the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 04 Feb 2022
Relating snowfall observations to Greenland ice sheet mass changes: an atmospheric circulation perspective
Michael R. Gallagher
Matthew D. Shupe
Hélène Chepfer
Tristan L'Ecuyer
Download
- Final revised paper (published on 04 Feb 2022)
- Preprint (discussion started on 02 Jun 2021)
Interactive discussion
Status: closed
-
RC1: 'Comment on tc-2021-150', Anonymous Referee #1, 02 Jul 2021
General Comments
The paper of Gallagher and colleagues touches upon an important aspect of the Greenland Ice Sheet (GrIS) mass balance variability. They provide a relationship between daily snowfall variability and atmospheric circulation, and advance from previous studies by linking the estimated mass contribution of snowfall to the GrIS mass balance for each classified circulation pattern. In doing so, Gallagher and colleagues are able to provide an estimate for mass accumulation versus ablation components, with a main interest in the active southerly pattern, that is a valuable step forward. Gallagher and colleagues use a novel approach to create the daily snowfall variability maps, but I do recommend that they elaborate and specify their method section in this regard, and possibly provide code, to the benefit of research reproducibility. Other minor comments that might be addressed are included below. Based on these comments I think the paper is well suited to merit publication in TC.
Specific comments
Abstract L. 15-18: I would argue that the mass values stated here, which stem from Fig.6, do not represent the “good overall agreement” between GRACE and CloudSat, which is implied here, due to the “relatively weak statistical correspondence” of Fig.6 (L. 365). Maybe the wording could be revised, or maybe values from either Fig.5 or Fig.7a could be included to argue the agreement between GRACE and CloudSat, as is done in the manuscript (see L. 372; L. 470).
Section 3.1: could you elaborate which different clustering algorithms have been tested and what the different performances and outcomes were? What where the differences in internal cluster variance, were there strong differences in the resulting identified regions, etc? How robust do you expect the clustering of the regions in Fig.1 is.
Figure 1: I suggest to add the 2km elevation line in Fig.1c
Section 3.2: Please introduce the naming convention of the nodes as represented henceforth in the manuscript, [a,1] to [e,4], with briefly stating which nodes are considered ‘southerly patterns, ‘northerly’ and ‘zonal’/’easterly’ etc. to prevent confusion later. For example, southerly patterns are described to be those “surrounding node [c,1]” which then later seems to be [b,1], [c,1], [d,1] based on Fig.7, thus only the horizontally adjacent nodes; but for zonal patterns “around node [c,3]”, also [b,2] is included (diagonally adjacent). It can be difficult to infer which nodes have been considered by the authors when describing effects of a certain circulation pattern.
Section 3.2: Why does the SOM algorithm need SLP anomalies as input rather than SLP fields (L. 280)?
Section 4.1: It could be interesting to mention the impact of negative snowfall anomalies, e.g. node [a,3]; when do they occur and what would that implicate?
Section 4.2.1: I think I am misinterpreting something in the discussion about dynamic mass loss. L. 328 states a bound on dynamic mass loss of 10 to 30 Gt/year, but as this is read from Fig.5 this should be 30 Gt/month? But if that is the case, the comparison to the estimated dynamic loss of 50 Gt/year from literature (L. 336) is no longer ‘realistic to the first order’. Could you please check and/or clarify the units presented in this part?
Figure 5-6-7: I suggest to add a bin count to the top histogram and/or mention the total number of points used for the regression, which is relevant as it changes for each of the figures. Furthermore, can you comment on why Fig.3 shows that patterns [b,1] [c,1] and [d,1] occur most often in ‘melt’ months but that the regression in Fig.6 and Fig.7 seem to have more ‘non-melt’ months data samples? Is the (top) histogram overlapping melt & non-melt bars or stacking them?
Section 4.2.2. It is mentioned (L. 363) that not all months are included in the regression of Fig.6. Could you specify which months that are and why not all months from the GRACE observations are utilized?
Acknowledgements: I feel the paper could benefit from more information in regards to reproducibility. Used packages are mentioned here, but maybe code could be made available as well?
