the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 21 Apr 2026
Review and Synthesis: operational prediction of ocean biogeochemistry in the Copernicus Marine Service. Main achievements and perspectives
Abstract. The operational prediction of marine biogeochemical cycles and ecosystems (the green ocean) has made significant progresses during the last decade. The green ocean is now routinely forecasted every day, and multi-decadal reanalyses are produced with an always increasing resolution and over longer periods. The quality of the green ocean products has increased thanks to the improved model formulations, resolution, data assimilation systems and increased availability of biogeochemical observations routinely delivered in near real time.
Here we review the advancements in our capabilities to predict the green ocean in the frame of the European Copernicus Marine Service (CMEMS) since its start in 2015 and for the five European seas, the Arctic and Global oceans. The evolutions of the prediction systems (e.g. model formulations, data assimilation, coupling with the physics and at the interfaces), delivered products (e.g. resolution, quality assessment, adequacy to support the development of indicators and the decision-making process), and computing resources requirements are reviewed. The predictability drivers and relevant time scales for marine biogeochemical cycles and ecosystems predictions are discussed. Recommendations for future developments are proposed based on a SWOT analysis of current CMEMS green ocean prediction systems and products.
- Preprint
(1736 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on egusphere-2026-813', Anonymous Referee #1, 12 Jun 2026
-
RC2: 'Comment on egusphere-2026-813', Anonymous Referee #2, 15 Jun 2026
Review of “Review and Synthesis: operational prediction of ocean biogeochemistry in the Copernicus Marine Service. Main achievements and perspectives.” By Grégoire et al.
This paper is a review and synthesis paper looking into the Copernicus Marine Environment Monitoring Service (CMEMS) operational biogeochemical ocean forecasting systems, covering the period 2015-2024 across seven regional forecasting centres spanning European seas, the Arctic, and the global ocean. It documents how the "green ocean" prediction systems have evolved in terms of model formulations, data assimilation, spatial resolution, and delivered products and demonstrates that overall forecast quality has improved over the decade. Using a Strengths-Weaknesses-Opportunities-Threats (SWOT) framework, the paper concludes that while significant progress has been made, persistent challenges remain around the scarcity of biogeochemical observations, the quality of river forcing, and the computational demands of future ensemble and high-resolution systems, with artificial intelligence identified as a potential opportunity that is currently under testing but not yet operational. The detailed description of the CMEMS “green ocean” service and the changes that have occurred over the past decade (2015–2024) of operation represents a genuinely valuable community resource. As mentioned in the paper, this resource could be particularly useful to provide knowledge to a wide range of users, to policy-makers and to the general public.
However, also coming from the same end arise my biggest concerns about the paper. This paper reads too technical, the writing style of this paper is more like a project report to a funding agency rather than a review and synthesis trying to reach the broader audience, which is supposed to be the biggest strength of the paper. The current version of the review will for sure lose the broader audience, addtional to the list of technical details, it contains too many unnecessary details that eventually bury the truly valuable information.
- There are way too many abbreviations in the paper, which are definitely not reader friendly. Some of the important abbreviations are even used without explanations, e.g., SWOT in the abstract and later in the paper; NRT and MY, throughout the paper, etc.
- Unnecessary details in the paper, for example, Section 2.7 Computing resources. Number of cores, storage in TB, simulation times are operational logistics, not relevant to the general audience.
- The writing style and level of details are not consistent. For example, in tables A1 and A2, each row corresponds to one centre and reads as a self-contained technical description. The formatting, level of detail, and writing style vary noticeably between centres.
- The general text reads more like a list of enormous amount of information. This could be useful as technical documentation, but it is very difficult to extract scientific insight from, especially given the need to jump between pages in the main text and the appendix tables.
Content-wise, I’m hoping to get some information about inter-model consistency and disagreement across the seven systems. Considering the review is about an overview and updates on the 7 models, it’s natural to include what aspects the models behave the same and what aspects are different and ideally also why and let downstream users know which products are most reliable for which applications.
