the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 07 Mar 2025
Brief Communication: Correction of Fundamental Errors in the EVP Sea Ice Dynamics in ICON
Abstract. Since its implementation, EVP sea ice dynamics in the ice-ocean model ICON-O contained severe errors and thus all ICON simulations coupled to it. Two errors prevented convergence of the EVP solver, while a third miscalculated the quadratic drag law using an incorrect relative ice velocity. This caused excessive ice drift, overly mobile ice, and large open water patches, distorting ocean-atmosphere exchanges—worsening at higher resolutions. Correcting them improved the sea ice drift, aligning it with observations and yielding a realistic ice cover. This study marks a turning point in ICON's sea ice representation, ensuring significantly improved simulations at all resolutions.
- Preprint
(4059 KB) - Metadata XML
- BibTeX
- EndNote
Status: closed
-
RC1: 'Comment on egusphere-2025-906', Anonymous Referee #1, 14 Mar 2025
Review of Correction of Fundamental Errors in the EVP Sea Ice Dynamics in ICON by Oliver Gutjahr, egusphere-2025-906
The manuscript describes bugs in the sea ice dynamics code of the sea ice component of the general circulation model ICON-O, and how fixing them improves the sea ice simulations with this model. The manuscript is well written and the figures are clear.
My main concern is the scope of the manuscript and the format. As it is, I do not learn anything from the manuscript other that there were bugs in the sea ice part of ICON-O (and possibly also in FESIM) and that they have been fixed with a beneficial effect. This is interesting only for a very specific group (scientists using ICON or ICON results) and I don’t see why this should be the subject of a peer reviewed paper, but it should show up in the bug-reporting system of ICON, and disseminated to the uses. Therefore, I do not recommend to publish this manuscript in the present form.
To me, it would make a lot of sense to make the manuscript more interesting to a larger group of sea ice modeller. With some small additions the author could give the manuscript a different spin to make it more general (and hence publishable). I have these suggestions:
- There are two bugs: (1+2) The EVP solver could not converge because the dynamics variables velocity and stress were reset to zero after each iteration, and (3) there was a bug in the formulation of the ice-ocean stress that lead to an overestimation of the ice-ocean stress. The effects of these totally different bugs are assessed together. I think that it would have been more instructive to isolate these effects, i.e. by comparing additional simulations where only one of the two bugs are fixed. I assume that the solver error (1+2) is less important than the ice-ocean stress error (3).
- The manuscript claims that the EVP solver now converges, but there’s no “proof” or illustration of this. Kimmritz et al (2015, doi:10.1016/j.jcp.2015.04.051 and 2016, doi:10.1016/j.ocemod.2016.03.004) and many others found that the convergence rates of the EVP solver are always very slow, and that true convergence is only reached with O(1e5) iterations. It’s important to check the convergence and if the solver would converge if enough iterations are allowed (see, e.g. Kimmritz 2016, but also others).
- It would be interesting, if the ice-ocean stress is treated explicitly or semi-implicitly in the ICON code. If it is the later, then it is much easier to make this mistake of forgetting the implicit sea ice velocity and that would be instructive for others to find similar mistakes, but the manuscript does not describe this aspect.
Minor comments:
page 1
l13: “internal rheology determines its deformation” rewrite, because “sea ice” as used here is not a realistic medium, but we parameterise it as a quasi-continuous non-normal fluid with a nonlinear (here viscous plastic) rheology connecting stress and strain rates, i.e. here you need to talk about the rheology of the quasi-continuum. “Sea ice” does not have a rheology but consists of floes that may have a rheology (probably very different from what we use for sea ice models).
page 3
l38: “The last error is also present in the recent version of FESOM2.5 (Rackow et al., 2025) and originated in ICON-O when the FESIM ice dynamics (Danilov et al., 2015) were adapted.”
I hope that the authors of FESIM etc have been informed.
page 4
l50: “Since the EVP solver did not converge in previous versions, a tuning was necessary. Increasing the number of subiterations NEVP = 500 was found to be an appropriate balance between computational time and solution quality.”
