Review of
Defining Coastal Antarctic polynyas in satellite observations and climate model output to support ecological climate change research
by
Landrum, L., et al.
Summary: This is a complex paper with the aim to cover quite some ground in our understanding and our capabilities to detect and map Antarctic polynyas - both in observations and in results from numerical modeling - with the overarching goal to arrive at an improved quantitative knowledge about how marine net primary production (NPP) is connected to the circum-Antarctic polynya area now and in projections of the future climate conditions. Besides the many objectives of this contribution I distilled that the authors want to come up with recommendations about how to reliably detect polynyas as a sub-grid scale phenomenon in coarse resolution climate models. The authors chose a simple sea-ice concentration threshold based approach to delineate polynyas both in observational and modeling data. They investigate the influence of different grid resolution and of different temporal sampling and come up with a rich set of results from correlation and trend analyses, partly only based on observational data, partly mixing both worlds.
General Comments:
GC0: I have one (first) general comment on this manuscript. I am hesitant to think that the way the study has conceptualized is delivering the results the authors may have wanted to achieve. I understand that the authors wanted to use satellite observations to test how well ESM models delineate polynyas and how accurate the (integrated) polynya area is compared to observations. One focus of the investigation was on the metrics of how to delineate polynyas in the model grid cells. With the then selected and/or optimized metric you want to get reliable polynya area estimates from the ESM data and transfer that information to estimates of the NPP - now and in the future. The conceptual drawback I see is that the paper does not convince with the generation of a well-analysed, well-evaluated polynya location, occurrence and area data set based on observations. The threshold based method used is a poor-man's approach to identify polynyas. While I understand that for ESM analysis it possibly has to be that approach, this is not the state-of-the-art using satellite observations. Hence, the estimates that are based on observations and are presented here are not fit-for-purpose. The "processing chain" could have been (if you need to stick to the thresholds approach): Select a number of key polynyas around Antarctica. Get high-resolution estimates of their area for a reasonable time period. Apply different SIC thresholds to ONE SIC product (You could use OSI SAF because it is a state-of-the-art algorithm / product with physically based uncertainty estimates and the possibility to go down to SIC = 0% if need be; but it might also be a good solution to seek for SIC products at considerably finer grid resolution, e.g. ASI-AMSR2, offering 3.125 km grid resolution.) and test with which threshold you achieve the best result in comparison to the independent data. Generate data sets (maps + time series) of the polynya location, occurrence and area based on SIC observations and the selected threshold. Get an estimate of the uncertainty by playing around with the SIC threshold (+/-10% and +/- 20% for instance). Upscale the maps obtained to the ESM grid resolution and use these as the "truth". Use daily ESM SIC data and apply different SIC (and SIT) thresholds to develop an approach of how to best identify polynyas in the ESMs. Again develop a feeling how sensitive the obtained polynya area is to changes in the applied SIC threshold. Once you have that you could - if need be - generate monthly data sets of polynya location / area based on the obervations and based on the ESM data. But, if I am honest, if the model allows daily output I would also use it - last but not least because of your overarching goal to look into NPP dynamics which might be dominated a lot by shortlived (a few days) blooming events which you would lose in the monthly context.
GC1: What are the different methods to delineate polynyas in satellite remote sensing data? What do they have in common? Where are these different from each other? How have there results been evaluated? What are the different SIC thresholds that have been used during the past decades to define polynyas? What are the conclusions of all this for your work? What is your definition of a polynya (used in this manuscript) and how (if so) does it differ from the definition given in the WMO sea ice nomenclature? To give an answer to these question seem a pre-requisite to me to use the output of polynya "retrieval" algorithms applied to satellite remote sensing data to kind of guide which method to use to define polynyas in ESM data.
GC2: Please motivate better your choice of the SIC products used - especially with regard to spatial resolution and with regard to the need to operate with a long time series. Please also take into account that the two SIC products you used are provided on different grids. I did not find any notion about the NSIDC polarstereographic grid of the NOAA/NSIDC SIC CDR and how your treated that in comparison to the EASE grid. Also into this topic falls my impression that your manuscript would benefit from a more careful description of the skills and limitations of passive microwave SIC retrieval algorithms to derive SIC values for low sea ice concentration and thin sea ice. See my specific comments in this regard.
