the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 24 Oct 2023
Linear Ice Fraction: Sea Ice Concentration Estimates from the ICESat-2 Laser Altimeter
Abstract. Sea ice coverage is a key indicator of changes in the global climate. Estimates of sea ice area and extent are primarily derived from satellite measurements of surface microwave emissions, from which local sea ice concentration (SIC) is derived. Passive microwave (PM) satellite sensors remain the sole global product for understanding SIC variability. Using a dataset of more than 27,000 high-resolution airborne optical images, we first examine biases in commonly-used products that emerge from challenges in sampling thin sea ice fractures and melt ponds on the sea ice surface. We show that the ICESat-2 (IS2) laser altimeter effectively samples these surface features and we develop a new, independent SIC product, which we term the Linear Ice Fraction (LIF). On monthly timescales, we show using an emulator that the LIF product offers an independent estimate of sea ice concentration over 60 % of the Arctic sea ice cover with similar-or-better error qualities compared to PM data. IS2 and its high-precision measurements of the sea ice surface should be considered for augmenting PM-SIC measurements in the future.
This preprint has been withdrawn.
-
Withdrawal notice
This preprint has been withdrawn.
-
Preprint
(4078 KB)
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2023-2312', Anonymous Referee #1, 28 Nov 2023
In this study, the authors utilize data from ICESat-2 to develop a new observational estimate of sea ice concentration, aiming to validate existing estimates derived from the widely-used passive microwave sensing method. This new 'linear' concentration estimate is initially validated using imagery from the Operation IceBridge airborne campaign.General comments:The objectives of this study appear quite sensible to me, and it is evident that considerable effort has been invested in the analysis. However, I have significant reservations about the methodology, which cause me some concern, as i detail here:1. The study attempts to cover a broad range of objectives in a single paper:a. Validating ice concentration estimates from various passive microwave data products using Operation IceBridge imagery during winter and the more challenging summer period.b. Validating ice concentration estimates from ICESat-2 with Operation IceBridge imagery.c. Employing ICESat-2 to generate new ice concentration estimates and to understand linear profiling sampling errors as a function of beam crossings.d. Creating a new global gridded Sea Ice Concentration (SIC) product from ICESat-2 data.My confidence in the conclusions drawn and the progression from step a to step d is not particularly strong. The validation efforts are commendable but seem somewhat limited. It appears the approach was driven by the desire to demonstrate clear PM biases based on the OIB data, thereby necessitating ICESat-2 data, but this was not always convincingly presented. I wonder if a more effective strategy might involve using more high-resolution imagery for validation and potential bias correction of PM data, rather than introducing the complexities of ICESat-2 data? The inclusion and description of a monthly gridded ICESat-2 SIC product felt overly ambitious, particularly considering the later admission that ICESat-2 struggles to distinguish between melt ponds and leads.2. The results in Table 2, showing the difference between IS-2 and imagery compared to IS-2 and PM data (2.4 vs 2.9-4.5), are interpreted by you more optimistically than I would. Given the significantly higher resolution of IS-2 data compared to PM, these results are somewhat disappointing. The much better performance for 'Best' (1.0) is intriguing and may indicate issues with IS-2 data that need addressing, considering other uncertainties (like sampling) and potential unknowns in using altimetry data for this purpose.3. Regarding PM products overestimating concentration in winter, I believe it's important to distinguish between actual open water and re-frozen leads, and how these are represented by different sensors. The effectiveness of ICESat-2 in classifying surfaces under various conditions was not thoroughly explained, and the discussion on specular versus dark leads was confusing. It's unclear how many leads are classified as dark and whether any of these were included in the OIB or S2/WV validation.4. The explanation of PM data spanning pages 3 and 4 was difficult to follow. A more focused and detailed discussion on PM data, its production, and a comprehensive account of its uncertainties and biases would have enhanced the paper.5. I was surprised that the imagery comparisons did not include any potential drift correction or optimal correlation approaches. Even a minor shift (the size of a lead) could result in misclassification in a given scene. This issue seemed to be dismissed too easily.Specific Comments:L40: Sub-meter scale? At L129, you discuss segment lengths of 15 m and 60 m. Could you clarify this? I am somewhat perplexed about the resolution and what can realistically be resolved with ICESat-2.L40: Could you elaborate on what you mean by 'snagging' in this context? I believe it still plays a role, does it not?L58: Would you mind adding a reference for DMS imagery and providing more details? I