Improvements to the Met Office's global ocean–sea ice forecasting system including model and data assimilation changes

Mignac, Davi; Waters, Jennifer; Lea, Daniel J.; Martin, Matthew J.; While, James; Weaver, Anthony T.; Vidard, Arthur; Guiavarc'h, Catherine; Storkey, Dave; Ford, David; Blockley, Edward W.; Baker, Jonathan; Haines, Keith; Price, Martin R.; Bell, Michael J.; Renshaw, Richard

doi:https://doi.org/10.5194/gmd-18-3405-2025

Articles | Volume 18, issue 11

https://doi.org/10.5194/gmd-18-3405-2025

© Author(s) 2025. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/gmd-18-3405-2025

© Author(s) 2025. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 18, issue 11

Model description paper

|

13 Jun 2025

Model description paper |

| 13 Jun 2025

Improvements to the Met Office's global ocean–sea ice forecasting system including model and data assimilation changes

Davi Mignac, Jennifer Waters, Daniel J. Lea, Matthew J. Martin, James While, Anthony T. Weaver, Arthur Vidard, Catherine Guiavarc'h, Dave Storkey, David Ford, Edward W. Blockley, Jonathan Baker, Keith Haines, Martin R. Price, Michael J. Bell, and Richard Renshaw

Download

Final revised paper (published on 13 Jun 2025)
Supplement to the final revised paper
Preprint (discussion started on 14 Oct 2024)
Supplement to the preprint

Interactive discussion

Status: closed

RC1:
'Comment on egusphere-2024-3143', Anonymous Referee #1, 06 Dec 2024
The manuscript describes the recent advances in the global ocean and sea-ice operational forecasting system by Met Office, widely known as FOAM. Improvements concern both the model and data assimilation (DA) scheme, however the changes in DA scheme are responsible for most of the improvements in terms of RMSE and BIAS metrics, as the Authors shown. Multiple one-year-long experiments are performed to compare the skills between the new and old system using different set up and two different model resolutions: 1/4° and 1/12° (DA is performed always at 1/4°). The largest improvement is seen when only surface data are assimilated, the old system presents un-realistic drifts in OHC and AMOC that are resolved in the new system, demonstrating the possibility to use the new system during the pre-argo era for reanalysis purposes.
FOAM is a well-known system that is largely used for different ocean-related applications. The manuscript is well written, results are robust and critically presented. Beside minor corrections, I can support the manuscript for publication however I am concerned about the fact GMD strongly recommends the dissemination of the code. In the present case, the main improvements come from the DA code that is not free. Ocean Science can be a good alternative.
MINOR COMMENT
GENERAL
There is a sort of confusing nomenclature throughout the manuscript that I encourage the Authors to clarify for readability purposes. To my understanding, GO6/GOIS9 are generally used to point previous and new system while ORCA12 /ORCA025 are used to identify the resolution.
In the Figures GO6 /GOI9 are used to differentiate the experiments, while the resolution (ORCA12 or ORCA025) is written in the caption. This is however confusing since ORCA12 parametrisation is different between GO6 and GOI9. I can suggest expanding the table 2 by introducing an different name for each experiment, something like GOIS9e12, GOIS9e025 etc.

The use of ORCA12 DA is to be avoid since DA is performed at ¼° to my understanding. If two different DA set up at 1/4 ° are used, a description can be added to table 2.

"FOAM" is used an attribute to both the system or the resolution, i.e. FOAM GO6 or FOAM ORCA12, etc., my impression is that you can safely remove "FOAM" everywhere.

