the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 21 Jun 2024
The unicellular NUM v.0.91: A trait-based plankton model evaluated in two contrasting biogeographic provinces
Abstract. Trait-based models founded on biophysical principles are becoming popular in planktonic ecological modeling, and justifiably so. They allow for slim, efficient models with a significant reduction in parameters, well suited for modeling the past and future climate changes. In their simplest form, trait-based models describe the ecosystem in one set of parameters defined by first principles, rooted in physics, geometry, and evolution. The result is an emerging ecosystem defined by physical and chemical limitations at the cell level. At present, however, a significant part of these parameters is not fully constrained, which potentially introduces a considerable uncertainty to the model results. Here, we investigate how these parameters influence the ecosystem structure of one of the simplest trait-based models, the Nutrient-Unicellular-Multicellular (NUM) model. We describe the unicellular module of the NUM model and through an extensive parameter sensitivity analysis, we demonstrate that the model – with a large span in parameters – can capture the general features of the pico-, nano-, and micro planktonic ecosystem at the southern California Current. We show that it is possible to narrow the range of parameters to get a stable, acceptable, solution. Finally, we show that the model responds correctly to a change in oceanographic setting.
Our analysis demonstrates that the unicellular module of the NUM model is accessible for the general non-expert without intimate knowledge of the parameter settings, and that the first-principal approach is well suited for modeling poorly resolved region and ecosystem evolution during current and deep time climate change.
- Preprint
(2379 KB) - Metadata XML
-
Supplement
(1131 KB) - BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on gmd-2024-53', Anonymous Referee #1, 24 Aug 2024
#Comment/review for: Hansen et al. "The unicellular NUM v.0.91: A trait-based plankton model evaluated in two contrasting biogeographic provinces." GMD Discussion. 2024.
Summary:
Hansen et al conduct a thorough analysis of an existing trait-based plankton model (NUM), adding some new components to the existing model (DOC and POM). They first compare model output using default parameters to size-partitioned biomass data at two ocean locations. They then conduct a detailed analysis of the likely parameter ranges and sensitivities. They then take this optimized model (with optimized parameters from calibration against the first ocean site) and test to see if it better simulates the second site. This test was successful in that it resulted in a better fit to the data than the ensemble with the full parameter range. Optimizing for this second site as well showed that the optimized space for three parameters (out of 23 parameters) did not overlap between the two sites.
General comments (overall quality):
The paper is well-written, well-organized, and easy to follow. The methods are clearly explained and many of the questions I had about the model were answered in the Supplement. The analysis has been conducted thoroughly and logically. The results are interesting not just for NUM users but also for those who might be interested in what aspects of a trait-based model might be the most uncertain. It should be published. My only major critique is to suggest that the authors go beyond just the description of the model uncertainty and parameter sensitivity and add some discussion of why the model might be more sensitive to some parameters than others, and what, for example, those three model parameters without overlapping optimized space might mean. Does this suggest areas in which model structure itself might be uncertain? Section 5.1 ("areas of improvement") basically lists areas in which the model could be made to include more processes or modify its descriptions, and in some places this discussion is linked to the sensitivities uncovered by the analysis, but this could be extended. For example, one of the parameters that is significantly different between the two optimizations is gamma_2 (discussed on line 599), which controls DOC and N supply, and so it would be interesting to develop a connection between the uncertainty of this parameter and the discussion of how DOC in the model may be described differently on lines 643-653. In short, the paper could use a bit more high-level synthesis so that it can be more useful in assessing the certain vs. uncertain processes (rather than simply parameters) in trait-based modeling more generally. If this really aims to be a universal, first-principles model, what needs to be changed so that the optimized parameter space is the same for both environments? What aspects are not yet universal?
Specific comments (individual scientific questions/issues):
L. 18: "Simplest" was confusing to me. Perhaps "in their idealized form" rather than "simplest"? Since below you say that many parameters are not constrained, to me that means that these models aren't yet exactly defined only by first principles (which by definition are not uncertain).
L. 19: "physics, geometry, and evolution." Should you add metabolism/chemistry here? You mention chemical limitations in the next line.
L. 25-27: I would be more specific about the change in oceanographic setting, and perhaps you could be less specific about the SCC, and rather call them (as you do in the conclusions) a more productive upwelling vs. a more oligotrophic downwelling system.
