the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 27 Jul 2026
Physical‑Biogeochemical Coupling in the Baltic Sea: How Joint Assimilation of Chlorophyll‑a and SST Reshapes Thermal Structure and Ecosystem Representation
Abstract. The coupled physical–biogeochemical dynamics of the Baltic Sea are challenging to simulate due to its complex bathymetry, strong stratification, and high optical turbidity. This study evaluates a multi-source data assimilation framework using a three-dimensional coupled model, comparing univariable (chlorophyll‑a only) and multivariable (chlorophyll‑a plus sea surface temperature) assimilation strategies. Our results quantify the vertical and dynamical limits of surface-constrained assimilation in stratified marginal seas. Satellite chlorophyll‑a assimilation markedly improves the spatial and seasonal representation of surface phytoplankton biomass, but in situ profiles are necessary to accurately reconstruct the subsurface chlorophyll maximum. Physical and biological constraints are complementary, and their joint assimilation yields a synergistic improvement, substantially reducing thermohaline biases and enhancing overall model consistency. Although chlorophyll‑a increases can modify subsurface heating through bio‑optical feedback, deep‑water oxygen profiles at monitoring stations remain insensitive to surface‑focused constraints and are governed instead by physical ventilation and remineralization. Crucially, the joint configuration provides a cross‑variable regularization that prevents noisy biological updates from degrading the physical state. These results demonstrate that, in optically complex marginal seas, joint physical–biogeochemical assimilation is essential to overcome the limitations of univariable frameworks and to ensure that surface observations translate into an accurate representation of the full water column.
- Preprint
(15371 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on egusphere-2026-3657', Anonymous Referee #1, 30 Sep 2026
-
RC2: 'Comment on egusphere-2026-3657', Anonymous Referee #2, 02 Oct 2026
This paper describes a set of data assimilation experiments for the Baltic Sea, using a two-way coupled physical-biogeochemical ocean model and a weakly-coupled LSEIK data assimilation system. Compared are a free run (REF), a run assimilating chlorophyll from ocean colour (CHL_SAT), a run additionally assimilating chlorophyll from in situ observations (CHL_3D), and a run additionally assimilating satellite SST (CHL_SST). Compared with the assimilated data, all runs reduced errors in near-surface chlorophyll, with the addition of chlorophyll profiles further improving subsurface chlorophyll. Assimilating SST improved model SST and brought some improvement to model chlorophyll. Assimilating chlorophyll had a small but generally beneficial impact on the SST. Little impact was found on deep water oxygen. The manuscript concludes with a discussion on the interpretation of these results and their implications for modelling and assimilation for the Baltic Sea and similar waters.
The paper is generally well written, interesting and relevant to the community, and a good fit for Biogeosciences. I have various concerns I would like to see addressed before recommending publication though. Some of the statements made, while likely accurate, are speculative and not backed up by corresponding analysis of the results; this analysis would be relatively straightforward to add and would strengthen the paper greatly. Furthermore, greater clarification would be useful on how some of the data is used, particularly the OLCI data, and any use (or not) of independent observations for validation.
General comments:
1. Use of OLCI data
The study assimilates both the multi-sensor and OLCI-only products, despite OLCI data already being included in the multi-sensor products, essentially meaning the OLCI data are double-counted. Was this accounted for in the assimilation methodology?
The manuscript justifies using both due to the different resolutions of the products:
“The combined assimilation of these datasets enhances spatial coverage and representation of variability across scales relative to a single-product approach. Differences in sensor characteristics and resolution result in complementary sampling from basin to coastal scales. Specifically, the 1 km dataset constrains basin-scale variability, while the 300 m dataset captures finer-scale features, particularly in coastal and heterogeneous waters. Their joint use enables consistent representation of large-scale patterns while improving the characterization of small-scale phytoplankton dynamics.”
This seems reasonable on the face of it, except that the model being assimilated into is 3.7 km resolution, so cannot resolve processes even on the scale of 1 km, let alone 300 m. Accordingly, “When multiple observations are available within the same model grid cell and depth level, they are averaged to provide a representative value.” This seems to negate the argument made above?