See also main comments: Can you show convergence rates? According to Kimritz etal 2015/2016, convergence is linear and generally slow. What was the criteria for N=500?
l52: “The constant ice-ocean drag coefficient was raised to Cw = 12 ×10−3 (up from 5.5×10−3)”
Not clear why this is necessary. Wouldn’t it make more sense to reduce the atmospheric drag on ice if too strong winds are the problem? What about the atmospheric drag on the ocean? That should be too strong, too. How was that handled?
l56: “The first impact is on the sea ice concentration, which is exemplarily shown in Fig. 1 for a date (22. March 2012 at 6 UTC) near the end of the simulation.”
See also main comments: It would have been very instructive to understand if the unconverged EVP solver or the ocean stress error are more important. I.e. have a simulation only with error (3), and (1+2) correct and vice versa. This would add more information than just the bug report.
l66: “friction” -> I believe the technical term is ice-ocean stress
page 4
l67: “Since ERA5 overestimates the wind stress over sea ice, using a higher ice-ocean drag coefficient further slows the drift, which is in better agreement with NSIDC (Fig. 2c).”
Again, since ERA5 has been identified as the culprit with too high wind speeds, either correcting the wind speed (no clue how) or reducing the atmosphere-ice stress seems to me to be a better method than counteracting the high momentum input into the ice by a large ice-ocean stress. In this way I would expect that the too fast ice (see figure 2c) imparts too much momentum to the ocean and the too high wind speeds lead to too high ocean surface velocities.
Citation: https://doi.org/10.5194/egusphere-2025-906-RC1 -
CC1: 'Comment on egusphere-2025-906', Sergey Danilov, 27 Mar 2025
The manuscript states that there is an error (listed as Error 3) in the sea-ice model code used by both ICON and FESOM. According to the author, the expression for the drag between the sea ice and the ocean omits the contribution of the sea ice velocity (see line 35 of the manuscript).
This statement is incorrect, as the author failed to see that the drag between the sea ice and ocean is treated implicitly in the code. Its contribution is accounted for on the left hand side, where it is combined with the terms from the time derivative and the implicit Coriolis force. This is specifically mentioned in the FESIM description (Danilov et al. (2015), see expression (55)).
The remaining errors discussed in the manuscript were caused by carelessness in porting the code. As such, they are a fact of someone's biography, but not a scientific message. I do not see why this manuscript should be published.
Good scientific practice would be to consult the developers of the sea ice code used by ICON-o first, but this opportunity has been missed here.
S. Danilov
Citation: https://doi.org/10.5194/egusphere-2025-906-CC1 -
RC2: 'Comment on egusphere-2025-906', Thomas Rackow, 06 Apr 2025
This is my initial review of the study titled “Brief Communication: Correction of Fundamental Errors in the EVP Sea Ice Dynamics in ICON” by O. Gutjahr. The author aims to demonstrate that three bug fixes to the ICON ocean model's source code taken together significantly enhance the realism of its sea ice simulations. The evidence presented includes typical comparisons of sea ice concentration snapshots and sea ice drift against NSIDC data. For the ICON family of models, this represents a potentially significant milestone worthy of documentation in the literature.
However, despite the overall improvement in ICON-O due to these changes, there are several major issues with the current manuscript that need to be addressed. Please refer to the detailed points and line-by-line comments below, along with suggestions for enhancing the paper's impact.
##############
Major comments:
- The author fails to document the specific issues within the ICON code. Consequently, the benefit for the broader community, beyond the ICON family, may be limited. To improve clarity, it is recommended to include code snippets from ICON, explicitly detailing the bugs (e.g., setting velocities to zero at specific points) and the precise improvements made. Given the author's assertion that these issues affect other models and may have been inherited elsewhere, it is crucial to document these occurrences comprehensively. This aspect of the study is potentially valuable, as it highlights how misunderstandings of variable meanings across different model components from different institutes may lead to persistent biases in coupled models.
- Upon reviewing the FESOM2.5 code in detail, I found no issue with the implementation of the quadratic drag law (“error 3”). For numerical stability, this term is discretized implicitly, with the “missing” velocity evaluated in the subsequent timestep. Although it may appear that the relative velocity between u_w and u_ice is not accounted for, it is indeed considered. Therefore, I see no implementation error in FESOM2.5 or earlier FESIM versions. Consequently, any claims of errors in FESOM2.5, erroneous implementations carried over from FESIM to ICON, etc., should be removed unless convincingly proven otherwise. I urge the author to double-check all statements or consult the relevant developers of the sea ice code for clarification.