L. 59: "Hereafter we address the parameters" was confusing to me. Could you make it more clear here that you first evaluate the model's ability using default parameters, and then second, variation in the parameters?
Table 1:
--A default assimilation efficiency of 0.8 is high for heterotrophic metabolism that uses organic C for energy -- is that different than what is being presented here?
--solubilization length scale is blank for the default parameters -- ?
--Light attenuation by POM also blank for default and same value for min and max -- ?
--POM sinking coefficient -- seems the max value is wrong.
--I don't know what m+/m- means.Model Description (section 2): If the DOC and POM module is new here, I think this could be better emphasized in the description of the model. Describe this first (rather than last in 2.2.4), and in more detail, since the remainder is a summary from earlier publications (Serra-Pompei etc.). I was very confused about the discussion of "nutrients and DOC" in the sections above (starting on line 106).
General: DOC and N: I am a bit confused by why DOC and N are produced -- are you implicitly assuming that DON is rapidly converted into inorganic form? Why is POM assumed to have N content (I assume this is why you call it POM instad of POC) while DOC isn't? Why not call it DOM? Are you representing heterotrophic osmotrophy, using DOC as the energy source? I could use a clearer description of this module overall, what microbially driven processes are explicitly vs. implicitly resolved.
Table 2 and general: I am guessing that "m" is the mass of each size group, but I couldn't find this stated anywhere in the main text.
L. 179: Again mention that you are doing the initial evaluation of just the default parameters.
L. 193 and more broadly: I'm not quite sure why we would be surprised that varying the parameters among only the more restricted parameter set would do anything but improve model performance.
L. 195 and more generally: Why only 7? What criteria were used? Apologies if I missed this somewhere.
L. 313: "no model results fulfill both criteria for all biomass size categories." I really want to know why this is! Could you comment in discussion what this might mean? With your intimate knowledge of the model, could you comment on what part of the model might not yet be universal?
L. 327: But the majority of the 60 uM DOC is recalcitrant. How would the model solutions change if initialized DOC were at 1 uM? How much DOC is remaining in the steady-state model solutions?
L. 383/Fig. 4: Why might the model produce these distinct groups? Could it be something to do with predator-prey oscillations? Can you comment on what part of the model structure/equations might lead to this behavior?
Fig. 6 panel c: It is neat that the seven solutions capture this peak behavior -- is there any insight into what in the model might produce this? Again, getting at a bigger synthesis of these insights.
L. 495: "parameter tuning may not be necessary." This doesn't quite make sense to me, because it seems that indeed you have tuned the parameters already by selecting the 7 best models. It's not surprising then that sampling within this more restricted space produces better results.
L. 524 (and more general topic of remineralization): What does it mean that some of the dead matter is "directly remineralized back to nutrients"? Are you referring to DOC as a nutrient here? What process is causing this remineralization? Are these microbial types that are not explicitly resolved? How would this affect biomass distribution? Are you assuming just that the POM consumers are implicit? This links back to setting up the reader with a clearer description of the new DOC and POM modules.
L. 571: "only 7 optimal" -- again, how are these 7 determined? Why are there 7? What was the threshold?
L. 598-601: You are in the discussion now, so rather than just again repeating the same results, can you discuss or even speculate about what it means that these are the most uncertain parameters. Are these in line with what we think are the most uncertain processes? What insights have been revealed? Does it say anything about what the field should be studying or observing more closely?
Technical corrections:
L. 25: "in the" rather than "at the" SCC.
L. 28: I woul not start another second paragraph within the abstract, but perhaps this is a typo. Also, perhaps just change "accessible for the general non-expert" to "broadly accessible."
L. 181: I would start the new paragraph at "The investigation.."
L. 194: I don't know what "with outset" means (here and the next line).
Fig. 2: At first glance it seemed as if the x and y axes were exactly the same for plots a and b. Could the y axis be changed to something like "size class biomass"?
L. 277: Should this be B_A-micro?
L. 301: Instead of "identify", do you mean "define"? Also, "are" instead of "is" for model results.
L. 303: take out "its" from "its STDs" and just write out "standard deviation"
L. 380: grammar problem: "size groups all overestimate" does not make sense.
L. 395: "lack of model results with.." doesn't make sense -- something like "model does not capture the biomass concentrations at ..."
