the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 07 Nov 2023
Weak relationship between remotely detected crevasses and inferred ice rheological parameters on Antarctic ice shelves
Abstract. Over the past decade, a wealth of research has been devoted to the detection of crevasses in glaciers and ice sheets via remote sensing and machine learning techniques. It is often argued that remotely sensed damage maps can function as early-warning signals for shifts in ice shelf conditions from intact to damaged states and can serve as an important tool for ice sheet modellers to improve future sea-level rise predictions. Here, we provide evidence for Filchner-Ronne and Pine Island ice shelves that remotely sensed damage maps are only weakly related to the ice rate factor field A derived by an ice-flow model when inverting for surface velocities. This technique is a common procedure in ice flow models, as it guarantees that any inferred changes in A relate to changes in ice flow measured through observations. The weak relationship found is improved when investigating heavily damaged shear margins, as observed on Pine Island Ice Shelf; yet, even in this setting, this association remains modest. Our findings suggest that many features identified as damage through remote sensing methods are not of direct relevance to present-day ice-shelf flow. While damage can clearly play an important role in ice-shelf processes and thus be relevant for ice-sheet behaviour and sea-level rise projections, our results imply that mapping ice damage directly from satellite observations may not directly help improve the representation of these processes in ice-flow models.
-
Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
-
Preprint
(1228 KB)
-
Supplement
(1171 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(1228 KB) - Metadata XML
-
Supplement
(1171 KB) - BibTeX
- EndNote
- Final revised paper
Journal article(s) based on this preprint
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2023-2362', Anonymous Referee #1, 31 Dec 2023
This paper presents a classification study on the relationship between remotely detected crevasses and spatial variation of ice rheology. The study concludes that the relationship is weak and that the use of remotely detected crevasses to directly map damage in ice flow models may not be a good way of improving ice flow models. This study provides valuable information for other ice modelers attempting to represent fractured ice in their simulations. The work is overall well designed and presented. I would recommend it for publication after addressing the minor revisions and questions listed below (organized in a few broad categories).
Improve the image quality and readability of some figures.
- Figures 1 and S2 can be difficult to read and would benefit from higher resolution.
- Figures 1 and S2 have difficult to read legends partially due to resolution and partially due to the low contrast between text color and background color.
- Figure S1: I recommend changing the top left box in each matrix to a lighter background color with better contrast. It can be difficult to read the numbers on a printed copy.
Include some additional technical details to better understand the study's methods.
- Line 160: you interpolate the field "A" onto the regular crevasse mask coordinates. Could you briefly name or explain the interpolation method?
- Are results robust to changes in the other regularization parameter? The paper focused on sensitivity in one parameter, but not the other.
- Is there a numerical threshold or method for determining the "consistent agreement" mentioned at the end of the Figure 4 caption?
- In Figure S3(b) for the highly damaged case, the ROC curves become close to random classifiers at high values of the regularization parameter. However, the opposite is true for the "all crevasse" case in Figure S3(a). What is the reason for that behavior?
Provide some additional context/clarity for some background information and conclusions.
- Lines 35-38: Can you provide citations for the impact of surface crevasses in different situations?
- I was confused by the "(see details below)" note in line 48 and the "(described further below)" in line 56. Can specific sections or subsections be referenced to make that information more clear?
- I find it interesting that for the analysis on Pine Island Ice Shelf including only heavily damaged crevasses (Figure 3b), the relationship no longer behaves like a random classifier. I would be interested to know what attributes "heavily damaged crevasses" have that other crevasses lack in the input data. I think that some brief discussion of the input data's classification method could provide valuable context to the "heavily damaged" result.
- Is there a different well-motivated choice in sliding law or sliding law exponent (e.g. regularized Coulomb friction) that would affect the results of this study? How dependent is the spatial variation of "A" dependent on the sliding law for this study?
- Lines 365-366: You state "We find that for any threshold value, the performance of this predictor is like that of a random classifier". That seems inconsistent with the "high damage" study in the manuscript. I don't think that the "high damage" study weakens the paper's overall conclusions, but is an interesting result that perhaps merits some additional discussion.
- Lines 382-384: you mention the possibility of surface crevasses penetrating to the water line. Can you give additional context to the correlation between surface crevasses reaching the water line and inferred variation in ice rheology?
- I quite like Figure S6 as a visualization for the classification. Can a similar figure for Filchner-Ronne ice shelf be included as well?
