RAU-Net++: River Channel Extraction Methods for Remote Sensing Images of Cold and Arid Regions
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsDear Authors,
The manuscript "RAU-Net++: River Channel Extraction Methods for Remote Sensing Images of Cold and Arid Regions" presents an interesting topic concerning the detection of water bodies in arid and cold places. The topic addressed by his article, which deals with the detection of rivers in arid and cold places, presents several challenges that need to be better explored.
The structure adopted in your article is appropriate and the writing needs occasional revision. Parts of the text with confusing wording were pointed out in the comments on the digital file.
The problem and the review of current solutions have been adequately presented. However, the issue of the presence of rivers in arid and cold places is not adequately highlighted in the presentation of the datasets.
I have highlighted a total of sixty-three comments which can be consulted in the digital file and I would ask you to pay attention to the following issues:
1) Abstract: in addition to presenting the results obtained with the proposed methodology also cite the improvement obtained when compared to current methods. Improve the wording as the text is confusing when presenting what was done.
2) Line 105: "available professional remote sensing". What is professional remote sensing? Are you referring to the characteristics of the available data? Periodicity?
3) Table 1: review the positioning of the lines, especially with regard to the first column;
4) Line 347: "we set the dynamic factor α to the default value of 0.25 and the adjustable hyperparameter 𝛾 to the default value of 2". Please explain how these two parameters were set. If applicable, please provide a bibliographic reference.
5) Line 385: please provide the correction level of the Landsat images, for example, L1TP? Please provide the dates on which the images were acquired;
6) Band 8 of the OLI sensor is panchromatic and has a higher spatial resolution than selected bands. Wouldn't it have been better to merge the selected bands with band 8? Please explain.
7) From the previous question, explore the issue of the time of acquisition. Is this an appropriate time because this is an arid region?
8) Throughout the text it is commented that the methodology is applicable to arid and cold areas. In this case, these characteristics are present in the study site. The arid characteristic is obvious, but what about the cold?
9) Lines 404 to 406: were the parameters used for the network obtained empirically? Please detail this step.
10) Chapter 4.3, line 440: I suggest you consult this article to see other interesting metrics for evaluating the results obtained in your study: https://link.springer.com/article/10.1007/s10462-020-09830-9. With this, it is possible to add other metrics not normally covered in the segmentation evaluation.
11) Table 2: highlight the best result in each column by formatting the text with bold. It would be interesting to contextualize the choice of the different methods tested, as some of those presented are not included in Chapter 1.
12) Figure 9: Please detail the methodology used to obtain the Ground Truth shown.
13) It would be interesting to present the results obtained by the authors cited in the references and who use Deep Learning-oriented techniques;
14) Analysis of the results shows that the proposed method is somewhat superior to alternative methods. However, the difference between them is small. The inclusion of processing time is an interesting variable that can highlight the choice of the method developed.
15) The visual similarity of the results shown in Figure 9 is considerable. Considering the entire study area, what did it actually mean numerically: area and length of riverbanks extracted?
16) The conclusions could be improved to adequately respond to the proposed objectives. What would be the recommendations for future studies?
I end my review by congratulating you on the experiment carried out and the version of the manuscript presented.
Respectfully,
Comments for author File: 
Comments.pdf
Punctuation needs to be revised, especially the comma. When presenting acronyms in the text, please present their meaning. Avoid long paragraphs, as they tend to make reading confusing. I have pointed out in the digital comments the passages where the writing should be improved.
Author Response
"Please see the attachment."
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for Authors1. Explain the need for computing the weighted loss ?
2. What is the size of the feature ap at the input and output of the RAFF module?
3. Why Relu is used? Justify. What are the alternative activation functions?
4.what is the significance of improving F1 score from 99.51 to 99.75? Is it not possible to ignore the loss?
Author Response
"Please see the attachment."
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThis paper presented a new neural-network based method for the extraction of river channels from remote sensing images of cold and arid regions. Results are presented for images of the Tarim river. It is not clear, but it is possible that this is the only dataset used for training and testing. If so, it is necessary to use at least a few more datasets. It is also important to test the performance of the algorithm when it is trained on one dataset and tested on another to make sure it is more generalizable. In any case, the title suggests that the proposed method works for all remote sensing images, so either the title has to be qualified to make it more specific to those types of images tested in the paper, or more testing needs to be done. In general, more information should be provided about the dataset.
Equations 7-10 include a summation over an index i, but then the index does not appear anywhere else in the expressions in those equations. I assume that p and q should instead be p_i and q_i. This should be corrected properly.
Also, it would be good if the introduction had a table listing all previous methods and their characteristics for the sake of comparison.
Comments on the Quality of English Language
There are some parts of the paper where the English can be improved. For example the phrase on line 287-288 does not make sense to me: "which is performed regional category profiles of river remote sensing images". It seems one or more words are missing from that phrase.
Author Response
"Please see the attachment."
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsDear Authors,
The manuscript "RAU-Net++: River Channel Extraction Methods for Remote Sensing Images of Cold and Arid Regions" presents an interesting topic concerning the detection of bodies of water in arid and cold locations. This methodology is important in the management of water resources and has wide application due to growing demand and the drastic alteration of the water regime.
I have read the manuscript in detail and, comparing this version with the previous one, I see that you have made several changes. As a result, reading has become more fluid and the study is fully understood.
Consider repositioning table 12, which is in the conclusions chapter, to the results and discussions chapter.
Thank you for sending the cover letter.
Respectfully,
Comments on the Quality of English LanguagePlease check the punctuation in the text once again.
