the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 19 Jun 2023
Review Article: Earth observations of Melt Ponds on Sea Ice
Abstract. Melt ponds are pools of open water that form during summer on sea ice surface, playing a key role in the Arctic climate and sea ice energy budget. Due to their lower albedo, melt ponds absorb more radiation than un-ponded ice, spurring further ice melt and enhancing the positive ice-albedo feedback. This feedback is expected to increase as the Arctic ice cover is moving towards a regime where multiyear ice (MYI) is being replaced by first-year ice (FYI), which presents wider melt pond coverage as well as more energy absorption with profound consequences from an energy balance perspective. Nevertheless, the lack of knowledge or inclusion of melt pond fraction (MPF) on global climate and sea ice models, is pointed as their main source of uncertainty and disparity of predictions results. This, along with the recent conclusions on the potential of MPF for enhancing the forecasting ability of summer sea ice extent, underscores the importance of accurately obtaining large and spatiotemporal scale of MPF, across the Arctic. However, observations of melt ponds are far from adequate for the Arctic ocean and for both MYI and FYI, and on a large scale this is only possible through satellite-based Earth observations (EO). This paper provides an overview of efforts in EO remote sensing studies of melt ponds, for both optical sensors and radar-based approaches. The main algorithms used for melt pond identification and the different methods for MPF retrievals are outlined, ranging from the early traditional techniques to the increasingly prevalent use of Artificial intelligence (AI), namely machine and deep learning. The current large-scale optical-based pan-Arctic MPF datasets are intercompared along with the main advantages and disadvantages of various optical and radar data-based methods for MPF retrievals. The potential of radar, namely Synthetic Aperture Radar (SAR) technical abilities to the enhancement of reliability is analysed, since optical approaches, despite being more used, are hampered by cloud cover, spectral representativeness and resolution. Finally, current gaps in melt pond knowledge and MPF retrievals are discussed and summarised leading to the outline of further directions of research development.
- Preprint
(2470 KB) - Metadata XML
- BibTeX
- EndNote
Status: closed
- RC1: 'Comment on tc-2023-75', Anonymous Referee #1, 08 Aug 2023
-
RC2: 'Comment on tc-2023-75', Anonymous Referee #2, 22 Aug 2023
This is a review of: Earth Observations of Melt Ponds on Sea Ice. The objective of the article is to provide a comprehensive review of satellite retrievals of melt pond coverage over Arctic sea ice. Such a review article is timely given the scientific community’s growing interest in melt ponds on Arctic sea ice. The content within the article spans background information and the variety of passive and active remote sensing methods to detect melt ponds. I have several major concerns about the manuscript that will require substantial time and effort to address. The changes would fundamentally change the paper. Therefore, I recommend rejection with the opportunity for the author to resubmit.
My major concerns are the following:
Content:
- There are numerous errors and contradictory sentences throughout the article; I am concerned readers would interpret these as the truth, leading to greater misunderstanding of melt pond processes and properties. I’ve listed many of these errors below, but it is too exhaustive to list all of them.
- While I have not checked every single reference, I see an unfortunate pattern in this article of references being cited when they are unrelated to the topic of the sentence. Several particular references are cited prolifically despite their focus being a specific, niche topic. I’m also concerned about the references to one study that did not pass review and several conference presentations that lack a DOI. One of the key purposes of a review article is to provide an overview and large breadth of reliable resources. To do this, it is crucial to use references thoughtfully and appropriately so that readers will know where to go to find more information on a specific topic.
- The review is not all-encompassing of the melt pond literature. I appreciate that it may not be possible to include every melt pond paper in existence into a review article, but there are several landmark pond papers missing from the review (for example, Perovich et al., 2002; Holland et al., 2012; Petrich et al., 2012). I suggest doing a more thorough literature review to include these and other noteworthy melt pond studies. There are also several new melt pond studies from the MOSAiC campaign which would be relevant to this manuscript’s topic.
- I find the content and language describing different remote sensing methods to be biased and repetitive towards favoring SAR techniques over visible imaging techniques, favoring artificial intelligence methods over “traditional” methods, and favoring “new” over “old.” As a review article, neutral language is essential for sharing knowledge in an unbiased manner. It is also important to note that this favoritism is not well substantiated: e.g., several traditional approaches have reported 97% accuracies and higher, the oldest melt pond product has very high (1-m) resolution, and so on.
