Greetings authors.

First of all, great job. You all have provided a timely and necessary synthesis of marine primary production (PP) modeling, tactfully bridging the gap between satellite-based algorithms and ecosystem models. I appreciate framing the current discrepancies not just as a "multitude of theories" but as a "multitude of languages" (I love the title), and the manuscript does a great job clearly laying out the mechanistic sources of divergence and commonality. The figures and tables are excellent resources; they nearly tell the story of the paper on their own and provide a clear and visually engaging way to compare approaches. Overall, this is a substantial contribution that will serve both specialists and broader audiences alike. 

I have a few minor suggestions that I think could help strengthen the arguments further and/or provide clarification. These should not be considered conditional for publication, take or leave as you all see fit. 

• While the focus of this synthesis is understandably marine, a brief acknowledgment of the contribution of coastal and estuarine systems to global and annual NPP could provide useful context, especially in light of choosing to include Figure 3. These regions are ecologically and socioeconomically important, disproportionately impacting fisheries and food security under climate change. Resolution issues are mentioned briefly, as well as sparse coverage, but perhaps a sentence or two would add helpful perspective, noting that different environmental forcing, model parameterization, input uncertainties, validation biases, etc. are all compounded in these complex, dynamic systems. 
• Section 3.1 provides a robust technical comparison of model structures. To further engage readers beyond the specialist community, it may be useful to briefly explain why models differ in specific regimes a bit more explicitly. For instance, in oligotrophic gyres, the Eppley function captures potential higher photosynthetic efficiency at elevated temperatures, whereas the standard VGPM adjusts efficiency based on nutrient limitation in high-SST waters. Other mechanistic considerations, like estimating NPP below the mixed layer or incorporating fluorescence, might also be briefly mentioned to help readers interpret the practical consequences of different parameterizations, particularly among satellite-based models. Because these models often rely on a common, highly correlated set of input variables, differences in assumptions and internal formulations tend to play an outsized role in shaping their divergent behavior, especially across distinct oceanographic regimes.
• The brief mention of AI could be expanded slightly to address common skepticism. Beyond predictive applications, AI can be extremely useful in the creation of “numerical laboratories” for assessing (or re-tuning) parameter sensitivity under varying environmental forcings, enabling process-based insights rather than treating models as black boxes. This framing aligns well with the broader comparative focus of the manuscript.
• In the conclusion, it would be helpful to clarify (even philosophically) if "success" is defined by model convergence (all models better agreeing on a NPP number), by increased mechanistic realism, by reduced uncertainty in common input parameters, or something else altogether. If different models converge on the same result for the wrong reasons, we haven't actually improved our predictive foundation. How do we deem something “fit-for-purpose” in the end? This paper is a good opportunity to assert some of those absolute metrics of improvement with confidence :-)

Again, great job everyone. This was a pleasure to read and review. Cheerio.

In the paper “Modelling primary production: multitude of theories, or multitude of languages?”, Skákala and colleagues confront the issue of the large range in current estimates and projected trends in oceanic net primary production.  They call attention to the potential contributions of fixed model parameters and limited parameter testing in this uncertainty, and suggest that the “proliferation of both satellite and in situ observations of relevant variables at global scales and the availability of powerful statistical tools in data assimilation and machine learning” provides an opportunity to systematically examine parameters and gain insights into spatiotemporal variations that aren’t resolved by current models.  They imply that understanding spatiotemporal patterns may provide pathways to unifying structurally different biogeochemical and satellite-based primary production models and reduce discrepancies between primary production estimates and trends.

The paper is rich in detail and makes numerous important points.  I feel, however, that a clearer presentation of key points more concisely delivered, would make the paper more impactful.  I have outlined some specific thoughts on this below and hope that they are useful to the co-authors as they consider revisions.

First and foremost, in my opinion, the single largest challenge in improving estimates of primary production and its changes is the difficulty of reliably measuring primary production.  Any attempt to use data assimilation or machine learning tools to understand productivity patterns requires reliable observations, or at least observations with a reliable error model to extract those insights from. 

The authors allude to the challenge of NPP measurement at several points spread across the text.  In their discussion of net versus gross primary productivity (lines 97-101), for example, and BGC-Argo floats (lines 207-209).  Later, on lines 322-324, they mention that no clear winners emerged from many rounds of Primary Production Algorithm Round Robin (PPARR).  This may indeed have partly reflected an intent to avoid selecting winners, but the fact that no clear winner emerged after so many comparisons of strikingly different models suggests a larger challenge: Different NPP data sets, and even the same data set from different times, support different models.  It is difficult to know which data to trust, or whether all models were simply wrong in substantial ways in different places and times.

