the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 08 Aug 2025
Sea Ice Albedo Bounded Data Assimilation and Its Impact on Modeling: A Regional Approach
Abstract. We conducted a perfect model experiment using Icepack, a one-dimensional single-column sea ice model, to assess the potential of data assimilation (DA) to improve predictions of the mean sea ice state through the incorporation of sea ice albedo (SIAL) observations. One ensemble member is designated as the TRUTH, and synthetic observations drawn from it are assimilated into the remaining ensemble members. DA is carried out using the Data Assimilation Research Testbed (DART) with a bounded Quantile Conserving Ensemble Filtering Framework (QCEFF), which accounts for the bounded nature of sea ice variables. Icepack ensembles were spun-up for four Arctic locations based on small perturbations to atmospheric forcing. Results show that assimilating SIAL yields comparable or superior performance to more commonly assimilated observables such as sea ice concentration (SIC) and thickness (SIT) in three-quarters of the Arctic regions studied, and across all regions when observational uncertainty in SIAL is reduced below estimates from the current literature. These findings underscore the value of leveraging existing SIAL observations and expanding their temporal and spatial coverage in the Arctic. Furthermore, the study highlights the critical need to better constrain the observational uncertainty of SIAL. Enhanced observational networks would provide the necessary validation data, enabling more accurate uncertainty characterization and improved sea ice forecasts in a rapidly evolving polar climate.
-
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
(14615 KB)
-
Supplement
(968 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(14615 KB) - Metadata XML
-
Supplement
(968 KB) - BibTeX
- EndNote
- Final revised paper
Journal article(s) based on this preprint
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2025-2540', Anonymous Referee #1, 10 Sep 2025
- AC1: 'Reply on RC1', Joseph Rotondo, 21 Oct 2025
-
RC2: 'Comment on egusphere-2025-2540', Anonymous Referee #2, 19 Sep 2025
This study investigates the impact of assimilating sea ice albedo (SIAL) on Arctic sea ice prediction using a series of perfect model experiments. The work is well-structured and addresses a relevant topic in data assimilation using the Quantile Conserving Ensemble Filtering Framework (QCEFF). The objective to evaluate the impact of SIAL assimilation on sea ice forecasts is clearly stated. The authors selected four diverse Arctic regions and use Icepack model for the assimilation with spin-up period (2000-2010) and experimental timeframe (2011-2015). The manuscript presents interesting results, particularly regarding the complementary benefits of multi-parameter assimilation. Several aspects of the experimental design and presentation could be clarified to strengthen the manuscript.
Â
Specific comments
- Ensemble generation
The authors mentioned how to construct the ensemble but only in the abstract and L74 without details. It would be valuable to know the details on the ensemble generation, and elaborate in the Method section.
- Observation uncertainty
The assumptions for observation uncertainty are highly consequential for the DA results (Line ~170). The chosen setup—a parabolic distribution for SIC, 10% for SIT, and zero uncertainty at the bounds (0% and 100% SIC, 0m SIT)—is a common simplification but has significant implications. Setting uncertainty to zero forces the model to exactly match the observation at these bounds. This can lead to overconfidence and an artificial reduction in ensemble spread, potentially skewing the results. The authors could clarify this choice in the Methods or Discussion section.  A strong recommendation for future work would be to adopt a more realistic uncertainty that avoids zero uncertainty, for instance, by specifying a minimum uncertainty floor (e.g., 1-2%) for SIC at 0% and 100%.
- Interpretation of SIAL vs. SIC assimilation results
The central conclusion that SIAL assimilation outperforms SIC assimilation under low uncertainty is compelling. However, this advantage may be partially confounded by the differential uncertainty settings applied to each variable. SIAL's assigned uncertainty is likely low (though not explicitly stated in the provided text), while SIC's uncertainty is structurally defined by a parabola, which is higher everywhere except the bounds. A fairer comparison would require testing SIC assimilation under similarly low uncertainty assumptions. The authors may state this potential confounding factor in the discussion (e.g., around Line 306).
- Multi-parameter assimilation
The finding that the simultaneous assimilation of SIC, SIT, and SIAL yields the best performance is a key result. The complementarity between these variables, as likely illustrated in Figure 4, is a highly valuable insight. This multi-parameter complementarity deserves to be highlighted and discussed in greater depth as a major takeaway of the study, perhaps exploring the physical reasons behind why the constraints provided by these variables are non-redundant.
- Region-specific behavior of albedo (Section 4.3)
The identified unique albedo evolution for category `n=1` ice in the Chukchi Sea is interesting. Is there any reason for that? the formation of melt ponds? why is it pronounced in this region and not others.
