the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 22 Jul 2021
Changes in Supraglacial Lakes on George VI Ice Shelf, Antarctic Peninsula: 1973–2020
Abstract. High densities of supraglacial lakes have been associated with ice shelf instability and collapse. 2020 was a record melt year on George VI ice shelf with ~12 % of its northernmost portion being covered by lakes. We use 208 Sentinel-2 and Landsat-1-8 satellite images from the past 47 years, together with climate data and firn modelling, to assess the long-term presence of lakes on George VI, thus placing 2020 within a historical context. We find that the ~12 % lake coverage observed in 2020 is not unprecedented and similar to previous high lake years; events of similar magnitude occurred at least five times previously. Secondly, we find lake coverage is controlled by a combination of melting, accumulation, firn air content and firn build-up strong melting alone does not entail high lake coverage. Instead, while melting contributes positively to lake formation, we find accumulation to act as a limiting factor on the formation of lakes in response to melt, introducing new frozen material to the surface, thus cooling and storing meltwater. We find accumulation’s ability to limit melt to be further enhanced by its build-up, increasing available firn air content, and thus meltwater storage capacity. Our findings are supported by comparative analysis, showing years such as 1989 to have 55 % less melt, but similar lake coverage to 2020. Finally, we find that climate projections suggest future temperature increases, but steady snowfall in this region. Thus, in future there will be a greater propensity for higher lake densities on North George VI ice shelf, and associated risk of instability.
- Preprint
(1076 KB) - Metadata XML
-
Supplement
(233 KB) - BibTeX
- EndNote
Status: closed
-
RC1: 'Comment on tc-2021-214', Anonymous Referee #1, 06 Aug 2021
General assessment:
The authors of this paper attempt to synthesize the satellite imagery record, local weather station data, firn model output, climate reanalysis model output, and CMIP future climate model output in order to better understand the future expansion of the George VI’s surface meltwater drainage system. This is motivated by the recent findings of unprecedented melt in the 2019-2020 melt season. The motivations behind this study, to use the available data to give an assessment of present-day conditions of the surface drainage system to better predict its future evolution, is well-thought out, valuable, and promising for future study. Other studies have done similar multi-year assessments of Antarctic ice shelf surface hydrology systems (Langley et al., 2016, Stokes et al., 2020, Spergel et al., 2021). However, the authors present many different means of showing a disconnect between climate and surface conditions with lake coverage, but do not discuss that lake coverage, i.e where melt is observed ponding in satellite imagery, is mainly controlled by surface topography. The pre-existing surface depressions (discussed in Reynolds, 1981) must be filled to overspilling before water can drain over the surface into new areas/depressions. I am not familiar with the surface topography of GVIIS, but other ice shelves have more-or-less U-shaped depression cross-sections, so the addition of more meltwater does not change the surface area of the water body. It is only when all available space is filled with water that the surface area of water coverage expands via over-surface drainage. What the authors seem to describe with their analysis of similar meltwater lake coverage between 1989 and 2020 is not a dampened response to climatic forcing, but the fact that meltwater pond coverage increases as water flows and partially fills depressions nearby to meltwater production, but meltwater coverage plateaus as the partially-filled depressions fill, and only once overspilling occurs does lake coverage increase again significantly.
I would recommend refocusing this paper on the changes observed in melt pond distribution on GVIIS between the 1980s, as described in Reynolds, 1981, and where melt is observed today. There is a lot of value to giving a base-line and a detailed description of the inter-annual variability in the ice shelf’s hydrology. I would recommend a thorough search of the literature to give a broader context to the authors’ findings on GVIIS.
Originality:
The purpose of the paper, to assess the decadal trends in a persistent surface drainage network has a lot of merit.