Technical Corrections
- The acronym OPTtimal operating PoinT (OPTPT) defined in the text (lines 184 and 197) differs from the acronym used in Figures 2 and 3 (OPT-PNT), assuming the two have the same meaning like I believe they do.
- Table T2: Column 2, Row 3 has an extra "tab" space that should be deleted.
- Notation for the equation in section 1.3 of the supplementary information should be made consistent. Most of the terms use forward slashes but "Cost(P|N)" uses a vertical bar in the text.
Citation: https://doi.org/10.5194/egusphere-2023-2362-RC1 -
RC2: 'Comment on egusphere-2023-2362', Anonymous Referee #2, 18 Jan 2024
General comments:
This is a well written manuscript on an important problem in glaciology: The influence of brittle-failure damage on ice flow. I commend the authors for submitting a manuscript that is a delight to review. I had a lot of fun reviewing this and it is a fascinating study. I think that this study and manuscript make important contributions to the science and that the community will be very interested in reading this.
One key point: In the history of looking at “damage” (specifically, reflections and noise in radar data caused by crevassed zones at the edges of ice streams) in ice shelves, the shoe started out on the other foot: Signs of damage that were buried under un-damaged firn (specifically at Kamb Ice Stream, then called ice-stream C), and their extensions out on the Ross Ice Shelf were of great significance in telling the story of past change in the flow conditions. In this case, the change was the “shut down” of an ice stream (which has less press value in today’s world, but which needs to be studied, especially if there is ever going to be hope that ice-stream flow at Thwaites or Pine Island Glacier will self-limit). It might be worth touching on this point in the introduction. I believe that the original literature on it (from the late 1970’s and early 1980’s) is easy to find.
Generally speaking: the word on the “street” (I got this from an editor’s meeting of a related glaciological journal) is that use of acronyms can be an impediment to getting papers to be cited. In some ways, I think this is intuitively obvious; but apparently it is also a result of doing careful quantitative analysis with specific metrics for measuring acronyms and citations. This paper does not have a lot of acronyms, however, I still wonder if the paper would be easier to read (and thus more likely to be cited) if acronyms were minimized. The ones that I had to struggle with were: CNN, NeRD (that one appears to be a kind of subtle joke, which I like, as the word NeRD in English refers to a “smart” but slightly “dull” person), MOA, ROC, FPR, TPR, AUC, OPTPT, … This is an online journal, hence there is no cost in paper to write out the words in full. I think that the authors should consider this. The authors might additionally find it works more simply to assign actual variables to elements that are now a kind of hybrid acronym, for example: AUC-mean2009. These long, strung out variable names that incorporate an acronym make the reading of the paper a bit harder. With harder reading, there is then the possibility of fewer citations.
I notice that the analysis makes a distinction between 2019 and 2020 velocities in the AUC for Pine Island Ice Shelf (and different years for different regions). What specifically (as a reminder) is changing? Is it the detected crevassing fields or is it the velocity field?
Specific comments:
Abstract: “Wealth of research”? I’m not sure it adds precision to use the term wealth as a qualifier. Maybe another word would be more appropriate. “Wealth” appears as the first word in the discussion as well.
A minor point: I see that variables that appear in the text are italicized (as they should be), however, this italicization needs to be checked. For example, x- and y- around line 115. Ditto for the R that appears near there. A good double check just for this would be useful.
L-curve method (here the L should not be italicized, as I think that the “L” denotes a shape more than a variable). Also, I’ve never heard of this method before, so I wonder if the reference to it should appear right away. Also, I would find it helpful to possibly say in a few sentences how a user would “walk through” a problem following the L-curve method.
Something to check: numbers in the text sometimes appear in scientific notation (where there is a “times ten to the power of something”) and sometimes in digital computer notation or floating point notation, e.g., 1.3e-4… Journal style should be checked. I suppose it would be a bit pedantic to say so, since nobody even thinks about this any more: but it would be cool if every now and then people would report whether they are doing single or double precision computations (I don’t suppose there are single precision computations any more, but what the heck, I might as well bring it up).
Figure 4: I note that the vertical axes have a notation that is “ #10^n “ Is this standard notation (I usually see “x” replace the “#”)? Also, would it be better to have the scale (ten to the power of) in the axis label rather than perched on top of the axis frame?
