the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 21 Jul 2025
Subseasonal Forecast Improvements from Sea Ice Concentration Data Assimilation in the Antarctic
Abstract. This study evaluates the impact of sea ice concentration (SIC) data assimilation (DA) on subseasonal forecasts of Antarctic sea ice by comparing reforecast experiment suites initialized from two sets of initial conditions (ICs): one with SIC DA and the other without. The two ICs are evaluated against NSIDC SIC observations. Results show that the SIC DA significantly improves the climatology and interannual variability of the SIC IC. The improvement in sea ice ICs is more considerable in the Antarctic than in the Arctic. The sea ice thickness (SIT) field is mostly thinner after SIC DA except for the interior Weddell and Ross sectors. The results from reforecast experiments show that SIC DA improves the subseasonal forecast skill of Antarctic SIC in almost all initialization months except December and January, where the initial improvement is soon overtaken by the bias likely linked to the thin SIT bias. We also demonstrate that SIC DA improves the probabilistic prediction of the sea ice edge position at subseasonal time scales. The subseasonal reforecast skill of Antarctic SIC and the sea ice edge is improved the most in spring, followed by winter and summer, and has minor differences in autumn. The skill improvement associated with SIC DA is more significant in the Antarctic than the Arctic, consistent with the IC improvement. Our study demonstrates the critical role of SIC DA in the subseasonal prediction of Antarctic sea ice.
-
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
(18859 KB)
-
Supplement
(14629 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(18859 KB) - Metadata XML
-
Supplement
(14629 KB) - BibTeX
- EndNote
- Final revised paper
Journal article(s) based on this preprint
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2025-2807', Kenneth Hughes, 06 Aug 2025
-
AC2: 'Reply on RC1', Yong-Fei Zhang, 03 Nov 2025
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2025/egusphere-2025-2807/egusphere-2025-2807-AC2-supplement.pdf
-
AC2: 'Reply on RC1', Yong-Fei Zhang, 03 Nov 2025
-
RC2: 'Comment on egusphere-2025-2807', Anonymous Referee #2, 01 Sep 2025
In this study, the authors investigated the impact of assimilating SIC data on improving sea ice initial conditions (ICs) and prediction skill. Their results demonstrated that SIC assimilation generally improves both the IC and subsequent predictions, although the improvement exhibits strong regional and seasonal dependence. A slight decrease in skill occurs in the Weddell Sea and the Ross Sea for prediction initialized from December and January. The possible reasons accounting for this were also discussed. In addition, the authors compared the influence of SIC assimilation between two polar regions and concluded that SIC assimilation has a larger impact in the Antarctic than in the Arctic, which is of broad scientific interest to many researchers.
Overall, I found this paper is well organized, clearly expressed and addresses a cutting-edge research topic. I only have some minor comments and suggestions listed as below. I recommend minor revision for this manuscript.
Specific Comments:
1) Line 45, a recent study also demonstrated that SIT is a strong source of predictability for summer sea ice in the Weddell Sea, and it can be well constrained through atmospheric initialization.
Reference: Xiu et al. Impact of Ocean, Sea Ice or Atmosphere Initialization on Seasonal Prediction of Regional Antarctic Sea Ice. Journal of Advances in Modeling Earth Systems 17, e2024MS004382 (2025)
2) Line 60, please correct “C2S” to “C3S”.
3) Section 2.2, please consider to add some descriptions on how other variables (e.g., SIT) adjust in response to SIC DA since SIT is later used for analysis.
4) Line 103, (Bushuk et al., 2021).
5) Section 2.3, what is the frequency of the hindcast?
6) Line 123, it could be misleading to claim that GIOMAS can reasonably represent the Antarctic sea ice thickness climatology, as it shows large discrepancy with the satellite-based observation (e.g., Figure 3 in Liao et al. (2022)).
7) Lines 140-141, just for clarification: do you first compute the ensemble-mean SIC and then calculate SIE from the ensemble mean?
8) Lines 165-166 and lines 180-185, do you have any hypotheses on the seasonally dependent improvements from DA?