The very different behaviour seen in the OLCI data compared with the multi-sensor data in Fig. 5 is intriguing and worrying. Further assessment of the differing product characteristics, or at least references to any studies which have looked at that, would be very useful. I am not an expert in Baltic Sea phenology, but the OLCI data does not show the kind of seasonal cycle I would expect. The discussion that:
“None of the DA experiments are capable of replicating this early-season peak in Sentinel-3 product. This persistent mismatch originates from a hard-coded structural thresholding within the biogeochemical formulation rather than a failure of the assimilation scheme itself (Eilola et al, 2009; Ruvalcaba Baroni et al., 2024). Specifically, overly restrictive thresholds for minimum light availability and background phytoplankton biomass artificially suppress primary production under cold, low-light, early-season conditions. Therefore, this limits the model’s intrinsic ability to initiate the late-winter to early-spring bloom, regardless of the strength or configuration of the observational constraints applied.”
is potentially very useful and interesting to the community, but needs further interrogation. Firstly, it assumes the OLCI data are correct (implying the multi-sensor data are in error) which needs to be demonstrated. Secondly, to conclude that the thresholds are “overly restrictive” it would be useful to know what they are and how they compare to other models and lab studies. Suppression of primary production in low-light conditions is deliberately built into models based on empirical understanding, and so use of the word “artificial” could be disputed. Perhaps SCOBI does use more restrictive values than other models or observational studies would suggest, in which case this has revealed a model tuning issue. But if the understanding on which the model (and probably others) is built is flawed, then this is a valuable result for the wider community that merits greater discussion.
2. Validation
It’s not clearly stated in the manuscript, but it looks like all the SST and chlorophyll data used for validation is the same data that’s used for assimilation, meaning it’s not independent. When data is sparse this can be unavoidable, but it does mean the validation results come with caveats. The purpose of a data assimilation scheme is to bring the model closer to the observations, so comparing to the assimilated data essentially just demonstrates that the assimilation scheme is coded correctly. That doesn’t mean the validation presented isn’t meaningful, but in the absence of independent data sets the validation is not as robust as it could be and that caveat should be clearly stated in the manuscript.
The oxygen data used is independent which is great. I’ll discuss this more in my next comment, but I would like to see further analysis of unassimilated variables, whether compared to observations or not. Where this is the case, as with the oxygen data, please introduce these observations in Section 2 alongside the SST and chlorophyll data, making clear which data sets are used for assimilation, which for validation, and which for both.
Finally on this point, validation is performed using metrics such as bias and RMSE. These work best for variables with a Gaussian distribution, whereas chlorophyll is highly non-Gaussian. Robust equivalents such as median absolute error may be more appropriate.
3. Non-assimilated variables and vertical structure
Further assessment of non-assimilated variables, particularly nutrients and 3D temperature and salinity and density, would bring interesting insight into the impact of the assimilation on the wider model state, and help support some of the statements and hypotheses made. Ideally this would include validation against observations, but simply presenting model results would be sufficient.
In particular, presentation of the vertical density structure. Much is discussed about the importance of the pycnocline depth for instance, and the impact of the assimilation on the physical environment that governs nutrient entrainment and phytoplankton growth. But the only physics variable presented is SST, and nutrients are not presented. This means much of the discussion feels speculative, rather than grounded in results.
This feels especially important given that it’s repeatedly pointed out that the Baltic is primarily salinity-stratified, which presumably limits the impact that could be achieved by just changing the thermal structure through SST assimilation and feedbacks from chlorophyll.
I’ll highlight examples of where this would be useful as part of my “specific comments” below.
Specific comments:
L53: “Assimilation of biogeochemical variables, including nutrients, dissolved oxygen, and occasionally Chl‑a, has recently emerged”. Chl-a from ocean colour is by far the most commonly assimilated biogeochemical variable, so I’m puzzled by “occasionally”. “Recently” is a relative term, but the first chl-a assimilation study I’m aware of was Ishizaka (1990) (https://doi.org/10.1029/JC095iC11p20183) so 36 years ago. Although it has only become established operationally much more recently.
Section 2.1: Given its importance to the study, please add details of the two-way coupling and light penetration scheme.
