Barents-2.5km v2.0: an operational data-assimilative coupled ocean and sea ice ensemble prediction model for the Barents Sea and Svalbard

Röhrs, Johannes; Gusdal, Yvonne; Rikardsen, Edel S. U.; Durán Moro, Marina; Brændshøi, Jostein; Kristensen, Nils Melsom; Fritzner, Sindre; Wang, Keguang; Sperrevik, Ann Kristin; Idžanović, Martina; Lavergne, Thomas; Debernard, Jens Boldingh; Christensen, Kai H.

doi:https://doi.org/10.5194/gmd-16-5401-2023

Articles | Volume 16, issue 18

https://doi.org/10.5194/gmd-16-5401-2023

© Author(s) 2023. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/gmd-16-5401-2023

© Author(s) 2023. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 16, issue 18

Development and technical paper

|

22 Sep 2023

Development and technical paper |

| 22 Sep 2023

Barents-2.5km v2.0: an operational data-assimilative coupled ocean and sea ice ensemble prediction model for the Barents Sea and Svalbard

Johannes Röhrs, Yvonne Gusdal, Edel S. U. Rikardsen, Marina Durán Moro, Jostein Brændshøi, Nils Melsom Kristensen, Sindre Fritzner, Keguang Wang, Ann Kristin Sperrevik, Martina Idžanović, Thomas Lavergne, Jens Boldingh Debernard, and Kai H. Christensen

Download

Final revised paper (published on 22 Sep 2023)
Preprint (discussion started on 06 Mar 2023)

Interactive discussion

Status: closed

RC1:
'Comment on gmd-2023-20', Benjamin Barton, 19 Apr 2023

Review of "Barents-2.5km v2.0: An operational data-assimilative coupled ocean and sea ice ensemble prediction model for the Barents Sea and Svalbard"
General Comments
This paper presents the development methods and validation of the Barents-2.5km v2.0 forecast model. The new version of the Barents-2.5km model has been upgraded to include a 24 member Ensemble Prediction System. The data assimilation has also been upgraded from only sea ice concentration to additionally including sea surface temperature and hydrographic profiles.
Overall the paper shows the forecast system has promise as a useful tool for various applications that require predictions of ocean variables in the Barents, Norwegian and Greenland Sea. This is particularly important for the marginal ice zone areas and the paper shows the challenge and skill required for accuracy. Some of the description of the model forecast skill is hard to follow. Analysis and validation of currents are presented in a separate paper which means there is not a complete picture here. There is also no comparison to v1.0 of the Barents-2.5km model. There are a few points that need attention that I detail below.

Specific Comments
Line 7-9: Currents are only briefly mentioned in the paper and seems to be a larger part of another paper. The abstract should be used to discuss what is in the paper.
Line 108-110: Have you considered using the elastic-plastic anisotropic (EAP) rheology scheme in CICE (Heorton et al., 2018)? Why was EVP chosen?
Line 291: You should mention the SST RMSE July peak.
Lines 293-296: The model is compared against the same observations that are entrained through the data assimilation system. I understand that the observations are considered independent because they are compared with a previous forecast. This has value in quantifying the quality of the prediction for assimilated variables. However, I think the quality of the paper could be improved if you make comparison with un-assimilated data, perhaps sea surface height which could give an indication of current quality.
Line 300-305: Please add the SIC ice charts dataset to the methods and reference.
Line 312-314: Sea ice concentration tends to increase rapidly when there is no data for assimilation. It seems to be reliant of the SIC assimilation data. I see you've made a point about why this happens in Line 422 but I think it would be better to make that point here or rephrase the bit around Line 422.
Line 330-334: Why does the model have greater RMSE for SST up to 12h lead time than the persistence, deterministic or trend with observations?
Line 347-358: Describe the validation/analysis technique in the methods. Same goes for other validation methods. What do the bins represent?
Line 374-377: What are analysis increments? This needs to be clearer.
Line 380: How is the spread calculated? Is it the standard deviation of the ensemble?
Line 392-393: What are spread increments? It also does not say what they are in Figure 12 caption.
Line 442-446: I do not understand Figures 10 and 11 enough. Why is the spread in SST satisfactory? Why is the spread in SIC too low?
Line 471-472: How does the Barents-2.5km v2.0 of the model compare against original Barents-2.5km? Can you quantify the improvement in accuracy that users can expect?
Table 1: This could be quoted in the text instead of a table.

Technical Corrections
Line 307-308: Isn't it 3 periods? January, March and May 2022 in Figure 8 (not Figure 7).
Line 423-424: "looses skill within few days without DA. Deterioration of model skill within few days" typo "a few" for both "few".
Table 2: caption typo ")".
Line 478: "each satellite swath is compare to the model" typo "compared".
Figure 6: Please note the figures that the regions are used for in the caption.
Figures 7, 8 and 9: the text in the figure is too small to comfortably read. Please add ME and RMSE to the respective y-axis labels.
Figure 10: I do not understand what this is showing. The caption is too brief and the axis labels are unhelpful.
Figure 12: Please improve the caption with more detail. What are the black regions in g and h?.
References
Heorton HDBS, Feltham DL, Tsamados M. 2018. Stress and deformation characteristics of sea ice in a high-resolution, anisotropic sea ice model. Philos Trans A Math Phys Eng Sci. 28;376(2129):20170349. doi: 10.1098/rsta.2017.0349.