I do like section 4 and section 5, which are what I originally expected from this review. I would expect the AI part to particularly attract the public attention and would deserve more text than a short paragraph now. I think this part can be better received by the broader audience, potential users and policy makers. However, due to the issue above, it would be fairly difficult for a non-specialist reader to make it this far.
Due to the above-mentioned issues. I don’t think the paper should be published in its current form. I would suggest the authors to choose a clear direction, either developing it as a technical paper or a broader-audience-reaching review and synthesis paper. If the authors choose the technical-paper route, the paper should include more comprehensive model comparisons, and the appendix tables should be moved into the main text. However, in this case, I wonder whether Biogeosciences is the best choice, as the strong technical emphasis of the paper may not fit the typical scope of the journal. If the authors decided to go with more of a review and synthesis paper, I would recommend removing a substantial portion of the technical details in Section 2 or relocating them to the appendix. The authors should focus on synthesizing the insights from the observed changes, including their potential causes and consequences, rather than describing each change individually on a model-by-model basis. At the same time, Sections 4 and 5 should be expanded and given greater emphasis to better engage a broader audience.
As the manuscript suffers from significant structural issues, I therefore focus my review on general comments. Detailed line-by-line comments are not provided at this stage, as substantial reorganising of the manuscript would likely be required before such comments would be useful.
Citation: https://doi.org/10.5194/egusphere-2026-813-RC2 -
RC3: 'Comment on egusphere-2026-813', Anonymous Referee #3, 15 Jun 2026
The manuscript provides an overview of the marine biogeochemical forecasting systems of the Copernicus Marine Service for the seven forecasting centers producing operational products for the 5 European Seas, the Arctic Ocean and the global ocean. Next to describing the model status as of 2024, the manuscript discusses the advancements of the systems compared to the initial systems from the year 2015. Further, the development of the quality of the operational products is discussed. A selection of particular applications, from published or unpublished work, is included. The status of the operational systems is finally discussed in form of a SWOT (strengths-weaknesses-opportunities-threats) analysis. The manuscript relies heavily on tables to provide the overview over the operational systems of the 7 forecasting centers at the status of the years 2015 and 2024. The long SWOT analysis concludes the manuscript and represents the conclusion. There is no actual summary and outlook.
The manuscript fits into the scope of the journal Biogeosciences in the category 'Review and Synthesis'. I think that it can be interesting for general readers to get an overview of the biogeochemical forecast systems for the Copernicus Marine Service, in particular in their current state and their planned further development. However, I see several significant weaknesses in the manuscript, which the authors would need to resolve before a publication of the manuscript. In general, the manuscript shows little scientific rigor (e.g. superficial or incorrect descriptions, lots of missing references, unsupported claims), but reads more like a technical report, and is in this regard not suited for a publication in a scientific journal. With the comparison of advancements of systems in 2024 compared to the year 2015 it is also directed into past developments, which should be a less interest for general readers. Here, a clear forward-directed focus, e.g. on planned and coordinated upcoming developments should be more interesting. I have the impression that a suitable revision of the manuscript goes beyond a major revision, but rather requires a full reconsideration of the presentation approach of this 'review and synthesis' manuscript.