- The reporting of retrieval uncertainties is problematic and suggests that there may be one best method or product (i.e., AI-based SAR techniques). The referenced studies use different validation methods from one another, making their uncertainties an apples-to-oranges comparison. I suggest de-emphasizing their uncertainties but instead, focusing on the different sources of biases amongst products since this is relevant to better understanding strengths and weaknesses of different remote sensing techniques.
- There is inconsistent information across figures, tables, and in the text for the different products described.
Organization:
- I’m struggling to follow the organizational structure of the manuscript. The content jumps between visible and SAR background information, methods, and results. It would be helpful to restructure and streamline the content based on the remote sensing techniques (e.g., perhaps all content on SAR retrievals in one main SAR section).
- Sections 1-2 and 5 would greatly benefit from the guidance from other melt pond experts or even their co-authorship. The descriptions of physical processes in these sections are lacking, with contradicting statements, speculations stated as facts, incorrect referencing, and errors. I suggest revisiting the background information and considering what is truly needed in order to give readers a complete picture of melt ponds from a remote sensing perspective. Rather than having multiple background sections, consider combining these with Sections 3-4 according to topic (for example, Section 2.1 could be combined with the visible remote sensing section).
- Section 3.1 could be omitted, as it provides background information on remote sensing that is overly basic. Perhaps a question to ask is: who is the intended audience? The answer should help guide what level of technical information to include for background information and, more so, what kind of background information to provide (pond processes vs. remote sensing basics).
- The topics in Section 5 are too broad, and the content too thin and error-ridden. I suggest simplifying these subsections down into a single section more closely aligned to the topic of the remote sensing of melt ponds.
Detailed comments:
Somewhere, there needs to be a clear definition of the spatial scales associated with high-, medium-, and low-resolution imagery.
L7 Here and throughout the manuscript, please consider replacing “open water” with “melt water” since open water is often interpreted as leads between floes rather than melt ponds.
L9. Here and elsewhere: I suggest rewording positive ice-albedo feedback to positive albedo feedback, since the former refers to the loss of sea ice.
L10-11. It is repeated several times that melt pond coverage on first year ice is greater than that of multiyear ice. There are numerous studies showing contradictory results, which indicates this is a finding that is not solid enough to state as a fact, but should be discussed.
L13 Here and elsewhere, this statement is not true. Melt pond fraction is simulated by global climate and sea ice models.
L41. Sea ice is already parameterized in global climate models.
Figure 1. I recommend removing figure 1b as it is easily mistaken for sea ice melt ponds.
L54-57. This is unclear. Consider rewriting.
L59. Melt pond fraction can be defined in multiple ways, and it would be important to acknowledge this in the manuscript. Melt pond fraction can be the ponded ice fraction, as shown in the manuscript, or as the fraction of melt ponds within a satellite scene. Different melt pond products define this in different ways.
L60. In this sentence and other sentences: Melt ponds are not necessarily darker if they are deeper. This is why melt ponds can be light blue, but very deep, on thick ice.
L72-74. This is unclear. Consider rewriting.
L78-80. This is incorrect. Melting snow does not span a broader range of albedo values than sea ice, which spans ponded ice to bare sea ice (~0.15 to ~0.65).
L80. This is incorrect. Albedo is not an optical property.
L81-82. It is not clear how albedo can be used for visible and infrared light.
L83. Surface albedo on Arctic sea ice was not first studied in the 1960s. There are earlier studies of this going back to the 1950s.
L91. It would be important to clarify that melting sea ice often has a surface scattering layer.
L94-95. 0.4 seems too high of an albedo in the red wavelength range.
L109. What is meant by the albedo bottom? This is not common terminology.
L109-119. This paragraph is vague and confusing.
L122 “The first quantitative measurements of melt pond colour…” This is incorrect. There have been earlier studies on melt pond color before the Istomina et al. 2016 study.
L132. This is not wholly representative. Topography, microstructure, and salinity also influence melt pond evolution and distribution.
Figure 3. I don’t understand how one figure can have 4 references.
L156. Mid-August is not always associated with freeze-up.
L161-162. Here and throughout the manuscript: Melt pond coverage is not always greater on first year ice than multiyear ice.
L164. Reference needed for ice thickness estimates.