GC3: One aspect which I thought would be high up on the agenda of this contribution but which I could not find treated in a convincing manner is the optimization of how to delinear polynyas in global climate models. I may repeat a bit of what I stated in GC0 here ... but what I am missing is the description of why the authors ended up at the SIC and SIT values chosen; while I acknowledge that there is an attempt to fit thresholds used in the modeling world to those used in the observations I have problems to find clearly stated information about the sensitivity of the approach, how accurate polynya area derived the method chosen is, and how and with which independent observations you carried out your evaluation. I am aware that a number of your figures point into that direction but I have problems to distill quantitative information out of these and I would not be able to repeat your approach to check its validity or to ask one of my students to repeat the study using a different observational SIC product and a different GCM.
Specific comments:
L13: "no significant impact ..." --> This is an interesting result, suggesting that at daily scale, variations in the overall area of coastal polynyas cancel out? I am not so sure I believe this is true.
L31-33: "Open water ... more sea ice." --> Three comments here. You could already now mention on which type of polynya you are focussing in your contribution. You could mention whether some of the polynyas are perhaps of mixed type where both upwelling and katabatic winds play a role. You could come up with 1-2 more explaining sentences about persistence of coastal polynyas and their role in deep water modification (in more detail and more concrete then you have done so far), and also about the fact that especially in the Eastern Antarctic there is a very close interplay between land-fast sea ice and icebergs on the one hand and the generation of coastal polynyas on the other hand. There exist quite some publications into this direction which I invite you to find and cite.
L45-47: "Defining polynyas ... functioning." --> I agree on the polynya functioning part here but with respect to the "general" definition, I invite you to take a look at the WMO sea ice nomenclature where you can find a definition of a polynya. I also recommend to cite this nomenclature here. I note however now, after having read you paper completely, that I am not sure that you developed your own definition of what a polynya is.
L65-71: "Furthermore ... 2015)" --> These sentences about sea ice concentration retrieval algorithms and products based on satellite observations are not sufficiently well put into context with polynyas. There are 2 - 3 methods to define polynyas / infer the polynya area from satellite remote sensing data and I would have thought it is more straightforward to first introduce those before eventually moving on to sea ice concentration products.
L71-82: Also the remainder of this paragraph is still detached from the main topic of this contribution: polynyas. In this part there are several statements with respect to satellite remote sensing (products) for sea ice concentration and sea ice thickness that do not necessarily hold the way written. I don't "buy" the general statement that algorithms using passive microwave data generally underestimate sea ice concentrations at low concentration levels. What is correct is that the uncertainty of the retrieval is considerably larger due to the reasons laid out by you already and thanks to an often elevated fraction of thin ice and/or quite diverse ice types which both complicate the retrieval.
"to very thin thicknesses" needs to be quantified. I am also not sure whether this general statement holds because I would think ESMs use sea ice thickness categories and might also provide a grid cell mean sea ice thickness rather than the thickness of the ice that is actually present with that grid cell.
Your references for SIT retrieval focus on the Arctic. It might be good to learn something specific for the Antarctic in this regard.
The statement that SIT from satellite products has limited capacity for identifying thin sea ice is not correct. First of all there is the SMOS sensor which has been used successfully in the Arctic but recently also in the Antarctic to derive SIT of thin sea ice. Secondly, there are at least 2 approaches to use either thermal infrared or passive microwave satellite remote sensing to compute the sea ice thickness - especially in polynyas.
Satellite products of the SIC are typically daily or twice daily, that is true, but again, when it comes to polynyas there are a few studies that use swath data, so data from single overpasses, to delineate polynyas and hence get an improved and more accurate estimate of the actual polynya area change during the day. Polynyas are quite dynamic and can open and close during substantially shorter time spans than a day.
L84: The information given in this paragraph needs to come before you go into details about sea ice concentration and sea ice thickness products so the reader understands that "Ahaaaa ... people have used sea ice concentration data or thickness data to detect polynyas and to compute the polynya area." I invite you to search for contributions mentioning the Polynya Signature Simulation Method (PSSM), for contributions where polynyas are identified using SAR (e.g. COSMO-Skymed), and for contributions by Nihashi and Oshima // Fraser et al. // Nakata et al. to learn more possibilities to infer polynya area and polynya dynamics from satellite observations.
