Large-Scale Retrieval of Coloured Dissolved Organic Matter in Northern Lakes Using Sentinel-2 Data

Round 1

Reviewer 1 Report

I had a chance to revise this manuscript again. Generally, a lot of improvements were introduced to the original versions which is resulted in clearer presentation of Methods and Results. However, the Discussion part still needs to be further improved. Mostly this part requires more clearer statements. This is also valid for Conclusions part that now contains some general and vague statement often not in the scope of the paper (see my comment below). Some sentences still require rephrasing.

There are some examples of what reader expects from the paper:

The reader might think: why authors retrieve band ratio values from lakes that are covered by clouds in sat images? It is obvious that they will be outliers here More clear statements on spatial variability of CDOM: which landcover type and geographical settings explain the CDOM variability? Why f.e. lakes in Vaermland have large spatial variability (between 0 and 10)?

Moreover, there are still many small inaccuracies and omissions in the text, e.g. data matching in text and tables.

 

P1 L21 suggest removing “Multispectral Imager”

P1 L23-24 Suggest using: “Values of a(420)cdom were obtained for 46 lakes… by analyzing the water samples..” These values were not directly measured in the field by optical sensors etc.

P1 L26-27 Authors mention that B2/B3 showed poor relationship (in case of both 1C and 2A products) with R2=0.22. And at the same time authors define the R2 of 0.28 in case of B3/B4 as “relatively stronger”. From my point of view, these relationships are equally poor.

Perhaps, if authors remove “cloud-affected” lakes from B2/B3 dataset, they will obtain better results and the conclusion will be that the ratio of blue and green atmospherically not corrected bands better explains a(420)cdom variations.

(It is in fact what reader may think reading this). Maybe rephrase this.

P2 L41-42 better to use “lakes are generally small in size”

But how the size of lake affects its monitoring? Here, rather the benefit of the resolution of MSI should be addressed.

P3 L110 suggest replacing “chose” to “have decided”

P4 L126 When I’m looking into the Table 4, I can find differences between acquisition and sampling dates more than 10 weeks (e.g. n=37). Please check the table carefully or correct the statements on time differences.

P146-147 Suggest: “…different bands and are delivered as scenes with UTM Zone 33N projection (WGS-84 datum). Images with 10..”

P6 L151-152 Authors mention that cloud-free pixels are chosen for the analysis, and further mention “remove lakes from the analysis, that are affected by clouds and haze”. That might sound strange for readers. Indeed, cloud covered pixels might be excluded right from the beginning.

P6 L157-164 This part seems to better fit to the Discussion part (or Introduction?)

P6 L172 “well” change to “accurately”

P9 L210 delete extra bracket

Figure 3. Why do not authors show results of PLSR analysis for L1C product, too?

P10 L222 the correlation is rather weak

P10 L226 “a(420)cdom variation” instead of “variation a(420)cdom”

Table 5 I think there is a mistake in the caption: results AFTER atmospheric correction are grey shaded

P11 L244 delete “in”

P11 L247 mistake: replace “..than B3/B4 and B3/B5” to “..than B2/B3 and B3/B5”

P11 L248 Table 5 appears in the text after the Table itself.

And where is the reference to Figure 5?

P10 L222 – Figure 4D appear in the text earlier than Figure 4A

P12 L261 Again, why not to exclude these “outliers” from the beginning as they “strongly affected by atmospheric conditions”? The reader might think that authors attempt to manually improve the model by excluding the data that do not fit into it.

P12 L262 I have already mentioned, that water quality is not assessed in this study. Here assessed is CDOM absorption coefficient at a certain wavelength

Equation 1. I think it would be better to introduce real numbers in the equation (instead of K and m)

P12 L263-264 Why 46 lakes? “A420cdom cannot be adequately assessed as pixels used in the analyses are affected by clouds”.

P12 L271 I think it is not correct to write “a(420)cdom rich lakes”

Figure 6 I do not understand why it is necessary to again show the modelled and measured a420cdom values within this graph? Perhaps, it is more reasonable to assess a420cdom in more other lakes in these areas using the tested algorithm. And further compare a420cdom value ranges?

P13 L283 “small” instead of “smaller”. And for me small lakes (or ponds) have the size <1000 m2. Lake with the area of 68400 m2 is rather big. It is not an advantage of Sentinel-2 to assess CDOM in such big lakes. Please rephrase it.

P13 L285 “higher” than what? Landsat?

P13 L286-287 This sentence belongs to the Discussion part

P13 L287-288 It is a repetition of the previous statement (P12 L271)

Figure 7 a(420)cdom within the figure (and further in Supplementary)

P15 L303 Suggest replacing “green portion for the electromagnetic spectrum” to “green range of the electromagnetic spectrum”

P15 L306-307 I’m not sure that the red range is more influenced by high concentration of SPM. In turbid lakes, the whole visible range will be influenced (scattering). For example, red range may introduce more uncertainty in a CDOM estimation in case if the chlorophyll-a concentration is high. Chlorophyll has a peak of absorption in a red range.

