This is a very interesting and valuable contribution to the present scientific literature addressing the impact of (DEnKF) assimilation of physical and biogeochemical data, including Argo floats, on the simulated state of the ocean. More specifically, the paper addresses an important question of the impact of (mostly) surface data assimilation on sub-surface physics and biogeochemistry, focusing on the Gulf of Mexico. The simulated physical and biogeochemical tracers are validated with an independent BGC-Argo data-set. The paper is well written and I have mostly only minor comments to address, which can be found below. One remark: please note that in the downloaded .pdf version of the manuscript the first digits of the three-digit line numbers are running off the left side of the page, so beneath the section 1 I avoid referring to the line numbers and refer to the paragraph and section number instead.

Section 1:

The introductory section is nicely written and the only comments that I have are regarding the use of references:

- in particular on line 52 when the paper talks about assimilation of optical properties, the included references are very far from exhaustive. On top of my head I could add to the list the following papers: 1. Jones et al (2016), Biogeosciences, https://doi.org/10.5194/bg-13-6441-2016, 2. Gregg and Rousseaux (2017), Frontiers in Marine Science, https://doi.org/10.3389/fmars.2017.00060, 3. Skakala et al (2020), Journal of Geophysical Research: Oceans, https://doi.org/10.1029/2020JC016122. I completely understand that sometimes there simply are too many references and you cite only the first ``pioneering’’ papers, but in such case it would be good to put “e.g.” in front of the citations, or add after the citations “with few more recent follow-ups”, or something similar. Otherwise this might look like the cited papers are all that has been published on the topic and it might lead to the other papers being omitted when someone new-to-the-area uses your references as part of literature review on the topic. Similarly to this, I would make sure you somehow emphasize that the citations aren’t exhaustive also in the other cases, e.g. the phytoplankton size-class chlorophyll has been assimilated also in Ciavatta et al 2019, Journal of Geophysical Research: Oceans, https://doi.org/10.1029/2019JC015128 and for total surface chlorophyll there’s many references left out.

- more importantly the text on the lines 65-70 suggests that the subsurface validation was done only in the 2 cited studies and only on climatological, or basin-wide scale. Firstly, I believe it is appropriate to include here the Cossarini et al 2019 reference that is cited elsewhere in the paper (which looks at vertical profiles along BGC-Argo locations), but secondly there is a new study of Skakala et al (2021), Journal of Geophysical Research: Oceans, https://doi.org/10.1029/2020JC016649 that validates both free runs and surface data assimilative runs along glider trajectories (in the North-West European Shelf) to determine the impact of surface data assimilation (and some other DA as well) on the simulated sub-surface tracers. This study, although it uses a very different, 3DVar, system, is of direct relevance here, since it addresses similar questions to this submitted manuscript, and, as mentioned before, it does model validation along a specific glider 3D transect (which is of course spatially limiting), rather than on basin-wide scales. Btw. please note that there is another study that assimilated BGC-Argo oxygen data that perhaps deserves to be cited in this paper: Verdy and Mazloff, Journal of Geophysical Research: Oceans, https://doi.org/10.1002/2016JC012650.

Section 2:

I find it would be perhaps beneficial for the reader to provide slightly more information on the system set-up and maybe a bit rearrange the text. In particular:

- could you please mark the exact model domain? The Fig.1 shows the Gulf of Mexico region with its bathymetry, but I guess what’s in Fig.1 is not precisely the spatial model domain (?) Or is it the red rectangle in Fig.1, which however isn’t explained (?) Maybe you can mask out the ``irrelevant’’ regions?

- a very minor (bottom of first paragraph in the section 2.1): could you please provide a rough km scale for the 1/8 degree resolution. Of course this is slightly variable depending on the latitude, but is it something like ~12km?

- again extremely minor: just beneath Eq.2 “measurement operator” should be “observation operator”?

- I would perhaps suggest to put the first two paragraphs in section 2.4 behind the section 2.1, or maybe behind the section 2.2, so you say clearly how the ensemble is generated, how many members are run (and so on), around the same time when you talk about ensemble DA. I got firstly confused how little information is provided about the ensemble generation until I realized it’s been put behind the “Observations” section.

- regarding the ensemble generation: how was the parameter sensitivity determined, in order to perturb the ``right’’ parameters to account for the model uncertainty? Physical model parameters weren’t perturbed (e.g. vertical diffusion parameter values)? If so, why? I know 20 member ensemble is itself limiting, but perhaps some more reasoning on how the ensemble was generated would be desirable..