am not overly familiar with this data.L60: The comparison between OIB and PM is not entirely clear to me. Do you conduct this comparison for each OIB scene, even when multiple scenes span a single PM scene? This approach seems somewhat unconventional. Also, when focusing solely on the 'lead' scenes, wouldn't the locations be effectively identical?Figure 2: This figure is somewhat challenging to interpret. Have you considered presenting it as a violin plot for better clarity?L127: Regarding the use of dark leads, I was under the impression that these are no longer employed in determining SSH in ICESat-2 sea ice products. Could you clarify?L129: The sentence here is confusing. How does segment length correlate with resolution?Figure 3: If I understand correctly, it is quite surprising that in several instances, the difference between the SIC from the 'best' profile and the imagery SIC is zero. Does this imply that the linear beams are accurately capturing the entire scene? For instance, in Figure 3, the IS2best mean concentration in that scene is 97.5, identical to the mean concentration from the image.L140: For S-2/WV, it appears the scenes are significantly larger than those of OIB, thus offering comparable resolutions. Could you possibly illustrate the PM box in the regions shown in Figure 3?Figure 4: I found this figure quite complex to decipher. I would strongly recommend enhancing the caption and layout for better readability. The transition between OIB and WV, as well as the mixture of legends and plain text in the plots (particularly the PM data), was somewhat disorienting.Figure 4: Why does the Crossing Number start at 0? Should it not be 1?Figure 4: Regarding the initialization of the line in Figure 4b, it's somewhat unexpected how accurate the first one appears, especially considering that it seems plausible to find a crossing without a lead. Could you explain this?Figure 4: So, the cumulative SIC converges towards an accurate value as you continue averaging, but doesn't this also apply to the PM datasets if averaged? In practice, would you not be taking just 1 or 3 crossings and using that raw value? In such a case, might the error be similar to that of the PM data, or am I misunderstanding something?Figure 4: The explanation of 'winter bias' is a bit unclear to me.Figure: Why opt for using the median error?L201: What exactly do you mean by 'high precision'?L215: Are you suggesting that if the concentration on the day IS-2 passes over a grid-cell matches the monthly mean concentration, you assume there are no sampling biases in that grid-cell at that time? This seems like an unusual approach, especially if you believe there are biases in the SIC data that need addressing.Figure 5: Referencing my main points above, I remain unconvinced about the value of this, given my significant concerns regarding the interpretation of the 'validation' efforts.L242-244: "Due to the absence of surface meltwater, these non-specular returns are less likely to be contaminated by misclassification error in winter, potentially indicating a true PM-SIC overestimation bias." Could you elaborate on this? What about the influence of clouds?Citation: https://doi.org/
10.5194/egusphere-2023-2312-RC1 - AC1: 'Reply on RC1', Christopher Horvat, 12 Feb 2024
-
RC2: 'Comment on egusphere-2023-2312', Anonymous Referee #2, 06 Dec 2023
In this manuscript the authors develop a Linear Ice Fraction (LIF) product from the ICESat-2 satellite ATL07 sea ice height product. The LIF product is designed to be independent of the Passive Microwave (PM) products currently available, and preliminary analysis shows comparable results to those from PM with the benefit of similar-or-better error qualities. It’s really exciting to see people thinking outside the box for ICESat-2 and developing new and much-needed applications for the data. However, I have some significant issues with the methodology that should be addressed before publication.
Those significant issues are outlined below, followed by some Specific and Technical Comments.
The comparison of OIB imagery to various SIC products (Section 2) was unconvincing
I appreciate the author’s wanted to justify the need for a non-PM SIC product, especially during summer months. But the vastly different scales of the comparison limit its effectiveness. If this section is to remain in the manuscript, more analysis is needed on the spatial variability of SIC on the two different scales, to confirm that such a comparison is meaningful.
Emulator design
The emulator is developed by randomly intersecting imagery with straight lines, at various intersection angles. This is not representative of how ICESat-2 beam geometry actually samples the sea ice surface. The angle of crossovers is surely a key consideration due to lead geometry – most regions have a typical lead orientation (e.g. Brohan and Kaleschke 2014), meaning leads will be somewhat consistently aligned along or across a given ICESat-2 track in a given region, rather than the random alignment depicted by the emulator. The first paragraph of Section 3.1 refers to “the orientation errors discussed in Sec. 2.3.” but I can only see a brief mention of biases associated with lead orientation in the first sentence of Section 2.3. And nothing was discussed regarding the orientation of crossovers and how representative (or not) this makes the emulator.