L99: “A coupled ocean–biogeochemical modeling system was configured for the Baltic–North Sea” – if that was specifically done in this study, more details of the model need to be presented. If the modelling system was pre-existing, please provide references where more information can be found.
L99: “Baltic–North Sea” – presumably this means observations were only assimilated in part of the model domain (the Baltic)? In the next section on use of observations, this would be worth clarifying.
Fig.1 does not seem to be referenced in the main text.
L133: “statisticaled” – I’m not sure what is meant here.
L147: “all data were filtered using quality flags” – please provide more details.
L159-161: “(b) Basin-wide spatial sampling density mapping derived via smooth Gaussian kernel density estimation. The color scale highlights the geometric footprint of high-frequency undulating cross-basin transects overlaid on traditional fixed monitoring stations.” – I think I understand the general message being conveyed, but I don’t understand the details of the method or the plot; please rephrase/expand the explanation to be clearer.
L174: “Propoer” – Proper
L197-8: “The LSEIK filter is implemented in an offline configuration, where ensemble statistics are derived from a long-term free running model simulation rather than from a dynamically evolving ensemble.” Please signpost here to the later section where more details are given.
L201-2: “physical and biogeochemical variables are included in a common state vector” – please specify which variables are included. Just temperature and chlorophyll? Is chlorophyll a single model state variable?
L229: “20 samples per year” – please be clearer what is meant by this.
L245-6: “this window provides sufficient integration time for the model’s radiative transfer scheme to respond to the updated chlorophyll fields.” – I don’t understand why this needs to be accounted for in the choice of assimilation window. Please expand on your rationale.
L246-8: “chlorophyll concentration updates modify the light attenuation coefficient, which alters the vertical distribution of shortwave radiation and consequrently contribute to changes in the temperature tendency (Ciavatta et al., 2018, Skákala et al., 2021)” – I don’t think either of these studies address this, better references could be found.
L247: “consequrently” – consequently
L249: “Kang e al,” – Kang et al.
L251-4: “This weakly coupled assimilation framework circumvents…” – please list some references which discuss these issues.
L257-8: “the faster temporal variability of physical processes compared to biogeochemical dynamics” – surely this depends on which processes are being considered, I’m not convinced it’s true as a general statement. Plus, it seems to contradict L242-3: “The window is sufficiently short to capture the rapid onset and peak of phytoplankton blooms”.
L261-2: “excluded, along with duplicate records” – is this just for in situ data, or does it cover OLCI and multi-sensor chlorophyll?
L272: “corresponding to an approximate relative uncertainty of 10%, reflecting a stable measurement uncertainty.” – this seems low, especially for fluorescence data (it’s not stated if the in situ measurements are fluorescence, HPLC, or a mixture, this would be worth clarifying in the previous section). And representation error seems to be neglected completely. Please provide justification for this choice, it seems to me like it would lead to overfitting the observations in the assimilation (one reason why it’s important to validate against independent data, but that’s an aside).
L276: “εrel is a sub-basin–specific relative error coefficient” – please list these somewhere, either in a table or plotted on a map. Either in the main text or an appendix.
L290: “three numerical experiments” – four including REF.
L323-4: “This indicates that, in this weakly coupled framework, biological updates do not propagate directly into surface temperature fields.” – is that not by design/definition? In which case, “indicates that” is not the best phrasing.
L327-8: “provides a physical consistent baseline for interpreting the biogeochemical state.” – this relates to my third general comment. You conclude that CHL_SST provides a sound physical baseline for the biogeochemistry, but have only looked at SST, when parameters such as mixed layer depth and mixing have a more direct impact on biogeochemistry.
L334: “The domain-averaged value for each experiment is indicated in the upper-left corner of each panel.” – lower-right, not upper-left. Furthermore, for b-d this looks like the average of the difference in TSS rather than of the TSS itself, which in my view would be more informative. Also, please express it as e.g. 0.977 rather than in scientific notation.
L336-7: “CHL_SST allows for the developement of a more accurate thermal environment” – again, a conclusion is drawn about the whole water column based only on the surface.
