Biological data assimilation for parameter estimation of a phytoplankton functional type model for the western North Pacific

Hoshiba, Yasuhiro; Hirata, Takafumi; Shigemitsu, Masahito; Nakano, Hideyuki; Hashioka, Taketo; Masuda, Yoshio; Yamanaka, Yasuhiro

doi:https://doi.org/10.5194/os-14-371-2018

Articles | Volume 14, issue 3

https://doi.org/10.5194/os-14-371-2018

© Author(s) 2018. This work is distributed under
the Creative Commons Attribution 3.0 License.

https://doi.org/10.5194/os-14-371-2018

© Author(s) 2018. This work is distributed under
the Creative Commons Attribution 3.0 License.

Articles | Volume 14, issue 3

Research article

|

01 Jun 2018

Research article |

| 01 Jun 2018

Biological data assimilation for parameter estimation of a phytoplankton functional type model for the western North Pacific

Yasuhiro Hoshiba, Takafumi Hirata, Masahito Shigemitsu, Hideyuki Nakano, Taketo Hashioka, Yoshio Masuda, and Yasuhiro Yamanaka

Download

Final revised paper (published on 01 Jun 2018)
Preprint (discussion started on 29 May 2017)

Interactive discussion

Status: closed

AC: Author comment | RC: Referee comment | SC: Short comment | EC: Editor comment

- Printer-friendly version

- Supplement

RC1: 'Review comments on "Biological data assimilation for parameter estimation of a phytoplankton functional type model for the western North Pacific" by Hoshiba et al.', Anonymous Referee #1, 21 Jun 2017
RC2: 'Biological data assimilation for parameter estimation of a phytoplankton functional type model for the western North Pacific', Anonymous Referee #2, 26 Jun 2017
AC1: 'Errata for Figure 3', Yasuhiro Hoshiba, 29 Jul 2017

Peer-review completion

AR: Author's response | RR: Referee report | ED: Editor decision

AR by Yasuhiro Hoshiba on behalf of the Authors (01 Dec 2017) Author's response Manuscript

ED: Referee Nomination & Report Request started (19 Dec 2017) by Eric J.M. Delhez

RR by Anonymous Referee #2 (22 Jan 2018)

Suggestions for revision or reasons for rejection

I have read the revised version of the Hoshiba et al manuscript as well as their answers to my criticisms. The authors significantly improve the manuscript: they add some text that clarify a lot of issues, produce new figures in order to reinforce the performances of the model with parameters estimations and most importantly provide a new version of Figure 3.
Even if I am still not fully convinced by the significance of the improvement of the data assimilation experiment on the quality of 3D model results (I would have liked to see more error skills like bias, RMS), I think that the authors did a lot of efforts for improving their work and answering our criticisms. I still have some comments.

Line 220: what do you mean by effective nutrient? Limiting nutrient?
Line 271: please clarify what is the mean iron growth rate and give the units. Iron uptake rate? So, the production increases because you change the parameters describing the iron limitation which results that your phytoplankton is not anymore limted by iron. How realistic are your calibrated parameters? (I am not an iron expert and so I cannot realize if the iron uptake efficiency is overestimated or not)
Please clarify how the parameters are set at the interface between the “KNOT” and S1 regions? Are you imposing a sharp gradient or are you using some smoothing? From the figures it seems that it is the first option and this results in sharp transition of the simulated biogeochemical fields (e.g. figure 6c). You mention in section 3.5 that the version of the parameters that is smoothed with SST does gives worse results compared to the other one. DO you know why?
Figure 10: 1D model results are not represented here in Figure 10 b and 10c.
Figure 11: Please clarify what is represented in this Figure: maximum growth rate * limitation function by either nitrogen, silicate and dissolved iron? Uptake rate of nitrogen, iron?

Hide

RR by Anonymous Referee #1 (26 Jan 2018)

Suggestions for revision or reasons for rejection

A review on the revised manuscript, "Biological data assimilation for parameter estimation of a phytoplankton functional type model for the western North Pacific", by Hoshiba et al.

Having look the revised Fig. 3 and associated results, I found the parameter estimation by genetic algorithm gives a reasonable result and the manuscript is now ready for scientific discussion and review. As a whole, the manuscript was substantially improved in comparison to the one for the first submission, while there are some important points, for which more explanations/clarifications are necessary; many figures need to be improved; some part of the text should be rewritten for readability. For these reasons another round of revision is necessary.

- Introduction gives concise and good summary for previous work, and points out the necessity of the current work.

- line 56. ".. classical Michaelis-Menten equation .." needs reference.

