Referee comments on “Deep chlorophyll maximum and nutricline in the Mediterranean Sea: emerging properties from a multi-platform assimilated biogeochemical model experiment” by Teruzzi et al., currently in discussion at Biogeosciences.

Teruzzi et al presents the results from a data-assimilating biogeochemical model of the Mediterranean Sea. This work provides a novel case-study of the dual use of remotely-sensed (ocean color) and in situ (BGC-Argo) biogeochemical constraints in reconstruction of one year of the biogeochemical state of the Mediterranean Sea. Firstly, the data-informed model solution demonstrates fidelity with observations for a number of surface and depth-resolved biogeochemical metrics, such as the vertical position and longitudinal gradients of the deep chlorophyll maximum and its co-variance with the nutricline. Secondly, the study presents compelling evidence for the synergistic benefits of assimilating estimates of remotely-sensed chlorophyll concentrations in tandem with in situ, depth-resolved estimates of chlorophyll.  Thirdly, the study discuses some of the nuanced differences between the impacts of assimilating satellite versus BGC-Argo chlorophyll upon the model solution. The most striking difference discussed is the strong seasonal signatures the different observational streams have upon the solution, with remotely-sensed chlorophyll providing stronger constraint during winter months, and in situ chlorophyll providing stronger constraint during summer months. Finally, implications for optimized sampling strategies, such as the recommendation that BGC-Argo increase sampling frequencies during the ‘influential’  summer season, are discussed.

This work represents a timely contribution to the underway community efforts towards optimizing the design of a global, autonomous biogeochemical observing network for use in constraining reconstructions of the evolving ocean state. I recommend this manuscript for publication after a few minor clarifications and elaborations are incorporated, as I outline below.

Points of clarification:

1. The model setup describes the biogeochemical model as being coupled offline to the dynamical model. Coupled implies two-directional influence, i.e. that the biogeochemistry is both impacted by and impacts the dynamical fields. Is this the case?
   1. If it is the case that the model set up is “coupled” in the true sense of the word, then the impact on the dynamical fields should also be presented.
   2. If it is not the case, and in fact the biogeochemical model or assimilation of biogeochemical fields does not feed-back or modulate the underlying dynamical fields, then a more appropriate term to use is “forced” or “driven” – i.e. Line 135 should read “the MedBFM was forced or driven offline with output from MENO3.4-OceanVar…”
2. To improve accessibility, an additional paragraph of description or context of Eq. 1 should be included. This would involve, for instance, expalaination of the “innovation” term (Line 159), and discussing the significance or intuitive purpose of the different covariance vectors.
3. Regarding the “Impact indicator” metric:
   • (a) It needs to be explained that the impact indicator given by Eq 2, although it does not contain an explicit ‘directionality’ of the impact (i.e. towards better or worse agreement with the data-constraints), that the assimilation methodology works to push the model solution towards the data, therefore any non-zero value of I(t) represents a nominal improvement in model-data fit. From the equation alone, and without sufficient prior understanding of the methodology, this is not obvious, making it difficult to interpret if the additional streams of assimilated data are merely influencing or in fact improving the solution.
   • (b) The maps of Fig. 6 and Fig. 7 present a very derived / abstract metric. These figures present the 50^th and 90^th percentile ‘impact’ of assimilated the given field during the given season. Interpretability of this abstracted metric would improve if an additional subpanel was included in each figure which showed a representative distribution of the impact-indicator for a single grid-cell (or averaged over a region), with markings at the 50^th and 90^th percentile. This could also provide opportunity to contrast the summer vs. winter distributions of the indicator. For instance, for chlorophyll, the winter distribution of the indicator would be shifted toward "1" while the summer distribution would be closer to "zero".  See a mock-up below. This will help orient the reader as to what the maps are presenting.
   • 
4. The conclusions section would benefit from a final sentence that poses the significance of the study within the context of future advances in biogeochemical data assimilation within basin and global domains, i.e. something like “ The multi-platform assimilation yielded improvements in model representation of large-scale (hundreds to thousands of kilometers) bio-dynamical features and is suggestive of the applicability of this advancement to reconstructions of other ocean regions and the global domain.”

Minor editorial and stylistic suggestions:

L7-10:  Rewrite to something like: “Data assimilation has lead to advancements in biogeochemical modelling and scientific understanding of the ocean. The recent operational availability of data from BGC-Argo floats, which provide valuable insights into key vertical biogeochemical processes, stands to further improve biogeochemical modelling through assimilation schemes that include observations from floats in addition to traditionally assimilated satellite data." (bold is new)

L16: “maximum depth, intensity and nutricline depth” (added a comma and removed the first ‘and’)

L39: Add the following reference, which also provides motivaiton for the direct use of optical properties in data assimilation:

Dutkiewicz, Stephanie, Anna E. Hickman, Oliver Jahn, Stephanie Henson, Claudie Beaulieu, and Erwan Monier. 2019. “Ocean Colour Signature of Climate Change.” Nature Communications. 10 (1). https://doi.org/10.1038/s41467-019-08457-x.

L40: Ocean-colour observation assimilation takes advantage of the frequent, large-scale satellite observations of ocean properties related to the microbial biology of the upper ocean. 

L42: “deeper ocean layers requires approximations and assumptions.”

L45: “localization” – what does this mean?

L66: “ideal” – replace with “suitable” or “appropriate” – as it could be easily argued that a region with less coastal margins and more open-ocean conditons, where ocean color is more reliable, would be a more ‘ideal’ location to do a multi-platform assimilation.