L336: “developement” – development
L337: “thermal environment, which is a prerequisite for correctly representing the stratification-dependent processes” – at some point in this manuscript, more discussion is needed about how true this statement is for primarily salinity-stratified waters.
L338: “SCM” – this hasn’t been defined yet.
L338-40: “The subsequent thermal response to biological corrections is therefore interpreted as a secondary, nonlinear adjustment within the model rather than a basin-scale physical reorganization.” – please explain further what you mean by this and why you conclude it.
L343: “systematically underestimation” – systematic
L344-5: “The spatial distribution of model errors shows a clear south–north gradient (Fig. 4).” – this isn’t clear to me.
Section 4.2.1 – some of these RMSE values seem quite high, it would be useful to add some more discussion of what typical chlorophyll magnitudes are. Or at least say that this is looked at more in Section 4.2.2.
L355: “efficienicy” – efficiency
L356: “correponding a” – corresponding to a
L359: “reduces” – reducing
L362: “complementary rather than redundant” – rephrase for clarity
Fig. 5 – OLCI is spelt “ocli” in the key.
L372: “approxmiately” – approximately
L382: “The impact of assimilation on the spatial distribution of simulated surface Chl-a shows a clear seasonal divergence” – this and subsequent text just seems to be describing the seasonal cycle of Chl-a in the model, rather than in the impact of assimilation.
L387-8: “Notably, assimilating SST data yields an error reduction approximately 1.8 times greater than that achieved through the biological constraints alone” – assuming this statement is comparing Fig. 6m and 6q, then this error reduction seems to be limited to a small geographic area and is still rather low in absolute terms, which should be noted.
L389: “implying that correcting the thermodynamic state is more effective under extreme light-limited conditions” – following on from the above, this seems to be making a bolder conclusion than the evidence implies. I may be wrong about that, but in that case please provide more details about the likely mechanisms at play here.
L390: “bloom” – blooms
L392: “CHL_SAT” – surely CHL_3D is the appropriate experiment to compare CHL_SST to here?
Fig. 6 – given the results in Fig. 5, it would be interesting to plot OLCI here too.
L411-2: “Physical constraint is most important in winter and spring, while biological constraint becomes more important in summer and autumn” – true, but I think this is just reflecting the underlying oceanic processes, which is worth clarifying.
Fig. 7: “Note the enhanced representation of the Subsurface Chlorophyll Maximum (SCM) in the CHL_3D and CHL_SST experiments following the inclusion of vertical profile data.” – the results from CHL_SAT and CHL_3D look extremely similar to me. And I’m struggling to see much evidence of SCMs in the observations in the first place. What am I missing?
Section 4.3.1 and Fig. 8. The caption for Fig. 8 describes a different figure than the one actually shown. I think much of the text in Section 4.3.1 is based on the figure described rather than the one shown, hence why it seems to be drawing conclusions not backed up by the plot.
L441-2: “the basin-mean bias in the CHL_SAT run worsens” – but the RMSE improves, which seems worth noting.
L443-4: “However, the seasonal mean profiles at representative offshore stations show that the satellite assimilation better resolves the observed mixed layer depth relative to the REF run(Fig. 9)” – Fig. 9 shows chlorophyll, not mixed layer depth, so I would hesitate to draw conclusions about mixed layer depth purely based on that plot.
L455: “The largest improvement” – since this is just based on comparison against the assimilated data, some caution is needed.
L459: “reducing localized profile RMSE reduction” – rephrase
L461-2: “This adaptive capability is further evident in their precise responsiveness to transient hydrodynamic forcing. During episodic mixing, such as early 2018 at station BY29, these constraints promptly induce sharp downward shift in SCM following a mixing event (Fig. 8).” – this statement does appear to match the version of Fig. 8 shown. But I’m sceptical about the cause and effect described. The text is suggesting that the assimilation changed the wider model environment which led to the model reproducing the SCM as an emergent property. But is it just that the chlorophyll was assimilated and that’s the simple reason for the change in chlorophyll? This is an example where analysis of the vertical physics and nutrients would greatly help elucidate.
Fig. 9: “correct the timing” – how do these plots show that?
