the frequency of split or displacement SSWs. Their main tool of analysis are rank histograms created with models from the Multi-Model Large Ensemble Archive against ERA5 data.

I find the manuscript well written and the structure has a logical flow. I have only minor comments which I hope will help clarify some of the discussion.

Detailed comments:

- Abstract: I would suggest expanding the second last paragraph to clarify what is meant. I did not understand this until after reading the manuscript. Then, I would suggest to remove the last sentence of the Abstract, or at least clarify: It is somewhat misleading as no attempt at improving the representation of SPV dynamics is made, and there is not enough discussion of what could be the cause for the observed biases.

- Introduction/Methods: It is argued that one has to use large ensembles to assess wintertime NH variability - which is fine. However, ERA5 is assumed the truth, even though it only represents one ensemble member (the one we call “reality”). I am not suggesting that’s a problem, but I would welcome a short discussion on why evaluating large ensembles to one single realisation still makes sense.

- It is not clear - neither from Methods nor Code availability - whether the actual code from Seviour et al (2013) has been used to compute the aspect ratio and centroid lat/lon. Please clarify.

- Section 3:

- It would be helpful to provide a sense of how large the biases are. If one is not accustomed to look at rank histograms, it is difficult to judge whether the the shown biases are small or large - that is, whether they are concerning or within acceptable limits. For instance, what does it mean if in Fig. 1c one of the bars peaks around the number 200 but the adjacent bars are below that value? The authors say on line 147 “such strong biases” but I don’t know if I should consider them strong for the work I would like to do with those models.

- I know it’s not the focus of this paper, but some interpretation of results would be great. For instance, models seem to have similar biases at all three levels in centroid latitude, but they have different biases in aspect ratio at different levels. Is there something we can induce or learn from that?