Major weaknesses:
1. The information in the manuscript is outdated. It describes the status of the operational system as of the year 2024 and compares them with the status of the year 2015. However, the manuscript was submitted in the year 2026. Discussing a system status of 2 years ago hardly looks reasonable. In lines 477/478, the manuscript also states that the systems 'are continuously updated'. Some advances seem to be included in the provided illustrative examples, but they don't confirm the continuous update. The manuscript also appears to be outdated with regard to using the acronym 'CMEMS'. This was used for the 'Copernicus Marine Environmental Monitoring Service'. However, to my knowledge, using 'CMEMS' is no longer recommended by the leadership of the Service since its renaming to 'Copernicus Marine Service'.2. Many of the tables are inconsistent, incomplete and practically unreadable. Using 9-12 table columns, each column is so narrow that many line breaks occur even at grammatically invalid locations. A particularly bad example is Table A7, where the use of bullets amplifies the problem since in some lines only 2-3 letters fit into a column. However, the readability is only slightly better in Tables 1, A1, A2. Also the additions in red text (in contrast to orange-colored background exclusively in Tab. A4) are hardly readable. Further, the red information, which is intended to mark the additions or changes in 2024 as compared to 2015 are sometimes listed before the black text (the status of 2015) and sometimes after it. For some columns, at least in tables A1 and A2, the information is incomplete. E.g. there is a column "Data Assimilation: Code, scheme, Assimilated Obs, Corr.cted Obs., assim. Cycle, Indep data for validation" (please not that 'Corr.cted' is not a typo of me, but actually in the table; in any case this column title already requires some creativity to understand it.), but in the fields for the different systems, the information is not completely given. When it is provided, it often also not understandable for general readers. E.g. In Tab A1, for the MED-MFC, the table on data assimilation mentioned '3DVarBio' and 'OC Chl'. Now, '3DVarBio' is nowhere else mentioned in the manuscript, let aside being cited. Overall, the authors seem to attempt to put too much information into the tables and do not have enough space for a proper formatting or well understandable inputs. For 'OC Chl' one needs creative thinking to interpret 'OC = ocean color' (I think this should rather titled as 'satellite chlorophyll' as is used for other forecast centers (MFCs)). With regard to Tab. A3, the authors also do not seem to know how many MFCs there are. Table A3 states that, e.g., 8 MFCs deliver chlorophyll as an operational product. However, from the previous descriptions and e.g. Tab. A2, there are only 7 MFCs. (I stop here describing more details; looking at the tables it should just be evident that they are a mess - I'm sorry to be harsh here. The authors would need to re-think how to present the MFCs' systems in a clear way; changing just to a landscape formatting is likely not enough.)
3. The authors seem to see advances in data assimilation rather in the assimilation of additional observations instead of the actual algorithms. However, advancements in the DA methodology, in particular the use of dynamic ensembles and perhaps also nonlinear DA, and perhaps 4-dimensional smoother methods are expected to make a large impact in the assimilation skill and hence the forecast quality. In this regard, the discussion on DA should include the aspect of the algorithmic advancements, beyond simple statements that ensembles will provide uncertainty estimates or that machine learning might help.
The authors also give questionable statements about the use of ensembles. E.g. they argue about the cost of ensemble systems and state that they require 'usually at least 10' (line 542) members. There is no reference, but this number does not leave the impression that the authors consulted with their co-author actually running a dynamic ensemble (for the ARC-MFC). I'm not aware of any successful ensemble DA system for biogeochemical models what would run with only 10 members. Pradhan et al., JGR Oceans 124 (2019) 470 were successful using 20 members, but they used a rather coarse global model and the ensemble size seems to be at the lower end of such DA applications. Actually, in the ARC-MFC, as the only MFC-system running a dynamic ensemble, the TOPAZ4 system is described in Sakov et al., Oce. Sci. 8 (2012) 633 to use an ensemble of 100 members. Apart from this, the authors should take into account that a realistic uncertainty estimate from the ensemble can only be expected if the ensemble spread is carefully tuned.
The authors should also provide a sufficient credit to the data assimilation software, in the cases that the MFCs use software that is published itself. Perhaps, it could be meaningful to use a separate table relating to the DA methodology, given that it has an essential role, similarly to the models and observations, for the skill of the forecast systems.4. The authors extensively use the jargon of the Copernicus Marine Service, which is inaccessible for readers not affiliated with the Service. As an independent reader one seems to need a glossary to to understand e.g. 'EIS2024' (in the caption of Tables A1 and A2), NRT (line 81), MY (first occurrence in line 193). Actually, the manuscript misses also a proper explanation of what the NRT and MY products are; this is particularly critical since these are the main product types. However, the are also expressions that I consider a 'marketing'. E.g. using 'green ocean' (first appearance in line 17) is not a scientifically established term, but seems to be extensively and exclusively used by the Copernicus Marine Service. To make the manuscript accessible to readers not involved or experienced with the Copernicus Marine Service, the authors would need to reduce the use of this specific jargon and clarify the terms in the remaining cases.
5. There are also various acronyms used in the text and the tables that are never explained and in many cases also miss a citation.