Â
Â
Citation: https://doi.org/10.5194/egusphere-2025-2540-RC2 - AC2: 'Reply on RC2', Joseph Rotondo, 26 Nov 2025
-
RC3: 'Comment on egusphere-2025-2540', Anonymous Referee #3, 21 Oct 2025
 Review: Sea Ice Albedo Bounded Data Assimilation and Its Impact on Modeling: A Regional Approach
The paper presents a perfect model experiment comparing the prediction improvement caused by assimilating albedo in comparison to sea ice concentration and sea ice thickness in one dimensional Icepack simulations, for four regions in the Arctic. The study finds that dependent on the error of the assimilated albedo and the region, that albedo assimilation outperforms SIC and SIT assimilation, except for one region. The work is relevant to the scope of The Cryosphere. Overall the paper is well structured and the experiments well designed and I would recommend minor revisions.
Main comments:
1) The results emphasize the importance of the observation uncertainties. It is reasoned well why the SIC uncertainty varies with SIC, but should the same consideration not also be applied to SIT uncertainty and albedo uncertainty? I am unfamiliar with error dependencies in satellite albedo retrievals, but SIT errors are typically greater for thinner ice/thicker ice, dependent on the product used. Since the errors are such a central part of the outcome of this study I would expect a higher focus on this.
2) One major issue in the paper, is that the perfect model experiment is conducted regions in which dynamics do play a non neglectable role in the ice grows. All experiment locations are located in regions which would probably yeld different results, if conducted in a 3-D set up. This is not possible to fully address, but it would add to the significance of the study to include a site located in a region which is typically covered by land fast ice. If this would be the case it will be easier to relate future studies conducted in a 3-d set up to the findings in this study.
3) Currently the temporal set up is unclear. It reads as if the spin up is run for 11 years, the reference run for 5 and the assimilation for 7 month? A common description of which runs are run for how long would be nice for readability.
4) The albedo is typically a parameter over which sea ice models are tuned. To ensure that the results are useful for future studies it would be desirable to attach the relevant namelist parameters. How would the authors expect the results to change if different values for, for example, snow grain size would be used?
5) Overall the methods are well structured, including both the description of the experiment set up and the evaluation methods. It would further improve the structure of the paper to move the description of the category wise DA to the method section (4.3 to to example 2.5).
Minor comments:
row 15: AA is only used once. No acronym needed.
Line 27-31: missing reference
Some statements seem unnecessary for example line 41-46 and line 59-60. If the statements are not relevant to the study, why mention them. If the authors want to anyways include them, maybe these could be summarized to a motivation to study Arctic albedo processes.
Line 260: how the TRUTH was constructed should be in the methods
Citation: https://doi.org/10.5194/egusphere-2025-2540-RC3 - AC3: 'Reply on RC3', Joseph Rotondo, 26 Nov 2025
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2025-2540', Anonymous Referee #1, 10 Sep 2025
- AC1: 'Reply on RC1', Joseph Rotondo, 21 Oct 2025
-
RC2: 'Comment on egusphere-2025-2540', Anonymous Referee #2, 19 Sep 2025
This study investigates the impact of assimilating sea ice albedo (SIAL) on Arctic sea ice prediction using a series of perfect model experiments. The work is well-structured and addresses a relevant topic in data assimilation using the Quantile Conserving Ensemble Filtering Framework (QCEFF). The objective to evaluate the impact of SIAL assimilation on sea ice forecasts is clearly stated. The authors selected four diverse Arctic regions and use Icepack model for the assimilation with spin-up period (2000-2010) and experimental timeframe (2011-2015). The manuscript presents interesting results, particularly regarding the complementary benefits of multi-parameter assimilation. Several aspects of the experimental design and presentation could be clarified to strengthen the manuscript.
Â
Specific comments
- Ensemble generation
The authors mentioned how to construct the ensemble but only in the abstract and L74 without details. It would be valuable to know the details on the ensemble generation, and elaborate in the Method section.
- Observation uncertainty
The assumptions for observation uncertainty are highly consequential for the DA results (Line ~170). The chosen setup—a parabolic distribution for SIC, 10% for SIT, and zero uncertainty at the bounds (0% and 100% SIC, 0m SIT)—is a common simplification but has significant implications. Setting uncertainty to zero forces the model to exactly match the observation at these bounds. This can lead to overconfidence and an artificial reduction in ensemble spread, potentially skewing the results. The authors could clarify this choice in the Methods or Discussion section.  A strong recommendation for future work would be to adopt a more realistic uncertainty that avoids zero uncertainty, for instance, by specifying a minimum uncertainty floor (e.g., 1-2%) for SIC at 0% and 100%.