Scientific Quality:
The oversight of topography controlling where melt forms, and what that means for measuring meltwater lake coverage with satellite imagery, makes a lot of the analysis done in the paper unsuccessful in proving any climate-surface hydrology mechanisms. I also have many questions about methods that are not addressed in the paper. The results are reliant on the threshold of NDWI, the comparison of imagery coverage across the years, and uncertainty in manual mapping. These three issues and the uncertainty they contribute should be discussed.
Significance:
As the paper is now, the points that are presented successfully (persistent, widespread melt on GVIIS; inter-annual variability in melt production leads to variability in meltwater lake coverage; meltwater being divided between ponds and firn pore spaces) are not novel enough to be significant. In its current form, the paper is unsuccessful in supporting a new mechanism for climate-surface hydrology interaction, the “dampening effect” of increased firn air content on lake coverage.
Presentation Quality:
Much of the paper is well written, but there are a few issues in the paper’s presentation: 1) there remains a number of passages that could use scientific, quantitative terminology instead of conversational. 2) quantities such as averages, sums, etc, should be precise in what they are describing to avoid ambiguity. 3) The figures should be revised to be more readable, especially the time-series plots. 4) Much of the material presented in the supplementary materials is critical to assessing the paper, and should be brought into the main text.
Main Comments:
- The paper needs to be reassessed after considering how surface topography affects where water pools. Several of the proposed mechanisms and causal relationships between climate, firn, and meltwater lake coverage need to be reconsidered, and revised if still true or removed if no longer true.
- The methods by which lake pixels are selected need to be further explained. Moussavi et al. (2020) would be a good reference if the new method is to be kept, but I would recommend using Moussavi et al.’s available code for Landsat 8 imagery and discuss the process used to select the NDWI thresholds for Landsat 1-7. I also don’t understand what the scaling of lake pixels derived from non-Landsat 7 imagery includes, but the uncertainty introduced by this needs to be discussed.
- Many of the assertions about climate effects on meltwater lake coverage presented in the discussion/conclusion need to be supported by data or citation of the literature. The choice of MAR is discussed in the supplementary materials, but the authors also seem to use MAR output as a single point rather than a spatially-varying raster dataset.
- Some sections need to be rewritten to clear up ambiguity in what was done, what is being extrapolated, etc.
Line-by-line comments are included in the attached pdf
-
AC1: 'Reply on RC1', Thomas Barnes, 23 Sep 2021
Review response for reviewer 1 is below. Full response to both reviewers added as a supplement.
Review 1
- The paper needs to be reassessed after considering how surface topography affects where water pools. Several of the proposed mechanisms and causal relationships between climate, firn, and meltwater lake coverage need to be reconsidered, and revised if still true or removed if no longer true.
To approach this point we intend to look at a high resolution DEM to find out whether surface topography and water pools show a connection. Based on the results of this investigation, we will identify whether or not they support the above hypothesis.
- The methods by which lake pixels are selected need to be further explained. Moussavi et al. (2020) would be a good reference if the new method is to be kept, but I would recommend using Moussavi et al.’s available code for Landsat 8 imagery and discuss the process used to select the NDWI thresholds for Landsat 1-7. I also don’t understand what the scaling of lake pixels derived from non-Landsat 7 imagery includes, but the uncertainty introduced by this needs to be discussed.
This point will be approached by comparison of the results of Banwell et al., (2021) where Moussavi et al (2020)’s methods were used to compare an alternative method to the results of this study. This was done previously through correspondence with A. Banwell, but not included in the manuscript. Additionally, reasoning behind thresholding will be discussed. The final part of this comment is addressed in the response to point (8) from Reviewer 2.
- Many of the assertions about climate effects on meltwater lake coverage presented in the discussion/conclusion need to be supported by data or citation of the literature. The choice of MAR is discussed in the supplementary materials, but the authors also seem to use MAR output as a single point rather than a spatially-varying raster dataset.
Effort will be made to improve referencing in relation to climatic effects on lake coverage. Additionally, clarification will be made for the use of MAR, as it was initially tested as a point source, however over the course of the study this evolved into a gridded use, and may not have been fully updated in writing as an oversight.