Technical corrections
L. 5: the term ‘daily maps’ gives the wrong impression of the type of map. I’d recommend stating “maps of the daily spatial variability…”, as is done elsewhere in the manuscript (L. 479)
L. 9: ‘is contributes’ to ‘contibuting’
L. 42: ‘surface mass balance’ should be ‘mass balance’, I think
L. 102: replace comma after ‘accumulation’ to be after ‘region’
L. 108: unnecessary apostrophe at “time series’”.
L. 162: define abbreviation of SLP here instead of L. 165
L. 306: I think pattern [e,2] is meant instead of [e,1] (corresponding to 0.5 Gt snowfall)
L. 376; L. 378; L. 380; L. 384: gigatonnes to Gt for consistency with rest of manuscript
L. 378; L. 380: the use of ‘additional’ in ‘additional occurrence’ is somewhat confusing, and not consistent with elsewhere (e.g., L. 356 and others); I suggest to remove it
L. 462: “dependent on”
L. 468: “Because of the novel nature combined methodologies presented in this paper…”: What is meant with ‘novel nature’? Possibly revise the start of this sentence?
L. 471: mass values are switched from what is stated previously: “Every 1.0 Gt observed by CloudSat corresponds with GRACE mass increase of 1.19 Gt”
L. 472; L. 476: ‘overestimated’ should be ‘underestimated’ in agreement with said mass value switch. This has also been stated as 'underestimated' previously in section 2.1, L. 73-76 and section 4.2.1 L. 315-320.
Figure 2: add the node naming conventions [a,1] to [e,4] to the figure similar as in Fig.3
Figure 3: I’d suggest using the same coloring convention for (non)-melt months as used in Fig.5-7
Figure 8: add the node naming conventions [a,1] to [e,4] to the figure similar as in Fig.3
Citation: https://doi.org/10.5194/tc-2021-150-RC1 -
AC1: 'Reply on RC1', Michael Gallagher, 15 Sep 2021
Kind thanks to the reviewers and editors for their time and effort in examining our work, the critiques and comments provided were detailed, helpful, and have resulted in a significantly improved manuscript. Almost all changes suggested by reviewer comments were integrated into the revised submission, figures have been modified, and the authors have provided responses directly to comments from individual reviewers where necessary. Thank you also for the additional time it will take you to read these responses, any further commentary is of course welcome and we are happy to clarify as necessary.
-------------------------------------------------------------------------
Reviewer one, thank you for your supportive comments on the methodology and results of the study. We agree with the emphasis on reproducibility and accountability and we have revised the manuscript to reflect this. Clear and open presentation of methodology is in the best interest of research and progress, and we've made a clear statement that the code is available to anyone who should be interested. Thank you also for providing detailed comments and technical corrections.
All of the specific comments from reviewer two were integrated into the revised texts, as these critiques were insightful and have improved the text. Below are a few specific comments where important or requested:
Reviewer one: "can you comment on why Fig.3 shows that patterns [b,1] [c,1] and [d,1] occur most often in ‘melt’ months but that the regression in Fig.6 and Fig.7 seem to have more ‘non-melt’ months data samples? Is the (top) histogram overlapping melt & non-melt bars or stacking them?"
The histogram is stacking the melt and non-melt bars. The reasons these southerly patterns occur more often in non-melt months is simply because of the larger portion of the year, 5 vs 7 months. While this is a small difference, it's enough to approximately even out the number of both blue and orange points. Although the long-tail in the histogram is entirely orange.
Reviewer one: "Figure 1: I suggest to add the 2km elevation line in Fig.1c"
We had included this in a prior revision, but it was rejected during internal review for being too 'busy'. If the editor would like, we can add this to the figure.
Reviewer one: "It is mentioned (L. 363) that not all months are included in the regression of Fig.6. Could you specify which months that are and why not all months from the GRACE observations are utilized?"
This was poor wording on the part of the author, the idea to be communicated was that the trends quantify the linear relationship between a specific circulation pattern(s) and the mass change that month. While there are clear relationships, the mass change in any given month is also impacted by the circulation patterns that occured in that same month. Meaning a month with 7 [c,1] occurrences but 20 [a,4] occurrences would have a very different relationship to mass balance than a month with 7 [c,1] occurrences but 20 [e,1] occurrences. All available data was included in the figures.
Reviewer one: "Section 4.2.1: I think I am misinterpreting something in the discussion about dynamic mass loss. L. 328 states a bound on dynamic mass loss of 10 to 30 Gt/year, but as this is read from Fig.5 this should be 30 Gt/month? But if that is the case, the comparison to the estimated dynamic loss of 50 Gt/year from literature (L. 336) is no longer ‘realistic to the first order’. Could you please check and/or clarify the units presented in this part?"
This was confusing wording on the part of the author, the 21 Gt/month is the correct number from the mass loss in the non-existant year with zero snowfall. However, the average snowfall above 2km is approximately 17 Gt/month. Thus the net mass loss due to dynamics above 2km is ~4 Gt/month according to our methodology. In a year, this would be 48 Gt, the wording in the text has been changed to be this clear.