9) Lines 221, I think the ‘negative SIC biases’ only hold for the spring, while in other seasons, the positive biases dominate (Figure 2).
10) Lines 238-239, Figure 1 suggests that some regions are ice free in summer. Considering this, where does the SIC anomaly in the summer originate, especially for the western Antarctic? Can you add a spatial pattern of detrended ACC to support this claim?
11) Lines 241-242, It is somehow unexpected that the Damped Persistence is weaker in the Arctic than in the Antarctic because the Arctic SIT is thicker than Antarctic counterpart overall, as also mentioned in Lines 38-42. Is this conclusion strongly dependent on the assessment metric used?
12) Lines 247-248, Please specify the season and lead times more precisely. I think it should write as ‘Figure 8 also shows that SPEAR is generally less skillful in the Antarctic than the Arctic in autumn and winter, suggesting a larger room for correction in the Antarctic. In contrast, SICDA shows slightly better skill in the Antarctic than in the Arctic in the first two weeks in winter and summer.’
13) Lines 297-299, The statement “The co-location of an initial negative SIC bias and the faster decay in skill for SICDA suggests that the SIC-based sea ice albedo plays a role in exacerbating the low SIC bias” maybe need to be further clairfied. Specifically, what is the role of model error v.s. initial error in degrading prediction skill. For example, I notice that the SPEAR has no negative bias along the Weddell Sea coast (Figure 12g), but its prediction bias turns out negative. This obviously can’t be solely explained by the ice albedo feedback. So I’m wondering how model error versus initial error contributes to skill degradation?
Comments on Figures:
Figure 5, please add the units to y-axis label
Figure 8, SICDA
Citation: https://doi.org/10.5194/egusphere-2025-2807-RC2 -
AC1: 'Reply on RC2', Yong-Fei Zhang, 03 Nov 2025
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2025/egusphere-2025-2807/egusphere-2025-2807-AC1-supplement.pdf
-
AC1: 'Reply on RC2', Yong-Fei Zhang, 03 Nov 2025
Peer review completion
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2025-2807', Kenneth Hughes, 06 Aug 2025
This paper deals with applied problem of subseasonal prediction of Antarctic sea ice concentration/extent/volume. It looks in detail at two reforecast experiments, one with and one without data assimliation, and compares this to an observational sea ice record (plus a few other reference datasets). Results are discussed in terms of seasons and sectors of the Southern Ocean. A comparison to similar existing studies for the Arctic show a few interesting differences. For example, that sea ice predictability as a function of initialization season is not the same in the Antarctic as it is in the Arctic.
I don't know enough about the immediately adjacent literature to comment on the novelty or need for this study, but the paper seems solid. With a few exceptions (my comments below), the paper is organized well, the writing is easy to follow, the figures are clear, and the length seems appropriate.
Detailed comments:
Defining the two main datasets. This paper is focused on SPEAR vs SICDA, but this isn't made explicit early enough. It only started to become clear as I made my way through Section 2. The last paragraph of Section 1 hints at this comparison being important. And the end of Section 2.3 reiterates it a bit. But I'd suggest being more explicit. One option would be to move a couple of sentences from end of Sections 1 and 2.3 to the start of Section 2 (before Section 2.1 even). Make it very clear in one place. Another option would be to add a Table or Figure that summarizes Section 2 (what models will be used, what quantities will be compared, etc). Although it wouldn't add anything new, this Table/Figure would be a quick way for the reader to refer back and check the details quickly.
Add sector lines to all map figures. Figures 2, 3, 4, 6, 7, 11, and 12 would benefit from having five lines radiating out from the south pole that separate the maps into the five sectors defined in Section 2.6. To make sense of Section 3, I found that I had to draw these lines on myself (at least onto the top left map in each figure). Consider labeling the sectors in Figure 2a, then just repeating the lines thereafter.