This one is not an essential comment, and is motivated by the fact that Chris Borstad, one of the pioneers of damage mechanics in glaciology passed away in November of 2023. I see that Copernicus no longer provides an “acknowledgment” section in its articles where a dedication (if the authors were to want to make one) would normally appear. Instead, I see that Copernicus prefers to replace the general acknowledgement section with specific (and seemingly less noble) sections like “Funding”. This comment is not a criticism, just something that came to my mind (and heart).
Citation: https://doi.org/10.5194/egusphere-2023-2362-RC2 - AC3: 'Reply on RC2', Cristina Gerli, 07 Mar 2024
- AC2: 'Reply on RC1', Cristina Gerli, 07 Mar 2024
-
RC3: 'Comment on egusphere-2023-2362', Adrien Gilbert, 09 Feb 2024
This study examines whether the surface crevasse field observed from remote sensing is related to the flow rate factor derived from inverse methods constrained by surface velocity. A poor relationship is found, suggesting that surface crevasse observations may not be a good proxy for quantifying ice damage affecting ice rheology over a thickness relevant to ice flow dynamics.
The study addresses an important topic, as observational constraints on ice damage are critical for assessing ice shelf dynamics and stability in ice sheet models. The methodology is rigorous and clearly described, and the results are convincing and well presented. I recommend the paper for publication in The Cryosphere after major revisions.
General Comments
My main concern is that the inferred flow rate factor from surface velocity is consistently presented in the manuscript as the truth that the observed crevasse field should match. I think this way of presenting the study is not fair as it could be turn differently. For example, the crevasse field could be presented as the truth that the inferred flow rate should match, and one could conclude that the model does not capture the effective viscosity and stress field well due to its simplification, inappropriate physics or inaccurate ice shelf three dimensional geometry. The manuscript clearly lacks of a discussion about the model assumption and the reliability of the inferred flow rate factor. Even if the model is strongly constrained by surface velocity observations, it does not sound right to question the utility of crevasse field observations without even mentioning that the weak relationship could be due to the model lacking the relevant physics. Are we sure that the inferred flow rate factor is not affected by neglecting the elastic stress field or the depth-dependent variation of the flow rate factor or other things? The whole study could also conclude that the SSA does not capture the stress field of the ice shelf well, because the inferred value of the flow rate factor is weakly related to the observed damaged areas. This would give the opposite message to the community ...
I suggest that the authors should also consider the case where the SSA approximation is the cause of the discrepancy, and provide strong arguments if they think this cannot be the case.
Specific comments
You will find a small list of correction and specific comments embedded in the annotated PDF in attachment.
- AC1: 'Reply on RC3', Cristina Gerli, 07 Mar 2024
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2023-2362', Anonymous Referee #1, 31 Dec 2023
This paper presents a classification study on the relationship between remotely detected crevasses and spatial variation of ice rheology. The study concludes that the relationship is weak and that the use of remotely detected crevasses to directly map damage in ice flow models may not be a good way of improving ice flow models. This study provides valuable information for other ice modelers attempting to represent fractured ice in their simulations. The work is overall well designed and presented. I would recommend it for publication after addressing the minor revisions and questions listed below (organized in a few broad categories).
Improve the image quality and readability of some figures.
- Figures 1 and S2 can be difficult to read and would benefit from higher resolution.
- Figures 1 and S2 have difficult to read legends partially due to resolution and partially due to the low contrast between text color and background color.
- Figure S1: I recommend changing the top left box in each matrix to a lighter background color with better contrast. It can be difficult to read the numbers on a printed copy.
Include some additional technical details to better understand the study's methods.
- Line 160: you interpolate the field "A" onto the regular crevasse mask coordinates. Could you briefly name or explain the interpolation method?
- Are results robust to changes in the other regularization parameter? The paper focused on sensitivity in one parameter, but not the other.
- Is there a numerical threshold or method for determining the "consistent agreement" mentioned at the end of the Figure 4 caption?
- In Figure S3(b) for the highly damaged case, the ROC curves become close to random classifiers at high values of the regularization parameter. However, the opposite is true for the "all crevasse" case in Figure S3(a). What is the reason for that behavior?
Provide some additional context/clarity for some background information and conclusions.
- Lines 35-38: Can you provide citations for the impact of surface crevasses in different situations?
