Water Transparency Prediction of Plain Urban River Network: A Case Study of Yangtze River Delta in China
Round 1
Reviewer 1 Report
The article presents the relationship between water transparency and related impact factors in Suzhou civil river network - An extremely important topic in the context of urban development and the negative impact of this development on the aquatic ecosystem.
The paper is well organized and coherent. The context is supported by the bibliographical references, the research methods are explicitly presented, the results are relevant and the conclusions are in agreement with these results.
Author Response
Dear editors and Reviewers:
Thank you for your letter and for the reviewers’ comments concerning our manuscript entitled “Exploring the most effective parameter for water transparency prediction of plain urban river network: a case study of Yangtze River Delta in China”(ID:1231468). Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our researches. We have studied comments carefully and have made correction which we hope meet with approval. Revised portion are marked in red in the paper. The main corrections in the paper and the responds to the reviewer’s comments are as flowing:
Responds to the reviewer’s comments:
1. Response to comment (Reviewer 1):(
The title loos fine，However，the study does not address the most effective parameter. Also, the parameters are not ranked in the machine learning model .Therefore ,only an increase of accuracy of including another variables does not necessarily is most effective.
Response: Considering the Reviewer’s suggestion, we have change the title to"
Water transparency prediction of plain urban river network: a case study of Yangtze River Delta in China.
2.Response to comment(Reviewer 1):( Line17, The sentence is not clear ; please re-phrase it. The study focuses on predicting water transparency, so it should respond to hydrodynamic conditions and aquatic environment indicators to water transparency.)
Response：Considering the Reviewer’s suggestion, we have rewrite this part.
- Response to comment(Reviewer 1):( Line18/19, Especially given the complicated nonlinear relationship between them.” The sentence is not complete in itself. Also, between them is unclear.)
Response: Considering the Reviewer’s suggestion, we have rewrite this sentence.
- Response to comment(Reviewer 1):( Line19/20, related impact factor ==> what does it mean.)
Response: We are very sorry for our incorrect writing, and we have already revise it.
- Response to comment(Reviewer 1):( Line20-24 please consider re-writing. It is very difficult to follow the sentence structure. I suggest authors use short sentences with better clarity.)
Response: Considering the Reviewer’s suggestion, we rewrite this part.
- Response to comment(Reviewer 1):( The correlation from ANN is not shown in the result .ANN is used for regression. Please explain.)
Response: The ANN was used to explore the relative sensitivity of 5 parameters to SD, and we have revise this part in the paper.
- Response to comment(Reviewer 1)( Line25-27 It is redundant that you start the sentence with compared to the MLR model, and again you end the sentence with than the MLR model. Please consider reducing redundancy in the sentence.)
Response: Considering the Reviewer’s suggestion, we have rewrite this part.
- Response to comment(Reviewer 1):( Line 27-31:Please consider re-writing the sentence. For example, a verb is missing. I guess it should be improved instead of improvement.)
Response: Considering the Reviewer’s suggestion, we have rewrite this part in the paper.
- Response to comment(Reviewer 1):( Line 30: The correlation coefficient is mentioned, but it’s unclear what the correlation is between observed and predicted? Please make it clear.)
Response: The coefficient shows COD play a great role in affecting the transparency of river networks in plain cities, and we have revised this part in the paper.
- Response to comment(Reviewer 1):( Line 50 and line 77: Previous studies==>citation of more than one paper is necessary to support the world Previous studies)
Response: Considering the Reviewer’s suggestion, we have added several more reference here in the paper.
- Response to comment(Reviewer 1):( Line 72: Please mention the satellite sensors which can measure water transparency.)
Response: Considering the Reviewer’s suggestion, the satellite sensor was mentioned in this part.
- Response to comment(Reviewer 1):( What color sensors mean？How can remote sensing be combined with color sensors?)
Response: Previous study showed a regionally tuned model from the Geostationary Ocean Color Imager (GOCI) data over the Huang hai sea, the water transparency can be achieved from the data of ocean ribbon differences.
- Response to comment(Reviewer 1):( Line:77 Please describe hydrodynamic conditions, what are these?)
Response: The hydrodynamic conditions includes velocity and water level, and it is described in this part of the paper.
- Response to comment(Reviewer 1):( Reference need for predicted SD might be inadequate)
Response: Considering the Reviewer’s suggestion,, we have changed the reference here.
- Response to comment(Reviewer 1):( You only talk about simple regression model and ANN, are there studies about other machine learning models, e.g, random forest, logistic regression, support vectors machines? It would be better to mention them.)