- it seems that the observation uncertainties don’t come with the observation product (?), but based on what you estimate the 35% observational uncertainty for total chlorophyll, or the physical observation errors? This might be of importance, since you are aiming at estimating model variances quite accurately with the ensemble, so the true state estimate can be quite sensitive to the observation error estimates.

- btw do you assimilate chlorophyll, or log-chlorophyll? If the former, how do you treat the problem of non-Gaussianity?

– what exactly is the 7 day window assimilation at the end of third paragraph of section 2.3, and how exactly does it help with data sparsity, should be explained more in detail.. E.g. is this that you assimilate more data that have been otherwise merged in the OC CCI products?

- I am not particularly great expert on EnKF, but maybe you could say for a general readership a bit more about how exactly are the correlation length-scales (especially the vertical) calculated from the ensemble? There is something on the spurious correlations and localisation, but perhaps you can say a bit more? Also I understand correctly that SSH, T & S are updated without updating also the horizontal velocities? Could that be also discussed slightly more?

- the skill metrics at the end of section 2: why don’t you also include the overall bias? I find that calculating bias always carries worthy information (at least worthy enough to be mentioned somewhere in the paper).

Section 3

Nice, just some relatively minor clarifications would help:

- section 3.2: I find it surprising that addition of Argo (physical) data to the assimilation (Daargo) has such impact on surface chl (Fig.6), especially if it didn’t change much T & S (Fig.4)? Could you comment a bit more?

Section 3.2 paragraph 2: where are the nitrate data coming from? Maybe I’m getting confused, but nitrate data weren’t mentioned between the BGC Argo data in section 2.

specific Figure-related comments:

Fig.3: although I understand what the Figure is trying to demonstrate, it might be better to show the model – observation differences for the different runs, or at least show them for the free run next to the existing panels?

Fig.4: its interesting how little difference there is in subsurface T between sat DA and argo DA! Can you discuss how much of this is due to the assimilation length-scales and how much due to the model dynamical adjustment? Btw what was the relationship between the spatial positions of Argo floats and BOEM floats? This hasn’t been shown..

Fig.7: perhaps in this case there is no need to reproduce exactly Fig.4, since the nitrate concentrations are very similar between the three simulations? Maybe it’s better to have the first row and then differences between the DA and the free run, since this would show more clearly the changes introduced by DA? Also why there aren’t the observed nitrate concentrations similarly to T in Fig.4? It’s interesting that nitrate and surface chlorophyll aren’t correlated as much as nitrate with temperature (Fig.10), I understand that’s because of non-linearity, but in my region and model of experience this non-linearity still produces overall strong correlations, just highly variable in time (including changing signature)..

The manuscript by Wang and Coauthors investigates the impact of the assimilatio of satellite surface observations to both physical and biogeochemical variables in a coupled model of the Gulf of Mexico physical and biogeochemical dynamics. Independent (non-assimilated) profiles form five BGC-Argo floats are used for validation. Results provided in the manuscript highlight interesting aspects on the capability of DA (data assimilation) to effectively correct ocean simulations, on the difficulties arising in coupled physical and biogeochemical assimilation, and on the relevant role played by prior model calibration in DA.

The manuscript is well written and of interest for the scientific community considering recent and foreseen upgrades in physical-biogeochemical ocean DA. My review is limited to few points (most of them minor) that I think will further improve the manuscript quality.

Since line numbers are corrupted in the manuscript file, in the comments hereafter they are indicated by # followed by the part of line number visible in the manuscript. Page numbers are also provided together with line numbers.