- line 69-71. Again I have a question/concern about the use of physical field obtained from data assimilation, particularly since a 3D-Var system is used in this study. In reply to referee #1, the authors wrote "The physical field used in the offline ecosystem model (NSI-MEM) does not satisfy the mass conservation, but the passive tracers of the NSI-MEM in the offline setting do not have artificial sink/source without other boundary forcing and correction terms". I don't really understand how the authors achieve such a set-up, since the conservation of passive tracers depends on the conservation of the physical field (i.e., volume). The passive tracers in LTL models are generally represented by concentration in a grid cell. If there is an increment of SSH by assimilation, the volume of the corresponding cell changes, which automatically leads to change of passive tracer amount (if you don't change the concentration). How did the authors handle this issue? If the authors implemented a scheme which preserves amount of passive tracers even with the volume change, then the concentration of passive tracers undergoes artificial change. How much impact do you see in this case? I'm asking this question because the observed and modeled vertical section of T (Fig. 7) exhibits difference around the inter-gyre boundary (approx. 40 degree N), implying not a small amount of SSH increment might be added in this frontal zone. I think this is an important point to be checked and discussed somewhere in the manuscript before going LTL model analyses, since a use of assimilated physical field for ecosystem modeling is (probably) one of the way to go in the future.

- line 82. The authors describe "iron supply was only from the dust in the model setting", while in introduction they pointed out that "The source of iron for the WNP region is not only from air-born dust but also from iron transported in the intermediate water from the Sea of Okhotsk to the Oyashio region" (line 31-33), which seems to me contradicting. A justification for the model setup or discussion on the effect of missing iron source is necessary.

- line 95-102. The authors conduct a model run with SST-dependent physiological parameters, while at the same time they also stated physiological parameters (may) change with other conditions e.g., nutrient abundance, light intensity (line 303- ). An explanation is necessary here, why the authors chose SST field to smooth the transition between the two assimilated stations.

- line 107. "The parameters values" --> "The parameter values".

- line 107-108. Why the authors selected the specific year 1998? I asked this question in my first review but the authors did not provide reasonable answer. I'm asking this question because a parameter estimation by multi-year condition gives more reliable result. If you use the data from short period, the estimated parameters may be deformed so as to make best much for specific condition. In such a case you lose generality of the estimated parameters and difficult to apply them for interpretation of plankton physiology.

- line 167-184. Now I found the result is reasonable and the GA optimization works well.

- line 203-209. What can we learn from this analysis? It looks to me just giving a duplicated information with Fig. 9. If the authors intend to keep this part, more explanation is needed for the purpose and necessity of this analysis.

- line 230. The construction of the sentence seems strange.

- line 233. "maximum" --> "the maximum".

- line 229-249. Some sentences are tedious and preventing easy-reading. For instance, line 233-236 "At St. S1, the timing of (the) maximum phytoplankton ... " can be shorten as "At St. S1, OPT case reasonably captures the timing of the phytoplankton bloom, although the amplitude is slightly overestimated." Readers already know that you compare CTRL and OPT in this section, therefore you don't have to repeat "compared to CTRL case" many times.

- In relation to the above comment, I suggest to use short abbreviations, e.g., control case --> CTRL, parameter-optimised case --> OPT, SST-dependent case --> T-OPT etc. for easy-reading.

- line 263-265. This is one of the interesting results of this study (from my point of view), since you quantitatively showed how much difference occurs on the simulated biomass due to effect of horizontal advection, using 1D and 3D models which have exactly the same LTL model function. I suggest to briefly mention this in conclusion.

- line 263 and error bars in Fig. 8 and 10. Why the authors employ 0.3 degree range to define the uncertainty? Is this an autocorrelation scale of observed chlorophyll (Fig. 8) or ocean structure (Fig. 10)? An explanation or justification is needed (you can easily find such scale estimates in literature).

- line 286-287. I don't understand the meaning of this sentence. What is the verb of this sentence?

- line 322-325. I am skeptical to the statement in this paragraph. If we see the prescribed range (i.e., min and max) and estimated values of parameters in Table 2, many parameters go to its upper or lower prescribed bounds, indicating the optimization result (i.e., optimized parameter set) is strongly constrained by the prescribed bounds. In other words, the GA optimization did not search the entire parameter space freely due to the bounds. This means that the consistency between the physiological parameters and those obtained from in-situ observation (line 323) is not necessarily guaranteed by the current experiment setup (the consistency is already imposed in the experiment design). If the authors really intend to confirm the consistency, the prescribed ranges should be widen beyond the current ranges, and see whether the parameters still stay within the range of in-situ estimated values. From this point of view, I suggest to revise the concluding sentence.

- The quality of the figures is very poor and is not suitable for publication (except Fig. 2). They should be totally redrawn.

-- Fig. 3. Use a log-scale for the ordinate, otherwise it is difficult to distinguish PL lines in panel (b).

-- Fig. 6, 7, 10 and 11. Use the same vertical range for consistency.