The paper uses a small number of acronyms, but uses them a lot. I'm not sure it's a problem that has a good solution, but it does make a lot of the sentences stilted. A couple of egregious cases are the definition of EAFK on L97 (which is never subsequently used) and DART on L67 and L96 (which is only used twice, but defined in full both times). Also, avoid acronyms in headings (subsections of 3.1). Just write out each phrase in full.
Vague quantification in the Abstract: There is an overuse of vague adjectives in this part, rather than concrete numbers. Examples include 'significantly' (L5), 'considerable' (L6), 'mostly' (L6), 'improved the most' vs 'minor differences' (L11), 'more significant' (L12). Since many readers only interact with papers via the abstract, make sure to include at least some key metrics.
There is also some vague quantification in other parts as well:
- 'skillful predictions' at L48 (how skillful?)
- 'improved' at L49 (by how much?)
- 'outperformed' at L64 (by how much?)
- 'reasonably represent' at L122
- 'underestimate' at L124 (by how much)
- 'noticeably' at L16415. 'Steady increase' is not the correct description of Antarctic sea ice from late 70s to 2015. It didn't decrease over this period, but it wasn't clearly increasing, and it certainly wasn't doing so steadily.
26, 30. Unclear what 'perfect' means here.
161. Define the months for each season earlier in Section 3. You do define these later (at L169), but at least for L161-167, I was left wondering exactly what, say, 'summer' corresponded to.
194. Section 3.1.2 seems just as much about Thickness as it is about Volume. Perhaps this should be reflected in the heading.
197. 'thickest' seems like a weird adjective for describing sea ice volume.
215. Replace 'from the West Weddell Sea to the East Weddell Sea' with just 'across the Weddell sector'?
243. 'first month' seems a weird description. To me, 5–10 days would be a more appropriate description. Yes, some cases extend beyond 10 days, but a month is too much of an upper bound to be relevant for this sentence, right?
261. If I'm reading this sentence right, you're inferring that SICDA has the 'most considerable skill improvement' because it is better than SPEAR. But you're only comparing two cases. If so, the word 'most' is inappropriate as it should be reserved for comparing three or more cases.
282. Not clear what 'This' refers to. Is it the underprediction in Dec–Jan? Or is it referring to the merging of skill values (from an earlier sentence).
303. Unclear how Figure 12j has 'no noticeable bias'. Are you implying that, say, the 0.07 average value is ≈0?
310-317. The last paragraph before the Conclusion only cites results that are shown in the Supplementary material. Why? One would think the last paragraph of a 'Results and Discussion' section is a good place for a key result. But relegating it to the Supplementary gives the opposite impression. Is there a 12-figure limit for this journal, so you decided putting the figure in the Supplementary was a good workaround? If so, it isn't.
377-379. Remove the last three sentences. They're a repetitive and weak way to finish the paper. Finish with what you showed, not what didn't work.
Figures
-------Figures 1 and 5: set the bottom axis at exactly zero
Figures 5 and 6: what is sea ice volume measured in? Is it actually a thickness?
Figures 1 and 5: remove redundant decimal places in y axes (e.g., 6.00, 4.00, 2.00 → 6, 4, 2)
Figure 3:
- Why is the colormap so saturated? Increase color limit to 0.5, say?
- Redefine the acronym ACC here, since it isn't an intuitive acronym.Figure 8:
- Add x label ('Time (days)'?) directly to panel e, rather than just stating it in the caption.
- Make the 'DP' line for the Arctic a black dashed line. This would follow the pattern of the red and blue lines.
- Then consider making the legend two columns by three rows to make it really explicit how the line color/style system works.
- Consider replacing 'red dot' and 'red circle' with 'filled red circle' and 'open red circle'? since the open circles aren't obvious. I couldn't initially figure out how a 'dot' was different to a 'circle'.
- Remove the '(red dashed lines)', '(red solid lines)' etc from the caption, which aren't necessary since these details are in the legend already.Figure 9: Same as for Figure 8
Figure 5 error bars:
The error bars for this figure aren't appropriate. Each error bar is derived from five values. You've chosen to use ±2 standard deviations. But datasets of five are awfully small for standard deviation to be meaningful. Further, ±2 standard deviations is typically taken to be approximately equal to a 95% confidence interval (1.96 standard deviations). Of course, with only five values, you can't estimate a confidence interval that well. Could you solve all this by just using the minima and maxima of the five cases for the error bars?Figure 10:
- Different colored lines are unnecessary here since it's obvious which months are relevant from where the lines begin.