L348: “In our results, this hypothesis was supported by higher nutrient uptakes in the western…”

L409: Remind readers of what Vv means, i.e. “ Vv (vertical co-varience error)”

L415: Remind readers of what Vb means, i.e. “ Vb (biogeochemical co-varience error)”

Review of “Deep chlorophyll maximum and nutricline in the Mediterranean Sea: emerging properties from a multi-platform assimilated biogeochemical model experiment” by Teruzzi et al. 

The manuscript addresses the performance of a 3D-Var biogeochemical data assimilation system, constrained with both chlorophyll data from satellite and chlorophyll and nitrate data from BGC-Argo floats, that is applied to a realistic simulation of the Mediterranean Sea for the year 2015. After demonstrating the validity of the method, the authors use their product to investigate the spatial and seasonal variability in the vertical structure of chlorophyll and nitrate fields. The problem is introduced clearly, methodology appears sound, and results are compelling. I recommend the manuscript for publication in Biogeosciences after revisions. Below are a couple of important results that need to be clarified, a list of issues to be addressed, and some minor comments. 

1) Fig 2: With the exception of the “LEV” region, it seems that assimilating only float data yields a better fit with chl obs than assimilating floats + satellite. It seems like it would be best to ignore satellite data, even at the surface. On the other hand (Fig 3), assimilating satellite Chl yields a better fit with nitrate obs, which is counter intuitive, and assimilating float Chl barely has an impact. Please comment! 

2) L. 242-244, 260-261: In Fig 4 I don’t see a reduction in RMSD, but instead an increase in levels 4-6. So data assimilation does reduce the model skill in fitting oxygen? It looks like the pink and red curves are on top of each other, suggesting that satellite Chl is responsible for degrading the O2 solution. 

Other comments:

3) Abstract and L. 68: is “semi-independent data” a common way to refer to observations used in model-data comparisons before and after assimilation? I suggest changing “semi-independent data (before assimilation)” to “assimilated data (before and after assimilation)”. 

4) When data is excluded from assimilation based on a DA criterion, as stated in L. 98-100, is that a “quality check” on the data? Or is it removing observations that can’t be fit because of model inadequacies? Is there a reference for the threshold values (5 mg/m3 for Chl and 1 or 2 mmol/m3 for NO3)? Similarly on L. 87-88 “A further quality check on satellite values before the assimilation resulted in the exclusion satellite chlorophyll observations whose mismatch value with respect to the model was higher than 10 mg m-3.” - Is this a check of the quality of the data or a DA-based exclusion criterion?

5) L. 96-97: “chlorophyll profiles were checked for negative values (rejection)“ - I imagine that a bias correction is applied to the Chl data, resulting in sometimes negative values at some depths. Does that necessarily mean the profile can’t be used? Maybe negative values could be viewed as “zero”? 

6) L. 115-116: “3DVarBio is the data assimilation scheme for the correction of phytoplankton functional type and nutrient variables (i.e., nitrate and phosphate)” - It sounds like phytoplankton type is a control variable, which to me suggests that the type can be changed by assimilation. However, I believe you mean that the biomass in each phytoplankton class can be optimized. It would be useful to know how many functional types there are (it’s mentioned later, but at this point I was wondering). Also, this is one of a few places where phosphate is mentioned. Why control phosphate and not other bgc variables, like oxygen? Does the model include other nutrients?

7) Section 2.2 is hard to follow, with so many models and acronyms. I would start by describing MedBFM, then OGSTM and BFM, then 3DVarBio. What does it mean that the transport model is “fully consistent with the off-line coupling of the NEMO3.6 vvl”? On L. 135 the “MENO3.4-OceanVar model” is mentioned: should it be NEMO? If so, why mention NEMO3.6 earlier when you’re using version 3.4?

8) L. 141-145: What does it mean for climatological profiles to “integrate” data? Is it different from initializing a model from data and spinning it up for a couple years? What is a “Newtonian dumping term”?

9) L. 145: Tracer concentrations need to be relaxed to climatology even though the simulation is only 1-year long? Is that typical for the Med Sea?

10) Comment on the fact that RMSD increases at the surface in TYR when only satellite chlorophyll is assimilated?

11) L. 281: What is the significance of the 0.3 value?

12) Fig 7: Why does the y-axis stop at 300m? There is a large change at 300m in summer, TYR. Caption: “5 left panels” should be “5 right panels”. It would be helpful to label figures panels “a”, “b”, “c”, etc. Why are some panels empty?

13) L. 429-430: The results are consistent with previous studies. Can you elaborate on whether the results are quantitatively different? Or do they simply agree with previously known features of the Med Sea?

Minor comments:

- L. 84-86: “Original products […] were reviewed for spikes excluding observations whose anomalies were higher than 3 times the daily climatology standard deviation” - The sentence is not clear. Maybe rephrase to “In the original products […] observations whose anomalies were higher than 3 times the daily climatology standard deviation were excluded in order to remove spikes”.

- L. 122-123: Is it only the optical model this is not included in the present application (but the other 2 are)? Why isn’t it included?

- Fig 2: The gray lines hard to see; I suggest removing other grid lines.

- L. 286 p5th = 95th?

- L. 384-385: “Maintaining the diagonal the observation error covariance matrix” - remove second “the”?

- L. 393: “Thus, while it is desirable an increase of nitrate sensors number” - should be “while it is desirable to increase the number of nitrate sensors”?

- The Verdy and Mazloff reference is missing from the bibliography.