- The sensitivity analysis of the “code fixes” has been conducted collectively. It would be more informative to present the impact of each fix separately, especially since “error 3” does not appear to be an error. It is possible that explicitly accounting for u_w – u_ice, thus effectively considering it twice, may benefit ICON for incorrect reasons.
My comments focus mostly on the correctness of the paper. This brief communication is shorter and less detailed than some technical reports I have reviewed. At a minimum, the three errors and their "fixes" need to be better defined to provide the community with useful insights.
###################
Line by line comments:
Abstract: “while a third miscalculated the quadratic drag law using an incorrect relative ice velocity.”I don’t think this is correct (maybe in the ICON world/implementation?).
“severe errors”My understanding is this is not a fundamental problem of EVP at all – in the end just bugs are presented.
L.20 “The impact of these errors is assessed” -> “The combined impact of these errors is assessed”
L.24-26 “adopted from the Finite-Element Sea Ice Model (FESIM, version 2) (Danilov et al., 2015), contained significant errors that prevented convergence of the EVP solver and misrepresented the relative sea ice velocity in the quadratic drag law of the ice-ocean stress term.”
As explained above, after checking the code, I don’t think that any significant error is present in FESIM (at least in this part of the code) that was adopted, and this reads as if ICON-O errors come from here. Better to refer to FESIM in a separate sentence.
L. 35/36 “whereas the last term should have been u_w − u_ice.” Not true in my opinion. The last term is u_w − u_ice in FESIM/FESOM, as explained above.
L. 38/39 “The last error is also present in the recent version of FESOM2.5 (Rackow et al., 2025) and originated in ICON-O when the FESIM ice dynamics (Danilov et al., 2015) were adapted.”
Again, while there may be other issues in that model, I don’t think that FESOM2.5 has this error (see above). Please double-check and delete the first part of the sentence.
About the second part of the sentence: Were errors inherited, or were errors newly introduced in ICON at the time when correct FESIM dynamics (correct in the wider FESOM world) were adapted? I think it is the latter, and should therefore be made very clear.Citation: https://doi.org/10.5194/egusphere-2025-906-RC2
Status: closed
-
RC1: 'Comment on egusphere-2025-906', Anonymous Referee #1, 14 Mar 2025
Review of Correction of Fundamental Errors in the EVP Sea Ice Dynamics in ICON by Oliver Gutjahr, egusphere-2025-906
The manuscript describes bugs in the sea ice dynamics code of the sea ice component of the general circulation model ICON-O, and how fixing them improves the sea ice simulations with this model. The manuscript is well written and the figures are clear.
My main concern is the scope of the manuscript and the format. As it is, I do not learn anything from the manuscript other that there were bugs in the sea ice part of ICON-O (and possibly also in FESIM) and that they have been fixed with a beneficial effect. This is interesting only for a very specific group (scientists using ICON or ICON results) and I don’t see why this should be the subject of a peer reviewed paper, but it should show up in the bug-reporting system of ICON, and disseminated to the uses. Therefore, I do not recommend to publish this manuscript in the present form.
To me, it would make a lot of sense to make the manuscript more interesting to a larger group of sea ice modeller. With some small additions the author could give the manuscript a different spin to make it more general (and hence publishable). I have these suggestions:
- There are two bugs: (1+2) The EVP solver could not converge because the dynamics variables velocity and stress were reset to zero after each iteration, and (3) there was a bug in the formulation of the ice-ocean stress that lead to an overestimation of the ice-ocean stress. The effects of these totally different bugs are assessed together. I think that it would have been more instructive to isolate these effects, i.e. by comparing additional simulations where only one of the two bugs are fixed. I assume that the solver error (1+2) is less important than the ice-ocean stress error (3).