Citation: https://doi.org/10.5194/gmd-2023-20-RC1
- AC1: 'Reply on RC1', Johannes Röhrs, 25 May 2023
  
  Thank you very much for datailed comments on our manuscript, and suggestions for improving this work.
  
  Please find our answers to each question in the attached file.
  Wind kind regards
  
  Johannes Röhrs and collegues
  
  Citation: https://doi.org/10.5194/gmd-2023-20-AC1
RC2:
'Comment on gmd-2023-20', Anonymous Referee #2, 23 Apr 2023

Please see the comments attached.

Citation: https://doi.org/10.5194/gmd-2023-20-RC2
- AC2: 'Reply on RC2', Johannes Röhrs, 26 May 2023
  
  We would like to thank the reviewer for studying our manuscript and providing usefull feedback. Please find answers to your questions in the attached document. We look forward to present a revised manuscript taking account for the issues raised by both reviewers.
  With kind regards
  
  Johannes Röhrs and colleagues
  
  Citation: https://doi.org/10.5194/gmd-2023-20-AC2

Peer review completion

AR: Author's response | RR: Referee report | ED: Editor decision | EF: Editorial file upload

AR by Johannes Röhrs on behalf of the Authors (16 Jun 2023) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (05 Jul 2023) by Shu-Chih Yang

RR by Anonymous Referee #2 (15 Jul 2023)

RR by Benjamin Barton (19 Jul 2023)

ED: Publish subject to technical corrections (09 Aug 2023) by Shu-Chih Yang

1. The section of methodology should be improved. Particularly, the description of Deterministic EnKF should be better organized. Although the DA system is using the package of enkf-c. The meaning of different parameters should be clearly explained.
 The deterministic version of the EnKF (Sakov and Oke, 2008, SO08) should be briefly explained. “ In the EnKF, …. Based on the assumption of a smack KH term”. I don’t think this explains how SO08‘s deterministic EnKF works.
 The added paragraph does not explain how K=1.5 works?
 Please clarify why only the horizontal localization is used in this study? (P8 Line 223, 579)
2. Although it is good that the authors tried to explain the evaluation metrics in details, the explanations should be concise and can be referred to proper references, such as (Wilks, D.).
 The reliability diagram is often used for the verification of meteorological ensemble forecast. The paragraph for explaining the reliability diagram can be shortened and referred to the proper reference.
3. It should be noted that the designed ensemble forecast in this study does not represent “model uncertainty” since the ensemble members in this study only have differences in the initial states. The model uncertainty is usually referred to the ensemble with perturbed model physics. Please avoid to use “model uncertainty”.
4. The discussion regarding the non-Gaussianity of SIC should be more organized. Also, why is the small ensemble spread of SIC attributed to non-Gaussianity? The small ensemble spread may simply because of the model bias.
 On P19, Line 556: there is no direct argument to indicate that the small spread of SIC is related to non-Gaussianity. The issue of non-Gaussian SIC was mentioned later at Line 593-599. The discussions should be re-organized.
5. There are several unclear sentences in the revised manuscript. I suggest the authors should make the clarification.
 There are many acronyms that the readers may not be familiars with. Please provide a list to give the definition of the acronyms, such as TOPAZ, OSTIA…etc.
 P8, Line 239: I suggest to remove “some”.
 Section 4.3: Is there any QC procedure for SIC?
 P11, Line 325: Since a failure of the EnKF update is more common to occur, please clarify “the poor performing statistic representations in the EnKF algorithm”. Is this a too small ensemble spread or any kind of situation lead to filter divergence?
 P14, Line 410: the availability of the “observations”?
 P18, Line 506: The argument that “the model analysis and forecast is thought to contain skill if the model error grows with lead time” is not reasonable. Even the forecast error keeps growing, the forecast can be not be regarded as skillful compared to the reference.
 It is unclear why there is another forecast by combining the model trend with the past observation. What is the purpose?
 P16, Line 469: why does the small ensemble spread partly owes to the observation error? Do the authors suggest to inflate the observation error? The small ensemble spread may also be related to the model bias. Please give a brief explanation. Also, rescaling the forecast probability to match the observed occurrence frequencies is also a solution to remove the impact of model bias on forecast probability (Chang et al. 2015, MWR).
 P19, Line 567-569: Do the authors expect that the adaptive covariance inflation method can mitigate the issue of non-Gaussianity?
 P21, Line 624: “transform observation operators” is not clear. Did the authors mean “transform the observation variable from a skewed to a quasi-Gaussian distribution?

Hide

AR by Johannes Röhrs on behalf of the Authors (17 Aug 2023) Author's response Manuscript

Short summary

A model to predict ocean currents, temperature, and sea ice is presented, covering the Barents Sea and northern Norway. To quantify forecast uncertainties, the model calculates ensemble forecasts with 24 realizations of ocean and ice conditions. Observations from satellites, buoys, and ships are ingested by the model. The model forecasts are compared with observations, and we show that the ocean model has skill in predicting sea surface temperatures.

Barents-2.5km v2.0: an operational data-assimilative coupled ocean and sea ice ensemble prediction model for the Barents Sea and Svalbard

Download

Interactive discussion

Peer review completion

Suggestions for revision or reasons for rejection