L109: My immediate take on these five "primary" goals which are followed by a "we also seek to ..." is that this is far too much to be dealt with within one contribution. The sole intercomparison of the various existing polynya identification and area retrieval methods based on satellite observations is a topic which would deserve ONE paper on its own.
I cannot speak about the ESM world but I would assume that the diversity of approaches to define a polynya in ESMs is much smaller and is mainly a function of model resolution and time step as well as whether and how polynyas are parameterized. The spatial and temporal resolution of the wind forcing - driven by the resolution of the topography - plays a role, the representation of land-fast sea ice and, of course, how realistically the ESMs represent the oceanic heat flux and/or ice shelf cavity processes that modify water masses near ice shelf fronts. Putting all this on the table I am not convinced that this contribution can come up with a credible "optimal polynya metrics".
The 2nd primary goal also reads like it would deserve an own paper.
The 3rd primary goal reads like an analysis of ESM data ... hence it is not entirely clear why you need all the exercises related to the satellite observations and the metrics (i.e. goals 1 and 2) to come up with this part of your contribution.
Finally, the 4th and 5th goal seem to be linked more closely to goal 2 and seem not to be related to either observations or biogeochemistry.
So, in short, looking at the goals I see 3 papers here. One targets goal 1 (but possibly without the model component); one targets goal 3; and the final one targets goals 2, 4 and 5 and the part how do I define polynyas in ESMs.
L110: Given the multiple possibilities that exist to delineate polynyas in satellite remote sensing data that - mostly if not all - work without "time series" I am not sure I understand the incentive to look into time series data. This needs to be motivated better.
L135 / section 2.1: Okay, so you only focus on using a sea ice concentration threshold as the metrics to "define" what a polynya is. This does not result sufficiently well from your introduction / motivation.
I was wondering whether there are more recent studies dealing with an intercomparison between the NOAA/NSIDC sea ice concentration CDR and the one from OSI SAF. I am sure there are and I am sure in those you might find useful information about the limitations of the two products you used.
I invite you to clearly spell out which product version you are using. I assume OSI SAF is OSI-450a / OSI-430a, i.e. the CDR and the iCDR, but I am not sure whether you used v3 or v4 of the NOAA/NSIDC CDR.
I invite you to also clearly motivate why you are using SIC products that come at a comparably coarse grid resolution. As far as I have understood you (from your introduction), the main investigation about long-term trends will be based on model data and you will only use the observations to define an improved metrics to define polynyas across ESMs. Therefore, I was wondering whether it would not be considerably better to look into sea ice concentration products that come at a finer grid resolution, i.e. at 12.5 km grid resolution based on AMSR-E / AMSR2 or at 3.125 km grid resolution based on AMSR2. This is not made sufficiently clear.
L152/153: The OSI SAF SIC product is indeed provided on an EASE 2.0 grid. The NOAA/NSIDC CDR is not - neither v3 nor v4 nor v5; these are all on a polarstereographic grid so that whenever you compute an area in squarekilometers you need to take into account the actual size of the grid cells which changes with latitude.
L183-184: "compare well" --> I was wondering whether you would like to specify this a bit better. I was also wondering whether correct representation of the SIE is really such a good measure when it comes to assessing the skills of a model which sea ice concentration is subsequently used together with a threshold approach to delineate polynyas. What are the differences in SIC for the different SIC categories - especially around the threshold that I assume will be used and introduced later in this paper?
L243/244: "Passive microwave ... 10%" --> This statement is not true as it stands. What is true is that at lower sea ice concentrations the uncertainty budget increases but nevertheless the emissivity difference between open water and sea ice still allows adequate SIC retrieval. There have been publications about this by Soren Andersen et al. in 2006 and/or 2007, I guess. So, this cutoff that is applied by NOAA/NSIDC is a simple approach to discard those SIC values that might have a larger error budget. And one can argue whether one should use 10%, 15%, 30% or whatsoever ... Also, most algorithms apply special filters to filter out grid cells affected by weather - based on the gradient ratio of certain brightness temperatures; these filters cause low sea ice concentration values to be filtered out as well.