- I was confused by the "(see details below)" note in line 48 and the "(described further below)" in line 56. Can specific sections or subsections be referenced to make that information more clear?
- I find it interesting that for the analysis on Pine Island Ice Shelf including only heavily damaged crevasses (Figure 3b), the relationship no longer behaves like a random classifier. I would be interested to know what attributes "heavily damaged crevasses" have that other crevasses lack in the input data. I think that some brief discussion of the input data's classification method could provide valuable context to the "heavily damaged" result.
- Is there a different well-motivated choice in sliding law or sliding law exponent (e.g. regularized Coulomb friction) that would affect the results of this study? How dependent is the spatial variation of "A" dependent on the sliding law for this study?
- Lines 365-366: You state "We find that for any threshold value, the performance of this predictor is like that of a random classifier". That seems inconsistent with the "high damage" study in the manuscript. I don't think that the "high damage" study weakens the paper's overall conclusions, but is an interesting result that perhaps merits some additional discussion.
- Lines 382-384: you mention the possibility of surface crevasses penetrating to the water line. Can you give additional context to the correlation between surface crevasses reaching the water line and inferred variation in ice rheology?
- I quite like Figure S6 as a visualization for the classification. Can a similar figure for Filchner-Ronne ice shelf be included as well?
Technical Corrections
- The acronym OPTtimal operating PoinT (OPTPT) defined in the text (lines 184 and 197) differs from the acronym used in Figures 2 and 3 (OPT-PNT), assuming the two have the same meaning like I believe they do.
- Table T2: Column 2, Row 3 has an extra "tab" space that should be deleted.
- Notation for the equation in section 1.3 of the supplementary information should be made consistent. Most of the terms use forward slashes but "Cost(P|N)" uses a vertical bar in the text.
Citation: https://doi.org/10.5194/egusphere-2023-2362-RC1 -
RC2: 'Comment on egusphere-2023-2362', Anonymous Referee #2, 18 Jan 2024
General comments:
This is a well written manuscript on an important problem in glaciology: The influence of brittle-failure damage on ice flow. I commend the authors for submitting a manuscript that is a delight to review. I had a lot of fun reviewing this and it is a fascinating study. I think that this study and manuscript make important contributions to the science and that the community will be very interested in reading this.
One key point: In the history of looking at “damage” (specifically, reflections and noise in radar data caused by crevassed zones at the edges of ice streams) in ice shelves, the shoe started out on the other foot: Signs of damage that were buried under un-damaged firn (specifically at Kamb Ice Stream, then called ice-stream C), and their extensions out on the Ross Ice Shelf were of great significance in telling the story of past change in the flow conditions. In this case, the change was the “shut down” of an ice stream (which has less press value in today’s world, but which needs to be studied, especially if there is ever going to be hope that ice-stream flow at Thwaites or Pine Island Glacier will self-limit). It might be worth touching on this point in the introduction. I believe that the original literature on it (from the late 1970’s and early 1980’s) is easy to find.
Generally speaking: the word on the “street” (I got this from an editor’s meeting of a related glaciological journal) is that use of acronyms can be an impediment to getting papers to be cited. In some ways, I think this is intuitively obvious; but apparently it is also a result of doing careful quantitative analysis with specific metrics for measuring acronyms and citations. This paper does not have a lot of acronyms, however, I still wonder if the paper would be easier to read (and thus more likely to be cited) if acronyms were minimized. The ones that I had to struggle with were: CNN, NeRD (that one appears to be a kind of subtle joke, which I like, as the word NeRD in English refers to a “smart” but slightly “dull” person), MOA, ROC, FPR, TPR, AUC, OPTPT, … This is an online journal, hence there is no cost in paper to write out the words in full. I think that the authors should consider this. The authors might additionally find it works more simply to assign actual variables to elements that are now a kind of hybrid acronym, for example: AUC-mean2009. These long, strung out variable names that incorporate an acronym make the reading of the paper a bit harder. With harder reading, there is then the possibility of fewer citations.
I notice that the analysis makes a distinction between 2019 and 2020 velocities in the AUC for Pine Island Ice Shelf (and different years for different regions). What specifically (as a reminder) is changing? Is it the detected crevassing fields or is it the velocity field?
Specific comments:
Abstract: “Wealth of research”? I’m not sure it adds precision to use the term wealth as a qualifier. Maybe another word would be more appropriate. “Wealth” appears as the first word in the discussion as well.