L205-206. It is misleading to state the results of study as fact. There has not been consensus amongst the melt pond studies that melt pond fractions are increasing as a result of there being more first year sea ice in the Arctic.
L241. This is incorrect. Melt ponds are much smaller than 10 m2 at onset.
L243. I don’t understand this sentence. Please consider rewriting.
L244. The first studies done on melt ponds occurred much earlier than the cited references.
L312. Dry firn doesn’t apply to the sea ice environment.
L313-317. A reference is needed here.
L403. Here and elsewhere: it is unclear what is meant by melt pond dynamics.
L409. I believe this would still be considered medium resolution.
L414. Do you mean *not* well suited here since the pixel size is greater than the size of individual melt ponds?
L428-431. I’m concerned that the biases discussed here are from only one study, and are described as if they are common biases across all products from optical medium-resolution data. Please rewrite with the consideration that the biases will differ across products.
Figure 2. This figure has errors. For example, in the text, MEDEA imagery is described as panchromatic, but in the figure, NTM has 1-15 bands listed. Sentinel-2 has n.a., but this seems wrong. This and other sensor information needs to be cross-checked. I also suggest rephrasing Digital photos and DMS as RGB imagery since this is clearer. There is Operation IceBridge imagery from 2017 that appears to be missing from the figure. I don’t know if the Operation IceBridge in 2018 is a mistake or not, but it needs to be checked.
L506. This is incorrect. Traditionally, melt pond coverage was measured in situ from the surface, not a ship or by aerial photography.
L508-511. I find these statements concerning because the reported errors are not representative of MPF retrievals from all optical data, which can be misleading.
L512. In order to be detected… this is not true. A melt pond can be smaller than the pixel size to still have an effect on the mixed pixel signature.
L517. I don’t know what traditional refers to here. Pixel unmixing was not the approach used by the earliest remote sensing studies of melt ponds.
L553-554. This is incorrect. Melt ponds do not have a higher reflectivity than bare sea ice.
L868-900. This content meanders. It’s unclear what the key takeaways are. Please consider rewriting.
Table 4 has numerous errors.
L1009-1012. I suggest not including descriptions of products that are not publicly accessible or following FAIR principles.
L1029-1030. This is unclear. Please consider rewriting.
Section 5.1 has outdated information and errors. Here and elsewhere, there is an error: melt ponds are not the main source of uncertainty in climate models, as suggested multiple times in the manuscript.
L1105-1115 are repetitive and vague. It’s unclear what the main takeaways should be.
Section 5.3 is repetitive.
L1209 “The lack of standardized surface type definitions also adds up to the challenge of quantitatively monitoring sea ice characteristics”. I would argue that the surface classes are explicitly defined in each product described thus far. They may not have a common definition, such as ponded ice + submerged ice vs. ponded ice, but these definitions are still well described in the literature.
L1221-1222. This is incorrect. Climate models do simulate melt ponds.
L1245-1246. This is incorrect. As one case example, the traditional approach involving the MEDEA imagery has a reported accuracy of 96%-99%.
There is no acknowledgements section.
Citation: https://doi.org/10.5194/tc-2023-75-RC2
Status: closed
- RC1: 'Comment on tc-2023-75', Anonymous Referee #1, 08 Aug 2023
-
RC2: 'Comment on tc-2023-75', Anonymous Referee #2, 22 Aug 2023
This is a review of: Earth Observations of Melt Ponds on Sea Ice. The objective of the article is to provide a comprehensive review of satellite retrievals of melt pond coverage over Arctic sea ice. Such a review article is timely given the scientific community’s growing interest in melt ponds on Arctic sea ice. The content within the article spans background information and the variety of passive and active remote sensing methods to detect melt ponds. I have several major concerns about the manuscript that will require substantial time and effort to address. The changes would fundamentally change the paper. Therefore, I recommend rejection with the opportunity for the author to resubmit.
My major concerns are the following:
Content:
- There are numerous errors and contradictory sentences throughout the article; I am concerned readers would interpret these as the truth, leading to greater misunderstanding of melt pond processes and properties. I’ve listed many of these errors below, but it is too exhaustive to list all of them.
- While I have not checked every single reference, I see an unfortunate pattern in this article of references being cited when they are unrelated to the topic of the sentence. Several particular references are cited prolifically despite their focus being a specific, niche topic. I’m also concerned about the references to one study that did not pass review and several conference presentations that lack a DOI. One of the key purposes of a review article is to provide an overview and large breadth of reliable resources. To do this, it is crucial to use references thoughtfully and appropriately so that readers will know where to go to find more information on a specific topic.