L471: “Seasonal mean profiles (Fig. 9) show the greatest RMSE reductions occur in spring and summer” – perhaps, but this isn’t obvious just from looking at them.
L473: “successfully guide the simulated SCM” – this seems to be referring to Fig. 9d, which does not show any SCM?
L474: “However, the REF and CHL_SAT runs fail” – again, is this referring to Fig. 9d? If so, CHL_SAT and CHL_3D look extremely similar. Or is it referring to Fig. 9a (winter)? This does have a clear distinction between REF/CHL_SAT and CHL_3D/CHL_SST, but I still don’t see any SCM or change in the depth of vertical gradients. All of this paragraph, it’s not clear what I’m meant to be looking at.
Fig. 10 only plots REF and CHL_SAT, but the corresponding text in Section 4.4 repeatedly describes features “across Chl-a–only assimilation experiments” as if all experiments were shown.
Fig. 10 caption references “deep-water warming” but the plot only shows near-surface warming?
L502: “TSS change < 0.01%” – is this actually %, or should it just be 0.01?
L502-3: “This near-neutral response suggests that the bio-optical feedback primarily redistributes heat vertically rather than driving large-scale surface temperature anomalies” – it’s perhaps worth noting that since the ocean model is forced by atmospheric fluxes, this will limit the extent to which SST can vary.
L503-5: “In salinity-stratified marginal seas like the Baltic Sea, buoyancy and horizontal gradients are primarily influenced by wind mixing and riverine inputs, which dominate the weaker thermodynamic signatures of biological light absorption.” – as per my previous comments, how much does this limit what you can expect SST assimilation to achieve?
L517: “negative values represent the equivalent hydrogen sulfide (H2S) debt” – what would you expect observed values to be in these circumstances? Zero?
L522: “by current” – “by the current”
L526: “observational uncertainty” – please provide a reference if possible.
L527-8: “increasing model bias and observational uncertainty within the oxycline (50–100 m)” and L530: “The persistently wide observational uncertainty in the 50–100 m oxycline zone” – the increased observational uncertainty seems to be confined to the steepest gradients, whereas the model bias increases over a larger depth range. It’s worth being clear on this distinction.
L531-2: “These results delineate the boundaries of what surface-focused, weakly coupled DA.” – this sentence is incomplete.
L542-4: “Here, we extend these findings by separating physical control on ecosystem dynamics from the thermal response to biological corrections, allowing the direction and strength of biophysical coupling to be evaluated explicitly in a single modeling framework.” – the results presented here are valuable, but only the impact of assimilation is assessed, not the strength of the coupling itself, so I think this sentence is over-selling things.
L549-50: “indicates a redistribution of heat rather than a substantial change in upper-ocean heat content.” – it would be worthwhile and presumably straightforward to calculate upper-ocean heat content to confirm this.
L551-2: “they may still modify the local density gradients near the thermocline and influence the timing of seasonal mixed-layer deepening” – again, presumably it would be relatively straightforward to check this?
L552-3: “implications for nutrient entrainment” – again, a first-order check on how nutrients differ between experiments would be informative here.
L554-5: “ecological sensitivity may exceed thermal sensitivity, with small thermal perturbations influencing nutrient supply, oxygen ventilation, and the timing of seasonal transitions” – the two parts of this sentence seem to contradict each other.
L555: “These mechanistic insights” – without having actually checked if the model is responding as suggested above, this is just speculation rather than insights.
L558: “state(e.g.,” – add space
L565-6: “improves the physical environment governing phytoplankton dynamics” – again, this is purely based on SST, which is not the most important physical factor affecting phytoplankton dynamics. Even just looking at mixed layer depth would be helpful.
L570-2: “Biophysical coupling in the Baltic Sea is therefore not weak but strongly asymmetric, with physical forcing dominating biogeochemical variability and biological feedbacks exerting only a secondary influence on the thermal state” – maybe I’m misunderstanding, but this just seems to be stating well-known and basic oceanography that’s true almost everywhere, rather than providing novel insight.
L572-3: “provides a mechanistic basis for designing observation-sensitive assimilation strategies in halocline-dominated coastal seas.” – please elaborate.