Response: We have talked about ANN, MLR and SVM in the paper, and we believe machine learning is just a tool, instead of tangled with machine learning itself. What we study is the most important thing.
- Response to comment(Reviewer 1):( Line 88-92: It’s good to know that machine learning(ML) is used in many other disciplines. However, the information is not relevant to the paper-it can go into the discussion section in short.)
Response: Considering the Reviewer’s suggestion, we have moved this part to another place of the paper, and make a short for it.
- Response to comment(Reviewer 1):( Line 102: Please provide the list of each category for physical, chemical, and biological. It would be easy to recognize them.)
Response: Considering the Reviewer’s suggestion, we have made a list for each category in the paper.
- Response to comment(Reviewer 1):( Line 113: Machine learning includes a wide range of models, as mentioned above. It would be better to clarify the objectives that 1) the ability of ANN,2) between ANN and regression model. )
Response: Considering the Reviewer’s suggestion, we have clarified the objectives that 1) the ability of ANN,2) between ANN and regression model. )
- Response to comment(Reviewer 1):( Line 163: ANN is used for prediction in your study. Maybe better to start with the term prediction than explore the relationships.)
Response: Considering the Reviewer’s suggestion, we have rewrite this part in the paper.
- Response to comment(Reviewer 1):( Line 167: Why do you use the sigmoid activation function? Please mention.)
Response: We chose sigmoid as activation for it is often used in environmental science and biology because of its smoothness and easy derivation.
- Response to comment(Reviewer 1):( Line 189:we also attempt to investigate==> we investigate)
Response: We are very sorry for our incorrect writing, and we have revise it.
- Response to comment(Reviewer 1):( Line215: The SVM suddenly appears here, but it’s not clear how it can be used to explore relationships. Please explain it.)
Response: Considering the Reviewer’s suggestion, we have added a SVM part in the Methods, we use SVM model to clarify the feasibility of flow velocity as an input parameter of prediction model here.
- Response to comment(Reviewer 1):( Line 238-240: Please remove it. Instead, simple write, we know the results in different subheadings.)
Response: Considering the Reviewer’s suggestion, we have removed this part and simplify these sentences.
- Response to comment(Reviewer 1):( Line 241-245: This is a method, not results.)
Response: Considering the Reviewer’s suggestion, we have move this part to the Results.
25 Response to comment(Reviewer 1):( Line247-248: the verb is missing.
Figure 3: Why do you not show all three lines in a single plot.)
Response：We are very sorry for our incorrect writing, and we have revised it. We thought that putting together the three trend lines will interlace each other, which is not conducive to reflecting the trend of each line over time.
- Response to comment(Reviewer 1): ( Line249-252: It is not good to report the range only. The range is easily visible to all readers. Instead, the range can also be tabulated in Table 2.)
Response: Considering the Reviewer’s suggestion, we have added more content to this part, for better illustrating the significance of these results.
- Response to comment(Reviewer 1):( Line 266-267：Is it a good way to simply discard 1.5*Q1 values and 1.5*Q3 values? I suggest checking the outlier using other approaches. How many outliers were discarded(can you show in %).)
Response: According to the principle of statistical analysis, a sample with a mean square error value exceeding 5% of the total data volume is regarded as an abnormality, and the two parts in the article are just outside 5%, and the percentages have been given in the article.
- Response to comment(Reviewer 1):( Line 280-284: It seems like a method rather than results.)
Response: Considering the Reviewer’s suggestion, we have revise this part to match the results.
- Response to comment(Reviewer 1):( Line 286: Please present your results of trial and error. It’s important to know. Explain the error in the model. How did you measure that? How many iterations are done? And how did you know 15 neurons work best for you? How many times did you test the model? Did you perform k-fold validations? Maybe 15 neurons were better only one time for that samples?)
Response: In fact, the trial-and-error process and results are numerous and complicated, with more than 100 times. We feel that writing this part into the article will add unnecessary amount of space. The most important thing is the research result itself. After a lot of trial and error, and comparing the literature of previous studies, we decided to set the number of neurons to 15 in the entire final study.
- Response to comment(Reviewer 1):( Line 292: Maybe the setup can be made at least three times, considering that the test dataset should at least once train the model.)
Response: According to the suggestion, we train the model at least three times, and also made corresponding changes to model results.
- Response to comment(Reviewer 1):( If you combine training and verification for training, why did you spilt it first? Please explain. And please run the model at least three times and take the average of the result.
Merge figures 4 and 5 into a single figure with six subplots.)
