the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 13 Nov 2024
Improved Arctic Melt Pond Fraction Estimation Using Sentinel-2 Imagery
Abstract. Melt ponds play a vital role in determining the Arctic energy budget by accelerating the rate of sea ice loss aided by their lower albedo. Therefore, an accurate long-term estimate of Arctic Melt Pond Fraction (MPF) is necessary to forecast summer Arctic ice-free conditions. Earth Observation (EO) satellite systems provide ideal tools to monitor the evolution of melt ponds, both spatially and temporally, in near-real time. However, the MPF estimates from these studies are affected by the presence of small, fragmented ice floes called brash ice, and submerged ice. An improved workflow is necessary to remove the effects of the aforementioned sea ice features from the MPF estimate. Here, we estimate MPF using Sentinel-2 imagery, by coupling a Random Forest (RF) model with mathematical morphological algorithms – morphological dilation and morphological reconstruction – which improves the estimate of MPF by reducing misclassifications from nilas, submerged, and brash ice. Further, we present an inter-seasonal MPF time-series from 2018 to 2021 and show that employing morphological operations after the RF reduces the mean MPF by greater than 40 %. Our results show that the MPF exhibited considerable intra- and inter-seasonal variations, with the maximum MPF reaching as high as 57 %.
- Preprint
(6582 KB) - Metadata XML
-
Supplement
(11150 KB) - BibTeX
- EndNote
Status: open (until 23 Jan 2025)
-
RC1: 'Comment on egusphere-2024-3315', Anonymous Referee #1, 13 Jan 2025
reply
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2024/egusphere-2024-3315/egusphere-2024-3315-RC1-supplement.pdf
-
RC2: 'Comment on egusphere-2024-3315', Anonymous Referee #2, 17 Jan 2025
reply
"Improved Arctic Melt Pond Fraction Estimation Using Sentinel-2 Imagery" proposes to estimate melt pond fraction with a random forest classifier followed by morphological operations. The fundamental concept of the paper, that morphology can be exploited in melt pond retrievals, is sound and the paper is generally well-written. However, there are critical deficiencies in both the execution of the study and how the findings are communicated that need to be addressed.
The first word in the title is "Improved", but the manuscript does not clearly document an improvement relative to published work. The improvement that is demonstrated is for the final retrieval (after the morphological operation) relative to an intermediate processing step (after the random forest classification). The manuscript extensively discusses Niehaus et al. (2023), which also presents a retrieval of melt pond fraction from Sentinel-2 (S-2) imagery. Presumably the authors could run their algorithm on the same S-2 data as Niehaus et al. and compare to see whether an improvement is realized. It is also not clear whether the misclassification that the manuscript documents is a general feature of random forest classifiers or the specific one used herein. For example, the manuscript does not reference the random forest classifier for melt ponds from optical imagery from Wright and Polashenski (2018), which seems like another natural comparison since Wright and Polashenski published the code for the algorithm itself. I found this article by simply searching 'melt pond random forest' in google scholar, so the omission surprises me. More generally, the validation presented in the manuscript is insufficient. The manuscript nicely summarizes that: "the influence of misclassifications may be more pronounced in certain regions than in others, depending on ice characteristics" (lines 188-189). But the validation is performed with just 2 high-resolution images, both of which were from a similar time in the season. Two validation points is insufficient to justify the use of such an algorithm pan-Arctically.
Second, the title, abstract and much of the text implies that what is being retrieved by the proposed algorithm is the 'Melt Pond Fraction'; whereas it requires a closer reading to identify that what is being estimated is the fraction of melt ponds with areas of greater than 100 m2. This is a substantial difference, as the manuscript nicely demonstrates: "Melt ponds with areas less than 100 m2 constituted 38% and 39% of the total area in the WV-3 images acquired on June 27, 2020, and July 11, 2022" (lines 255-256). This is a major bias (note that it's higher than the bias in the Webster et al. (2022) paper that is presented as motivation), and needs to be clearly communicated throughout the article, including in the title and abstract. Using MPF or Melt Pond Fraction throughout the article is likely to mislead many readers. I am unaware if there is a generally accepted terminology for melt ponds greater than a certain area, but I think defining an acronym like 'MPF>100' early in the article and then using it throughout would be effective.
Wright, N. C. and Polashenski, C. M.: Open-source algorithm for detecting sea ice surface features in high-resolution optical imagery, The Cryosphere, 12, 1307–1329, https://doi.org/10.5194/tc-12-1307-2018, 2018.
Specific points below (by line number):
8: Late in the melt season, when submerged ice is most prevalent, it is not clear to me how or why one would distinguish melt ponds near the ice edge from submerged ice. Consider a melt pond connected to the edge of the floe by a channel. If we imagine widening the channel, at some point the original pond area would stop looking like a 'pond' and start looking like 'submerged ice', but the ice in the original pond itself would not have changed nor would the impacts on the thermodynamics.
11: Concluding the abstract with this estimate of maximum melt pond fraction gives it an emphasis that is unsupported by the text. Specifically, lines 181-182 of the results state: "we acknowledge the possibility of the influence of disintegrated ice and fully melted-through melt ponds on MPF estimates during the late-melt stage." Either the abstract needs to have similar caveat or the numerical estimate should be removed from the abstract.The introduction generally would benefit from a more thorough description of the summertime sea ice cover (ponds, brash ice, nilas, etc...)
35: this oversimplifies Webster et al. 2015 and implies that it evaluated melt ponds in CICE specifically. Melt pond fraction in coupled climate simulations like CESM2 may differ from observations for reasons other than the melt pond scheme (e.g., excessive surface melt, too thin ice, etc...)
48: Define 'S-2'
61-2: It feels weird that this statement is included without having a subsequent comparison of your pan-Arctic results with Niehaus et al.
67: 'will' feels like to strong of a statement given the evidence presented. One might alternatively hypothesize that for brash ice to persist it requires strong floes to be around
70: typo? 'imageries'
129-130: if the optimal threshold is site-specific how does that impact pan-Arctic application?
153: define WV-3
162: Removing ponds that are smaller than 100m2 means that this is no longer computing a melt pond fraction!
175: This should read Fig.'3'
198: I typically think of 'flooding' as due to runoff from land or ice dynamics or snow loading depressing the ice surface below sea-level. Is that what's going on here?
214-222: This paragraph seems to tie itself into knots to avoid making a comparison with Niehaus et al. (2023). Why not just do a direct comparison with that work? Presumably their S-2 imagery is available. Also, is this work's study area landfast ice? If so that needs to be communicated more clearly earlier in the paper. If not, I don't see how that is a relevant detail. Unless the suggestion is that pan-Arctic estimates are a poor estimate of pack-ice melt pond properties? Landfast ice is such a small area fraction of the Arctic I would doubt that suggestion.
Figure 4: please use actual values for the x-axis (they can still be logarithmically spaced).
Section 4.4: This section includes a lot of percentage differences in MPF that can be a bit hard to follow. I think it would be better to put the numbers in a table and then shorten the text.
Figure 6: please use actual values for the x-axis.
272-276: the errors across size classes happened to balance out for these two cases, but I don't see why that would be guaranteed to be the case.Citation: https://doi.org/10.5194/egusphere-2024-3315-RC2
Data sets
Improved Arctic Melt Pond Fraction Estimation Using Sentinel-2 Imagery K. Sivaraj et al. https://doi.org/10.5281/zenodo.12802216
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 114 | 24 | 9 | 147 | 18 | 5 | 2 |
- HTML: 114
- PDF: 24
- XML: 9
- Total: 147
- Supplement: 18
- BibTeX: 5
- EndNote: 2
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1