- Some sections need to be rewritten to clear up ambiguity in what was done, what is being extrapolated, etc.
Line by line comments will be addressed to improve the manuscript and lessen any ambiguity in writing. Additionally, effort will be made to include much of the supplemental information in the main body of the manuscript so as to avoid further ambiguity with methods.
-
RC2: 'Comment on tc-2021-214', Anonymous Referee #2, 22 Aug 2021
The comment was uploaded in the form of a supplement: https://tc.copernicus.org/preprints/tc-2021-214/tc-2021-214-RC2-supplement.pdf
-
AC2: 'Reply on RC2', Thomas Barnes, 23 Sep 2021
Response to reviewer 2 below. Full response to both reviewers added as a supplement.
Reviewer 2
- The abstract of the text needs to be re-written. When compared to the introduction it is of a lower standard of writing, and it doesn’t convey the key findings well.
The abstract will be rewritten to cover findings more comprehensively, once all other comments are addressed.
- When describing George VI Shelf (Lines 47-55), the authors need to do some wider reading of literature. For example, they should use past work here to describe the different glaciological settings of the north and south GVIIS. A study area figure is also required.
A study area figure will be produced and included, further efforts to discuss the differing settings of the north and south end of George VI ice shelf will be made. References such as Smith et al., (2007), Holt et al., (2013) and Hambrey et al., (2015) will be used among others to improve the quality of this discussion.
- It makes little sense to me to use the NDWI Green and Near Infrared method over the NDWI blue and red method, given that the majority of literature would use the latter, and this has been well justified in many previous papers. I am not convinced as to why the authors chose to use this alternative thresholding method, and the text in S2 still does little to convince me. It would be interesting to see some maps showing the differences between the two thresholding approaches.
Inclusion of information from the supplement and associated MSc project will be added to this work. Comparisons between each NDWI methodology were made at the initiation of the study on a test region, where Green-NIR served to produce lake polygons which were more ‘strict’ towards lake shorelines, and therefore less likely to pick up slush and saturated snow surrounding lakes.
- The lakes in some imagery were manually delineated, yet there is no mention of the error that should be considered when comparing these manually delineated lakes to lakes found using the thresholding method. Overall, the authors should consider the errors associated with all methods, and reference these where appropriate.
Consideration of error will be made for manual delineation, with the generated error being subject to discussion within the research group. This was tested again at the initiation of the project, however it was found that due to the subjectivity of manual delineation, a true value of error was hard to ascertain if multiple people carried out manual delineation, or the same person carried out the process over several days. Hence, further testing may be necessary.
- The authors state that they use a different threshold for Landsat 1 because the bands do not correlate with the other Landsat instruments. But I question whether the Sentinel-2 bands correlate? If not, why did you not use a different threshold for that too?
Sentinel 2 and Landsat 8 bands correlate closely, and therefore the same threshold value was used for each satellite. This will be made clearer in the text. Landsat-1 was an anomaly as it includes many fewer bands of differing width to more modern instruments.
- Is there full ice shelf coverage for every data point investigated? If not, how much of the ice shelf is ‘missing’?
Full ice shelf coverage is found for all data points other than those specified. A diagram including this information was produced but not included in the text due to initial constraints on manuscript length. However this will be included with the changes made in response to comments.
- The authors only show satellite imagery of GVIIS in maximum melt years, however they comment (Line 167) on the spatial organisation of surface meltwater in low melt years too. It would be useful to see some figures showing this, to allow the reader to see the changes that occur over time.
Included in the original MSc thesis was a series of diagrams showing the full lake coverage across GVIIS for each year during the study period. This will be included in the supplement in future submission as per this comment. However, it would not be appropriate as part of the main text due to the size of the diagram.
- The authors suggest that they convert the areas for all data that wasn’t affected by the Landsat-7 scan line failure, ultimately reducing the areas? This is a questionable decision as it broadly means the data presented is not representative of the true area of melt on GVIIS, which is an important statistic to have. I suggest the authors present both the converted and unconverted data.