- The manuscript claims that the EVP solver now converges, but there’s no “proof” or illustration of this. Kimmritz et al (2015, doi:10.1016/j.jcp.2015.04.051 and 2016, doi:10.1016/j.ocemod.2016.03.004) and many others found that the convergence rates of the EVP solver are always very slow, and that true convergence is only reached with O(1e5) iterations. It’s important to check the convergence and if the solver would converge if enough iterations are allowed (see, e.g. Kimmritz 2016, but also others).
- It would be interesting, if the ice-ocean stress is treated explicitly or semi-implicitly in the ICON code. If it is the later, then it is much easier to make this mistake of forgetting the implicit sea ice velocity and that would be instructive for others to find similar mistakes, but the manuscript does not describe this aspect.
Minor comments:
page 1
l13: “internal rheology determines its deformation” rewrite, because “sea ice” as used here is not a realistic medium, but we parameterise it as a quasi-continuous non-normal fluid with a nonlinear (here viscous plastic) rheology connecting stress and strain rates, i.e. here you need to talk about the rheology of the quasi-continuum. “Sea ice” does not have a rheology but consists of floes that may have a rheology (probably very different from what we use for sea ice models).
page 3
l38: “The last error is also present in the recent version of FESOM2.5 (Rackow et al., 2025) and originated in ICON-O when the FESIM ice dynamics (Danilov et al., 2015) were adapted.”
I hope that the authors of FESIM etc have been informed.
page 4
l50: “Since the EVP solver did not converge in previous versions, a tuning was necessary. Increasing the number of subiterations NEVP = 500 was found to be an appropriate balance between computational time and solution quality.”
See also main comments: Can you show convergence rates? According to Kimritz etal 2015/2016, convergence is linear and generally slow. What was the criteria for N=500?
l52: “The constant ice-ocean drag coefficient was raised to Cw = 12 ×10−3 (up from 5.5×10−3)”
Not clear why this is necessary. Wouldn’t it make more sense to reduce the atmospheric drag on ice if too strong winds are the problem? What about the atmospheric drag on the ocean? That should be too strong, too. How was that handled?
l56: “The first impact is on the sea ice concentration, which is exemplarily shown in Fig. 1 for a date (22. March 2012 at 6 UTC) near the end of the simulation.”
See also main comments: It would have been very instructive to understand if the unconverged EVP solver or the ocean stress error are more important. I.e. have a simulation only with error (3), and (1+2) correct and vice versa. This would add more information than just the bug report.
l66: “friction” -> I believe the technical term is ice-ocean stress
page 4
l67: “Since ERA5 overestimates the wind stress over sea ice, using a higher ice-ocean drag coefficient further slows the drift, which is in better agreement with NSIDC (Fig. 2c).”
Again, since ERA5 has been identified as the culprit with too high wind speeds, either correcting the wind speed (no clue how) or reducing the atmosphere-ice stress seems to me to be a better method than counteracting the high momentum input into the ice by a large ice-ocean stress. In this way I would expect that the too fast ice (see figure 2c) imparts too much momentum to the ocean and the too high wind speeds lead to too high ocean surface velocities.
Citation: https://doi.org/10.5194/egusphere-2025-906-RC1 -
CC1: 'Comment on egusphere-2025-906', Sergey Danilov, 27 Mar 2025
The manuscript states that there is an error (listed as Error 3) in the sea-ice model code used by both ICON and FESOM. According to the author, the expression for the drag between the sea ice and the ocean omits the contribution of the sea ice velocity (see line 35 of the manuscript).
This statement is incorrect, as the author failed to see that the drag between the sea ice and ocean is treated implicitly in the code. Its contribution is accounted for on the left hand side, where it is combined with the terms from the time derivative and the implicit Coriolis force. This is specifically mentioned in the FESIM description (Danilov et al. (2015), see expression (55)).
The remaining errors discussed in the manuscript were caused by carelessness in porting the code. As such, they are a fact of someone's biography, but not a scientific message. I do not see why this manuscript should be published.
Good scientific practice would be to consult the developers of the sea ice code used by ICON-o first, but this opportunity has been missed here.