Not fully considering the implications of including dark leads from ICESat-2 in LIF calculation
The assumption that new/gray ice is considered ice for LIF calculations (L147-148) is too simplistic. Although the author’s reference the Petty et al. 2021 paper, they fail to consider that new/gray ice can be falsely classified as a dark lead in ICESat-2 data. New/gray ice being treated as open ocean will lower the LIF calculated from ICESat-2, and is therefore a critical consideration for the results and comparisons presented in the manuscript. I was disappointed not to see any discussion on this. How consistently is new/gray ice classified as ice rather than open ocean in ATL07? Have the authors verified this with multiple scenes? Do they expect any significant implications of incorrectly classifying new/gray ice as a lead in LIF calculation? The inclusion of an LIF (specular) product does not address the intricacies of the problem.
Specific comments
- L40: I don’t think it’s technically true that ICESat-2 can resolve leads at the sub-meter scale. The ICESat-2 footprint is stated to be 10 m in the manuscript, and the resolution of the ATL07 data is even more coarse, which is the value that’s applicable for this manuscript.
- L60: Considering the large number of scenes analyzed, what fraction are visually validated? Based on that, can this really be considered a “validation”?
- L65: It isn’t clear what’s meant by “equal to the maximum of the Bootstrap and NASATeam algorithms”
- Section 2, in general, was difficult to follow. For example, there’s a lot of jumping around between tenses when explaining the method. L61-73 especially would confuse someone who’s not very familiar with each product. I’d suggest re-wording how the products are introduced.
- L90: I’m not sure “worst-performing” is the best phrase here. At present there’s not enough in the text to convince me it’s the worst-performing, based purely on having the greatest differences with OIB.
- L32: I’d suggest also considering data quality flags
- Figure 3: I assume the 3 ICESat-2 lines show just the strong beams? State this in the figure caption, considering the analysis is for all 6 beams.
- L160: Do you mean April 7, 2022 rather than May?
- I’m struggling with the description of the product as “global”, when in the Arctic it only covers 25-65% of the sea ice zone
- State what release of the ICESat-2 ATL07 data you’re using. If you’re not using release 6 (the latest version of the data), please explain the reason for that.
Technical comments
- L48: Change ICESat-2 to IS2
- L176: Change if to is
- L58: Define DMS
- L61: Acronym has not been defined yet
- L104: Random italic font for “overestimate”
- L106: Change unit to integer
- L123: IS2 has already been defined. Check throughout.
- L218: Change cell to cells
Refs
Bröhan, D.; Kaleschke, L. A Nine-Year Climatology of Arctic Sea Ice Lead Orientation and Frequency from AMSR-E. Remote Sens. 2014, 6, 1451-1475. https://doi.org/10.3390/rs6021451
Citation: https://doi.org/10.5194/egusphere-2023-2312-RC2 - AC1: 'Reply on RC1', Christopher Horvat, 12 Feb 2024
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2023-2312', Anonymous Referee #1, 28 Nov 2023
In this study, the authors utilize data from ICESat-2 to develop a new observational estimate of sea ice concentration, aiming to validate existing estimates derived from the widely-used passive microwave sensing method. This new 'linear' concentration estimate is initially validated using imagery from the Operation IceBridge airborne campaign.General comments:The objectives of this study appear quite sensible to me, and it is evident that considerable effort has been invested in the analysis. However, I have significant reservations about the methodology, which cause me some concern, as i detail here:1. The study attempts to cover a broad range of objectives in a single paper:a. Validating ice concentration estimates from various passive microwave data products using Operation IceBridge imagery during winter and the more challenging summer period.b. Validating ice concentration estimates from ICESat-2 with Operation IceBridge imagery.c. Employing ICESat-2 to generate new ice concentration estimates and to understand linear profiling sampling errors as a function of beam crossings.d. Creating a new global gridded Sea Ice Concentration (SIC) product from ICESat-2 data.My confidence in the conclusions drawn and the progression from step a to step d is not particularly strong. The validation efforts are commendable but seem somewhat limited. It appears the approach was driven by the desire to demonstrate clear PM biases based on the OIB data, thereby necessitating ICESat-2 data, but this was not always convincingly presented. I wonder if a more effective strategy might involve using more high-resolution imagery for validation and potential bias correction of PM data, rather than introducing the complexities of ICESat-2 data? The inclusion and description of a monthly gridded ICESat-2 SIC product felt overly ambitious, particularly considering the later admission that ICESat-2 struggles to distinguish between melt ponds and leads.2. The results in Table 2, showing the difference between IS-2 and imagery compared to IS-2 and PM data (2.4 vs 2.9-4.5), are interpreted by you more optimistically than I would. Given the significantly higher resolution of IS-2 data compared to PM, these results are somewhat disappointing. The much better performance for 'Best' (1.0) is intriguing and may indicate issues