If you want to better understand this issue then I recommend you to read the Lavergne et al. (2019) paper in a bit more detail. There you will also find the information that the OSI SAF product actually contains the raw, unfiltered SIC estimates which allow a user to utilize SIC values down to 0%. You will also find information about physically-based (by the retrieval settings) and mathematical (by the gridding process) uncertainties of the OSI SAF SIC product you are using.
L274: How about sea ice that is even thinner than 5cm?
L251-253: I can follow your discussion about that SIT might be a better measure to "define" a polynya. But to do that and to further argue about the pros and cons you first need to come up with a clear definition of the polynya area. Is it i) only the area of open water? Well, then under very cold conditions the polynya is very small - because of the freezing. Or is it ii) the area of open water and frazil / grease ice and the thin ice downstream of the polynya up to a thickness of x cm? In that case you need to define x and depending on that definition a polynya might be of substantial size even under very cold conditions (at least for a day). Or is it iii) somewhere between i) and ii), following the polynya model of Pease et al., defining the polynya area via the so-called frazil ice collection depth? To my opinion, the action item for you that is coming out of these lines is to answer the question: What is your definition of a polynya? And why?
L258-266: I don't understand this step and I suggest not to do it. I would not modify any of the input data sets for your inter-comparison beyond regridding or the like. But applying the thresholds and filters that are used on the observational side to the model side does not sound overly credible to me - for the reasons I laid out earlier in my comments.
Also, you might want to bear in mind that the results shown in Ivanova et al. 2015 with respect to thin ice are based on data of the SMOS sensor; the thin ice values used in that paper were computed from SMOS data. This computation has its limitations in case the sea ice concentration is not near 100%. Also, the sea ice conditions that are covered by that investigation do not reflect the conditions in a polynya. There one has frazil and grease ice, small pancakes and then all sorts of young and new ice often co-existing with open water whereas the investigation shown by Ivanova et al. 2015 is mainly based on thin ice regions of larger extent (i.e. sheets of nilas / grey and grey-white ice) as they occur in the Arctic, e.g. towards Siberia, during fall freeze-up. Hence the applicability of that "filter" to your model data does not appear credible to me.
Finally, using just 50% of original modeled SIC in case the SIT is between 5cm and 20cm appears a far too large modification (I deliberately do not say correction). If you look at the Ivanova et al. 2015 paper, figure 4, you can see that in the two reasonably populated thin thickness bins (0.1m-0.15m and 0.15m-0.2m) OSI SAF takes values of 80% and 87% and Cal/Val, which you could see as a surrogate for the Comiso Bootstrap, takes values of 82% and 90%.
Overall, I would say this comparison of "degraded" model SIC values is of limited value and I propose to discard it.
L268 / Section 3.4 / the entire paper: I don't understand / did not find it laid out in a very transpararent way, how the thresholds that are used henceforth were selected. I could not find an investigation that would tell me exactly why, e.g., for OSI SAF 75% SIC are used (for all cases, i.e. daily, monthly, re-gridded) while for NOAA/NSIDC CDR the thresholds are 1st different to OSI SAF and 2nd also different for the cases used. Possibly I overlooked this very important part in your paper, I am sorry. In some sense this resonates with my general comment GC0.
L306: What made you to chose NOAA/NSIDC CDR? Isn't the choice made in the NOAA/NSIDC SIC CDR to use the maximum value of two SIC products inferior to the well-thought through concept used in OSI SAF to derive a SIC CDR in a physical consistent way, with physically based uncertainty estimates? OSI SAF offers more information to be used than the NOAA/NSIDC CDR.
Figure 1: Just for clarification: The shaded area around the NOAA/NSIDC CDR curve represent +/- one standard deviation of the mean over the 42-years period at the respective longitude. Okay. But the results from the model represent both: a mean over the 50 ensembles PLUS the mean over the 42 years? Did you first compute the ensemble mean for every year and then average over the years? In short: My impression is that the shaded areas around the 3 curves are not directly comparable because they are based on different statistics.
I was wondering in this context, whether - in general - it might be more informative to use the median instead of the mean.