S. Danilov
Citation: https://doi.org/10.5194/egusphere-2025-906-CC1 -
RC2: 'Comment on egusphere-2025-906', Thomas Rackow, 06 Apr 2025
This is my initial review of the study titled “Brief Communication: Correction of Fundamental Errors in the EVP Sea Ice Dynamics in ICON” by O. Gutjahr. The author aims to demonstrate that three bug fixes to the ICON ocean model's source code taken together significantly enhance the realism of its sea ice simulations. The evidence presented includes typical comparisons of sea ice concentration snapshots and sea ice drift against NSIDC data. For the ICON family of models, this represents a potentially significant milestone worthy of documentation in the literature.
However, despite the overall improvement in ICON-O due to these changes, there are several major issues with the current manuscript that need to be addressed. Please refer to the detailed points and line-by-line comments below, along with suggestions for enhancing the paper's impact.
##############
Major comments:
- The author fails to document the specific issues within the ICON code. Consequently, the benefit for the broader community, beyond the ICON family, may be limited. To improve clarity, it is recommended to include code snippets from ICON, explicitly detailing the bugs (e.g., setting velocities to zero at specific points) and the precise improvements made. Given the author's assertion that these issues affect other models and may have been inherited elsewhere, it is crucial to document these occurrences comprehensively. This aspect of the study is potentially valuable, as it highlights how misunderstandings of variable meanings across different model components from different institutes may lead to persistent biases in coupled models.
- Upon reviewing the FESOM2.5 code in detail, I found no issue with the implementation of the quadratic drag law (“error 3”). For numerical stability, this term is discretized implicitly, with the “missing” velocity evaluated in the subsequent timestep. Although it may appear that the relative velocity between u_w and u_ice is not accounted for, it is indeed considered. Therefore, I see no implementation error in FESOM2.5 or earlier FESIM versions. Consequently, any claims of errors in FESOM2.5, erroneous implementations carried over from FESIM to ICON, etc., should be removed unless convincingly proven otherwise. I urge the author to double-check all statements or consult the relevant developers of the sea ice code for clarification.
- The sensitivity analysis of the “code fixes” has been conducted collectively. It would be more informative to present the impact of each fix separately, especially since “error 3” does not appear to be an error. It is possible that explicitly accounting for u_w – u_ice, thus effectively considering it twice, may benefit ICON for incorrect reasons.
My comments focus mostly on the correctness of the paper. This brief communication is shorter and less detailed than some technical reports I have reviewed. At a minimum, the three errors and their "fixes" need to be better defined to provide the community with useful insights.
###################
Line by line comments:
Abstract: “while a third miscalculated the quadratic drag law using an incorrect relative ice velocity.”I don’t think this is correct (maybe in the ICON world/implementation?).
“severe errors”My understanding is this is not a fundamental problem of EVP at all – in the end just bugs are presented.
L.20 “The impact of these errors is assessed” -> “The combined impact of these errors is assessed”
L.24-26 “adopted from the Finite-Element Sea Ice Model (FESIM, version 2) (Danilov et al., 2015), contained significant errors that prevented convergence of the EVP solver and misrepresented the relative sea ice velocity in the quadratic drag law of the ice-ocean stress term.”
As explained above, after checking the code, I don’t think that any significant error is present in FESIM (at least in this part of the code) that was adopted, and this reads as if ICON-O errors come from here. Better to refer to FESIM in a separate sentence.
L. 35/36 “whereas the last term should have been u_w − u_ice.” Not true in my opinion. The last term is u_w − u_ice in FESIM/FESOM, as explained above.
L. 38/39 “The last error is also present in the recent version of FESOM2.5 (Rackow et al., 2025) and originated in ICON-O when the FESIM ice dynamics (Danilov et al., 2015) were adapted.”
Again, while there may be other issues in that model, I don’t think that FESOM2.5 has this error (see above). Please double-check and delete the first part of the sentence.
About the second part of the sentence: Were errors inherited, or were errors newly introduced in ICON at the time when correct FESIM dynamics (correct in the wider FESOM world) were adapted? I think it is the latter, and should therefore be made very clear.Citation: https://doi.org/10.5194/egusphere-2025-906-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 156 | 32 | 14 | 202 | 12 | 22 |
- HTML: 156
- PDF: 32
- XML: 14
- Total: 202
- BibTeX: 12
- EndNote: 22
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1