Citation: https://doi.org/10.5194/tc-2021-150-AC1
-
AC1: 'Reply on RC1', Michael Gallagher, 15 Sep 2021
-
RC2: 'Comment on tc-2021-150', Anonymous Referee #2, 03 Jul 2021
Summary
Gallagher et al. investigate the role of specific atmospheric circulation patterns on snowfall across Greenland. This is a good paper that deserves to be published in The Cryosphere. I was impressed by the novelty of the techniques which were mostly described thoroughly in the methods. Some of the writing and figures could be improved. I also remain a little skeptical about the technique used to quantify total snowfall for each month of the study period. Some more analysis about the intra-pattern snowfall variability is required before I can fully judge the robustness of this technique.
Specific comments
L2-3: This is pretty poor first sentence. I’ve read “…the spatial and temporal variability of its contributions to mass balance have so far been inadequately quantified” multiple times and still can’t figure out what it means. Please revise.
L8-9: “…with each daily occurrence of the most extreme southerly circulation pattern is contributes an average of 1.66 Gt of snow to the Greenland ice sheet” also doesn’t make sense. Contributing?
L13-14: On the surface this statement is confusing because surely there are some atmospheric regimes that’s produce any snowfall? I get that you always get a some snowfall if you sample thousands of atmospheric conditions. But that is not clear from just the abstract. Consider removing this sentence.
L26: Please consider capitalizing “ice sheet”. It’s the Amazon River, the Tibetan Plateau and should be the Greenland Ice Sheet.
L27: “recent years” will not be recent soon, could you be more specific about the timeframe over which Greenland contributed 0.47 mm to sea-level rise?
L30: Would be useful to add that snowfall is also difficult to model accurately.
L40-41: This part of the sentence is a little difficult to comprehend, replacing “observations of” with “observed”might help.
L40-41: I think this is too simplistic and not entirely true. McIlhattan et al. (2020) link ice-phase and super-cloud liquid water snow cases to atmospheric circulation patterns (see figures 14-17). Consider revising this statement.
L43-44: And Ryan et al. (2020) JGR
L45-46: Again, I don’t feel like this statement is entirely true since it looks like they do link snowfall to atmospheric circulation patterns. Can you be more specific about what McIlhattan et al. (2020) does not do and why that matters (i.e. what it prevents them from concluding). I feel that clear statements of literature gaps are very important in this era of mass publication. This would then set-up lines 47-57 better (which provide a good overview of the paper’s goals).
L60: Would read better if you reversed this sentence. Something like “We derived snowfall observations over the entire Greenland Ice Sheet using the…”
L62: Don’t think “taken” is the right verb here, “derived from”? “acquired”
L65-67: Consider mentioning that the battery anomaly prevented observations during night-time here since this is not common knowledge to non-CloudSat users.
L69: Consider removing the Palerme et al. (2017) reference here because they did not evaluate CloudSat with in situ observations like the other two studies.
L69-70: This sentence seems irrelevant, consider removing.
L73-74: Ryan et al. (2020) also estimate that this causes an 8% underestimation of snowfall at Summit Station, consider including this reference.
L77: Consider adding “rare over the Greenland Ice Sheet”. Also note that the reference for this statement is incomplete, please provide a peer-reviewed one.
L78: Capitalize “-Jackson”
L79-81: I don’t understand how you can assume that CloudSat snowfall retrievals “represent the mean snowfall on the day of observation” when earlier you state that only “CloudSat observations from June 1 st 2006 to April 16 th 2011 were used… because this analysis requires uniform data to avoid assimilating diurnal biases tied to seasonality in the Arctic…”. It’s either one of the other. If hourly snowfall observations do represent the daily mean then why not use all CloudSat snowfall observations to 2017? Surely more observations makes it more likely that CloudSat’s infrequent sampling represents mean conditions?
L82: Consider adding something like “All observations were initially gridded at 1 x 1 degree but later binned into eight regions identified by cluster analysis (Section X.X)” so that it is obvious to the reader that the Figure 1c represents the main spatial unit of analysis, not 1 x 1 degree.
L125: And Ryan et al. (2020)
L128-134: This paragraph is pretty vague and much of the text is repeated in the next paragraph (just with more detail). Consider removing some of this text or placing it below lines 135-143, once we know more about the details of the cluster analysis.
L132-133: What is meant by “…minimum of 90% daily sampling coverage…”
Figure 1a/b: Having different y-axis scales for these figures is confusing since it would nice for the reader to be able to compare snowfall rates in different regions. Also the caption does not make sense since the two of lines look like they have minima in March, not summer. I get that having eight lines on the graph is busy but the distinction between (a) and (b) is not really working here. Consider revising.