SPECIFIC
Line 25: “In the absence of profile DA [..]”
The sentence is difficult to follow and “profile DA” is not clear. Please rephrase it you can start with something like “Limited to the assimilation of surface data only [..]”
Line 96-101
I do not understand if those improvements are applied only to 1/12 ° or also to ¼°
Figure F3 is blurry please replace it.
Line 193 and Figure 2:
Those results seem not conclusive, is there a reason why statistics are calculated only between Jan-May 2019 without covering the full year? Do you expect the results to be the same by including more statistics?
Line 220
Is the minimization in GOIS9 always converging within 120 iterations? If not, can the Authors comment whether it is a noise problem, caused by the use of a first-order minimizer, or else.
Fig4
Similarly to my previous doubt, I am not sure the results in Fig4 are conclusive. There is a small difference among experiments and the statistics consider only 4 months. These results can be related to the specific position of few argo that do not fully span the depts and the area. Can the Authors add the number of observation, extend the statistics or provide some study on the significativity?
Line 288
How many cores per node? Which is the frequency of the cpu ?
Lines 310-315
Can the Authors specify whether there is some preprocessing on the drifter velocities? are the drifters without drogues, used?
325 “4.2 Impacts on the […] with all observation types assimilated”
Which is the full set of observation assimilated? It is written that it is similar to the operational system, but the latter may vary in time. For example it is not clearly whether the satellite SST is assimilated or not together with the drifter SST. Is the new rejection algorithm impacting the number of insitu profile assimilated significantly?
418: “Additionally, the same SI3 settings from […]“
If I understood correctly the sea-ice DA is not changed passing from the old to the new system. Is it so? Is the same parametrization used for ORCA12 and ORCA025 in Si3?. What about CICE (old system)? Was there a different parametrization for ORCA12 and ORCA025?
421 “Potential impacts […] pre-Argo reanalysis ”
I have not understood whether satellite sst are assimilated in the pre-argo era configuration.
422-426
The Authors missed to comment the importance of a good initial condition in pre-argo era. Especially in the southern hemisphere, the model state can be far from true state and the assimilation of few SLA measurements can exacerbate any hidden biases. Initial condition in 2019 are instead well spin up and the absence of drastic changes in OHC can partially come from that. Please add a comment on this.
430 : “GOSI9-NoTSProf run […] SLA RMSDs in comparison to its original […]”
I was expecting GOSI9-NoTSProf to be closer to SLA obs since the system is assimilating mainly SLA data. Why the RSME is similar to the one that ingest insitu data too?
493-497
Which is the method used to estimate the AMOC? Are the Authors using the RAPID decomposition? Probably this is the cause of the difference. I noticed that the assimilation seems to degrade the correlation with RAPID timeseries, do you have any reason why?
Citation: https://doi.org/10.5194/egusphere-2024-3143-RC1
- AC1: 'Reply on RC1', Davi Mignac, 04 Feb 2025
  
  The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2024/egusphere-2024-3143/egusphere-2024-3143-AC1-supplement.pdf
  
  Citation: https://doi.org/10.5194/egusphere-2024-3143-AC1
RC2:
'Comment on egusphere-2024-3143', K. Andrew Peterson, 10 Dec 2024

I am recommending this article be published with some minor revisions. A detailed review of the paper is attached. Summarizing: The manuscript is well written and rigorous in its scientific analysis. I have a few recommendations to include a review of important aspects of the NEMOVAR system, in particular the balance relationships applied, so as to better understand the effects and reasons for the system update. I also recommend some revision to the discussion of particular components of the update so as to better demonstrate their effectiveness.

Citation: https://doi.org/10.5194/egusphere-2024-3143-RC2
- AC2: 'Reply on RC2', Davi Mignac, 04 Feb 2025
  
  The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2024/egusphere-2024-3143/egusphere-2024-3143-AC2-supplement.pdf
  
  Citation: https://doi.org/10.5194/egusphere-2024-3143-AC2

Peer review completion

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

AR by Davi Mignac on behalf of the Authors (04 Feb 2025) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (05 Feb 2025) by Vassilios Vervatis

RR by K. Andrew Peterson (18 Feb 2025)

Suggestions for revision or reasons for rejection

The authors have successfully revised the manuscript as was recommended by my and another reviewers comments. I now find the manuscript acceptable for publication with only a few very minor clarifications of which the authors can choose whether they warrant further (but extremely minor) revisions:

L. 320: Validation experiment runs are initialized from a 2017-2018 run of GO6. Exact phrasing "They come from 2017-2018 ORCA12 and ORCA025 runs with the FOAM GO6 configuration (see Barbosa-Aguiar et al., 2024)." This is ambiguous. I believe what you meant is "They are initialized from the end of 2017-2018 ORCA12 and ORCA025 runs with the FOAM GO6 configuration on 1 Jan, 2019 (correct date?).