Response: Maybe we make some mislead in our first manuscript, and we didn’t spilt the part. We have revised these sentences in the paper, and merge figures 4 and 5 into a single figure with six subplots.
- Response to comment(Reviewer 1):( Figure 6: You describe ANN and regression in process, but the figure includes SVM and LR for regression. Please add a section in methods about SVM.
Overall most of the results seem to be the method.)
Response: According to the suggestion, We have added a SVM section in the method part, we propose the SVM to Introduce the rationality of the flow velocity as an input parameter in the prediction model.
- Response to comment(Reviewer 1):( It would be better to discuss the insights in the discussion section. For example, when you cite some tables or figures, provide the main outcomes in the result section.
Provide reasons behind why DO and TE are less sensitive to the model. If possible, try to figure out some physical dynamics behind the results.)
Response:DO and TE are mainly used as environmental factors to affect the chemical environment of the water body, which in turn affects the transparency of the water body, However, after years of water diversion and diversion in the river network cities of the Yangtze River Delta, the chemical environment of the water body tends to be stable, so transparency is not sensitive to DO and TE, but COD, which is closely related to sedimentation, is more sensitive. This content was added to paper.
- Response to comment(Reviewer 1):( Line393-395: very difficult to understand)
Response: Considering the Reviewer’s suggestion, we have rewrite this sentence.
- Response to comment(Reviewer 1):( Line 397 400: Remove them. You do not need to mention that these details are in figures and tables redundantly.)
Response: Considering the Reviewer’s suggestion, we have removed this part.
- Response to comment(Reviewer 1):( Line 401-404: move to the result section.)
Response: Considering the Reviewer’s suggestion, we have moved this part to result section.
- Response to comment(Reviewer 1):( Line 404-406: Try to answer why you got such results. These results should in the result section.)
Response: We try to use SVM results to illustrate the rationality of using flow rate as an input parameter of the transparency prediction model, and clarify that within a certain flow rate threshold, the increase in flow rate has a positive effect on the improvement of transparency.
- Response to comment(Reviewer 1):( Line 426: It would be better if you rank the variables for prediction than mentioning the inclusion of COD improves the model.)
Response: According to the suggestion ,we make a rank of the variables in the paper.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
This manuscript presents and analyzes empirical models used to predict water transparency (Secchi depth, SD) based on hydrodynamic (velocity) and water quality parameters in the Suzhou urban river network. The artificial neural network model performed better than the linear regression model. The correlation between water transparency and flow velocity is also explored. Results could inform management through development of a flow rate threshold to improve water transparency in the river network. However, there are some issues to be addressed to clarify the objectives and methodology, and clearly substantiate the conclusions. These issues are highlighted in the comments below, and specific comments included in the attached file.
1. Cleary identify and develop the goals/objectives of the study. The objectives listed in the Introduction (evaluation of machine learning for SD prediction and comparison of model type performance) do not align with the title, and the title does not adequately reflect the results.
The paper has a second main component that explores the correlation between water transparency and flow velocity, which is poorly integrated into the paper. This topic is not reflected in the objectives, its Methodology section (2.3.4) comes as a surprise, and it is not clear why the analysis of the correlation between SD and velocity uses a different methodology than the developed ANN model in Section 2.3.3. The conclusion of incorporating a flow rate threshold into hydrodynamic regulations is a welcome addition, and indicates the potential relevance of this paper. However, the paper needs to be substantially revised to substantiate and appreciate this conclusion, and the conclusion that SD has improved under long-term hydrodynamic control measures is also not supported by the paper in its present form.
2. Include essential text to frame and support research. The Introduction needs to provide more context on the Yangtze and its delta, and be better referenced. Use of water transparency as an indicator of water quality needs to be better explained and framed. The specific use that this study is targeting should also be clarified.
More information about the “large-scale clearing water regulation” needs to be included such as when it was implemented, its major policy/management components, and the relevance of this study.
There is too much non-essential information in the Methods section (including equations), and critical explanatory pieces are missing or included instead in the Introduction, Results, or Discussion sections. The Methods section (2.2) for water quality monitoring data is largely missing, and why/how the Flow data processing method is used is not clear.
3. Discussion. The discussion focuses too much on specific model performance results, and needs to elaborate on how results build on and align with other studies, and lead to clear conclusions.