- The caption is incorrect and appears to have been copy/pasted from a different figure.Figure S7: Same as for first comment for Figure 10.
Typos:
330: 'from' is repeated
Captions of Figures 7 and S6: 'forcast'
Citation: https://doi.org/10.5194/egusphere-2025-2807-RC1 -
AC2: 'Reply on RC1', Yong-Fei Zhang, 03 Nov 2025
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2025/egusphere-2025-2807/egusphere-2025-2807-AC2-supplement.pdf
-
AC2: 'Reply on RC1', Yong-Fei Zhang, 03 Nov 2025
-
RC2: 'Comment on egusphere-2025-2807', Anonymous Referee #2, 01 Sep 2025
In this study, the authors investigated the impact of assimilating SIC data on improving sea ice initial conditions (ICs) and prediction skill. Their results demonstrated that SIC assimilation generally improves both the IC and subsequent predictions, although the improvement exhibits strong regional and seasonal dependence. A slight decrease in skill occurs in the Weddell Sea and the Ross Sea for prediction initialized from December and January. The possible reasons accounting for this were also discussed. In addition, the authors compared the influence of SIC assimilation between two polar regions and concluded that SIC assimilation has a larger impact in the Antarctic than in the Arctic, which is of broad scientific interest to many researchers.
Overall, I found this paper is well organized, clearly expressed and addresses a cutting-edge research topic. I only have some minor comments and suggestions listed as below. I recommend minor revision for this manuscript.
Specific Comments:
1) Line 45, a recent study also demonstrated that SIT is a strong source of predictability for summer sea ice in the Weddell Sea, and it can be well constrained through atmospheric initialization.
Reference: Xiu et al. Impact of Ocean, Sea Ice or Atmosphere Initialization on Seasonal Prediction of Regional Antarctic Sea Ice. Journal of Advances in Modeling Earth Systems 17, e2024MS004382 (2025)
2) Line 60, please correct “C2S” to “C3S”.
3) Section 2.2, please consider to add some descriptions on how other variables (e.g., SIT) adjust in response to SIC DA since SIT is later used for analysis.
4) Line 103, (Bushuk et al., 2021).
5) Section 2.3, what is the frequency of the hindcast?
6) Line 123, it could be misleading to claim that GIOMAS can reasonably represent the Antarctic sea ice thickness climatology, as it shows large discrepancy with the satellite-based observation (e.g., Figure 3 in Liao et al. (2022)).
7) Lines 140-141, just for clarification: do you first compute the ensemble-mean SIC and then calculate SIE from the ensemble mean?
8) Lines 165-166 and lines 180-185, do you have any hypotheses on the seasonally dependent improvements from DA?
9) Lines 221, I think the ‘negative SIC biases’ only hold for the spring, while in other seasons, the positive biases dominate (Figure 2).
10) Lines 238-239, Figure 1 suggests that some regions are ice free in summer. Considering this, where does the SIC anomaly in the summer originate, especially for the western Antarctic? Can you add a spatial pattern of detrended ACC to support this claim?
11) Lines 241-242, It is somehow unexpected that the Damped Persistence is weaker in the Arctic than in the Antarctic because the Arctic SIT is thicker than Antarctic counterpart overall, as also mentioned in Lines 38-42. Is this conclusion strongly dependent on the assessment metric used?
12) Lines 247-248, Please specify the season and lead times more precisely. I think it should write as ‘Figure 8 also shows that SPEAR is generally less skillful in the Antarctic than the Arctic in autumn and winter, suggesting a larger room for correction in the Antarctic. In contrast, SICDA shows slightly better skill in the Antarctic than in the Arctic in the first two weeks in winter and summer.’