L. 325. Presented OmB statistics are calculated from ..., SST from in situ drifters, ... This is an accurate statement, but neglects to specify you are ignoring further OmB statistics from assimilated (but biased) satellite SST observations. Perhaps this is worth stating?

All OmB spatial plots: There is no warning/admission that zero OmB difference and no data are both signified by white colour mapping. In general this is not a problem (most observed areas have at least some non-zero shading), or as in the velocity plots, the rms error plots display the observation error holes.

Again OmB spatial plots: Presumably the OmB statistics have been binned into lat-lon bins or onto the model grid? The bin size is never stated, but looks relatively high resolution. Please state if possible in Figure 5 (assuming it remains the same in all figures).

LL 462-464: However, it is worth highlighting that FOAM SLA statistics in the central and eastern North Atlantic are slightly better in GOSI9-NoTSProf (Fig. 10g) than in GOSI9 (Fig. 10e). This suggests that there could still be minor issues in assimilating SLA and T/S profile data together, even after the substantial SLA improvements caused by GOSI9 DA changes.

This also appears to be true in the equatorial Pacific -- and possibly the whole of the Pacific, just not degradations from GO6 (deeper blue improvements). Were velocity statistics also performed on the no profile runs to further support SLA OmB statistics. Not suggesting to add plots -- just possibly state velocity statistics confirm better enhanced adjustment to SLA innovations without (potentially) conflicting T/S innovations.

Hide

RR by Anonymous Referee #1 (27 Feb 2025)

Suggestions for revision or reasons for rejection

The Authors clarified most of my concerns except for two points, one of which shall be solved by Authors before I can support the publication.

1) I previously proposed to move the manuscript to OS since GMD guidelines strongly recommend the dissemination of the code. The Authors answered that they have submitted to OS at the beginning, but OS Editor recommended GMD. I will not hold on that and leave the GMD Editor to decide.

2) The importance of present Figure 2 is still not clear to me. If I understood correctly, the removal of long-range temperature length-scale mitigates the problem with SLA assimilation, moreover it improves the BIAS and RMSE specifically in the Southern Ocean. The behaviour is opposite for salinity, the long-range length-scale can be kept since it does not enter the density calculation, moreover it improves the BIAS and RMSE especially in the Southern Ocean.

I found that this opposite behaviour represents an interesting point to discuss in some more details since previous papers on FOAM system stated that the use of the sole short-range length-scale goes in the direction of improving the RMSE. The inclusion of long length-scale is meant not to reduce the RMSE but rather to reduce the presence of noisy fields in the case of sparse data, and to constrain the bias of tracer fields when observations are sparse (Blockley et al. 2014,Water et al, 2013,2014).

Based on the results of Figure 2, are the Authors confident with this opposite behaviour in temperature and salinity, over a longer period says one or few years? If not, I can suggest to remove Figure 2. My impression is that it lacks information that confirm the significativity of such results within 5 months: are such OmB statistics stable in time or do the improvements come only from a short period within the 5 months? Are the number of observations, used to evaluate OmB, similar in the three experiments? A further check on the number of observations can be beneficial, since two mistakes have been already rectified in other figures.

Hide

ED: Publish subject to minor revisions (review by editor) (28 Feb 2025) by Vassilios Vervatis

AR by Davi Mignac on behalf of the Authors (18 Mar 2025) Author's response Author's tracked changes Manuscript

ED: Publish as is (19 Mar 2025) by Vassilios Vervatis

AR by Davi Mignac on behalf of the Authors (23 Mar 2025)

Short summary

We describe major improvements of the Met Office's global ocean–sea ice forecasting system. The models and the way observations are used to improve the forecasts were changed, which led to a significant error reduction of 1 d forecasts. The new system performance in past conditions, where subsurface observations are scarce, was improved with more consistent ocean heat content estimates. The new system will be of better use for climate studies and will provide improved forecasts for end users.