This sentence needs to be clearer.

P15 L312 I think it is not a right wording “a(420)cdom content”

P15 L319-320 Relationship between a(420)cdom and spectral indices is not a function of surrounding land cover and land use. I think it is obvious. Heterogeneous land cover may influence CDOM range among lake that are located in these different landscapes.

P15 L321 delete “takes”

P15 L324 Why May? In the Table 4 dates are between July and August only. I understand, that the selection of samples was made, but that is misleading the reader. It is not necessary to mention each set of samples here if the data have not been used in the final analysis.

In Table 4 mentioned are some images that were acquired in September.

P15 L326 Suggest deleting this sentence.

P15 L327-328 see my comment above concerning the time difference and data presented in Table 4. I would avoid discussing that such time difference “can be negligible” in this case, but rather as a potential source of uncertainty in the final model, even if the final determination coefficient was higher than found in other studies.

Moreover, the time difference effects cannot be a part of “Atmospheric effects” section.

Maybe move this to “Band ratio algorithms”?

P15 L343-344 There is not enough data within this paper to state this. Suggest deleting this sentence

P16 L350-351 How authors can assess the performance of Sen2Cor algorithm by comparing lake reflectance values among regions? I would understand if the TOA and BOA reflectance are compared to each other. Suggest rephrasing.

P16 L356 “occurred?”

P16 L359 “water reflectance data”

P16 L371-372 suggest replacing “water quality” to “lake CDOM”

P16 L375 Ecosystem cannot be “small”. Suggest rephrasing

P16 L378 suggest replacing “levels” to “values”

P16 L387-389 How can land cover influence the high shift of CDOM within the same lake? Suggest rephrasing.

P17 L398 See my comment above on “Water quality”

P17 L399 “cause”

CDOM is a direct proxy of lake colour

P17 L399-403 It is the repetition, suggest deleting

P17 L404 “wide variety of lakes” – what kind of variety? Maybe authors better use “lakes located in different landscapes across Sweden”? Perhaps, the algorithm will not work nicely for lakes located in tundra or other regions.

P17 L405 There is no need to refer to the Equation 1 in the “Conclusions” part

P17 L413-417 This paragraph is mostly repetition of the conclusion 1 (I would suggest excluding it)

P17 L418-424 That is not a conclusion out of the data presented here. Moreover, no significant quantitative analysis is done in order to explain how different environmental variables explain the differences in a(420)cdom values.

Developing a protocol for inland water monitoring is out of the scope of this paper.

How land cover types in Sweden influence a(420)cdom values? Conclusions require a number of clear statements that are supported by data presented.

P425-427 see my previous comment.

Author Response

"Please see the attachment."

Author Response File: Author Response.pdf

Reviewer 2 Report

Overall, this paper fits very well in the journal and is highly relevant concerning its practical application. The paper has a clear and relevant focus.

I think the paper needs to be improved concerning the following aspects:

ABSTRACT:

in the abstract, the range of the concentrations should be added, some other data could be removed

 

INTRODUCTION:

specify in a concise manner how DOM affects drinking water quality (odour, color, taste), and in what ranges this typically is problematic, and in what areas (if known), to indicate the general relevance of the approach

 

MATERIALS AND METHODS:

why was the colour provided, and not concentration of DOM? It is a pity to work at proxy level...

there is a good description of the areas where the research is executed. I miss a motivation for the (limited) number of sites, is there a statistical reason for this? Why not more sites? Why wasn't a specific (broad or specific) gradient of concentrations aimed for (it seems based on the outcomes this was at the end the case)?

Concerning the statistical model: why this technique was chosen? What is the motivation for the specifics of the approach mentioned in lines 175-181: 'A retrieval algorithm was then derived from the resulting regression model of different band ratios. Repeated k-fold cross-validation was performed in order to test the validity of these resulting coefficients models. During this procedure, the data were divided into k = 10 portions of equal size, where one of these were retained for validation while the rest (k-1) were used for training. The procedure was repeated 5 times to reduce variance. Model performance was assessed using the root-mean-square error (RMSE) and mean absolute error (MAE).': In particular, why K=10 for validation (why not 10 fold cross validation), and K-1 for training? This was repeated 5 times, but strikingly, these resulting data are not reported in the tables with outcomes, probably only averages. It would be good to at least in the appendix integrate the results of the variation found in these repeats.

 

RESULTS:

overall nice graphs, but I suggest not all results (large tables) need to be in the full text, and can be better put in appendix

 

DISCUSSION:

overall good and critical discussion. I miss however a fourth paragraph on application (challenges). Perhaps a fifth could be future research, and remove this from conclusions.

 

CONCLUSIONS

shorten these to clear messages of max one or two sentences per conclusion, as it will ease the reading

 

Author Response

"Please see the attachment."

Author Response File: Author Response.pdf