Comments for author File: Comments.pdf
Author Response
Thank you for your letter and for the reviewers’ comments concerning our manuscript entitled “Exploring the most effective parameter for water transparency prediction of plain urban river network: a case study of Yangtze River Delta in China”(ID:1231468). Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our researches. We have studied comments carefully and have made correction which we hope meet with approval. Revised portion are marked in blue in the paper. The main corrections in the paper and the responds to the reviewer’s comments are as flowing:
Responds to the reviewer’s comments:
- Response to comment (Reviewer 2):( Lines 42-43: State that “water transparency is a commonly used indicator of water quality” but this is not substantiated by the reference provided)
Response: Considering the Reviewer’s suggestion, we have changed the reference here.
- Response to comment (Reviewer 2):( Lines 61-62: Please explain what is meant by” water diversion projects.” What is water diversion for, or to?)
Response: For many years, the Yangtze River Delta region of China has used water diversion and diversion to ecologically replenish water in plain river network cities, aiming to improve the urban river network water environment.
- Response to comment (Reviewer 2):( Line 74: Please include brief description of color sensor. Is this ocean color remote sensing, or anther method. )
Response: Considering the Reviewer’s suggestion, we already make a description of color sensor and remote sensing in this part of paper.
- Response to comment (Reviewer 2):( Line93-100: Table 1 shows selected studies for SD predictions, but it is not clear that this table is necessary or adds much to the Introduction. The contents of the table are already summarized, which could be bolstered, and th table removed(or moved to Methods since ANN model builds on this).
Response: Considering the Reviewer’s suggestion, we have rewrite this part for better expressing its scientific significance.
- Response to comment (Reviewer 2):( Line86-93: Since ANN is being used in this paper, it should be more thoroughly introduced, including its benefits. It is used in many more environmental applications than just the ones listed.)
Response: Considering the Reviewer’s suggestion, we have a more thorough introduction to ANN, and more relevant reference for this part.
- Response to comment (Reviewer 2):( Line 109: TSS is usually total suspended sediment or total suspended solid concentration(not substance).)
Response: We are very sorry for our incorrect writing, and we have revised it.
- Response to comment (Reviewer 2):( Line 172: x_j in the equation should be also be defined. All variables in all of the future equations use in the paper too.)
Response: Considering the Reviewer’s suggestion, relevant equation have been defined.
8 Response to comment (Reviewer 2):( Line 204: Is the reference to the equation 6 accurate, or should it be equation7?)
Response: The reference should refer to the equation 7,and we have already revised it
- Response to comment (Reviewer 2):( Lines 196-208: These appear to be equations for the MLM, and if so ,consider including a reference instead. The MLM equations that need to be satisfied or the exact equations to obtain the estimators do not seem relevant here.)
Response：Considering the Reviewer’s suggestion，we have added relevant reference here. The MLM occupy a certain proportion in the nonlinear analysis of this article, and use this to lead to the SVM model for subsequent analysis。
- Response to comment (Reviewer 2):（Line 210: Isn’t y_o SD and not turbidity?）
Response：we are sorry for this mistake, and we have revised it.
- Response to comment (Reviewer 2):( Lines 217-236 : These are standard model performance assessment methods, and the equations do not need to be included.)
Response: Considering the Reviewer’s suggestion，we have rewrite this part.
- Response to comment (Reviewer 2):( Lines 238-240: Remove.)
Response: Considering the Reviewer’s suggestion，we have removed this part.
- Response to comment (Reviewer 2):( Lines 241-243: This text is about methods, and should be moved to the Methods section.)
Response: Considering the Reviewer’s suggestion, we have moved this part to the Method section.
- Response to comment (Reviewer 2):( Lines 259-263: It would be helpful to summarize the other 5 parameters investigated, either with figures and/ or tables similar to those of SD.)
Response: We thought that comparing all the other 5 variables into a chart will make the page too verbose and does not highlight the focus of transparency.
- Response to comment (Reviewer 2):( Lines265-271: This may be a better fit in the method section.)
Response: Considering the Reviewer’s suggestion, we have put forward the results of data processing here, and led to the subsequent model input parameters, so we think it is more appropriate to put this part of the content in the results.
- Response to comment (Reviewer 2):( Lines 294-295: Why do SD Xmax and Xmin not match to the values in Table 2?)
Response: In fact, the SD Xmax and Xmin match to the values in Table 2 for all the data is got once a week, and the abscissa of time may be elongated in the figure and it is not obvious to see.
- Response to comment (Reviewer 2):( Lines:287-291: Better fit in Methods section. Lines 296-297.Explain why is there a training, verification, and test dataset for Section3.2, and only training and testing datasets for 3.3. How this relates to the ANN model(validation dataset) should be explained.)
Response: Actually all models have the training, verification, and test section, it’s just that we didn’t write them all. We should not be too entangled with the method itself, use it as a tool, what we study is the most important
- Response to comment (Reviewer 2):( Lines 302: Doesn’t the M4 model have better performance than M3? )
Response: The CC and RMSE of M4 are all better than those of M3, so M4 model have better performance than M3. And we already revised it.
- Response to comment (Reviewer 2):( Line 305: I don’t see a large distinction among the MLR fits for the train, verify, and test cases. How is the MLR well fitted on the training dataset?)
Response: Maybe we didn’t express it well here, We make this statement because the MLR model has the best performance in the training phase, which is better than the verification phase and the test phase.
- Response to comment (Reviewer 2):( Lines 326-329: This explanation would have been helpful earlier when this method discussed. How is the flow controlled, are there certain water level or thresholds that trigger management action)
Response: Because the water level of all river sections in the plain urban river network is almost the same, the flow velocity is mainly the change factor. We made some minor changes here.
- Response to comment (Reviewer 2):( Lines 329-334: Better fit in Methods. How were the training and datasets determined?)
Response: Considering the Reviewer’s suggestion, we have made some revision here. We train 60% of the data set, test 20% of the data, and adjust the relevant parameters by comparing the results
- Response to comment (Reviewer 2):( Lines 335-336: Not sure how this table is informative, and it is not across-referenced in the text. It does not appear to go with this section(3.3)? Is it from section 3.2?)
Response: Considering the Reviewer’s suggestion, we changed the position of this table.
- Response to comment (Reviewer 2):( Line 348:Please explain how the SVM model was optimized by adjusting parameters.)
Response: When using a support vector machine, we only need to adjust the range of the regularization parameter C. The parameters we need to adjust are the regularization parameter C and the kernel function parameter gamma. In order to ensure the accuracy of tuning, we generally use the grid search method to determine the parameters. The grid search method is to give the adjustment range and adjustment step length of each parameter, calculate the possible value of each parameter, and then traverse all the combinations to return the best parameter value.
- Response to comment (Reviewer 2):( Line359-360: Why is the “Test dataset” shown first in the table? It seems out of order, should be training and then test to correspond with the methodology.)
Response: This is just a coincidence. When writing the article, the test data happened to be at hand, and the article has been modified.
- Response to comment (Reviewer 2):( Line 366: Isn’t the CC higher for ANN, not “lower”?)
Response: Considering the Reviewer’s suggestion, we have revise it. We are very sorry for this mistake。
26：Response to comment (Reviewer 2):（Line 433: This particular analysis/result/R2 of 0.793 is nowhere previously included in the paper. ）
Response：Considering the Reviewer’s suggestion, we have already revised it.
We tried our best to improve the manuscript and made some changes in the manuscript. These changes will not influence the content and framework of the paper. And here we did not list the changes but marked in revised paper.We appreciate for Editors/Reviewers’ warm work earnestly, and hope that the correction will meet with approval. Once again, thank you very much for your comments and suggestions.
Yours
Sincerely
Yipeng Liao with all authors
Author Response File: Author Response.pdf
Reviewer 3 Report
The authors present an interesting and comprehensive investigation of effective parameters for water transparency prediction of the plain urban river network. Though the study is a case study of a selected study area of the Yangtze River Delta in China, the methods applied are insightful and replicable to the regional and global scales. The research deserves publication and outreach. However, there are many locations where revisions and modifications are needed. I compiled a list including recommendations and questions to the authors.
Please see the attachment.
Comments for author File: Comments.pdf
Author Response
Thank you for your letter and for the reviewers’ comments concerning our manuscript entitled “Exploring the most effective parameter for water transparency prediction of plain urban river network: a case study of Yangtze River Delta in China”(ID:1231468). Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our researches. We have studied comments carefully and have made correction which we hope meet with approval. Revised portion are marked in blue in the paper.
Round 2
Reviewer 2 Report
The authors have fully addressed my minored comments, and largely incorporated my main comments. In combination with other changes, the paper has been clarified.
Reviewer 3 Report
Previous comments were incorporated. However, the article should be checked for English structure precisely. I could see lots of issues with grammar and language. I suggest the authors have a professional English check or via a native speaker before a final publication.
For instance,
Line 17: is still need explore
Line 17-18: use of word 'explore' three times
Line 20 - 24: very difficult to follow the English structure
Line 30 : SD (first occurrence but no full form)
Please improve the language through the manuscript. The above is an example while looking at the abstract section.
Line 264 : were shown == > are shown. Use consistent tense to show the figures and tables.
and methods in the past tense.
Texts used in figures are not readable, for instance, Figure 1.