We agree with the suggestion to approach Landsat-7 data in an alternative manner. Pre- and post- conversion data will be included in discussion in the updated manuscript. Another approach we have discussed would be to keep non-LS7 data as unaltered, and to convert Landsat-7 data using the 0.78 scaling factor. However, inclusion of both sets of values would show a more complete picture.
-
AC2: 'Reply on RC2', Thomas Barnes, 23 Sep 2021
- AC3: 'Comment on tc-2021-214', Thomas Barnes, 18 Mar 2022
Status: closed
-
RC1: 'Comment on tc-2021-214', Anonymous Referee #1, 06 Aug 2021
General assessment:
The authors of this paper attempt to synthesize the satellite imagery record, local weather station data, firn model output, climate reanalysis model output, and CMIP future climate model output in order to better understand the future expansion of the George VI’s surface meltwater drainage system. This is motivated by the recent findings of unprecedented melt in the 2019-2020 melt season. The motivations behind this study, to use the available data to give an assessment of present-day conditions of the surface drainage system to better predict its future evolution, is well-thought out, valuable, and promising for future study. Other studies have done similar multi-year assessments of Antarctic ice shelf surface hydrology systems (Langley et al., 2016, Stokes et al., 2020, Spergel et al., 2021). However, the authors present many different means of showing a disconnect between climate and surface conditions with lake coverage, but do not discuss that lake coverage, i.e where melt is observed ponding in satellite imagery, is mainly controlled by surface topography. The pre-existing surface depressions (discussed in Reynolds, 1981) must be filled to overspilling before water can drain over the surface into new areas/depressions. I am not familiar with the surface topography of GVIIS, but other ice shelves have more-or-less U-shaped depression cross-sections, so the addition of more meltwater does not change the surface area of the water body. It is only when all available space is filled with water that the surface area of water coverage expands via over-surface drainage. What the authors seem to describe with their analysis of similar meltwater lake coverage between 1989 and 2020 is not a dampened response to climatic forcing, but the fact that meltwater pond coverage increases as water flows and partially fills depressions nearby to meltwater production, but meltwater coverage plateaus as the partially-filled depressions fill, and only once overspilling occurs does lake coverage increase again significantly.
I would recommend refocusing this paper on the changes observed in melt pond distribution on GVIIS between the 1980s, as described in Reynolds, 1981, and where melt is observed today. There is a lot of value to giving a base-line and a detailed description of the inter-annual variability in the ice shelf’s hydrology. I would recommend a thorough search of the literature to give a broader context to the authors’ findings on GVIIS.
Originality:
The purpose of the paper, to assess the decadal trends in a persistent surface drainage network has a lot of merit.
Scientific Quality:
The oversight of topography controlling where melt forms, and what that means for measuring meltwater lake coverage with satellite imagery, makes a lot of the analysis done in the paper unsuccessful in proving any climate-surface hydrology mechanisms. I also have many questions about methods that are not addressed in the paper. The results are reliant on the threshold of NDWI, the comparison of imagery coverage across the years, and uncertainty in manual mapping. These three issues and the uncertainty they contribute should be discussed.
Significance:
As the paper is now, the points that are presented successfully (persistent, widespread melt on GVIIS; inter-annual variability in melt production leads to variability in meltwater lake coverage; meltwater being divided between ponds and firn pore spaces) are not novel enough to be significant. In its current form, the paper is unsuccessful in supporting a new mechanism for climate-surface hydrology interaction, the “dampening effect” of increased firn air content on lake coverage.
Presentation Quality:
Much of the paper is well written, but there are a few issues in the paper’s presentation: 1) there remains a number of passages that could use scientific, quantitative terminology instead of conversational. 2) quantities such as averages, sums, etc, should be precise in what they are describing to avoid ambiguity. 3) The figures should be revised to be more readable, especially the time-series plots. 4) Much of the material presented in the supplementary materials is critical to assessing the paper, and should be brought into the main text.