13) Lines 297-299, The statement “The co-location of an initial negative SIC bias and the faster decay in skill for SICDA suggests that the SIC-based sea ice albedo plays a role in exacerbating the low SIC bias” maybe need to be further clairfied. Specifically, what is the role of model error v.s. initial error in degrading prediction skill. For example, I notice that the SPEAR has no negative bias along the Weddell Sea coast (Figure 12g), but its prediction bias turns out negative. This obviously can’t be solely explained by the ice albedo feedback. So I’m wondering how model error versus initial error contributes to skill degradation?
Comments on Figures:
Figure 5, please add the units to y-axis label
Figure 8, SICDA
Citation: https://doi.org/10.5194/egusphere-2025-2807-RC2 -
AC1: 'Reply on RC2', Yong-Fei Zhang, 03 Nov 2025
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2025/egusphere-2025-2807/egusphere-2025-2807-AC1-supplement.pdf
-
AC1: 'Reply on RC2', Yong-Fei Zhang, 03 Nov 2025
Peer review completion
Journal article(s) based on this preprint
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 10,356 | 1,481 | 248 | 12,085 | 539 | 371 | 401 |
- HTML: 10,356
- PDF: 1,481
- XML: 248
- Total: 12,085
- Supplement: 539
- BibTeX: 371
- EndNote: 401
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Mitchell Bushuk
Michael Winton
William Gregory
Bill Hurlin
Liwei Jia
Feiyu Lu
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(18859 KB) - Metadata XML
-
Supplement
(14629 KB) - BibTeX
- EndNote
- Final revised paper
This paper deals with applied problem of subseasonal prediction of Antarctic sea ice concentration/extent/volume. It looks in detail at two reforecast experiments, one with and one without data assimliation, and compares this to an observational sea ice record (plus a few other reference datasets). Results are discussed in terms of seasons and sectors of the Southern Ocean. A comparison to similar existing studies for the Arctic show a few interesting differences. For example, that sea ice predictability as a function of initialization season is not the same in the Antarctic as it is in the Arctic.
I don't know enough about the immediately adjacent literature to comment on the novelty or need for this study, but the paper seems solid. With a few exceptions (my comments below), the paper is organized well, the writing is easy to follow, the figures are clear, and the length seems appropriate.
Detailed comments:
Defining the two main datasets. This paper is focused on SPEAR vs SICDA, but this isn't made explicit early enough. It only started to become clear as I made my way through Section 2. The last paragraph of Section 1 hints at this comparison being important. And the end of Section 2.3 reiterates it a bit. But I'd suggest being more explicit. One option would be to move a couple of sentences from end of Sections 1 and 2.3 to the start of Section 2 (before Section 2.1 even). Make it very clear in one place. Another option would be to add a Table or Figure that summarizes Section 2 (what models will be used, what quantities will be compared, etc). Although it wouldn't add anything new, this Table/Figure would be a quick way for the reader to refer back and check the details quickly.
Add sector lines to all map figures. Figures 2, 3, 4, 6, 7, 11, and 12 would benefit from having five lines radiating out from the south pole that separate the maps into the five sectors defined in Section 2.6. To make sense of Section 3, I found that I had to draw these lines on myself (at least onto the top left map in each figure). Consider labeling the sectors in Figure 2a, then just repeating the lines thereafter.
The paper uses a small number of acronyms, but uses them a lot. I'm not sure it's a problem that has a good solution, but it does make a lot of the sentences stilted. A couple of egregious cases are the definition of EAFK on L97 (which is never subsequently used) and DART on L67 and L96 (which is only used twice, but defined in full both times). Also, avoid acronyms in headings (subsections of 3.1). Just write out each phrase in full.
Vague quantification in the Abstract: There is an overuse of vague adjectives in this part, rather than concrete numbers. Examples include 'significantly' (L5), 'considerable' (L6), 'mostly' (L6), 'improved the most' vs 'minor differences' (L11), 'more significant' (L12). Since many readers only interact with papers via the abstract, make sure to include at least some key metrics.