L148 and L151 (and elsewhere): I don’t agree with “annual variability” here. Season to season differences (i.e. winter vs. summer) should be “seasonal variability”. In L426 and L449 you refer to this variability as “seasonal”.
L156-163: These sentences don’t add much, consider removing.
Figure 2. Would be useful to have some arrows on this figure. I also don’t think the colorscale works very well. The dark green is too similar to dark blue, consider using a more distinctive colorscale. Finally, consider adding labels (a, b, c and 1, 2, 3 etc.) so that the reader can easily refer to a specific panel after reading the text.
198-205: This should be moved to the results section
206-211: This is also results so should be moved
L222-224: Would be nice to reference figure 1 here.
L225-227: Irrelevant and repetitive test, consider removing.
Figure 4: Is this gigatonnes per year? Or during the whole GRACE record? Consider adding maps next to these panels to make it more obvious which regions of the ice sheet they are referring to.
L301-302: For the whole ice sheet? Or for a specific region? Please clarify.
Figure 5: I don’t understand why the dots are colored in orange for “melt months”. Earlier you stated that little melt occurs above 2 km elevation (L257: “…surface melt generally does not occur.”). Even if there is some melt at these high elevations, it’s unlikely that it happens in May and September. So the distinction between orange and blue does not make a lot of sense.
Figure 5: The statement “Each dot is colored by the time of year that it occurred” is misleading because it makes it sound like the dots are colored on a continuous scale when actually they are just arbitrarily divided into two groups.
Figure 5: please explain the histograms on each axis in the caption.
L301-307: I remain a little skeptical about the technique used to quantify total snowfall for each month of your study period. I get that it is a necessary (due to infrequent CloudSat sampling for each month) and potentially powerful approach. But I think a few more details need to be provided for me to fully judge how robust it is.
For example, does pattern c1 always produce 1.2 Gt of snowfall on average? Surely this is very much dependent on which grid cells CloudSat sampled on that day? Could the authors provide some more statistics for each pattern (e.g. standard deviation, histogram)? Pattern c1 may produce 1.2 Gt in July but does it also produce 1.2 Gt in April? This should be checked because, if not, you may consider not using all months when computing average cumulative snowfall for each pattern. This information does not have to go in the main paper but could be added to the supplementary materials.
Figure 6: Similar comments to Figure 5.
L361-L362: Agreed, would be nice to explore this intra-pattern variability a little more.
Figure 8: The black lines are a little thick making it hard to see how the coverage of the orange and red areas, consider making them thinner or removing.
L443: I think you did something a little more specific than “looking at CloudSat snowfall observations…” Please revise.
L447-453: It should be clarified that these findings are not original contributions. Please provide a little more attribution to previous studies to indicate this point. Something like “confirming previous modeling and observational studies by XXX”.
L463: These “anomalously extreme days” have not been discussed in the manuscript. Please provide some more information about these intra-pattern extremes as per previous comment.
L464-465: Repetition of L453-455
L465-466: These statements left me wondering how frequent these southerly and northerly events are. I realized that Figure 3 is barely discussed in the results or discussion section. Would it be fair to say that the increase in summer snowfall can partly attributed to an increase in southerly events during summer? I think this an important point that could be developed further. I don’t think this study needs to provide the final word but it could be interesting to touch on in regards to Clausius–Clapeyron vs. atmospheric circulation debates.
Citation: https://doi.org/10.5194/tc-2021-150-RC2 -
AC2: 'Reply on RC2', Michael Gallagher, 15 Sep 2021
Kind thanks to the reviewers and editors for their time and effort in examining our work, the critiques and comments provided were detailed, helpful, and have resulted in a significantly improved manuscript. Almost all changes suggested by reviewer comments were integrated into the revised submission, figures have been modified, and the authors have provided responses directly to comments from individual reviewers where necessary. Thank you also for the additional time it will take you to read these responses, any further commentary is of course welcome and we are happy to clarify as necessary.
--------------------------------------------------------------------------------------
Reviewer two, thank you kindly for the detailed commentary on both the scientific content as well as the presentation. Thank you also for your clear investigation and supportive commentary on the novel methodology developed for this analysis.