- The review is not all-encompassing of the melt pond literature. I appreciate that it may not be possible to include every melt pond paper in existence into a review article, but there are several landmark pond papers missing from the review (for example, Perovich et al., 2002; Holland et al., 2012; Petrich et al., 2012). I suggest doing a more thorough literature review to include these and other noteworthy melt pond studies. There are also several new melt pond studies from the MOSAiC campaign which would be relevant to this manuscript’s topic.
- I find the content and language describing different remote sensing methods to be biased and repetitive towards favoring SAR techniques over visible imaging techniques, favoring artificial intelligence methods over “traditional” methods, and favoring “new” over “old.” As a review article, neutral language is essential for sharing knowledge in an unbiased manner. It is also important to note that this favoritism is not well substantiated: e.g., several traditional approaches have reported 97% accuracies and higher, the oldest melt pond product has very high (1-m) resolution, and so on.
- The reporting of retrieval uncertainties is problematic and suggests that there may be one best method or product (i.e., AI-based SAR techniques). The referenced studies use different validation methods from one another, making their uncertainties an apples-to-oranges comparison. I suggest de-emphasizing their uncertainties but instead, focusing on the different sources of biases amongst products since this is relevant to better understanding strengths and weaknesses of different remote sensing techniques.
- There is inconsistent information across figures, tables, and in the text for the different products described.
Organization:
- I’m struggling to follow the organizational structure of the manuscript. The content jumps between visible and SAR background information, methods, and results. It would be helpful to restructure and streamline the content based on the remote sensing techniques (e.g., perhaps all content on SAR retrievals in one main SAR section).
- Sections 1-2 and 5 would greatly benefit from the guidance from other melt pond experts or even their co-authorship. The descriptions of physical processes in these sections are lacking, with contradicting statements, speculations stated as facts, incorrect referencing, and errors. I suggest revisiting the background information and considering what is truly needed in order to give readers a complete picture of melt ponds from a remote sensing perspective. Rather than having multiple background sections, consider combining these with Sections 3-4 according to topic (for example, Section 2.1 could be combined with the visible remote sensing section).
- Section 3.1 could be omitted, as it provides background information on remote sensing that is overly basic. Perhaps a question to ask is: who is the intended audience? The answer should help guide what level of technical information to include for background information and, more so, what kind of background information to provide (pond processes vs. remote sensing basics).
- The topics in Section 5 are too broad, and the content too thin and error-ridden. I suggest simplifying these subsections down into a single section more closely aligned to the topic of the remote sensing of melt ponds.
Detailed comments:
Somewhere, there needs to be a clear definition of the spatial scales associated with high-, medium-, and low-resolution imagery.
L7 Here and throughout the manuscript, please consider replacing “open water” with “melt water” since open water is often interpreted as leads between floes rather than melt ponds.
L9. Here and elsewhere: I suggest rewording positive ice-albedo feedback to positive albedo feedback, since the former refers to the loss of sea ice.
L10-11. It is repeated several times that melt pond coverage on first year ice is greater than that of multiyear ice. There are numerous studies showing contradictory results, which indicates this is a finding that is not solid enough to state as a fact, but should be discussed.
L13 Here and elsewhere, this statement is not true. Melt pond fraction is simulated by global climate and sea ice models.
L41. Sea ice is already parameterized in global climate models.
Figure 1. I recommend removing figure 1b as it is easily mistaken for sea ice melt ponds.
L54-57. This is unclear. Consider rewriting.
L59. Melt pond fraction can be defined in multiple ways, and it would be important to acknowledge this in the manuscript. Melt pond fraction can be the ponded ice fraction, as shown in the manuscript, or as the fraction of melt ponds within a satellite scene. Different melt pond products define this in different ways.
L60. In this sentence and other sentences: Melt ponds are not necessarily darker if they are deeper. This is why melt ponds can be light blue, but very deep, on thick ice.
L72-74. This is unclear. Consider rewriting.
L78-80. This is incorrect. Melting snow does not span a broader range of albedo values than sea ice, which spans ponded ice to bare sea ice (~0.15 to ~0.65).