1. I suggest to introduce the alternative parametrization of the light absorption in a different way. In the manuscript it is currently described as an alternative that has been considered after investigating the results of previous simulations. I think that presenting this formulation as an alternative since the beginning would better emphasize the role of prior model calibration. Thus, I suggest to describe the alternative formulation not in a temporal framework (i.e., without specifying that it has been applied after previous simulations) as it is currently done in the abstract and in the manuscript sections. In particular: i) the first paragraph of Section 3.3 could be moved and adapted to Section 2.4; ii) in Section 2.4 the Authors could indicate that five (instead of three) simulations were performed; iii) it can be further stressed through the manuscript that the alternative formulation for the light absorption was adopted to investigate the sensitivity of subsurface DA impacts to model calibration (and in particular to the light penetration formulation); iv) the abstract should be adapted accordingly.
2. I think that the comparison with independent BGC-Argo floats is a valuable aspect of the manuscript, however, it would be interesting to know the spatial-temporal distances of the non-assimilated profiles with respect to the assimilated ones. Did they cover similar areas of the gulf of Mexico? And in the same period? In my opinion clarification on this aspect would help to better understand and comment the relatively small impact of Argo profiles assimilation when compared to the independent BGC-Argo data. Moreover, this could help also in commenting the differences between the two maps of Fig. 6 (are the differences mainly located close to assimilated Argo profiles?).
3. I suggest to insert some comments in Sec. 3.3 about the impact of the alternative parametrization on RMSE with respect to satellite chlorophyll. Results of tab. 2 show that RMSE in Free_alt is slightly higher than in Free, but on the other hand the improvement due to the assimilation (DAsat_alt) is relatively higher. I think that this point could be further highlighted and commented in the manuscript.
4. From the sentence at lines 18-19 (p. 1) in the abstract it seems that the model was tuned using BGC-Argo in the present. However, the tuning was made in Wang et al. (2020) (lines #08-#09, p. 4). Tha Authors should consider to rephrase the sentence in the abstract.
5. 16 p. 1. I think that the correct term for biogeochemical Argo floats is BGC-Argo instead of BGC Argo. Please, check other occurrences in the whole manuscript.
6. 94 p. 4. Probably Figure should be abbreviated with Fig. (please, check other occurrences).
7. #09 p. 4. Concerning the re-tuning of the half-saturation constant of nitrate, was it is done similarly to Wang et al. (2020), i.e. based on BGC-Argo? If not, could you explain how the updated value of the parameter was obtained?
8. #32 and L. #36 p. 5. I suggest to check if Equ. Is the correct abbreviation for Equation in Ocean Science Journal.
9. #53-#54, L. #61-#62 p. 6 and L. #58 p. 7. How were the values of observation errors defined? I suggest to add some references or details on criteria used.
10. #88 p. 7. Probably an and is missing in the sentence: zero mean and variance of 1.
11. #74 p. 11. I am not sure that the term subsurface is fully consistent here. Indeed, I would say that the model fails to simulate the high spatiotemporal variability also close to the surface, more generally the whole euphotic layer is affected by the issue.
12. #94 p. 11. How were the two light attenuation parametrizations calibrated? Could you provide some details or references about?
13. #16 p. 12. Since the large use of BGC-Argo floats for model validation demonstrated in Salon et al. (2019), consider if it is relevant to add it in the listed references.
14. #62-#63 p. 14. As far as I know, in Goodliff et al. (2019) the muting of the multivariate update concerned not only chlorophyll but all the phytoplankton variables (whilst multivariate updated was maintained for nutrients and oxygen).
15. Fig. 10, I suggest to insert measurement units (at least in the caption).
16. Accordingly to comment 1, I think that in Discussion and/or in the Conclusion the need of an a priori well calibrated model could be further stressed by the results obtained using the alternative light attenuation parametrization, since it is an example of how DA benefits can be limited by a parametrization that it is not fully consistent with the modelled processes.

This study implements DEnKF-based data assimilation of satellite sea surface temperature, sea surface height and chlorophyll, and in situ temperature and salinity, with a physical-biogeochemical model of the Gulf of Mexico. The results are validated using the assimilated data plus profiles from five independent BGC-Argo floats, with a particular focus on subsurface biogeochemistry. Following the validation, a change was made to the light attenuation parameterisation to improve the fit to BGC-Argo chlorophyll.

The paper is well written and the experiments and results useful and well described. Subject to a few minor revisions, detailed below, I recommend publication in Ocean Science.

P2 L41 - "discretion schemes" should presumably be "discretization schemes"?

P3 L65-70 - while less focus has definitely been given in the literature to validating the subsurface than the surface, it's not as rare as this paragraph would suggest. Some of the studies already referenced, and more besides, perform some validation of the subsurface, including on a point-to-point basis. See e.g. Fontana et al., 2013; Ford and Barciela, 2017; Mattern et al., 2017; Cossarini et al., 2019. The paragraph should be rephrased accordingly.

P3 L74 - I think "Garcon" should be "Garçon".

Is any transformation performed for chlorophyll to deal with non-Gaussianity?

Do the observation errors used account for representation error? Is this likely to affect results?

P8 L04 - would additional metrics, even just bias and correlation coefficient, give additional information? Furthermore, chlorophyll and other biogeochemical variables are not normally distributed. Is RMSE appropriate here?

P10 L56 - can the uncertainties be quantified?

P11 L87 - how was significance calculated?

P11 L93-94 - what method was used to calculate the old and new parameterisations?