L337+ I am very sorry to have missed that but why are you also separately showing results derived from the NASA-Team SIC algorithm? As you indicate, it is very often biased low (often by 10-20%) compared to the Comiso Bootstrap algorithm SIC or independent estimates of the SIC and it is hence not overly surprizing that SIC fall below 85% and with that trigger the count of a polynya - either way, a coastal or an open water one. In order to keep the results and the text refering to the NASA Team SIC results, you need to motivate why you include this product in addition much better. The simplest solution (and more credible as well) would be to simply discard all results based on the NASA-Team algorithm.
L347/348: I am happy with your decision to focus on coastal polynyas because your results for what you call "open water polynyas" are not always convincing and are in particular not matching with the general idea that exists about what an open water polynya is (the one in the Weddell Sea near Maud Rise for instance or the Cosmonaut Sea polynya).
L356-358: I can understand that you need to use different thresholds when defining polynyas in the ESM models and in the observational data. But I don't understand the necessity and the scientific reasoning to use different thresholds for different observational data sets if - as is the case - these are based on the same satellite data with the same native and the same grid resolution. Therefore, for OSI SAF and NOAA/NSIDC the same SIC thresholds should be used. See also my comment made in general to section 3.4 further above.
L377/378: See my comment to Figure 3.
L384/385: "underscoring ..." --> Well, yes, also this is not an overly surprizing result. Especially when using the SIC and the SIT as a metric to define a polynya this will apply. While satellite remote sensing offers other possibilities to identify polynyas, this seems to be more limited in ESMs - especially because in ESMs polynyas are predominantly sub-grid scale phenomena while in observational data polynyas can be resolved when using the correct observations. I am not sure where you are heading for with these illustrations.
L385-388: "Additionally ..." --> Okay, this is all interesting ... but nobody (?) would regrid a 25 km (or even finer resolved observational product) to 1 degree grid resolution with the incentive to derive a trend. One would use the finest resolution possible to get the most accurate information. I guess what I want to state is that it is already clear from what you stated above that there are serious differences in the polynya statistics when deriving them on the native grid to deriving them after re-gridding and there is no need to expand on this also towards the trend analysis.
L403-419: I am sorry, but I don't see a convincing scientific rationale for this inter-comparison. First of all: defining polynyas in satellite observations by means of a SIC threshold is a simple approach with considerable limitations and not the state of the art. There are other means to do so and, an aspect that has not been mentioned in your paper so far sufficiently well, the sensitivity of the used SIC algorithms and products to the diversity of the sea ice types that are encountered in a polynya render usage of a SIC threshold to define a polynya a rather arbitrary approach.
Secondly, as stated already, nobody in the observational community would regrid such data to a coarser resolution to then do a trend analysis. Polynyas are so dynamic that the idea would rather be to go down to even finer grid resolutions and to swath data than to daily gridded data.
I propose to reduce this part of the paper considerably.
L500: I might have overlooked that but since you will be using the 1 degree outcome of the exercise presented in this chapter 4.2 to state something about the reliability and accuracy of the ESM output: Did you do any evaluation against independent, high-resolution estimates of the polynya location and polynya area? I mean, after all, you are applying a SIC threshold algorithm which in that extent has not yet been applied that extensively to delineate and map polynyas. Hence the credibility of the results need to be illustrated. Otherwise this data cannot serve as "a truth" for the ESM model output. This resonates with general comment GC0.
L572-580: I think talking about these complexities would be more credible if you would have i) evaluated your satellite-based products against independent data and ii) if you would focus on those regions of the Southern Ocean where the ESM is really capable to simulate the seasonal cycle of the polynya area reliably - which I assume is basically the Weddell Sea, the Ross Sea and the Amundsen Sea - whereas the number of sea-ice covered model grid cells along the East Antarctic coast is often so small that there is no proper delineation of polynyas possible.
L591/592: Isn't the argument of rapid freeze-over also applying to the Weddell Sea where most of the polynyas are situated quite far South and hence are experiencing similar if not even colder conditions than the polynyas along the East Antarctic coastline?