Main Comments:
- The paper needs to be reassessed after considering how surface topography affects where water pools. Several of the proposed mechanisms and causal relationships between climate, firn, and meltwater lake coverage need to be reconsidered, and revised if still true or removed if no longer true.
- The methods by which lake pixels are selected need to be further explained. Moussavi et al. (2020) would be a good reference if the new method is to be kept, but I would recommend using Moussavi et al.’s available code for Landsat 8 imagery and discuss the process used to select the NDWI thresholds for Landsat 1-7. I also don’t understand what the scaling of lake pixels derived from non-Landsat 7 imagery includes, but the uncertainty introduced by this needs to be discussed.
- Many of the assertions about climate effects on meltwater lake coverage presented in the discussion/conclusion need to be supported by data or citation of the literature. The choice of MAR is discussed in the supplementary materials, but the authors also seem to use MAR output as a single point rather than a spatially-varying raster dataset.
- Some sections need to be rewritten to clear up ambiguity in what was done, what is being extrapolated, etc.
Line-by-line comments are included in the attached pdf
-
AC1: 'Reply on RC1', Thomas Barnes, 23 Sep 2021
Review response for reviewer 1 is below. Full response to both reviewers added as a supplement.
Review 1
- The paper needs to be reassessed after considering how surface topography affects where water pools. Several of the proposed mechanisms and causal relationships between climate, firn, and meltwater lake coverage need to be reconsidered, and revised if still true or removed if no longer true.
To approach this point we intend to look at a high resolution DEM to find out whether surface topography and water pools show a connection. Based on the results of this investigation, we will identify whether or not they support the above hypothesis.
- The methods by which lake pixels are selected need to be further explained. Moussavi et al. (2020) would be a good reference if the new method is to be kept, but I would recommend using Moussavi et al.’s available code for Landsat 8 imagery and discuss the process used to select the NDWI thresholds for Landsat 1-7. I also don’t understand what the scaling of lake pixels derived from non-Landsat 7 imagery includes, but the uncertainty introduced by this needs to be discussed.
This point will be approached by comparison of the results of Banwell et al., (2021) where Moussavi et al (2020)’s methods were used to compare an alternative method to the results of this study. This was done previously through correspondence with A. Banwell, but not included in the manuscript. Additionally, reasoning behind thresholding will be discussed. The final part of this comment is addressed in the response to point (8) from Reviewer 2.
- Many of the assertions about climate effects on meltwater lake coverage presented in the discussion/conclusion need to be supported by data or citation of the literature. The choice of MAR is discussed in the supplementary materials, but the authors also seem to use MAR output as a single point rather than a spatially-varying raster dataset.
Effort will be made to improve referencing in relation to climatic effects on lake coverage. Additionally, clarification will be made for the use of MAR, as it was initially tested as a point source, however over the course of the study this evolved into a gridded use, and may not have been fully updated in writing as an oversight.
- Some sections need to be rewritten to clear up ambiguity in what was done, what is being extrapolated, etc.
Line by line comments will be addressed to improve the manuscript and lessen any ambiguity in writing. Additionally, effort will be made to include much of the supplemental information in the main body of the manuscript so as to avoid further ambiguity with methods.
-
RC2: 'Comment on tc-2021-214', Anonymous Referee #2, 22 Aug 2021
The comment was uploaded in the form of a supplement: https://tc.copernicus.org/preprints/tc-2021-214/tc-2021-214-RC2-supplement.pdf
-
AC2: 'Reply on RC2', Thomas Barnes, 23 Sep 2021
Response to reviewer 2 below. Full response to both reviewers added as a supplement.
Reviewer 2
- The abstract of the text needs to be re-written. When compared to the introduction it is of a lower standard of writing, and it doesn’t convey the key findings well.