6. Missing citations are overall a very weak point of the manuscript. There are many claims in the text without a reference or programs like CLIVAR are mentioned without a reference. There are also inconsistent citations. E.g. in line 40, the Topex/Poseidon mission is cited as (Le Traon and Dibaroure, 1999). However, the article is not about this mission, but about 'multiple-satellite Altimeter Missions'. Another example is in Tab. 1: Here, the ERGOM model is cited using a publication from 2021; this is certainly inconsistent with a model used operationally in 2015. In contrast on line 115, the authors cite Neumann (2000), which is referring to the original version of ERFOM, but likely also not describing the model version that is actually used. Also the citations referring to using 'spectrally resolved radiate transfer schemes are misleading. The authors state (lines 182/183): "Some models (BAMHBI, BFM) have recently adopted spectrally resolved radiative transfer schemes (e.g. Terzić et al., 2021; Macé et al., 2025)..." However, both cited papers do not show that use of spectrally resolved radiative transfer schemes in the model. Actually, Terzic et al (2021) does not mention the model BFM at all. Mace et al. (2025) have added a radiative transfer scheme as an observation operator. Thus, the model BAMHBI has not 'adopted' the radiative transfer, i.e. accounting for the radiation spectrum in its dynamics, but Mace et al., are able to compute spectrally resolved radiation from the non-spectrally resolved model dynamics. Overall, the authors would to carefully check the included references so that they are accurate and complete.
7. The introduction section is particularly weak and would need to be rewritten. While it contains a large amount of information (but with insufficient citations), it misses clarity and scientific rigor. Many aspects are sloppily described, like 'The 90's mark a paradigm shift in our capabilities of observing and forecasting the ocean.' (line 37) or 'the Argos' (line 49; meant are 'ARGO floats'). But also descriptions like "In this decade, remote and in-situ observing systems had developed sufficiently to make global, real-time observation possible." (lines 38/39) or "GODAE ended in 2008 and has allowed to achieve significant progress in our forecasting capabilities of ocean physics that has now reached a level of maturity comparable to that of the atmosphere." (lines 51-52) are made without any details or proper citations. Actually, it's not clear that e.g. the Topex/Poseidon mission allowed for 'real-time observation' (real time implies immediate access to data). Also the claim that "forecasting capabilities of ocean physics that has now reached a level of maturity comparable to that of the atmosphere" is hardly accurate. Even today, atmospheric forecasting capabilities seem to be more advanced than those for the ocean, as e.g. the data assimilation capabilities are still more rudimentary in the ocean than the atmosphere.
8. Section 5 provides a SWOT analysis. This is another particularly weak section. A SWOT analysis is a management method, but not a scientific method. In fact, I've never seen a SWOT analysis in a scientific paper. As common for SWOT analyses, also parts of the manuscript text is rather handwaving argumentation. It would be suitable for a manuscript aimed for publication in a scientific journal to perform a proper scientific discussion. In the manuscript there are also inconsistencies. E.g. the subsection 'Threats' misses to state any threats, but rather state the relevance of compute resources or observations. E.g., the text mentions possible insufficient compute resources 'to support the development of (ensemble) (high-resolution) prediction systems and their efficient integration in the European digital twin ocean' (lines 600-601). However, this is not a threat to the current systems, but for ideas on the future development. a threat would be the loss of compute resources. (Note, that the manuscript did not discuss before, e.g. 'the European digital twin ocean' or 'What-if scenarios' (line 590), and why they should be of relevance to the operational forecasting in the Copernicus Marine Service. Here the authors apparently open another discussion, with unclear relevance for the scope of the manuscript). In 'Opportunities', the relevance of the project 'NECCTON' is mentioned as 'aimed at enhancing our prediction capabilities of the biology in an operational framework'. However, as a research project, NECCTON is not part of the Copernicus Marine Service, and while it can enhance the modeling of the marine biology it cannot have direct impact on the 'operational framework'. Only after the research project, i.e. within the actual funding of the Marine Service, the impact on the operations can be realized (BTW: While the authors seem to find NECCTON relevant, they ignored the previous EU-funded project 'SEAMLESS' (https://www.seamlessproject.org/). This enhanced the data assimilation capabilities and e.g. assessed the sensitivity and controllability of the biogeochemical models that are used in the MFCs (see e.g. Ciavatta et al., Prog. Oce., 231 (2025) 103384). Further advancements can be found in other publications resulting from the project or the project deliverables. I can only suppose that some of the research developments of SEAMLESS are already transferred into operational use in some of the MFCs.).