L594-606: I have difficulties to understand the scientific rationale behind showing these regional trends - especially when looking at data with 1 degree grid resolution and monthly temporal resolution. Also, I thought that this intercomparison between CESM-JRA and the observations was done to check the reliability of the modeled against the observed polynya locations and areas. I see limited value to look into integrated polynya area. It would make, in my eyes, more sense to look into specific key polynyas: Ronne-Filchner Ice Shelf polynya, Cape Darnley Polynya, Mertz Glacier Polynya, Terra-Nova Bay Polynya and Ross Ice Shelf Polynya. Finding a good match in occurrence and size of these polynyas over time would help us much more to understand what the capabilities and difficulties are with the CESM-JRA model to come up with a reasonable estimate of polynya location and area. I am not convinced we learn a lot from this regional trend analysis which, to my opinion, is too much influenced by artefacts in the polynya identification method, both in the observational and the model data.
It might, in this context, also make sense to keep in mind that the grid cell area of a ESM grid cell is about 500 sqkm, hence when you talk about a trend in polynya area of 1000 sqkm / decade you are talking about 2 grid cells. What is your uncertainty estimate of how well you can delineate a polynya in the model?
L611: "Satellite imagery ..." --> This statement is not true as it stands because SIC retrieval algorithm can reliably provide SIC values of less than a few percent and they detect sea ice even when it is very thin (1-2 cm); The SIC estimates are biased low, yes, but the detection works. Whether and to which degree a SIC retrieval algorithm is biased low in the presence of thin sea ice depnds on the tie points chosen.
L643 "significantly more" --> can you elaborate and provide a number here? As a rule of thumb, from observations it is said that about 1% of the area bound by the ice edge is covered by polynyas (I guess this has been written down in one of the papers by Tamura et al.).
L651/652: "These results ..." I was wondering whether you can comment on whether CESM has a snow cover on sea ice. Snow on sea ice limits light availability considerably and I am not sure this is taken into account in your investigations sufficiently well.
L725: "differ only by retrieval algorithm" --> But the differences in the retieval algorithms are substantial - especially in light of that the NOAA/NSIDC CDR is in fact the maximum SIC value of two SIC products which often differ by up to 20% in absolute terms.
L752/753 "thus monthly data ..." --> When I look at Figure 7, i.e. the correlation between NOAA/NSIDC and OSI SAF CDR based polynya identification it is obvious that the correlation is highest for the daily data. For me this is an indication that an adequate coincidence in the derived polynya area requires daily temporal resolution.
When I look at Figure 6 and compare the polynya areas derived with the three different settings I can also see considerable differences between daily, daily regridded and monthly regridded data. From those I would not conclude that using monthly data may be "sufficient".
And: Phytoplankton blooms may occur only for a number of consecutive days and certainly there are peaks of activities that are linked to light availability not only because of the sea ice conditions but also because of weather and hence shortwave radiation conditions. Given that the polynya area quickly (aka within hours) responds to temperature and wind speed and direction changes and hence changes on daily if not even sub-daily scale, I am not convinced that the statement given here, that monthly data may be sufficient, fits well with your overarching aim to also say something about the NPP and its dynamics.
L810-812: I don't think that further investigations of "these contradictionss" as you state them are beyond the scope of your work. Rather on the contrary. To my opinion, your investigation of the different methods to delineate polynya area from observational data is not yet complete. See my general comment GC0.
Figure A1: I don't understand why you use two different SIC thresholds in your schematic illustration. It implies the question: What if the SIC is between 50% and 90%?
L907: Better: "and two polynya grid cells in (b)"
Figure B3 clearly advocates to use an as small grid cell dimension as possible because case b) results in an identified total polynya area that is closer to what you call "integrated polynya area" than case a)
Editoral comments / Typos:
L58/59: "will consequently ... form." --> This statement does only apply to open water polynyas, does it?
L64: What are "polynya-like" features? Are you also refering to leads?
L223: "25km²" needs to be "25 km x 25 km"
L226: "25/45 km" --> 45 km reads strange. Are you sure it is not 50km?
L248: Please specify what you mean by "very small fractions"
L264/265: The daily sea ice concentration data you are using are either based on daily average gridded brightness temperature maps (NOAA/NSIDC) or were computed at swath level and subsequently gridded (and averaged) to form a daily average gridded product (OSI SAF). With that there is no "instantaneous snapshot" - at least not more than what you have in a model.
L278: Is this a typo? "0.4 cm" aka 4 millimetres ...