with IS-2 data that need addressing, considering other uncertainties (like sampling) and potential unknowns in using altimetry data for this purpose.3. Regarding PM products overestimating concentration in winter, I believe it's important to distinguish between actual open water and re-frozen leads, and how these are represented by different sensors. The effectiveness of ICESat-2 in classifying surfaces under various conditions was not thoroughly explained, and the discussion on specular versus dark leads was confusing. It's unclear how many leads are classified as dark and whether any of these were included in the OIB or S2/WV validation.4. The explanation of PM data spanning pages 3 and 4 was difficult to follow. A more focused and detailed discussion on PM data, its production, and a comprehensive account of its uncertainties and biases would have enhanced the paper.5. I was surprised that the imagery comparisons did not include any potential drift correction or optimal correlation approaches. Even a minor shift (the size of a lead) could result in misclassification in a given scene. This issue seemed to be dismissed too easily.Specific Comments:L40: Sub-meter scale? At L129, you discuss segment lengths of 15 m and 60 m. Could you clarify this? I am somewhat perplexed about the resolution and what can realistically be resolved with ICESat-2.L40: Could you elaborate on what you mean by 'snagging' in this context? I believe it still plays a role, does it not?L58: Would you mind adding a reference for DMS imagery and providing more details? I am not overly familiar with this data.L60: The comparison between OIB and PM is not entirely clear to me. Do you conduct this comparison for each OIB scene, even when multiple scenes span a single PM scene? This approach seems somewhat unconventional. Also, when focusing solely on the 'lead' scenes, wouldn't the locations be effectively identical?Figure 2: This figure is somewhat challenging to interpret. Have you considered presenting it as a violin plot for better clarity?L127: Regarding the use of dark leads, I was under the impression that these are no longer employed in determining SSH in ICESat-2 sea ice products. Could you clarify?L129: The sentence here is confusing. How does segment length correlate with resolution?Figure 3: If I understand correctly, it is quite surprising that in several instances, the difference between the SIC from the 'best' profile and the imagery SIC is zero. Does this imply that the linear beams are accurately capturing the entire scene? For instance, in Figure 3, the IS2best mean concentration in that scene is 97.5, identical to the mean concentration from the image.L140: For S-2/WV, it appears the scenes are significantly larger than those of OIB, thus offering comparable resolutions. Could you possibly illustrate the PM box in the regions shown in Figure 3?Figure 4: I found this figure quite complex to decipher. I would strongly recommend enhancing the caption and layout for better readability. The transition between OIB and WV, as well as the mixture of legends and plain text in the plots (particularly the PM data), was somewhat disorienting.Figure 4: Why does the Crossing Number start at 0? Should it not be 1?Figure 4: Regarding the initialization of the line in Figure 4b, it's somewhat unexpected how accurate the first one appears, especially considering that it seems plausible to find a crossing without a lead. Could you explain this?Figure 4: So, the cumulative SIC converges towards an accurate value as you continue averaging, but doesn't this also apply to the PM datasets if averaged? In practice, would you not be taking just 1 or 3 crossings and using that raw value? In such a case, might the error be similar to that of the PM data, or am I misunderstanding something?Figure 4: The explanation of 'winter bias' is a bit unclear to me.Figure: Why opt for using the median error?L201: What exactly do you mean by 'high precision'?L215: Are you suggesting that if the concentration on the day IS-2 passes over a grid-cell matches the monthly mean concentration, you assume there are no sampling biases in that grid-cell at that time? This seems like an unusual approach, especially if you believe there are biases in the SIC data that need addressing.Figure 5: Referencing my main points above, I remain unconvinced about the value of this, given my significant concerns regarding the interpretation of the 'validation' efforts.L242-244: "Due to the absence of surface meltwater, these non-specular returns are less likely to be contaminated by misclassification error in winter, potentially indicating a true PM-SIC overestimation bias." Could you elaborate on this? What about the influence of clouds?Citation: https://doi.org/
10.5194/egusphere-2023-2312-RC1 - AC1: 'Reply on RC1', Christopher Horvat, 12 Feb 2024
-
RC2: 'Comment on egusphere-2023-2312', Anonymous Referee #2, 06 Dec 2023
In this manuscript the authors develop a Linear Ice Fraction (LIF) product from the ICESat-2 satellite ATL07 sea ice height product. The LIF product is designed to be independent of the Passive Microwave (PM) products currently available, and preliminary analysis shows comparable results to those from PM with the benefit of similar-or-better error qualities. It’s really exciting to see people thinking outside the box for ICESat-2 and developing new and much-needed applications for the data. However, I have some significant issues with the methodology that should be addressed before publication.