- Interpretation of SIAL vs. SIC assimilation results
The central conclusion that SIAL assimilation outperforms SIC assimilation under low uncertainty is compelling. However, this advantage may be partially confounded by the differential uncertainty settings applied to each variable. SIAL's assigned uncertainty is likely low (though not explicitly stated in the provided text), while SIC's uncertainty is structurally defined by a parabola, which is higher everywhere except the bounds. A fairer comparison would require testing SIC assimilation under similarly low uncertainty assumptions. The authors may state this potential confounding factor in the discussion (e.g., around Line 306).
- Multi-parameter assimilation
The finding that the simultaneous assimilation of SIC, SIT, and SIAL yields the best performance is a key result. The complementarity between these variables, as likely illustrated in Figure 4, is a highly valuable insight. This multi-parameter complementarity deserves to be highlighted and discussed in greater depth as a major takeaway of the study, perhaps exploring the physical reasons behind why the constraints provided by these variables are non-redundant.
- Region-specific behavior of albedo (Section 4.3)
The identified unique albedo evolution for category `n=1` ice in the Chukchi Sea is interesting. Is there any reason for that? the formation of melt ponds? why is it pronounced in this region and not others.
Â
Â
Citation: https://doi.org/10.5194/egusphere-2025-2540-RC2 - AC2: 'Reply on RC2', Joseph Rotondo, 26 Nov 2025
-
RC3: 'Comment on egusphere-2025-2540', Anonymous Referee #3, 21 Oct 2025
 Review: Sea Ice Albedo Bounded Data Assimilation and Its Impact on Modeling: A Regional Approach
The paper presents a perfect model experiment comparing the prediction improvement caused by assimilating albedo in comparison to sea ice concentration and sea ice thickness in one dimensional Icepack simulations, for four regions in the Arctic. The study finds that dependent on the error of the assimilated albedo and the region, that albedo assimilation outperforms SIC and SIT assimilation, except for one region. The work is relevant to the scope of The Cryosphere. Overall the paper is well structured and the experiments well designed and I would recommend minor revisions.
Main comments:
1) The results emphasize the importance of the observation uncertainties. It is reasoned well why the SIC uncertainty varies with SIC, but should the same consideration not also be applied to SIT uncertainty and albedo uncertainty? I am unfamiliar with error dependencies in satellite albedo retrievals, but SIT errors are typically greater for thinner ice/thicker ice, dependent on the product used. Since the errors are such a central part of the outcome of this study I would expect a higher focus on this.
2) One major issue in the paper, is that the perfect model experiment is conducted regions in which dynamics do play a non neglectable role in the ice grows. All experiment locations are located in regions which would probably yeld different results, if conducted in a 3-D set up. This is not possible to fully address, but it would add to the significance of the study to include a site located in a region which is typically covered by land fast ice. If this would be the case it will be easier to relate future studies conducted in a 3-d set up to the findings in this study.
3) Currently the temporal set up is unclear. It reads as if the spin up is run for 11 years, the reference run for 5 and the assimilation for 7 month? A common description of which runs are run for how long would be nice for readability.
4) The albedo is typically a parameter over which sea ice models are tuned. To ensure that the results are useful for future studies it would be desirable to attach the relevant namelist parameters. How would the authors expect the results to change if different values for, for example, snow grain size would be used?
5) Overall the methods are well structured, including both the description of the experiment set up and the evaluation methods. It would further improve the structure of the paper to move the description of the category wise DA to the method section (4.3 to to example 2.5).
Minor comments:
row 15: AA is only used once. No acronym needed.
Line 27-31: missing reference
Some statements seem unnecessary for example line 41-46 and line 59-60. If the statements are not relevant to the study, why mention them. If the authors want to anyways include them, maybe these could be summarized to a motivation to study Arctic albedo processes.
Line 260: how the TRUTH was constructed should be in the methods
Citation: https://doi.org/10.5194/egusphere-2025-2540-RC3 - AC3: 'Reply on RC3', Joseph Rotondo, 26 Nov 2025
Peer review completion
Journal article(s) based on this preprint
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 2,058 | 206 | 37 | 2,301 | 78 | 69 | 83 |
- HTML: 2,058
- PDF: 206
- XML: 37
- Total: 2,301
- Supplement: 78
- BibTeX: 69
- EndNote: 83
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Molly Michael Wieringa
Cecilia Marie Bitz
Robin Clancy
Steven Cavallo
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(14615 KB) - Metadata XML
-
Supplement
(968 KB) - BibTeX
- EndNote
- Final revised paper
Please see the attached file.