Reply on RC1

This manuscript examines the representation of polynyas in CMIP6 models as compared to observations. Some of these comparisons are not straightforward, due to lack of CMIP model variables, limited observations, and the different metrics that could be used to define polynyas, but the authors are transparent in these limitations and convey the information clearly. Modelled coastal polynyas are often too large, likely as a result of coarse horizontal resolution. Modelled open water polynyas are often too small compared to observations, and there is a large inter-model spread in the frequency of open water polynyas. The authors examine vertical ocean profiles in polynyas versus sea ice covered regions in a subset of the models and in float data. The Discussion contains a number of useful insights on the reasons behind the intermodel variation in polynya activity, relating to resolution, simulation of the ACC and overflow parametrizations.


Main comments
Section 5.2 -I found the arguments here a bit hard to follow. I would like to see additional subplots for the other relationships discussed here added to Figure A6. As you mention the results from this section in the abstract and conclusions, the figure should also be brought into the main paper. It seems like the results here would be interesting to a wide audience, so I think it is worth spending some more time on the presentation. → Thank you for this suggestion. We added the discussed relationships to the Figure A6 and brought it into the main paper. We also added some additional explanations to Section 5.2. Section 4.3 or Section 2.3 -Please give some more details on the domain of the SOCCOM float -e.g. time period, number of profiles etc. Please also describe how you extracted the profiles from the CMIP6 models -is this one profile per grid cell in the Weddell Sea region? Is there some time averaging?
→ We agree that this information was lacking and have now added a more detailed description of our method: "We concentrate on the models that form most OWPs (see Table 2) and show only the top three models (MPI-ESM1.2-HR, ACCESS-ESM1.5, BCC-ESM1) in Fig. 10. For comparison, we present the observed hydrographic data of a SOCCOM profiling float (Johnson et al., 2018), which was deployed in January 2015 and surfaced two times in the Maud Rise Polynya in winter 2017 (Campbell et al., 2019). To provide regionally and seasonally comparable data sets for the models and the profiling float, we chose to extract vertical profiles during the month of September from within a rectangle around the profiling float trajectory (see Campbell et al., 2019) with the edge coordinates 61°S-66°S, 0°E-6°E in the Weddell Sea. This region includes the northern flank of Maud Rise, where we found OWPs to be most common (Fig. 3, A3, A4).
For the SOCCOM profiling float, we use the information provided in Campbell et al. (2019) to differentiate vertical profiles when the float surfaces within an open water polynya from those sampled under the sea ice. For the models, we use our algorithm to differentiate and group the grid points by whether they are within an OWP or not. Based on this criteria, we extract and group the vertical salinity and conservative temperature profiles (monthly) and plot them in either blue (under sea ice) or red color (OWP) in Figure 10." L329: 'To evaluate the effect of OWPs on vertical stratification' -'To evaluate vertical stratification in OWPs' (also L371) -as there isn't a clear cause and effect relationship here.

→ Changed as suggested
Conclusions -I would like to see more of the polynya statistics summarised here. This could work well as a bulleted list. → Our conclusions now start with a bullet list summarizing important polynya statistics and findings: "In this paper, we evaluated the representation of Southern Ocean open water and coastal polynyas in CMIP6 climate models and their effects on the modelled Weddell Sea. We found that: All 27 analysed models have coastal polynyas around the Antarctic continent, while OWPs are present in only half of the models CMIP6 models show OWPs most commonly in either the Weddell or the Ross Seas The position of polynya formation is very similar for models of the same family and likely determined by the model properties In comparison to observations, nine models underestimate polynya areas based on thickness threshold but overestimate them if based on concentration threshold method Coastal polynyas in CMIP6 have a large annual variability of at least a factor of 2.5 With total polynya areas from 6.5 x 10^3 km2 up to 215 x 10^3 km2, CMIP6 models show a large intermodel spread"

Minor comments
L8 'presence or absence of OWPs are' > 'presence or absence of OWPs is' → changed as suggested L12 'that require to be addressed' > 'that should/must be addressed' → changed to "should be addressed" "With the aim of detecting polynyas, we start with the sea ice concentration or thickness (Fig. 1a). To mask out the open ocean beyond the northern sea ice extent, we use a "flood-fill" algorithm from the scikit-image library (Van der Walt et al., 2014). Starting from a grid cell with no sea ice, the seed, the algorithm detects similar cells below a specified sea ice concentration/thickness and masks them out, effectively "filling them" with ice (Fig. 1b). Afterwards, a maximum sea ice threshold filter returns all grid cells that are classified as polynyas". We discussed making the ocean dark blue in all the figures of the paper. While it would improve the clarity of the Figures, there is unfortunately no (easy) consistent way to mark the ice shelves in the CMIP6 models in another color than the ocean, as these areas are not differentiated in the sea ice output. In practice that means that we cannot easily color the ocean dark blue without coloring the ice shelves and (or) continent dark blue for some of the models. We estimate making the suggested color change work for all models would take some days of work on a relatively small visual only benefit, so we would prefer to leave it as it is. Fig. 4 -shouldn't this be 'equivalent ice thickness' not 'floe thickness'? → In Figure 4 we show that the floe thickness (CMIP6 variable name: sithick) cannot be used with our algorithm and therefore, we continue our analysis with the sea ice concentration and equivalent ice thickness variables (siconc and sivol). We find this naming scheme somewhat confusing, but want to stay as close as possible to the terminology used in the CMIP6 guidelines and documentation. L163 -Why doesn't the mean of daily data go from 1st May to the end of Nov? → During the ice melting phase in late November, observed and modeled polynyas often become an order of magnitude larger than during the winter (Figure 9). We did not want these polynyas to dominate our results. Moreover, when we provide a comparison of the maximum yearly polynya area (computed according to Eq. 2) in Figure 6

Additional changes
Changes that are not listed in this document can be found in our response to Carolina Dufour. Moreover, we received a helpful suggestion aside from the public review process, which we also want to address here. A reader of our preprint pointed out that we counted the total number of CMIP6 models in Table 2 of the cited Beadling et al. 2020 paper incorrectly, and that it would be preferable to stress which results were significant. Thank you for the corrections, we improved this part:

L436-438
Before: "However, Beadling et al. (2020) found that 30 of 35 CMIP6 models underestimated the ACC, 34 of 38 showed their wind stress curl minimum not sufficiently south and 33 of 38 underestimated the wind stress curl maximum. All these parameters are positively correlated with OWP activity (Campbell et al., 2019)" Improved: "However, Beadling et al. (2020) found that out of the 34 CMIP6 models they analysed, 29 underestimated the ACC (of which 12 significantly), 30 show their wind stress curl not sufficiently south (5 significantly) and 30 underestimated the WSC minimum (9 significantly)"