The Distribution of pCO_2W and Air-Sea CO₂ Fluxes Using FFNN at the Continental Shelf Areas of the Arctic Ocean

Round 1

Reviewer 1 Report

The study presents the approximation of CO2 concentration and CO2 sea-air fluxes using feed-forward artificial neural networks a.k.a. multilayer perceptrons.

The quality and the layout of the presented manuscript is low. Some figures are of very low quality, and the captions are barely visible due to their graphical nature. The figures 5,6 are barely visible themselves. The manuscript looks like a very preliminary draft, not a ready-to-show paper.

The English of the manuscript is understandable, however, it is sometimes difficult to grasp the authors' train of thought.

The method of the approximation is described for pCO2, however, the description is scattered across the manuscript. The description of the method for CO2 flux approximation is vague and should be clarified.

The authors claim some results being 95% significant, and there is also a discussion about the uncertainties. At the same time, there is no method for uncertainty estimation presented as well as no method described for significance assessment. Uncertainty estimation for neural networks is not an obvious task, and the authors need to present it clearly.

The coordinates and the dates of data points are used as predictors in the study. It is an arguable approach. With an ANN flexible and expressive enough, it may "remember" the dataset completely with its coordinates and dates. In this scenario, the network may demonstrate signs of overfitting. In the presented study, the authors did not mention any measures that would help assessing the overfitting rate. Thus, there is no way to get any idea whether the networks overfit or not. Thus, there is no way to understand whether the decision to include the coordinates and dates into predictors list is valid.

The models (FFNNs) used in the study are pretty weak, and the dataset is small. The models are then applied to the whole domain and some conclusions are drawn based on these weak models with unknown uncertainties. The conclusions drawn are invalid until the uncertainties are estimated correctly, since the differences visible in the figures may not be significant.

Author Response

Dear Editor and Reviewer,

We would like to thank you for your insightful, thorough, and meticulous review, for your time devoted to our work and for many valuable comments and questions that help us to improve the quality of our articles. We want to also apologize for all grammar and language mistakes and errors. 

Below we answer all the questions and suggestions and corrected the manuscript via MDPI English Editing Service. We hope the corrected version clears any remaining controversies.

We sincerely thank you and best regards,

On behalf of the co-author and myself,

Iwona Niedzwiecka

Author Response File: Author Response.pdf

Reviewer 2 Report

Wrobel-Niedzwiecka and colleagues present monthly average maps of pCO2 from a feed-forward neural network for the European sector of the Arctic Ocean. The alogrithm ingested common predictor variables such as chlorophyll, SST, SSS, etc. From this they calculate the monthly air-sea flux of CO2. I believe the methodology to be sound, but the paper has many deficiencies in presentation. The grammar is quite poor at times. It needs an English-language editor to revise it. Some of the figures are not publication quality, to the point of being illegible. 

The use of the Nightingale et al gas exchange velocity parameterization needs defending. The kw parameterization from that study was obtained from dual tracer experiments in the North Sea and in situ winds. N00 is valid for a comparatively limited range of wind speeds. In my opinion a better choice would be to use a parameterization such as Sweeney et al 2007 or Wanninkhof 2014, which has been tuned to bomb 14C, and is applicable to a much wider range of U10. Neither W14 or N00 is specifically implemented for ECMWF winds, however, so that is also a concern. This is not a show-stopper, but it needs to be discussed and defended in the context of the literature. It also bears mentioning that there are also better approaches which include bubble parameterizations which would likely be more accurate. I understand that this might be out of scope of the study, but at the least the fluxes need to be presented as estimated monthly averages...which leads me to the final point, that the treatment of uncertainty in the manuscript is weak, and some of the interpretation is a bit too positive when one considers the substantial errors in pCO2 compared with the observations. I cannot recommend publication until these issues are addressed.

 

 

 

MINOR COMMENTS

L24: "greenhouse gas that the atmospheric" -- should read something like "greenhouse gas whose atmospheric"

L27: I don't know what is meant by "positive effect", suggest to strike

L28: should read "decreases in the"

L39: "magnitude dictated by k" -- This is not correct as written, from the well-known flux equation k*K0*DpCO2 one can see that the magnitude does not solely depend on k.

L42-45: I don't think this is the right way to phrase this. It's important to distinguish between observations and interpolated or modeled data products--the data gap problem can only be solved (in future) by more observations. Rather it is applications which require gapless data could benefit from improved interpolation schemes.

L43-44: "as the uncertainty reduction associated with choosing an  appropriate parameterization and the factor affecting the k have been described in [9] and [7]" -- garbled sentence

L44-45: This sentence needs a citation. It also seems circular to me--isn't this what the authors are trying to show? I also do not think it is strictly correct. These variables together may help to predict pCO2 but it is controlled by air-sea exchange, net photosynthesis and respiration, changes in solubility, and atmospheric or ocean transport. The sentence should also lead with "Seasonality in pCO2w" if this is meant, since the authors say it is controlled by "seasonal changes" in the mentioned drivers.