9. The manuscript provides some examples of current developments or even 'preliminary results' (line 525). Overall, these examples seem to be aimed as 'illustrations'. It is unclear how these examples are selected and why particularly these examples are considered to be relevant. They are also partly inconsistently used. For example, Fig. 2 is stated to show the 'Impact of synthetic chl a profile assimilation'. However, the figure only shows the root mean square error when the assimilation is applied, but one cannot see any impact since the comparison to the case not assimilating profiles is missing. In any case, using 'illustrations' is commonly not suitable for a scientific publication.
10. In contrast to the included illustrations, the authors miss completely to give a consistent overview of the developments in the different MFCs. The manuscript type 'Reviews and Syntheses' requires to 'outline future directions'. (see https://www.biogeosciences.net/about/manuscript_types.html). Fulfilling this aspect would call for a consistent and complete description of the expected developments in the different MFCs (e.g. in the next release cycle, if such cycles are formally done) instead of relying on some examples, which seem to be rather arbitrarily selected.
11. The authors claim "The CMEMS biogeochemical modelling systems are grounded on state-of-the-art prediction capabilities and are continuously updated with the latest developments to improve the quality of the service and to meet user’s needs" and "The evolution of the modelling systems, the quality assessment protocols and documentation are coordinated across the seven forecasting centres..." (lines 477-479). However, the manuscript does not leave the impression that this is true.
First it is questionable whether the modeling systems are 'state-of-the-art'. While the authors claim that this is the case (also in line 96), they actually don't even include an information on how they define 'state-of-the-art'. Actually, the MFCs use a wide variety of the models ranging from simpler biogeochemical models with fixed ratios and diagnostic computation of chlorophyll (like ERGOM) to actual quota models in which the stochiometry is dynamically simulated. The models seem to be partly chosen based on what the members of the MFCs had at hand (lines 118/119 mention the chance of a model because of the 'change of partner'). While the simpler models are usable, their skill is likely lower and it is, at least, debatable whether these models can be considered to be state-of-the-art for the complex biogeochemical modeling for the MFCs. If the authors are convinced that their statement is accurate, they would need a proper citation to support it. Further, there is a wide range of sophistication in the used data assimilation methodologies. Using no data assimilation at all, as listed for the IBI-MFC and BAL_MFC in Tab. A1, can certainly not be considered as state-of-the-art. (It is actually irritating in the 'Copernicus Marine Data Store' that there are products denoted 'Analysis and Forecast' that do not include any data assimilation. This violates the definition of an 'Analysis' as the product is a free-running simulation). Further, assimilating satellite chlorophyll was established a long time ago (e.g. Nerger and Gregg, J. Mar. Syst. 68 (2007) 237 produced a reanalysis assimilating chlorophyll observations over 7 years; their DA system already used fixed ensemble perturbations similar to the GLO-MFC and BAL-MFC as described in the manuscript, but also a model of similar complexity as models used in the Copernicus Marine Service). As such it's rather surprising that it is not already common practice in all MFCs. However, also using fixed ensemble perturbations as in the GLO-MFC and BAL-MFC, nudging in the ARC-MFC NRT system, or variational data assimilation with parameterized covariances, can hardly be considered to be state-of-the-art given the known limitations of such methods as compared to using fully dynamic ensemble data assimilation methods (either variational or statistical filters).