Please comment further on why Fig. 9 shows limited improvement in the free run with the new parameterisation. Does this imply that the parameterisation is too tuned to the BGC-Argo data, or that state-parameter estimation might be the best long-term approach?

Fig. 1 - please state what the red square represents. Is this the model domain?

Fig. 3 - please include units where appropriate. (a-e) should probably be (a,b,d,e).

Fig. 7 - should the observations be included in a subplot here, as in Fig. 4?

Fig. 10 - what is meant by DCM increment? DCM is not one of the state variables included in the assimilation.

Fig. S3 does not appear to be referred to in the text. More generally, the figures in the supplement could reasonably be included in the main paper, but I appreciate the desire to limit the number of figures.

The study is investigated the ability of available satellite information on ocean surface physical and biological properties to constrain and improve simulated subsurface biogeochemistry in the Gulf of Mexico. The study also shows an example of using complementary Argo data. In this respect the paper nicely suits the frame of the journal. Generally, the manuscript is well structured and clearly written. The figures are of a good quality. I have got just minor comments (please see below) the authors might still want to consider before publishing the manuscript.

Title: an edit is required ”…assimilation of satellite physical and biological observations …”.

It is a bit confusing: the title emphasizing the use/role of satellite information in improving subsurface biogeochemistry, however Argo data are also assimilated. Could the title reflect the use of Argo data?

Specific comments:

Line 15:16. reads as BGC-Argo data are also assimilated complementary to the satellite data. Reads a bit contradictory to the title or vice versa the title reads contradictory to the statement.

P1, L18: “… into a three-dimensional biogeochemical model …”

P2, L30-31: How was the tuning done? To a certain extent it is still a kind of assimilation of the information.

P2, L41: “discretization and numerical schemes” instead of “discretion”

P2, L45: “.” is missing in the end of the sentence.

P2, L48: suggest to add “(e.g. Chla)” after “satellite data of ocean colour have been the major source of observations”

P2, L51: correct reference is Pradhan et al., 2020

P4, L85-86: are the mentioned five BGC-Argo floats really independent if used for the model calibration (model optimisation even though by "trails-and-error")?

P5, L31: “observational operator” instead of “measurement operator”

P6, L53-54: The specified (assumed) observational errors for SSH and SST are quite small, which could lead to overfit to the data with possible deterioration of the state for other model variables.

P7, L79-80: Inflation is normally introduced to account for uncertainties in approximation of model error (due discrepancies in the forcing or internal model parameter/ parameterisations), which consequently alters the ensemble spread.

P7, L89-91: It would be nice to provide a reference to a study on model sensitivity to these particular parameters? A motivation and a reference to a procedure of parameter perturbation would support. Please also make it clearer whether the parameters are perturbed just to introduce more stochasticity to the system (e.g. Pradhan et a. 2019, 2020) or the data assimilation experiment considers also parameter estimation (Doron et al., 2011, Simon et al. 2015).

P8, L1-3: Please consider rephrasing this sentence. The length of the state vector should not affect crucially the computational cast. Normally the time required for the analysis (independent on the length of the state vector) takes much less than the computational expenses required for running the ensemble itself. It is worth providing another argument for justification of the choice of model variables included to the state vector.

P8, L4-12: It is worth showing both criteria RMSD and unbiased RMSD (+ additional? bias). In this case it would be clearer for the reader for which model variables the solution deviate systematically or randomly from the observations, whether the data assimilation allows to reduce bias (if any) or random part of the obtained differences between model and observations.

P9, L24: "This figure shows" or "This comparison shows"

Figure 2 could be slightly increased.

Supplement, Figure S3: a legend or more detailed figure caption is required.

References:

Doron, M., Brasseur, P., & Brankart, J.-M. (2011). Stochastic estimation of biogeochemical parameters of a 3D ocean coupled physical–biogeochemical model: Twin experiments, Journal of Marine Science, 87(3), 194– 207. https://doi.org/10.1016/j.jmarsys.2011.04.001

Pradhan, H. K., Völker, C., Losa, S. N , Bracher, A. and Nerger, L. (2020): Global Assimilation of Ocean-Color Data of Phytoplankton Functional Types: Impact of Different Data Sets , Journal of Geophysical Research-Oceans, 125 , e2019JC015586 . doi: 10.1029/2019JC015586

Simon, E., Samuelsen, A., Bertino, L., & Mouysset, S. (2015). Experiences in multiyear combined state–parameter estimation with an ecosystem model of the North Atlantic and Arctic Oceans using the ensemble Kalman filter. Journal of Marine Systems, 152, 1– 17.