A minor point: I see that variables that appear in the text are italicized (as they should be), however, this italicization needs to be checked. For example, x- and y- around line 115. Ditto for the R that appears near there. A good double check just for this would be useful.
L-curve method (here the L should not be italicized, as I think that the “L” denotes a shape more than a variable). Also, I’ve never heard of this method before, so I wonder if the reference to it should appear right away. Also, I would find it helpful to possibly say in a few sentences how a user would “walk through” a problem following the L-curve method.
Something to check: numbers in the text sometimes appear in scientific notation (where there is a “times ten to the power of something”) and sometimes in digital computer notation or floating point notation, e.g., 1.3e-4… Journal style should be checked. I suppose it would be a bit pedantic to say so, since nobody even thinks about this any more: but it would be cool if every now and then people would report whether they are doing single or double precision computations (I don’t suppose there are single precision computations any more, but what the heck, I might as well bring it up).
Figure 4: I note that the vertical axes have a notation that is “ #10^n “ Is this standard notation (I usually see “x” replace the “#”)? Also, would it be better to have the scale (ten to the power of) in the axis label rather than perched on top of the axis frame?
This one is not an essential comment, and is motivated by the fact that Chris Borstad, one of the pioneers of damage mechanics in glaciology passed away in November of 2023. I see that Copernicus no longer provides an “acknowledgment” section in its articles where a dedication (if the authors were to want to make one) would normally appear. Instead, I see that Copernicus prefers to replace the general acknowledgement section with specific (and seemingly less noble) sections like “Funding”. This comment is not a criticism, just something that came to my mind (and heart).
Citation: https://doi.org/10.5194/egusphere-2023-2362-RC2 - AC3: 'Reply on RC2', Cristina Gerli, 07 Mar 2024
- AC2: 'Reply on RC1', Cristina Gerli, 07 Mar 2024
-
RC3: 'Comment on egusphere-2023-2362', Adrien Gilbert, 09 Feb 2024
This study examines whether the surface crevasse field observed from remote sensing is related to the flow rate factor derived from inverse methods constrained by surface velocity. A poor relationship is found, suggesting that surface crevasse observations may not be a good proxy for quantifying ice damage affecting ice rheology over a thickness relevant to ice flow dynamics.
The study addresses an important topic, as observational constraints on ice damage are critical for assessing ice shelf dynamics and stability in ice sheet models. The methodology is rigorous and clearly described, and the results are convincing and well presented. I recommend the paper for publication in The Cryosphere after major revisions.
General Comments
My main concern is that the inferred flow rate factor from surface velocity is consistently presented in the manuscript as the truth that the observed crevasse field should match. I think this way of presenting the study is not fair as it could be turn differently. For example, the crevasse field could be presented as the truth that the inferred flow rate should match, and one could conclude that the model does not capture the effective viscosity and stress field well due to its simplification, inappropriate physics or inaccurate ice shelf three dimensional geometry. The manuscript clearly lacks of a discussion about the model assumption and the reliability of the inferred flow rate factor. Even if the model is strongly constrained by surface velocity observations, it does not sound right to question the utility of crevasse field observations without even mentioning that the weak relationship could be due to the model lacking the relevant physics. Are we sure that the inferred flow rate factor is not affected by neglecting the elastic stress field or the depth-dependent variation of the flow rate factor or other things? The whole study could also conclude that the SSA does not capture the stress field of the ice shelf well, because the inferred value of the flow rate factor is weakly related to the observed damaged areas. This would give the opposite message to the community ...
I suggest that the authors should also consider the case where the SSA approximation is the cause of the discrepancy, and provide strong arguments if they think this cannot be the case.
Specific comments
You will find a small list of correction and specific comments embedded in the annotated PDF in attachment.
- AC1: 'Reply on RC3', Cristina Gerli, 07 Mar 2024
Peer review completion
Journal article(s) based on this preprint
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 381 | 135 | 34 | 550 | 52 | 22 | 14 |
- HTML: 381
- PDF: 135
- XML: 34
- Total: 550
- Supplement: 52
- BibTeX: 22
- EndNote: 14
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Cristina Gerli
Sebastian Rosier
Hilmar Gudmundsson
Sainan Sun
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(1228 KB) - Metadata XML
-
Supplement
(1171 KB) - BibTeX
- EndNote
- Final revised paper