Second, the descriptions and tables in the manuscript do not leave the impression of a coordinated development of the modeling systems, in contrast to what is claimed in the manuscript (e.g. lines 478/479). The only common, and perhaps coordinated, aspect seems that a carbonate module was added in all systems of 2024 (but not necessarily using the same formulation), which allows the MFCs to provide carbon-related products. Apart from this, the systems still assimilate a wide variety of different observation types, or even don't apply data assimilation at all, and apply a wide variety of different assimilation schemes. It is not visible that this variety was reduced between 2015 and 2024.12. The authors claim in the 'Author contributions' that 'All co-authors have reviewed, discussed, and agreed on the submitted version of the manuscript'. However, the many typos and grammatical errors in the manuscript, and the issues described above, leave the impression that each of the co-authors rather relied on the other co-authors on performing a careful reading and revision so that a well structured, consistent and readable manuscript would be submitted. From the viewpoint of the reviewer, this is quite unfortunate and the authors should not rely on the reviewers to fix this. To this end, I omit including concrete statements and e.g. typos or further details in inaccurate or unclear parts of the text. Apart form this, the roles of the different authors are not sufficiently clear (a careful reading should also have shown this before the submission): The authors state that co-authors are 'running the biogeochemical part of the MFCs': E.g. 'GC, for MED' and 'HM, LM for BAL'. Now 'running' likely means the daily duty to actually run the model systems. Now I'm wondering whether it is realistic that 'GC' (Gianpiero Cossarini) is really 'running' the model. He is the lead of the group at OGS, and I would be surprised if he really has time to do this daily duty (this might likewise hold for 'MG... for BLK'). For the BAL-MFC, 'LM' is mentioned. However, the is not even an author with these initials. Apart from this it is stated 'CV is responsible for the downstream service organisation, KV for the Ocean State Report in Mercator Ocean International.' Here, I'm wondering how these roles in the Copernicus Marine Serive relate to the manuscript. For outsiders it is certainly not clear what 'downstream service' is and how this might be related to the forecast systems of the MFCs. Given that the manuscript is not submitted as part of the
'Ocean State Report' it is also unclear what active and relevant role 'KV' had for the manuscript (BTW: 'KV' should be 'KvS' as the author's name is not 'Karina von').As a final critical aspect to consider for the authors (sorry for being harsh again):
Overall, a main impression one can obtain from reading the manuscript is that there is a wide variety of approaches and complexity throughout the different MFCs in both the biogeochemical modeling and the data assimilation and that such variety remained over the time period from 2015 to 2024, despite advancements in the different MFCs. A coordinated approach, i.e. an approach that would lead to more unification with a similar skill in the forecast systems, is not visible from the past developments and the manuscript does not leave the impression that one can expect a coordinated approach in the future. Professional operational forecasting requires rigor. This is, however ever not obvious given, e.g., the rudimentary approach to data assimilation in at least some MFCs and indicates lower ambitions than the operational forecasting of weather centers. I'm unsure whether the authors like to take the risk that readers of the manuscript leave with this impression.Citation: https://doi.org/10.5194/egusphere-2026-813-RC3
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 337 | 149 | 18 | 504 | 27 | 21 |
- HTML: 337
- PDF: 149
- XML: 18
- Total: 504
- BibTeX: 27
- EndNote: 21
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
The authors provide a review of the history of CMEMS along with a description of the biogeochemical models and the regions (monitoring and forecasting centers) where they are being deployed. They also present some example model results and evidence for improvements in biogeochemical modeling and data assimilation. The manuscript is not difficult to read, but the text could be improved in many places to make it more accessible and useful for a wider variety of readers.
General comments
I enjoyed reading about the history of CMEMS and bits about the more general history of operational oceanography and data assimilation. Yet, the introductory text sometimes feels repetitive and not very informative. For example, the description of the Global Ocean Data Assimilation Experiment (GODAE) contains 3 mentions of its time frame in the same paragraph: "1998-2008" (l 43), "GODAE was a 10-year international initiative" (l 44), and "GODAE ended in 2008" (l 51). It's international nature is mention twice and so are its achievements: "GODAE demonstrated the feasibility and value of a long-term global ocean observing system supporting real-time ocean forecasting and (re)analysis" (l 47) and "GODAE ended in 2008 and has allowed to achieve significant progress in our forecasting
capabilities of ocean physics". Here, the text could be tightened and more information could be added, making it a better and more informative read without using up more space. Similarly, some of the introduction reads like a list of projects (project X was followed by project Y) without providing much, sometimes any (e.g., GMES), information about the projects. Overall, the introductory texts in sections 1 and 2 could be improved structurally in many places to make them more informative and less repetitive for the reader.