L590-2: “For example, at station BY15, CHL_SST brought the SCM depth closer to the observed pycnocline during the 2018 winter transition (Fig. 8), indicating a more realistic coupling between surface forcing and subsurface biomass structure.” Please reference other results too, otherwise this seems cherry-picked.
L592-3: “This spatial correction helps align the simulated biological productive center with the observed nutricline and pycnocline (Teruzzi, et al, 2021)” – without providing any assessment of the observed nutricline and pycnocline, I’m not sure you can confidently conclude this. Teruzzi et al. (2021) is referenced, but that study assimilated nitrate and chlorophyll and no physics. It’s certainly worth referencing here and discussing how your results compare to theirs, but that needs expanding on.
L596-7: “SST constraints do not only modify the thermal field but also determine the effective vertical extent of biogeochemical corrections (Goodliff et al, 2019).” – similarly, does this statement apply to your study, to Goodliff et al. (2019), or to both? Given Goodliff et al. (2019) looked at a similar region, it would be interesting to discuss further.
L603-6: “The primary advantage of integrating SST lies precisely in its ability to dynamically re-adjust this hydrodynamic boundary. In this dynamic context, our results show that biogeochemical corrections affect the SCM only when the hydrographic environment retains sufficient vertical coherence to allow surface anomalies to penetrate below the mixed layer” – perhaps this is true, but it hasn’t been demonstrated in the results section.
L606: “(Kahru et al., 2012; Quartly et al., 2023).” – again, references to other studies are attached to a statement about your own results rather than separately discussed.
L608-9: “SST therefore defines the stratified domain” – are you sure?
L612-4: “The nearly invariant oxycline trajectories in our results (Fig. 11) reveal that the static benthic boundary layers in the model effectively seal the sediment-water interface, completely insulating deep-water redox renewals from any pelagic biological updates.” – please elaborate on why you conclude it’s due to the benthic boundary rather than lack of mixing across the pycnocline.
L627-8: “the assimilation system remains blind to the deep anoxic pool” – rephrase this, I think a different phrasing could get your intended meaning across better.
L631-4: “This suggests that surface biogeochemical corrections cannot be expected to reorganize the deep redox environment unless they are accompanied by independent changes in ventilation or benthic flux conditions. From this perspective, the remaining deep-oxygen bias should not be interpreted simply as a residual model truncation error of the DA algorithm, but rather as a physical manifestation of missing deep-water constraints.” – I don’t fully understand what’s being said here, is it basically: “surface biogeochemistry is weakly- or un-correlated with deep-water oxygen, so it is not possible for assimilation of surface biogeochemistry to correct deep-water oxygen errors, observations of the deep ocean are required”, similar to the conclusions of Fontana et al. (2013) (https://doi.org/10.5194/os-9-37-2013)?
L634-7: “The Baltic Sea should therefore be viewed as a system of partially connected compartments. The photosynthetic zone forms the upper compartment, where biomass, light, and the local thermal structure interact closely. Conversely, the deep basin operates as a separate domain. Its oxygen state is largely controlled by external renewal events and internal consumption processes.” – I’m not familiar enough with the Baltic Sea to know, but is this a novel insight or just reaffirming something that’s already known?
“Tabel 2” (change to Table) is very interesting, but in reference to my previous comments it doesn’t seem as well backed up by validation results as it could be.
L659: “correcting the seasonal pycnocline and mixed-layer depth” – this isn’t shown.
L674: “which may be adequate at the basin scale” – is this based on the results or just speculation?
L683: “The 30–39% reduction in Chl-a RMSE improves the reliability of bloom detection” – as a caveat, this is against the assimilated data. What you ideally want from a model used for monitoring is for it to be skilful in places you don’t have observations – by only comparing against assimilated data it is hard to make conclusions about that.
L684-5: “by providing a baseline that better separates anthropogenic nutrient forcing from natural variability” – this hasn’t been shown.