L80. This is incorrect. Albedo is not an optical property.
L81-82. It is not clear how albedo can be used for visible and infrared light.
L83. Surface albedo on Arctic sea ice was not first studied in the 1960s. There are earlier studies of this going back to the 1950s.
L91. It would be important to clarify that melting sea ice often has a surface scattering layer.
L94-95. 0.4 seems too high of an albedo in the red wavelength range.
L109. What is meant by the albedo bottom? This is not common terminology.
L109-119. This paragraph is vague and confusing.
L122 “The first quantitative measurements of melt pond colour…” This is incorrect. There have been earlier studies on melt pond color before the Istomina et al. 2016 study.
L132. This is not wholly representative. Topography, microstructure, and salinity also influence melt pond evolution and distribution.
Figure 3. I don’t understand how one figure can have 4 references.
L156. Mid-August is not always associated with freeze-up.
L161-162. Here and throughout the manuscript: Melt pond coverage is not always greater on first year ice than multiyear ice.
L164. Reference needed for ice thickness estimates.
L205-206. It is misleading to state the results of study as fact. There has not been consensus amongst the melt pond studies that melt pond fractions are increasing as a result of there being more first year sea ice in the Arctic.
L241. This is incorrect. Melt ponds are much smaller than 10 m2 at onset.
L243. I don’t understand this sentence. Please consider rewriting.
L244. The first studies done on melt ponds occurred much earlier than the cited references.
L312. Dry firn doesn’t apply to the sea ice environment.
L313-317. A reference is needed here.
L403. Here and elsewhere: it is unclear what is meant by melt pond dynamics.
L409. I believe this would still be considered medium resolution.
L414. Do you mean *not* well suited here since the pixel size is greater than the size of individual melt ponds?
L428-431. I’m concerned that the biases discussed here are from only one study, and are described as if they are common biases across all products from optical medium-resolution data. Please rewrite with the consideration that the biases will differ across products.
Figure 2. This figure has errors. For example, in the text, MEDEA imagery is described as panchromatic, but in the figure, NTM has 1-15 bands listed. Sentinel-2 has n.a., but this seems wrong. This and other sensor information needs to be cross-checked. I also suggest rephrasing Digital photos and DMS as RGB imagery since this is clearer. There is Operation IceBridge imagery from 2017 that appears to be missing from the figure. I don’t know if the Operation IceBridge in 2018 is a mistake or not, but it needs to be checked.
L506. This is incorrect. Traditionally, melt pond coverage was measured in situ from the surface, not a ship or by aerial photography.
L508-511. I find these statements concerning because the reported errors are not representative of MPF retrievals from all optical data, which can be misleading.
L512. In order to be detected… this is not true. A melt pond can be smaller than the pixel size to still have an effect on the mixed pixel signature.
L517. I don’t know what traditional refers to here. Pixel unmixing was not the approach used by the earliest remote sensing studies of melt ponds.
L553-554. This is incorrect. Melt ponds do not have a higher reflectivity than bare sea ice.
L868-900. This content meanders. It’s unclear what the key takeaways are. Please consider rewriting.
Table 4 has numerous errors.
L1009-1012. I suggest not including descriptions of products that are not publicly accessible or following FAIR principles.
L1029-1030. This is unclear. Please consider rewriting.
Section 5.1 has outdated information and errors. Here and elsewhere, there is an error: melt ponds are not the main source of uncertainty in climate models, as suggested multiple times in the manuscript.
L1105-1115 are repetitive and vague. It’s unclear what the main takeaways should be.
Section 5.3 is repetitive.
L1209 “The lack of standardized surface type definitions also adds up to the challenge of quantitatively monitoring sea ice characteristics”. I would argue that the surface classes are explicitly defined in each product described thus far. They may not have a common definition, such as ponded ice + submerged ice vs. ponded ice, but these definitions are still well described in the literature.
L1221-1222. This is incorrect. Climate models do simulate melt ponds.
L1245-1246. This is incorrect. As one case example, the traditional approach involving the MEDEA imagery has a reported accuracy of 96%-99%.
There is no acknowledgements section.
Citation: https://doi.org/10.5194/tc-2023-75-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 433 | 374 | 16 | 823 | 19 | 17 |
- HTML: 433
- PDF: 374
- XML: 16
- Total: 823
- BibTeX: 19
- EndNote: 17
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1