As I was reading through the manuscript, there were a few instances where I thought the level of detail provided was not very helpful to a typical reader. Readers familiar with biogeochemical ocean models were given statements that they knew by heart, while not enough detail was given for readers unfamiliar with the subject. A good example of this pattern appears in line 101: "All these processes are mathematically formulated. These models are NPZD-type models encompassing a wide range of complexities. The models differ by the levels of details in the description of the inorganic and organic (living and dead) components and interactions." These sentences state quite basic facts about biogeochemical ocean models and thus aren't really useful to readers who work with them. Yet, I agree that there is clearly a benefit to describing biogeochemical models to other readers. But what does "these processes are mathematically formulated" mean, what are "NPZD-type models" and how does that relate to "inorganic and organic (living and dead) components". The readers who would benefit from learning more about biogeochemical ocean models are not given enough information and provided with an inadequate description here and in other places. I recommend that the authors go through the technical descriptions included in the manuscript and improve them.
The manuscript text includes many abbreviations, some of which are not explained, even though that would be useful to the reader. I have tried to highlight a subset of those in my specific comments below, but I would recommend that the authors go through the text again with an eye on the abbreviations that are being introduced.
The manuscript makes a brief mention of SOCAT. Overall, there is now a variety of capable statistical models aimed at estimating carbonate system variables but also others, such as nitrate. These include ESPER (Carter et al., 2021, DOI: 10.1002/lom3.10461) and CANYON-B (Bittig et al., 2018, DOI: 10.3389/fmars.2018.00328), and also region-specific versions like CANYON-MED for the Mediterranean (Fourrier et al., 2020, DOI: 10.3389/fmars.2020.00620). These statistical models provide novel ways to assess or improve biogeochemical models and deserve a mention in the "Opportunities and perspectives" section.
Specific comments
Fig. 1 caption: "used in the CMEMS MFCs in 2024 in the seven MFCs": Double-mention of MFCs.
L 28: Mention what "SWOT" stands for.
L 37: Here the tense used switches between past and present: "The 90’s marks [...] In this decade, observing systems had developed sufficiently [...] The ocean surface is observed [...] GODAE was a 10-year international initiative".
L 48: The abbreviation "CLIVAR" is not explained, and neither is "WOCE" in the following sentence.
L 60: GMES started and ended; what happened in between?
L 67: "GODAE transitioned to GODAE OceanView (GOV)": It would be useful to provide a year here.
L 75: "Biogeochemical and biological ocean variables are more difficult to observe routinely using sensors on board satellite and in-situ autonomous platforms.": This sentence is not clear, in that it does not mention what the difficulty is: Is it more difficult to observe biogeochemical and biological ocean variables routinely compared to non-routinely? Is it more difficult to use the sensors on board of satellites and in-situ autonomous platforms to observe biogeochemical and biological ocean variables? I assume the authors mean to compare biogeochemical and biological observations to physical observations; this should be made more explicit and explained better.
L 76: "The situation is particularly critical for biological data that still require field sampling and laboratory analyses.": Well indeed, because these variables cannot be observed at all using remote sensing or autonomous platforms. I would suggest rephrasing to something like, "In fact, some biogeochemical and biological data such as X still require field sampling and laboratory analyses."
L 77: This is not a good use of "on the other hand", because it mentions yet another reason why observing and modeling the green ocean is more difficult than the blue ocean. "Furthermore" would be more suitable, and "on the other hand" or "however" could be used for the next sentence, which describes an advancement despite the difficulties described here. It might also make sense to use "physical circulation model" in this sentence to better make the distinction between physical and biogeochemical modeling.
L 94: "2.1 Model structure": This section title is very short and nondescript and makes the reader expect the description of the structure of a single model. "Structure of the CMEMS biogeochemical models" would be more informative and useful to the reader. A similar remark applies to the following section titled "2.1.1. Description" -- what is being described here?