Citation: https://doi.org/10.5194/egusphere-2026-3657-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 166 | 87 | 34 | 287 | 29 | 23 |
- HTML: 166
- PDF: 87
- XML: 34
- Total: 287
- BibTeX: 29
- EndNote: 23
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This manuscript assesses how the weakly coupled assimilation of Chl-a and SST into a coupled physical-biogeochemical model of the Baltic Sea, contributes to improve the representation of temperature, chlorophyll and dissolved oxygen, both in surface and in depth. The representation of the assimilated variables is improved in surface (spatially and temporarily). Whereas assimilating SST contributed to improve the biological state by improving the physical context, the assimilation of CHL impacts only minorly but noticeably the physical state. The transferring to depth, on the other hand, is limited for the observed variables given the vertical structure of the water column, and absent for the non-assimilated oxygen, which reveals different controls on shallow and deep water of oxygen content.
Main comment
I think the findings of the manuscript are interesting for the potential readers of Biogeosciences. It is timely, as the coupled assimilation of physical and biological variables is becoming more popular. Therefore, assessing to which extent the physical and/or biological state is controllable by the available observations is worthwhile. I find particularly interesting the discussion on the potential missing model constraints that could be explaining the non-controllability by DA. The ms is well composed and written, the figures are good-quality and informative. Below, I include some comments of specific aspects that I think would benefit from some clarification. Other than that, I think the manuscript reads well.
Specific comments
ABSTRACT
L20. “optically complex marginal seas” usually accounts for waters where the optical signal is not dominated solely by Chl-a. In waters defined like that, I do expect that the bio-optical feedback is driven by other optical constituents, as it is explained by the authors in the introduction (L64). In that case, I do not get why the hypothesis is that Chla assimilation could propagate to temperature though bio-optical feedback. In fact, other works have shown that, in CDOM-rich waters, this mechanism is driven by the high CDOM attenuation and not Chl-a. Therefore, the influence on thermal structure of Chl-a-induced variations in light absorption should be second order. I understand the authors justify using only Chl-a as the attenuating component because they validate extinction coefficients, am I right? Maybe that is something the authors could elaborate on a bit in the Discussion.
INTRODUCTION
L53. Some of the references provided indeed assimilate oxygen and nutrients, but on a broader view, is not Chl-a more commonly assimilated than nutrients and dissolved oxygen? Could the authors clarify if they refer to Baltic Sea applications?
L79. This is a continuation of the previous comment. Some of the references provided are not Baltic-specific studies, but the sentence seems to apply only to studies that have assessed the effectivity of joint assimilation in the Baltic Sea. Maybe the authors could clarify.
METHODS
L114 The bgc model is less detailed described than the physical model. I think how tracers “interact through local biogeochemical transformations” should be elaborated a bit more, especially those tracers and processes that are explored in this study. Maybe, the authors could include few words to clarify, for instance: is Chl-a a single tracer, or does it account for several plankton groups? how do Chl-a impact light propagation? which proceses produce/consume oxygen? Also, which is the origin of the “monthly climatological fields” prescribed for bgc in the open boundaries is not clear.
Figure 1. The figure legend does not mention that the satellite and in situ observations are Chl-a. It is clear in the main text, but at first read this was a bit confusing.
L168. Regarding the vertical coverage of the in-situ Chl-a. Since it is fluorometric Chl-a retrieved from CTD casts, I believe those are profiles taken every 1m in the first 50m. Maybe the authors could mention the 1m vertical resolution. Later, when reading the results of assimilating CHL_SAT and CHL_3D, the doubt came to my mind on how representative was the 3D coverage.
L215. The previous validation for the Baltic Sea of this solution is described in the references cited in the previous sentence? This is not clear.
L216. In this paragraph the authors discuss the uncertainty associated to the sensitivity of results to the choice of the localization radius. This is the first of several sentences through Methods discussing limitations related to pre-, pro or post-processing choices. I feel all these considerations belong to Discussion. Another example in L286 regarding the specification of observation errors.
L276. This “relative error coefficient” is the one displayed in Figure 1 panel b, right? Maybe the authors could point to this figure.
L292. Here the authors mention initial and restart fields taken from a previous reanalysis. I am assuming, therefore, that the authors have run a spinup period and later restart the proper runs? This is not clear from the simulations period stated earlier. If thats not the case, where those restart files are used is not clear from the text.