-- Fig. 3, 8, 10, 11. Apply a consistent manner for color and line type in these figures, e.g., dotted-line for 1D case, solid-line for 3D case; red line for CTRL, blue line for OPT, green line for T-OPT, black line for OBS. etc. The different rules between different figures makes readers confused.

-- Fig 8 and 10. Use color shade (with transparency) instead of the error bars. In some cases the error bars are stacked each other and not distinguishable.

-- Fig 11. "(/day)" --> "[day^{-1}]".

-- Fig. 3, 5, 6, 7, 8, 10, 12. Divisions and labels for abscissa, ordinate and color bars should be reconsidered for easy-reading.

-- Fig. 3, 8, 9, 10, 11. Put appropriate legend for line type and color with examples, in empty space in panels.

-- Fig. 1. Use the same panel size for (a), (b) and (c), since the practical information in each panel is nearly the same.

Hide

ED: Reconsider after major revisions (06 Feb 2018) by Eric J.M. Delhez

AR by Yasuhiro Hoshiba on behalf of the Authors (20 Mar 2018) Author's response Manuscript

ED: Referee Nomination & Report Request started (27 Mar 2018) by Eric J.M. Delhez

RR by Anonymous Referee #1 (16 Apr 2018)

Suggestions for revision or reasons for rejection

Review on the 2nd revised manuscript, "Biological data assimilation for parameter estimation of phytoplankton functional type model for the western North Pacific" by Hoshiba et al.

As a whole the manuscript becomes better than the previous version, while I think it still needs a further clarification at least the points mentioned below. In addition, I'm still frustrated to the discussion, for example, those in lines 311-319, which looks to me contradicting (or inconsistent) to the aim of this study described in introduction. Anyway, it might be a matter of taste, and I have no further comments contributing to improvement of the manuscript.

- Line 70-74: The description of the dynamical field used for the experiment is still ambiguous (or confusing). Additional sentence(s) are necessary for clarification. The manuscript describes that "... 3D variational analysis scheme that synthesizes the observed information such as temperature, salinity and sea surface height", while it also describes "... but the amount of water mass was conserved (Fujii and Kamachi, 2003)", without any reasoning. The following sentence in the response from the author also made me confusing: "Because the 3D-Var system used in this study only changes the temperature and salinity, and does not directly change the SSH and velocity". What does "directly" mean? Are SSH/velocity changed "indirectly"? The answer is "No", if I understand correctly the 3DVar system in Fujii and Kamachi (2003). I suggest to revise line 72-74 as follows,
"The T/S fields are obtained from a free model run plus T/S increments. These increments are derived from the 3DVar system so as to minimize model - observation misfits of T and SSH. Different from the T/S fields, dynamical fields (e.g., u, v, psi, SSH etc.) are not modified by the data assimilation (i.e., the physical field holds mass conservation, which is necessary to run the ecosystem model with a consistent manner)."
If I'm wrong and the authors disagree the revision, further explanation is necessary in this part.

- Line 210 - 216: Since the time series are composed of monthly mean of 12 data, the 2-month lag-correlation is calculated by only 10 data points, which looks to me not always sufficient for scientific robustness. I suggest to provide a limitation regarding robustness of the result due to the short time series.

- line 243 - 254: I agree that the diagram shown in Fig. 9 is a concise way to summarize the results, whereas it is misleading to show this figure without mentioning a limitation of this analysis. Since the current experiment deals with monthly mean data, the phase (angle of the arrows) have a resolution of 30 degree. Therefore "True" does not mean a perfect match to the satellite data, while it means a match with 1-month (30 degree) error range.

- Line 290: ".. in which much smaller gradients than the observed gradients are found". This is valid for silicate only. Isn't it?

- line 311-319: After reading the paragraph here, I'm wondering what is the benefit of the current parameter optimization. If the ecosystem model parameters are changing with the condition mentioned in the manuscript (e.g., SST, nutrient abundance, light intensity, ..), what can we learn about the mechanisms governing ecosystem behavior from the optimized parameter sets for a specific situation? It looks to me that the dependence of the parameters on such environmental condition reveals insufficiency of the model formulation. Maybe further discussion/explanation which is consistent with the aim of this study should be necessary here. I guess at least the last co-author of the manuscript has an idea explaining the background philosophy of this experiment.

Hide

ED: Publish subject to minor revisions (review by editor) (24 Apr 2018) by Eric J.M. Delhez

AR by Yasuhiro Hoshiba on behalf of the Authors (05 May 2018) Author's response Manuscript

ED: Publish as is (07 May 2018) by Eric J.M. Delhez

Short summary

We developed a three-dimensional lower-trophic-level marine ecosystem model (NSI-MEM) and employed biological data assimilation using a micro-genetic algorithm to estimate physiological parameters for two phytoplankton functional types in the western North Pacific. The NSI-MEM optimized by the data assimilation improved the timing of a modelled plankton bloom in the subarctic and subtropical regions compared to the model without data assimilation.