Those significant issues are outlined below, followed by some Specific and Technical Comments.
The comparison of OIB imagery to various SIC products (Section 2) was unconvincing
I appreciate the author’s wanted to justify the need for a non-PM SIC product, especially during summer months. But the vastly different scales of the comparison limit its effectiveness. If this section is to remain in the manuscript, more analysis is needed on the spatial variability of SIC on the two different scales, to confirm that such a comparison is meaningful.
Emulator design
The emulator is developed by randomly intersecting imagery with straight lines, at various intersection angles. This is not representative of how ICESat-2 beam geometry actually samples the sea ice surface. The angle of crossovers is surely a key consideration due to lead geometry – most regions have a typical lead orientation (e.g. Brohan and Kaleschke 2014), meaning leads will be somewhat consistently aligned along or across a given ICESat-2 track in a given region, rather than the random alignment depicted by the emulator. The first paragraph of Section 3.1 refers to “the orientation errors discussed in Sec. 2.3.” but I can only see a brief mention of biases associated with lead orientation in the first sentence of Section 2.3. And nothing was discussed regarding the orientation of crossovers and how representative (or not) this makes the emulator.
Not fully considering the implications of including dark leads from ICESat-2 in LIF calculation
The assumption that new/gray ice is considered ice for LIF calculations (L147-148) is too simplistic. Although the author’s reference the Petty et al. 2021 paper, they fail to consider that new/gray ice can be falsely classified as a dark lead in ICESat-2 data. New/gray ice being treated as open ocean will lower the LIF calculated from ICESat-2, and is therefore a critical consideration for the results and comparisons presented in the manuscript. I was disappointed not to see any discussion on this. How consistently is new/gray ice classified as ice rather than open ocean in ATL07? Have the authors verified this with multiple scenes? Do they expect any significant implications of incorrectly classifying new/gray ice as a lead in LIF calculation? The inclusion of an LIF (specular) product does not address the intricacies of the problem.
Specific comments
- L40: I don’t think it’s technically true that ICESat-2 can resolve leads at the sub-meter scale. The ICESat-2 footprint is stated to be 10 m in the manuscript, and the resolution of the ATL07 data is even more coarse, which is the value that’s applicable for this manuscript.
- L60: Considering the large number of scenes analyzed, what fraction are visually validated? Based on that, can this really be considered a “validation”?
- L65: It isn’t clear what’s meant by “equal to the maximum of the Bootstrap and NASATeam algorithms”
- Section 2, in general, was difficult to follow. For example, there’s a lot of jumping around between tenses when explaining the method. L61-73 especially would confuse someone who’s not very familiar with each product. I’d suggest re-wording how the products are introduced.
- L90: I’m not sure “worst-performing” is the best phrase here. At present there’s not enough in the text to convince me it’s the worst-performing, based purely on having the greatest differences with OIB.
- L32: I’d suggest also considering data quality flags
- Figure 3: I assume the 3 ICESat-2 lines show just the strong beams? State this in the figure caption, considering the analysis is for all 6 beams.
- L160: Do you mean April 7, 2022 rather than May?
- I’m struggling with the description of the product as “global”, when in the Arctic it only covers 25-65% of the sea ice zone
- State what release of the ICESat-2 ATL07 data you’re using. If you’re not using release 6 (the latest version of the data), please explain the reason for that.
Technical comments
- L48: Change ICESat-2 to IS2
- L176: Change if to is
- L58: Define DMS
- L61: Acronym has not been defined yet
- L104: Random italic font for “overestimate”
- L106: Change unit to integer
- L123: IS2 has already been defined. Check throughout.
- L218: Change cell to cells
Refs
Bröhan, D.; Kaleschke, L. A Nine-Year Climatology of Arctic Sea Ice Lead Orientation and Frequency from AMSR-E. Remote Sens. 2014, 6, 1451-1475. https://doi.org/10.3390/rs6021451
Citation: https://doi.org/10.5194/egusphere-2023-2312-RC2 - AC1: 'Reply on RC1', Christopher Horvat, 12 Feb 2024
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 270 | 122 | 24 | 416 | 15 | 11 |
- HTML: 270
- PDF: 122
- XML: 24
- Total: 416
- BibTeX: 15
- EndNote: 11
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1