L303. It is not clear which observations are used for validation. I understand the variables are Chl-a, SST and dissolved oxygen, but Chl-a and SST are assimilated. Are Chl-a and SST not used for assimilation being used for validation? This is not clear form Section 3.6
RESULTS
L338. I am not understanding this sentence. The thermal response to biology is observable in CHL_SAT and in CHL_3D. In CHL_SST is mixed with the SST assimilation reorganization. The use of “subsequent” suggest to me that the authors are referring here to CHL_SST, which is confusing. What I am understanding from this Section 4.1 is that assimilating SST improves SST, assimilating Chl-a doesn’t, at least in the grand scheme of things, but it is expected to have a residual effect. Maybe the authors could help the reader (me) by clarifying here how they are going to explore this secondary adjustment, by comparing which specific experiments?
L372 & Figure 5. The differences in Chl-a retrieval from the multi-sensor product and the olci-sensor product are intriguing. OLCI Chl-a shows a completely different (and a bit weird for a temperate to cold sea in the northern hemisphere…) seasonal cycle. Have the authors any idea on the reason for the discrepancy between the multi-sensor and the OLCI Chl-a’s?
L374. In Methods, the authors mentioned that both products are assimilated (L150), but somehow the model follows pretty closely the multi-sensor Chl-a. In these lines, the authors explain that this has to do with the strict limitations the model imposes to phytoplankton growth. Given the large difference between the multi-sensor and the OLCI Chl-a’s I do wonder which is the one that the model is supposed to capture if it wants to be realistic. Maybe it will be useful to add to Figure 5 the superficial in situ Chl-a. Looking at Figure 7, it looks like the bloom happens in late-spring, close to summer in the stations showed, which seems to agree more with the multi-sensor Chl-a product.
L382. Section 4.2.2 describes Figure 6 but the individual panels of the figure are not pointed thought the paragraph, which makes the explanations a bit difficult to follow.
L421. The same in Section 4.3.1, it seems to describe Figure 7 but where exactly in the figure we can see the facts the authors are describing is not clear.
L435. There are some figure legends (Figure 8 in L436, and Figure 9 in L466) that include description of results. I would suggest keeping results in the text and leaving in the Figure legend only the description of the Figure.
L461. It is not clear to me what the authors mean in: “This adaptive capability is further evident in their precise responsiveness to transient hydrodynamic forcing.”
L481. I guess the sentence “The direct SST constraints in Section 4.1 provide the large-scale physical baseline for the analysis” connects to my comment in L338, where it was not clear how the authors were going to explore the Chl-a induced changes in temperature. Indeed, in this Section 4.4 they explore CHL_SAT and CHL_3D which is reasonable, but still the meaning of the first sentence of the paragraph is unclear.
Figure 10. It is not clear what the authors refer to with deep-water when explaining this figure, because the 0.4 degrees difference seems pretty near-surface to me.
Section 4.5. The description of layers is a bit confusing. The oxic layer is defined both as the 0-40m (L509) and the upper 20m (L527). Deeper waters are both defined as below 75m (L515) and 50–100 m (L528 and L530).
DISCUSSION
L606. Not quite sure, but the logical order of the two last sentences of this paragraph seems to be the inverse. As I see it: SST defines the stratified domain within which biogeochemical signals can be expressed, but it does not directly generate those signals. Therefore, the added value of assimilating SST lies in its restoration of a physically plausible upper-ocean background…
L611. The discussion about the lack of biological state update in the deeper water column is very interesting. However, it is not clear to me how the authors justify that this is a signal of layer decoupling or layer-specific unrelated processes. The assimilation of surface Chl does not translate so well to the subsurface Chl-a, and to my understanding this does not mean that the processes that lead to the dynamics of Chl-a at surface and subsurface are decoupled, but that subsurface Chl-a is not as controllable by surface Chl-a as it might seem. Maybe the authors could elaborate a bit on why if subsurface Chl-a is not controllable from surface Chl-a, it is expected the deep water dissolved oxygen to be so.
Data availability
According to the data availability policy for BG, all datasets used in a published work must be publicly available in repositories or been made public at the time of publication. I believe listing datasets as “available upon reasonable request” does not complain with the journal policy.