L326: I guess it needs to be "Figs. 3-4" here?
L330: The text says "July 15".
L346/347 "This is likely ..." --> True, but July 15 is not the melt season.
L369: "are much quite small" ?
L761: "higher / lower resolution" --> I suggest to use "finer" (i.e. smaller grid cell size) and "coarser" (i.e. larger grid cell size) because, applying a fixed SIC threshold of, say 75%, will lead to a larger polynya area in the coarse resolution than the fine resolution grid because of the smearing effect.
L767-769: You are using "very high" two times here. Would you mind to be a bit more specific here? Maybe a sentence that details "typical" values used to identify polynyas in SIC data would put the first "very high" into context when you compare it with the values you used here? And the second "very high" could potentially be solved by writing "near 100%"?
L784-787: Also here "very high" and "very thin" could / should be replaced by more specific values. You also might want to take into account the comments I made earlier about your statement of the capabilities of the satellite SIC retrieval.
L803: Another aspect that you could mention in this paragraph is the dependency of polynya formation in the existence of fast ice / icebergs - as demonstrated through various publications. How realistic is the fast ice formed in CESM?
Also, the oceanographic setting in the model might play a role - depending on how realistically the model simulates processes related to ice shelf cavities and nearby polynyas.
Finally, many Antarctic polynyas are bound to ice shelves. These calve and consequently the boundary for the polynyas change over time - but at the same time very radpidly from one year to the next. How does your investigation takes this into account?
Comments to the supplements:
Figure S3 caption line 4: "The horizontal dashed lines ..." because there are also vertical dashed lines.
Figure S4: Do I need to understand why there are no blue dashed lines?
Fig. 7: I know, this is just a supplemental figure but it takes quite some time to digest what is in the panels. I was wondering whether it would make sense to more clearly explain which data sets are correlated in the panels on the left hand side. I have difficulties to understand the term "EASE2" since this does not apply to an algorithm. I also suggest to be more consistent and, e.g., remove the "CDR" so that the names of the products are the same throughout the figure. I suggest to use "Month" instead of "Month/Season" as the x-axis annotation because there are no seasonal data shown. I recommend to refer to the panels only by a) to f) without the name that is put behind currently. I recommend to find a different solution to express which products are correlated. The usage of ":" is not very intuitive in that regard. You could perhaps write at the Y-axis "Correlation OBS vs. JRA" in panel e) and fine similar solutions for the other occasions. I also suggest to try to avoid that data points overlap with the symbols used as a legend. The caption contains a typo in the name of the Bellinghausen Sea.
Most of the above-given comments also apply to other figures of the same type. |
I see four main results in this paper:
All these four are interesting and important areas of research.
Unfortunately, the first three are mixed together in a story that I struggled to follow. In particular, the representation of polynyas in the models is presented as a detection issue; the fact that ”there may be model biases” is only finally acknowledged line 618. Yes, one needs to know how to detect polynyas in models before being able to evaluate them, but that’s why you are doing all these tests on the CDR where you degraded the resolution and/or changed the low values to reproduce known model/observation differences: You have a reference to compare your tests to. I would recommend you set the detection method, using the one most adapted to the models' detection based on the many tests you did on the observations, and only then look at the models, using this on method only. The model comparison of the two temporal resolutions would remain useful though, since different communities use the monthly and daily output.
It is also confusing that for most of the analyses you present only the results of JRA-CESM, when 1. Figure 1 suggests that CESM-LE has a more accurate sea ice and 2. The NPP analysis at the end is done on CESM-LE. That is, the case for using JRA-CESM is not well motivated. As I said above, the comparison between JRA-CESM and CESM-LE would allow you to discuss the effect of forcing vs full-coupling, but you do not discuss this for now. Removing JRA-CESM from this paper could allow you to keep this discussion for a dedicated study. You could even consider using only the subset of ensemble members that are the most accurate for your NPP analysis; according to Figure 8b, bottom panels, some are ok-enough.
Despite how this may sound, I suspect that all the work is there already, and that it is only the text that needs re-arranging for your argument to be convincing. The rewriting would be substantial though, so I do not provide minor comments that could become irrelevant this time.
The figures need adjusting as well to increase readability: