Review Reports

Round 1

Reviewer 1 Report

This version of the manuscript is a signficant improvement compared to the original. The authors more clearly state what the objectives of the analysis are, and apply the empirical algorithm (which was identified as the best fit) to other lakes. As requested, they also examined the relationship between in situ SPM and in situ P values. 

I think this is still somewhat of a "regional" study, in that one of the main conclusions is that you need to generate new empirical algorithms for each waterbody type (i.e. if they have similar SPM, SPIM/SPOM, and Chla, the same algorithm can be used, but if those values change, then the algorithm needs to be retuned). 

A useful analysis from this study is the underlying mechanisms for why the empirical algorithms work at all. The fact that in situ SPM:P does worse than the MODIS-derived empirical relationship (in a statistical evaluation) strongly suggests that the MODIS band ratios are picking up on properties other than SPM, which improves the fit. While the authors argued that Chla and other parameters were not significant, I do wonder if a multiple regression approach would improve the semi-analytical algorithm (i.e. Chla by itself doesn't have much predictability, but it contributes to the residuals). 

My first recommendation was to reject the paper. This version provides more insight into the underlying mechanisms, and addresses some of the issues with broadly applying these algorithms to other waterbodies. As such, I think it is now a reasonable contribution to the literature.

For the first paragraph of the Discussion, I would also suggest rewriting it. Generally you don't want to start a paragraph with "however", since that implies that you are following some previous paragraph where points were raised. And the paragraph reads a bit like a laundry list of topics. An example of a better introductory sentence would be:

Based on the results from this study, several important points remain. 1) Is there an underlying mechanism that explains why empirical algorithms perform better than semi-analytical algorithms for Lake Hongze?

etc.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

The manuscript entitled “Remote sensing estimation of lake total phosphorus concentration based on MODIS: a case study of Lake Hongze” approaches an interesting theme, because phosphorus an important indicator of water quality, but, in the same time, it is not an optically active constituent. Consequently, it is expected that its estimation would not be directly possible by remote sensing. However, other studies have shown to be able of predicting it.

The document has remarkable problems in Methods and consequently, Results, Discussion and Conclusions. I disagree in dividing the two methods tested in empirical and semi-analytical. Both cases are empirical. The difference is at estimating directly or indirectly by reflectance. In the first method, TP is predicted directly by spectral band or band combinations and, in the second, TP is empirically estimated relies on SPM, which was empirically estimated by spectral index and statistical regression. Theoretically, the second method is more coherent. This part is the most important section in the text, but the explanation still is superficial. By the way, we can find methodology in Results.

Based on Figures, I cannot agree that ‘empirical’ algorithms exhibited better performance than ‘semi-analytical’ algorithms. MREs for the second methods were lower, which indicate better estimation. Although RMSE is slightly low for ‘empirical’ approach, maybe statistically it is equal to RMSE resulted by ‘semi-analytical’ approach. In addition, the good performance of this methodology is totally dependent on water quality condition, since P is not an optically active constituent, as shown in Discussion.

 

Specific comments

 

Line 43. Only ‘water’ rather than ‘water quality’.

 

Line 58. Is it really ‘founded’ or is ‘found’?

 

Lines 60-61. Use ; to separate the examples.

 

Lines 75-76. Explain better or rewrite ‘The empirical method process can

76 be simplified by semi-analytical method…’

 

Line 81-83. Do not use capital letters and separate the objectives using ;

 

Line 88. Rewrite ‘with water levels, at the water level of 12.5 m’.

 

Line 95. Is it an average rainfall value? Monthly? It does not look annual.

 

Line 111. Is datasets or samples?

 

Line 132-135. Rewrite. It sounds ambiguous.

 

Lines 137-139-8. Standardize Rrc.

 

Line 144. ‘code’ instead of ‘cod’.

 

Line 144. Spatial resolution.

 

Line 147. Section 2.4. This is main part of the work, but it is not well written. It is very superficial.

 

Line 148. How were every band combinations tested? There is something cited in Results, but it need to be better explained here.

 

Line 149 and 155. What it means 'existing research'? Is the in situ data collected in this work or some research in literature?

 

Line 150-151. The optimal band combination selection based on correlation coefficient.

 

Line 154. Which semi-analytical model is that? It is clear. Is SPM estimated? Were other OACs estimated?

 

Line 156. If the correlation between P and SPIM is higher than P and SPM, why was the model fitted using SPM?

 

Line 169. Explain why: ‘the mean value in winter was slightly higher than that in summer’.

 

Line 171. Explain better. What did the authors mean?

 

Table 1. I guess it is possible to fix these columns. This table is confuse.

 

Lines 174-175. This must be clear in Methods: ‘the band and algorithm

175 are selected directly according to the published research (Table 2)’. This is method and not results.

 

Line 177. Explain briefly why this bands are related to SPM.

 

Line 181-184. This must be in Methods, but it is not clear how it was done.

 

Line 194. Figure.

 

Line 196. Bias can show this better than only visual interpretation of a 1:1 graphic. RMSE and MRE do not explain that. The same in line 206.

 

Line 196. MRE of 47% is not low.

 

Line 203. Is Table 2 or 3?

 

Line 212. Figure 3. MRE is quite lower here. Why 'empirical algorithm' was considered better? In terms of RMSE, the results look to be almost similar. In addition, all these relationships are empirical.

 

Line 222. Question 1) The results do not show this.

 

Line 223. How are the algorithms semi-analytical.

 

Lines 232-234. Why this methodology is called semi-analytical? Semi-analytical algorithms rely on RTE.

 

Line 245-249. This state is not supported by results.

 

Lines 257-258. Correlation coefficients of 0.53 abd 0.61 are not considerable.

 

Line 260. It is possible to address that an average chla of 14 ug/L is low. It can be low compared with other lakes.

 

Line 266. So, why SPIM was not used to estimate TP?

 

Line 275. What it means to be classified in Type 2? What are the characteristics?

 

Line 311-312. Where is the verb in: Only 6 groups for Chla, and 8 groups for others.

 

Lines 312 and 319. Standardize R^2 in whole text.

 

Line 356-357. Rewrite.

 

Line 378-379. Why does this happen? Should not it be the opposite?

 

Line 384. ‘when’

 

Line 393. Plural.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

The authors changes the document significantly. However, I send other reviews.

As the models were calibrated with significantly wide dataset, representing different quality water conditions according to seasons, the best model could be applied in a time-series of MODIS image of 2016 -2018.

Reinforcing the comment above, as MODIS images were downloaded and processed, the best model could be applied in images. TP mapping will help in Discussion section. In addition, the spatial distribution the final product expected from remote sensing application.

 

Specific comments

 

Line 159. Remove the point in the beginning of paragraph.

 

Line 213. Figure 2. Let clear in Figures and text whether the model is direct or indirect.

 

Line 264. Subsection 4.2. I do not agree that SPIM is not a COA. SPIM is composed of optical constituents such as sand, clay, and other mineral, which interact with the light. And, of course, SPIM is a part of SPM. Therefore, SPIM is a COA.

 

Line 488. Format the references section. The font is not right. Standardize and include the available DOI for each article.

Author Response

Please see the attachment.

Author Response File: Author Response.docx