L 103 and following: The text here is a good example of information spread around and sometimes repeated that is not helpful to the reader: Paragraph 1 in Section 2.1.1 mentions "all the CMEMS models" , provides a general description of BGC models and introduces the two most complex CMEMS models (ERSEM and BFM) without citations. Paragraph 2 broadly describes the models' applications, including regions of application, without providing specifics. Paragraph 3 then finally mentions that there are 6 CMEMS models with citations, and gives a location for each. This is the information that the reader could have used at the start. I would suggest starting out with the six models, mentioning what they have in common, what makes them distinct, and why the large range of marine environments requires these different model formulations.
L 118: "Over the past decade, two MFCs changed their model, for a more suitable model or a change of partner. The ARC-MFC...": Could this information be moved to Section 2.1.2 "Evolution of CMEMS biogeochemical models over 2015-2024"?
L 125 and following: Another example of descriptions that are spread out and could be connected and perhaps shortened. Paragraph 1 in Section 2.1.1 briefly describes and lists all elements in CMEMS models; it then also describes how compartments/variables are used in the model. It mentions biogeochemical reactions/dynamics transfer elements from one compartment to another; it mentions that "these models are NPZD-type models" (l 101). This information is then expanded on and partly mentioned again in paragraphs 4 and 5, which describe the division of phytoplankton into groups and the elemental stoichiometry. A more structured approach could be to introduce the concept of NPZD-type models, including compartments, elements and dynamics, how phytoplankton compartments are representing PFTs, which elements are used, etc.
L 172: "To obtain the complete radiance distribution in time, spatial and spectral space requires solving a full radiative transfer model, which is computationally expensive...": Perhaps mention here that PAR is typically still treated as a single spectral band in many applications. That is, describe the simple approach, the spectral decomposition, and where the CMEMS models fall on the spectrum.
L 193: Explain what "MY" stands for.
Eq 1: Consider calling the "sedimentation" term "sinking and sedimentation".
L 234: "2.4 Spatial resolution": It would be useful to include some numbers -- which resolutions are used by the models?
L 262: "2.6 Data Assimilation": Here it would be useful to many readers to briefly describe data assimilation and its goals.
L 264: "(Ensemble) Kalman filters and smoothers and 3Dvar are the DA methods of choice.": If not too complicated, add which model is using which DA technique.
Fig. 2: Change x-axis labels to dates like in Fig. 3e.
Fig. 2 caption: The use of "synthetic chl a profile assimilation" and "The nudging of the synthetic profiles" is not helpful, because readers may not understand that assimilation and nudging refer to the same process.
Fig. 2 caption: "over the first 100m and 2015-2019 and for 5 depth ranges": I would suggest changing to "for 5 depth ranges in the top 100m in 2015-2019".
L 318: "Four of the seven CMEMS biogeochemical models": I presume these should be the seven MFCs.
Fig. 4: Why are observations shown using different markers (x's and o's)?
Fig. 5: Include "nitrate difference" in the 3rd row colorbar labels, just like the first and second rows also state what is shown. It is not helpful that the caption calls it "free run", but the labels say "no DA run" (or just "no DA").
Fig. 6: The colors are nice, but it would be useful to also include numbers in the center of each square where entries are non-zero.
L 394: It is interesting to see all RMSD values aggregated together into a single distribution. Which data products are mainly contributing to this ratio? Why not divide it up by observation type (as described in the lines above)? Later, I see that some of this is done in the following Fig. 8. I would suggest combining the two figures.
L 482: "carbonates": Does this include pH and pCO2? It might make sense to mention them explicitly.
L 495: "sound quality assessment": This makes it sound like "sound quality" is assessed; I would suggest rephrasing.
L 505: "A coordinated predictability analysis as such done in Leroux et al., (2022) would allow to set a limit to the minimum length scales and the time horizon that are predictable considering the different sources of error, with among them, some that cannot be reduced, at least currently." What exactly does this mean? That a predictability analysis could provide better forecast time windows and resolutions at which the model stays close to observations? Or indicate the key sources of model error? Some more explanation would be useful here -- and breaking up the long sentence, so it becomes easier to understand.