The most striking example of this for me was the PPARR work of Friedrichs et al. (2009).  This study focused on a commendably long time series of spatially-distributed primary production measurements in the equatorial Pacific.  In their abstract, they state that this “tropical Pacific database captures a broad-scale shift from low biomass-normalized productivity in the 1980s to higher biomass-normalized productivity in the 1990s, which was not successfully captured by any models”.  Essentially, the measurements suggested large differences in NPP without any clear changes in physical properties, nutrients or chlorophyll enabled the dozens of models included in the study to explain it. The authors interpreted this as a sign that both satellite- and BGC model-based NPP estimates may have trouble capturing climate-driven trends.  Chavez et al. (1996), however, acknowledged that continuous improvements/changes in measurement techniques could not be ruled out.  The same challenge holds across study sights: different laboratories with relatively subtle differences in measurement techniques can yield quite different NPP estimates.  These challenges are perhaps even greater for BGC Argo-based estimates, which are attempting to infer NPP from a rather complex series of assumptions.

The authors come back to the issue of NPP measurement uncertainty just before the conclusions (Fig. 10), but I think it is important to address it earlier, present it in a more cohesive way, and propose a more specific productive path forward to support the parameter-estimation and variation aspirations described elsewhere in the text.  Is the development of a comprehensive NPP error model across data types and times, for example, possible? Would this enable the data-driven extraction of the insights that you are aiming for?

My second observation relates to the potential synergies between satellite NPP models and ecosystem models that estimate NPP.  The paper outlines many differences and commonalities between these two types of models.  I think it would benefit, however, from a more concise distillation of the basic differences and pathways for productive interaction that leverage their relative strengths.  For example, on line 460, after a quite extensive review of many aspects of BGC model and satellite model details, the authors state: “Thus, the differences between satellite-based and ecosystem-based models of primary production are not clear cut”.  What I was hoping for was something like: “While the preceding review describes many aspects of satellite-based an ecosystem-based NPP models, several primary differences emerge:”,  Followed by something like: “These relative strength and limitations, together with commonalities in underlying principles, could be leveraged to gain new insights into spatiotemporal NPP patterns”.  In short, could the authors increase the visibility of the basic contrasts between ecosystem- and satellite-based NPP and how this can be leveraged for your goals?

On this topic, I found the short paragraph on lines 626-631 to be compelling:

“Since parameter spatio-temporal variability results from poorly resolved species types or ecosystem processes, interesting insights into its scale and patterns can be obtained by comparing models of different complexity.  For example, high complexity models (such as the Darwin ecosystem model) could be used in some cases to deduce parameter variability of simpler models.”

I see a tangible path forward here that could start with ecosystem models simpler than Darwin.  Even a biogeochemical model with a basic representation of allometric tradeoffs for nutrient uptake, grazing and light harvesting, for example, offers a potential step forward relative to most current size-agnostic satellite-based algorithms.  There are published examples of approaches to relate the more complex patterns emerging from ecosystems models to the emergent growth functions employed by satellite-based NPP models (Stock et al., 2019).  The utility of the ecosystem- and satellite-model synergy, of course, would be contingent on the capacity of the ecosystem model to skillfully capture observed chlorophyll, Chl:C, nutrient, zooplankton and other patterns, but fortunately skill is improving on this front (thanks in part to satellite-based products!). 

In short, this specific approach seems to provide a concrete strategy for how ecosystem- and satellite-based models could be integrated in a way that helps overcome NPP data challenges.  Systematic addition and removal of physiological and ecological tradeoffs, for example, could provide “fingerprints” that help explain and justify data-driven findings.  I would consider giving this potential interaction a more central place and reducing some background/descriptive elements that are less actionable.  A more specific call-to-action would make the paper more impactful.

In addition to the primary comments above, I have provided some more minor critiques below that I hope the authors find useful.  I have suggested a minor revision because I do not have any strong reservations about the paper aside from bringing the observational uncertainty to the fore.  A more concerted attempt to distill the main messages, however, would likely lead to a more impactful contribution.

Minor comments:

Line 39-40; 45-47: Consider this statement relative to my first primary comment.  It is important to acknowledge that the in-situ data is also highly uncertain.

Line 48-52: The concrete and achievable synergy between ecosystem and satellite-based models (lines 626-631) highlighted in my second primary comment seems central to these points.  I would consider a clearer and more prominent articulation of this synergistic path forward.