The abstract will be rewritten to cover findings more comprehensively, once all other comments are addressed.
- When describing George VI Shelf (Lines 47-55), the authors need to do some wider reading of literature. For example, they should use past work here to describe the different glaciological settings of the north and south GVIIS. A study area figure is also required.
A study area figure will be produced and included, further efforts to discuss the differing settings of the north and south end of George VI ice shelf will be made. References such as Smith et al., (2007), Holt et al., (2013) and Hambrey et al., (2015) will be used among others to improve the quality of this discussion.
- It makes little sense to me to use the NDWI Green and Near Infrared method over the NDWI blue and red method, given that the majority of literature would use the latter, and this has been well justified in many previous papers. I am not convinced as to why the authors chose to use this alternative thresholding method, and the text in S2 still does little to convince me. It would be interesting to see some maps showing the differences between the two thresholding approaches.
Inclusion of information from the supplement and associated MSc project will be added to this work. Comparisons between each NDWI methodology were made at the initiation of the study on a test region, where Green-NIR served to produce lake polygons which were more ‘strict’ towards lake shorelines, and therefore less likely to pick up slush and saturated snow surrounding lakes.
- The lakes in some imagery were manually delineated, yet there is no mention of the error that should be considered when comparing these manually delineated lakes to lakes found using the thresholding method. Overall, the authors should consider the errors associated with all methods, and reference these where appropriate.
Consideration of error will be made for manual delineation, with the generated error being subject to discussion within the research group. This was tested again at the initiation of the project, however it was found that due to the subjectivity of manual delineation, a true value of error was hard to ascertain if multiple people carried out manual delineation, or the same person carried out the process over several days. Hence, further testing may be necessary.
- The authors state that they use a different threshold for Landsat 1 because the bands do not correlate with the other Landsat instruments. But I question whether the Sentinel-2 bands correlate? If not, why did you not use a different threshold for that too?
Sentinel 2 and Landsat 8 bands correlate closely, and therefore the same threshold value was used for each satellite. This will be made clearer in the text. Landsat-1 was an anomaly as it includes many fewer bands of differing width to more modern instruments.
- Is there full ice shelf coverage for every data point investigated? If not, how much of the ice shelf is ‘missing’?
Full ice shelf coverage is found for all data points other than those specified. A diagram including this information was produced but not included in the text due to initial constraints on manuscript length. However this will be included with the changes made in response to comments.
- The authors only show satellite imagery of GVIIS in maximum melt years, however they comment (Line 167) on the spatial organisation of surface meltwater in low melt years too. It would be useful to see some figures showing this, to allow the reader to see the changes that occur over time.
Included in the original MSc thesis was a series of diagrams showing the full lake coverage across GVIIS for each year during the study period. This will be included in the supplement in future submission as per this comment. However, it would not be appropriate as part of the main text due to the size of the diagram.
- The authors suggest that they convert the areas for all data that wasn’t affected by the Landsat-7 scan line failure, ultimately reducing the areas? This is a questionable decision as it broadly means the data presented is not representative of the true area of melt on GVIIS, which is an important statistic to have. I suggest the authors present both the converted and unconverted data.
We agree with the suggestion to approach Landsat-7 data in an alternative manner. Pre- and post- conversion data will be included in discussion in the updated manuscript. Another approach we have discussed would be to keep non-LS7 data as unaltered, and to convert Landsat-7 data using the 0.78 scaling factor. However, inclusion of both sets of values would show a more complete picture.
-
AC2: 'Reply on RC2', Thomas Barnes, 23 Sep 2021
- AC3: 'Comment on tc-2021-214', Thomas Barnes, 18 Mar 2022
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 1,007 | 438 | 61 | 1,506 | 87 | 50 | 53 |
- HTML: 1,007
- PDF: 438
- XML: 61
- Total: 1,506
- Supplement: 87
- BibTeX: 50
- EndNote: 53
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1