L50: The use of "however" doesn't make sense to me here, please rephrase.

L54: ANNs can be used for a variety of applications, but as written it is implied that they are only used for pCO2w and pCO2a, please rephrase.

L57: "Biogeochemical" would be better than "water" here

L59: Define AO on first use

L59-60: "Their results show that both observed pCO2W and that estimated with SOM were lower in the Greenland and Barents seas" -- seas should be capitalized. This sentences also needs rephrasing for proper grammar

L59-70: I think there is too much detail here. The Introduction should contain the background necessary to explain the study. Suggest you cut most or all of this, it isn't necessary to give quantitative details on the performance of other studies here.

L75: Seasonal amplitudes, not annual?

L79: which are related. What is the thermodynamics effect?

L83-84: Rephrase, poor grammar

L85: Seas, plural

L86: "study extent"

L91: fulfill

L100: characteristics

L111: How many data points were retained for validation?

L120: "concentration", delete "_"

L125: interpolated

L127: I don't understand how exactly the two chl-a datasets are being merged here. The use of "seasonal" and "yearly" is confusing me. From the text it seems one is observed average monthly chl-a from satellite, and one is from a reanalysis product. Z1 has chl-a from reanalysis, and Z2 is observed. Is that correct? Or does Z2 use reanalysis data when there are few to no observations during winter?

Table 1: SST, SSS, etc should be capitalized in the caption. The second bracket should be a right bracket. The table is inserted in the middle of the text, please put on separate line.

L144: What technique was used to regrid the data?

fig2: please include a 1:1 line on each panel. You label each figure a-c, but these aren't referenced in the caption. A title would help, otherwise one has to look carefully at the tiny subscript in the y axis.

L253: I don't think this statement is supported by the figure. A climatology is presented, how does this relate to the interannual variability?

L263: What is "similar  satisfying  accuracy "?

L264: "closest"

L267: The monthly average is well mapped during *summer* months, please rephrase--this is a little misleading. The agreement is poor during winter (as stated earlier in L260)

Figure4: The figure is cut off--it needs an x-axis. Please plot the two quantities on the same scale. Please include either error bars or an envelope to indicate the spread of each point (1 sigma, for instance). Are these data deseasonalized or otherwise filtered? If so, it is not the interannual variability, it is the time series.

L270-272: This sentence is garbled and needs rephrasing for proper construction

L277: I don't agree with this statement. The model seriously underestimates the seasonal peak.

L279: "than observations"

L279: Right--if there is an errant trend, the temporal variation is not reproduced with good accuracy.

L282: "received from rare pCO2W" -- what does this mean?

L282: Grammar needs work here.

L284: solubility is high or low, not good or bad

L284: What do you mean, enhanced by DOM from sediments? How do you know this? This is a claim totally unsupported by data.

L287: "In winter was the greatest longitude amplitude " -- rephrase with proper grammar

L287: parallels of what?

L290: "dissolved in seawater carbon is converted into the organic matter by the photosynthesis reaction" -- DIC is converted to DOM through photosynthesis. Note also that this is not strictly true. Photosynthesis converts DIC into phytoplankton biomass. There are many steps before it becomes DOM.

L292: correct "intensive  carbon  consumption  by  algae  in photosynthesis"

L295-296: correct "as thermocline disappear and carbon from the sediments through upwelling reach sea surface"

Figure 5 and 6: These figures are illegible and not publication quality. They are also too small to be read.

L298: Citation needed

L299-303: "Carbon  is  the  most  important element of life on earth..." -- this is veering into nonsense. Why is this in your Results subsection?

Figure7: This figure is illegible, not publication quality, and the panels are too small to be read.

Discussion: This is mostly a summary of the paper, not a discussion. There are numerous grammatical errors.

L358: delete first "during"

L362: How was the uncertainty estimated?

L362-365: I don't think this comparison has much relevance without a rigorous discussion of how the uncertainty was constrained by both studies.

L364: something like "the entire Arctic Ocean, while we focus on a smaller domain"

Conclusion: This is the first time the cosine/sine thing has been brought up. It needs to introduced in the Results or Discussion first before it appears in the conclusions.

L380-381: How are you defining what "satisfactory accuracy" is? The errors in monthly average pCO2 are large and will lead to correspondingly significant errors in the flux estimates. This needs to be rephrased.

 

 

Author Response

We would like to thank you for your insightful, thorough, and meticulous review, for your time devoted to our work and for many valuable comments and questions that help us to improve the quality of our articles. We want to also apologize for all grammar and language mistakes and errors. 

Below we answer all the questions and suggestions and corrected the manuscript via MDPI English Editing Service. We hope the corrected version clears any remaining controversies.

We sincerely thank you and best regards,

On behalf of the co-author and myself,

Iwona Niedzwiecka

Author Response File: Author Response.pdf