Line 71-73: Is the purpose here only to review, or do you intend to suggest a path forward?

Line 90: Perhaps reframing this paragraph to address my first general comment is: “A central challenge in understanding primary production and predicting primary production is that it is difficult to measure”, then consolidating these issues here?  You would then just need to articulate how these challenges could be overcome.

Line 97-98: Most ecosystem models predict and report NPP, not GPP.  Respiration is often included in the growth function (e.g., Geider), basal respiration is either omitted or accounted for in reported NPP.

Line 132-133: I would argue that not including T-dependent parameters for a global algorithm, or even a large regional one across seasons, is not defensible given the data.  Can’t we at least eliminate this?

Line 156: I would call the 17-83 range the range, not the uncertainty.  The lower and upper bounds can clearly be rejected with existing constraints.  There seems to be an unwillingness to do so directly, but maybe the effort you are calling for could help us do so in an objective and constructive fashion.

Line 207-213: See general comment 1.  The difficulty of observing NPP deserves a more central, consolidated place in your narrative and needs to be more directly addressed in your call to action.

Line 232-234; 238-241: Is there enough data to do this, or would an ecosystem-/satellite-based model synergy be the primary driver of this process?  Would intermediate-/high-complexity ecosystem models be an intended target of such activity, or would it be a tool helping to understand and introduce this added structure into other models?

Line 322-327: See general comment 1.

Line 347-349: Not always ignored; Sometimes treated implicitly.

Line 404-405: Do you think it is defensible to not have a temperature dependence?

Line 434-440: This seems key to the synergy between these two model types, and comes up again on lines 626-631.  It would be nice to have a more cohesive presentation of these points (see general comments).

Emergent Patterns: The satellite-based models that do not have explicit nutrient and temperature dependencies, inherently contain those dependencies in the model parameter values (line 434-435); BGC models have these dependences and allow you to probe emergent outcomes.

Line 460: See general comments.

Fig. 6A: See Stock (2019) for an example of how to derive the equivalent photosynthesis-irradiance parameter from an ecosystem model.

Line 535-537: This flexibility allows models to account for the diversity of plankton and processes that are not explicitly represented in current models (BGC models capture quite a bit, but I could see a data-driven approach capture more.  The only problem is that we don’t have reliable measures of NPP to find emergent patterns).

Line 585: “Together, these considerations suggest that investigating parameter assignment and parameter variability may be an important route to understand and potentially reduce many of the apparent differences between marine PP models, and hence in the estimated magnitudes of production.” – I feel like the recommendations that follow could be more concrete, and worry a bit that this statement comes nearly 600 lines into the paper.

Line 628-631: This seems like a concrete path forward yet the treatment is very limited (see general comments).

Line 634-636: An considerable uncertainty in in-situ NPP observations?

Line 655: New paragraph?

Line 656-675, Fig. 10: Yes, data volume has increased rapidly, but has it yielded reliable, consistent NPP estimates across global ocean biomes?  If this hasn’t happened yet, is the time really right? Is there a way to address the fact that it hasn’t in a rigorous way using new techniques to provide a self-consistent data source for much of the analysis you’ve proposed.  I feel like this section should have come earlier in the text as one of the central challenges to executing what you are proposing, rather than just before the conclusion.

Line 690: I’m not sure what you mean by “assignment is quite rare”.

Line 693-695: Would the natural next sentence be: A combination of data-driven approaches and natural synergies with intermediate- and high-complexity ecosystem models could provide the means to fruitfully uncover these variations.”

References:

Chavez, F.P., Buck, K.R., Service, S.K., Newton, J. and Barber, R.T., 1996. Phytoplankton variability in the central and eastern tropical Pacific. Deep Sea Research Part II: Topical Studies in Oceanography, 43(4-6), pp.835-870.

Friedrichs, M.A., Carr, M.E., Barber, R.T., Scardi, M., Antoine, D., Armstrong, R.A., Asanuma, I., Behrenfeld, M.J., Buitenhuis, E.T., Chai, F. and Christian, J.R., 2009. Assessing the uncertainties of model estimates of primary productivity in the tropical Pacific Ocean. Journal of Marine Systems, 76(1-2), pp. 113-133.

Stock, Charles A.. 2019. Comparing apples to oranges: Perspectives on satellite-based primary production estimates drawn from a global biogeochemical model. Journal of Marine Research 77, (S). https://elischolar.library.yale.edu/journal_of_marine_research/480