There is also some vague quantification in other parts as well:
- 'skillful predictions' at L48 (how skillful?)
- 'improved' at L49 (by how much?)
- 'outperformed' at L64 (by how much?)
- 'reasonably represent' at L122
- 'underestimate' at L124 (by how much)
- 'noticeably' at L164
15. 'Steady increase' is not the correct description of Antarctic sea ice from late 70s to 2015. It didn't decrease over this period, but it wasn't clearly increasing, and it certainly wasn't doing so steadily.
26, 30. Unclear what 'perfect' means here.
161. Define the months for each season earlier in Section 3. You do define these later (at L169), but at least for L161-167, I was left wondering exactly what, say, 'summer' corresponded to.
194. Section 3.1.2 seems just as much about Thickness as it is about Volume. Perhaps this should be reflected in the heading.
197. 'thickest' seems like a weird adjective for describing sea ice volume.
215. Replace 'from the West Weddell Sea to the East Weddell Sea' with just 'across the Weddell sector'?
243. 'first month' seems a weird description. To me, 5–10 days would be a more appropriate description. Yes, some cases extend beyond 10 days, but a month is too much of an upper bound to be relevant for this sentence, right?
261. If I'm reading this sentence right, you're inferring that SICDA has the 'most considerable skill improvement' because it is better than SPEAR. But you're only comparing two cases. If so, the word 'most' is inappropriate as it should be reserved for comparing three or more cases.
282. Not clear what 'This' refers to. Is it the underprediction in Dec–Jan? Or is it referring to the merging of skill values (from an earlier sentence).
303. Unclear how Figure 12j has 'no noticeable bias'. Are you implying that, say, the 0.07 average value is ≈0?
310-317. The last paragraph before the Conclusion only cites results that are shown in the Supplementary material. Why? One would think the last paragraph of a 'Results and Discussion' section is a good place for a key result. But relegating it to the Supplementary gives the opposite impression. Is there a 12-figure limit for this journal, so you decided putting the figure in the Supplementary was a good workaround? If so, it isn't.
377-379. Remove the last three sentences. They're a repetitive and weak way to finish the paper. Finish with what you showed, not what didn't work.
Figures
-------
Figures 1 and 5: set the bottom axis at exactly zero
Figures 5 and 6: what is sea ice volume measured in? Is it actually a thickness?
Figures 1 and 5: remove redundant decimal places in y axes (e.g., 6.00, 4.00, 2.00 → 6, 4, 2)
Figure 3:
- Why is the colormap so saturated? Increase color limit to 0.5, say?
- Redefine the acronym ACC here, since it isn't an intuitive acronym.
Figure 8:
- Add x label ('Time (days)'?) directly to panel e, rather than just stating it in the caption.
- Make the 'DP' line for the Arctic a black dashed line. This would follow the pattern of the red and blue lines.
- Then consider making the legend two columns by three rows to make it really explicit how the line color/style system works.
- Consider replacing 'red dot' and 'red circle' with 'filled red circle' and 'open red circle'? since the open circles aren't obvious. I couldn't initially figure out how a 'dot' was different to a 'circle'.
- Remove the '(red dashed lines)', '(red solid lines)' etc from the caption, which aren't necessary since these details are in the legend already.
Figure 9: Same as for Figure 8
Figure 5 error bars:
The error bars for this figure aren't appropriate. Each error bar is derived from five values. You've chosen to use ±2 standard deviations. But datasets of five are awfully small for standard deviation to be meaningful. Further, ±2 standard deviations is typically taken to be approximately equal to a 95% confidence interval (1.96 standard deviations). Of course, with only five values, you can't estimate a confidence interval that well. Could you solve all this by just using the minima and maxima of the five cases for the error bars?
Figure 10:
- Different colored lines are unnecessary here since it's obvious which months are relevant from where the lines begin.
- The caption is incorrect and appears to have been copy/pasted from a different figure.
Figure S7: Same as for first comment for Figure 10.
Typos:
330: 'from' is repeated
Captions of Figures 7 and S6: 'forcast'