L. 523: here you mention "partly aF" and discuss, and then in other sections of the manuscript you just discuss gamma_2 and cF without aF. This led to my incorrect summary above (that I am just now realizing) that there are only two parameters that do not overlap. Just noting this. aF does overlap somewhat, and some other parameters (like r_D) also don't overlap much, so aF not qualitatively different than others in this light. I realize this is why you end up just discussing two. Perhaps make this clear here: "aF is one example of where there is some, but little, overlap"...
Citation: https://doi.org/10.5194/gmd-2024-53-RC1 -
RC2: 'Comment on gmd-2024-53', Anonymous Referee #2, 18 Nov 2024
Review of ms. gmd-2024-53, The unicellular NUM v.0.91: A trait-based plankton model evaluated in two contrasting biogeographic provinces. by Trine F. Hansen et al.
Overall: The manuscript presents a thorough overview of a newly developed generalized trait-based Nutrient-Unicellular- Multicellular (NUM) model of the size distribution of plankton, which accounts for a mixotrophy, i.e. both autotrophic and phagotrophic (heterotrophic) growth. The manuscript includes an evaluation of model performance based on 100,000 simulations with varied parameter values for each of two contrasting oceanic sites: 1) oligotrophic station ALOHA, with persistently low nutrients and biomass and overall smaller size distribution of plankton 2) coastal/shelf CCE, with relatively higher nutrients and biomass and accordingly relatively more larger plankton. Based on the best results so obtained, the range of parameter values was narrowed and further optimized (by randomly sampling the parameter space), and this was shown to yield further improvements in the agreement between model results and the observed size distribution of plankton. A one-way cross validation, using the parameter values obtained for CCE to simulate ALOHA, revealed reasonable agreement between model results and observations. However, the reverse cross-validation was not presented. The combination of a size-based model, formulated to be comparable to typical oceanic observations, with a representation of mixotrophy make this a promising model framework for exploring hyoptheses about observed patterns.
As the authors note in the Discussion, one important feature of the trait-based approach is its potential for general applicability, even to sites or scenarios where limited (or no) observations are available. To some degree, the cross validation presented supports this general applicability of the new NUM model. That case could be made stronger if the reverse cross validation (using parameters tuned to station ALOHA to simulate CCE) were presented. The formulation presented does seem to provide a generally applicable modeling framework, which could be applicable to a wide variety of oceanic sites/regions. The parameter sensitivities identified by the authors, and some insightful information about the how the model could potentially be modified to better capture the physiological response of prokaryotic phytoplankton, provide a useful basis for future studies to enhance the applicability of the model framework.
Specific Comments:
Line: 53: “how much tuning the model acquire…”. Should that be “require”, or perhaps better yet “how much the results depend on the specific values chosen for parameters” ?
Line 192-193: “This subanalysis allow us to determine how much model performance is increased by simply reducing the parameter span.” Does this refer to the fact that randomly smapling parameter values over a narrower, rather than a wider, range will most likely find parameter sets that better capture the observations? In any case, revise for to clarify the meaning.
Line 277: “autotrophic microplankton (B_T-micro).” Shouldn’t the subscript be “A”: i.e., “B_A-micro”)?
Also, a Table summarizing the notation for the various classes of biomass considered, as well as the different metrics (correlation and RMSE errors) would be helpful. It is a little complicated and not obvious what these differnet subscripts mean. I had to go back and search to refresh my memory while reading the manuscript.
Line 521: “Figure 11 show a set of simulations that performs particularly well for Station ALOHA. “ Were those parameter sets (and corresponding simulations) obtained from the narrowed search based on the parameter range identified from the intial simualtions of CCE?
Finally, although the authors have already clearly gone to an extensive effort, would it be possible to include a brief presentation of the cross validation using parameters tuned for station ALOHA (as the author have already identified) to simulate CCE? The would provide additional information about the portability of the model formulation. It could also be put into the supplemental and only briefly summarized in the main text, to avoid making the manuscript too long.
Minor: the manuscript includes numerous mis-spellings and minor grammatical errors, which can be corrected later, during production.
Citation: https://doi.org/10.5194/gmd-2024-53-RC2
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 172 | 42 | 16 | 230 | 25 | 14 | 12 |
- HTML: 172
- PDF: 42
- XML: 16
- Total: 230
- Supplement: 25
- BibTeX: 14
- EndNote: 12
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1