In the summary comments from reviewer two, the sole concern expressed was for the role of intra-pattern variability of snowfall and its relationship to the quantified snowfall magnitudes. The authors have spent significant time investigating sources intra-pattern variability prior to the submission of this paper and will address these concerns directly here. Before discussing, it is important to clarify that there data is aggregated into two categories for daily circulation patterns, cumulative (net) values and seasonally anomalous values. Sources of intra-pattern pattern variability are significantly different for each. For the cumulative values (Fig. 4) used to quantify net daily mass input, intra-pattern variability is significant and can be almost entirely attributed to the seasonal cycle of variability in snowfall across Greenland. In this way, the snowfall magnitude for each pattern is a spatial and seasonal average of the snowfall mass input. Whereas for anomalies (Fig. 2), having subtracted the annual snowfall cycle from the observations, the primary source of intra-pattern variability is simply only the spatial variance in pixel aggregation (Fig. 1). Thus, there is significantly more intra-pattern variability present in the net values than the anomalies.
Interestingly, to add to the scientific discussion, the authors have previously calculated the net snowfall magnitudes for each season for the circulation patterns and there are many fascinating results. For example: patterns with northerly and weak zonal advection have significantly lower intra-pattern variability than the 'active' southerly regime identified in the paper. Meaning occurrences of these northerly patterns in winter and summer result in approximately the same net magnitude snowfall. Whereas southerly patterns are shown to vary significantly between seasons (~ +/- 1Gt) with the most snowfall actually occuring in winter(!), although southerly advection in winter is exceedingly rare. These figures are posted along with this comment.
However, the primary issue in looking at the data this way is that there are significantly lower statistics for each season for some individual nodes due to the intra-annual variability in circulation node frequency. While some qualitative insight can be gained this way, the authors chose not to include these figures in the paper because there are severe questions about the sample size for slicing specific circulation patterns during specific seasons and thus the representativeness/statistical validity of these numbers values is in question. It is for all of these reasons that an analysis of intra-pattern variability, and the subtlety therein, would be a scientific analysis in its own right. In particular, this analysis would be framed as an exploration of the inter-annual variability of snowfall first and foremeost as opposed to quantifying the contribution of snowfall to GrIS mass balance (as was the goal of this paper).
All of the specific comments from reviewer two were integrated into the revised submitted text. We would like to thank them again for their effort in providing this valuable feedback. The authors contribute replies where further warranted.
Reviewer two: "Figure 5: I don’t understand why the dots are colored in orange for “melt months”. Earlier you stated that little melt occurs above 2 km elevation (L257: “...surface melt generally does not occur.”). Even if there is some melt at these high elevations, it’s unlikely that it happens in May and September. So the distinction between orange and blue does not make a lot of sense."
Melt months are highlighted in orange so as to complement the discussion in the text about the differences in mass input between summer and winter months. The fact that all large snowfall events and thus mass increases occur in summer (and that the number is quantified directly) is a new and important result. While it may not be relevant to this figure specifically, it's important contextually.
Reviewer two: "L26: Please consider capitalizing “ice sheet”. It’s the Amazon River, the Tibetan Plateau and should be the Greenland Ice Sheet."
While this is funny, true, and agreeable, we will refer this decision to the copy editors at The Cryosphere.
Reviewer two, asked about the trends and differences in frequency of occurrence for the southerly regime:
While the authors agree that understanding the long-term connections between trends in snowfall and circulation is a fascinating question, it's unfortunately untenable with the available data. The authors have tried using several methodologies to identify trends in the occurrence of important regional circulation patterns and have failed to reach convincing conclusions. While the 70 year reanalysis record of surface pressure provides enough data to identify changes in circulation beyond decadal cycles of the AO/NAO, connecting those trends to the <20 year record of snowfall is a difficult proposition. Unfortunately this is still the realm of modelers and is impossible to tie to available CloudSat observations. Further, the 'large' suite of circulation patterns (5x4 SOM) is compelling for its ability to provide a more detailed look into circulation variability. Even with the long reanalysis record, the authors have found that identifying trends still requires a more blunt tool and is statistically limited to a more granular search space of 6 (3x2) or less patterns. This means that any of the interesting spatial variability identified in this paper would be significantly less clear.
-
AC2: 'Reply on RC2', Michael Gallagher, 15 Sep 2021
-
AC3: 'Comment on tc-2021-150', Michael Gallagher, 15 Sep 2021
Kind thanks to the reviewers and editors for their time and effort in examining our work, the critiques and comments provided were detailed, helpful, and have resulted in a significantly improved manuscript. Almost all changes suggested by reviewer comments were integrated into the revised submission, figures have been modified, and the authors have provided responses directly to comments from individual reviewers where necessary. Thank you also for the additional time it will take you to read these responses, any further commentary is of course welcome and we are happy to clarify as necessary.
Citation: https://doi.org/10.5194/tc-2021-150-AC3
Peer review completion
