Identification and Monitoring of Irrigated Areas in Arid Areas Based on Sentinel-2 Time-Series Data and a Machine Learning Algorithm

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

All these comments are sown in the PDF file of the article in the Sticky notes

The article titled 'The identification and monitoring of irrigation areas in arid regions based on Sentinel-2' is an effort to identify irrigated areas in the arid region of Northwest China-Xinjiang Santun River using Sentinel-2 images and three machine learning methods: The Random Forest, CART Decision Tree, and Support Vector Machine algorithms.

In my opinion, the paper has the potential for publication, provided that the suggested comments are addressed as part of a major revision.

 

·       It is necessary for the article's title to be more comprehensive and to also mention machine learning in the title.

·       At the end of the introduction section, it is customary for authors to clearly present the research gap and the innovation of their study. However, in this article, I did not find such a point. The authors have provided a comprehensive literature review of other studies but have not specified what the research gap is and what innovation they are pursuing in this study. Instead, at the end of the introduction, they have defined their methodology.

·       -Line 119 : In my opinion, it would be better to change the title of the second section to 'Materials and Methods.

·       -Figure 1 could be better illustrated to show which province in China the study area is located in. It is not clear to me where the study area is. Use vector lines to show the area within the country of China, the province, and the region in different data frames. Additionally, a frame from a satellite image of the study area could also be included.

·       -In my opinion, the images in Figure 2 do not convey any specific information and only increase the length of the article.

·       In Figure 3, better colors should be used in part (a) to clearly distinguish the shapes. Additionally, the colors in (a) should match the colors in (b) and (c) for the years under study. Figures (b) and (c) can all be placed in a single frame to reduce the length of the article. Figures (b) and (c) also need further explanation.

 

·       Figure 4 has a lengthy caption. The text within the figure is also in small font and is not easily readable. The methods shown in the figure should be thoroughly explained in the text.

·       In my opinion, it is necessary to explain the tools used, such as the machine learning models, in a dedicated section within the Materials and Methods part of this study.

·       -Line 237: need to write these indices completely

· It is unclear what tools the researchers used to apply machine learning methods such as RF, CART Decision Tree, and SVM for classification. Specifically, what software was used in the research? In some studies, researchers use the Google Earth Engine platform and code within it. It is essential for the researchers to specify the type of software they used in the Materials and Methods section of their paper. Did they write their own code in a different environment?

· -Line 324: Overall Accuracy (OA)

· -Section 4.1 and Figure 5 : Understanding this section and Figure 5 was somewhat challenging for readers. Is there a reference for this analysis? The authors state, 'After conducting comparison experiments at various segmentation scales, a scale parameter of 70 was selected for use in this study (Fig. 5).' Does this mean they chose 70 because it represents the intersection of the local variance and rate of change curves? This section requires more explanation to clarify its relevance to the research objectives.

· -Line 346:  needs explanation and reference for Savitzky–Golay  filter

· In Figure 6, it is not clear what the dates are? Also, all Figures can be combined in one framework.

· In Figure 7, it is not clear what the dates are? Also, all Figures can be combined in one framework.

· -Line 400 : repeated sentence. It was written before.

·       Figure 8 : Legend in the Figure is not clear.  As well, it is better to show the layers in the period of 2019 -2023 from left to right.

·       Figure 9 :  all figures should be shown in one frame. They took two and a half pages, too long.

·       Figure 11:  all figures should be shown in one frame . They took two and a half pages, too long.

Comments for author File: Comments.pdf

Comments on the Quality of English Language

The quality of English is appropriate.

 

Author Response

Comments 1: [It is necessary for the article's title to be more comprehensive and to also mention machine learning in the title.]

Response 1: Thank you for your valuable comments. Based on your comments, we have changed the title to Identification and Monitoring of Irrigation Districts in Arid Zones Based on Sentinel-2 Time Series Data and a Machine Learning Algorithms (see the blue text in Line 1).

 

Comments 2: [At the end of the introduction section, it is customary for authors to clearly present the research gap and the innovation of their study. However, in this article, I did not find such a point. The authors have provided a comprehensive literature review of other studies but have not specified what the research gap is and what innovation they are pursuing in this study. Instead, at the end of the introduction, they have defined their methodology.]

Response 2: Thank you for your valuable comments on our article. In response to your question about the lack of a clear presentation of the research gaps and research innovations at the end of the introduction, we have adjusted the content at the end of the introduction accordingly (see the blue text in Lines 104–109).

 

Comments 3: [In my opinion, it would be better to change the title of the second section to 'Materials and Methods.]

Response 3: Thank you for your valuable comments, in response to which we have changed the title of this section to Materials and Methods (see Line 119).

 

Comments 4: [Figure 1 could be better illustrated to show which province in China the study area is located in. It is not clear to me where the study area is. vector lines to show the area within the country of China, the province, and the region in different data frames. Additionally, a frame from a satellite image of the study area could also be included.]

Response 4: Thank you for your valuable comments. We have revised the overview map of the study area according to your request and have laid it out according to the country, province, city, watershed, and irrigation district (see Figure 1).

 

Comments 5: [In my opinion, the images in Figure 2 do not convey any specific information and only increase the length of the article.]

Response 5: Thank you for your comment. Figure 2 is a drone image taken while we were conducting field sampling. As per your suggestion, we have removed the image.

 

Comments 6: [In Figure 3, better colors should be used in part (a) to clearly distinguish the shapes. Additionally, the colors in (a) should match the colors in (b) and (c) for the years under study. Figures (b) and (c) can all be placed in a single frame to reduce the length of the article. Figures (b) and (c) also need further explanation.]

Response 6: Thank you for your valuable comments on our article. In response to your questions, we have made the following changes. ① In order to further enhance the information communication effect and visual clarity of Figure 3(a) (now labeled Figure 2), we have carefully adjusted its color scheme and content layout. This improvement aims to ensure that the charts can present the key data features more intuitively, making it easier for readers to quickly capture the main points of information. ② We have clearly marked the sample points in Figure 3(a) (now labeled Figure 2). By clearly distinguishing them with symbols or colors, the differences and distributions among the samples can be seen at a glance, which enhances the persuasiveness and readability of the chart. (3) Considering the clarity of the pictures and the harmony of the overall layout, we decided to temporarily separate the two pictures that were originally planned to be displayed together. Although this decision is a slight adjustment, it aims to ensure that each image is presented in the best way, which not only ensures a detailed presentation of the information but also avoids the inconvenience of reading caused by too dense a layout. We sincerely hope that this adjustment will meet your requirements (see Figure 2).

 

Comments 7: [Figure 4 has a lengthy caption. The text within the figure is also in small font and is not easily readable. The methods shown in the figure should be thoroughly explained in the text.]

Response 7: Thank you very much for your detailed feedback. We have made corresponding changes to the article according to your suggestions. The following is a description of the revisions: ① Regarding the problem that the title is too long, we have streamlined the title (see Figure 3). ② Regarding the issue that the text font is too small, we have increased the size of the font in the figure. ③ The content of Figure 4 is explained in Lines 208 to 215 in the article.

 

Comments 8: [In my opinion, it is necessary to explain the tools used, such as the machine learning models, in a dedicated section within the Materials and Methods part of this study.]

Response 8: The problem you have raised is described in detail in Section 3.3 Object-oriented classification algorithms for Random Forest, CART, and SVM (see Lines 262–287).

 

Comments 9: [need to write these indices completely]

Response 9: Thank you very much for your detailed feedback. We have defined the acronym of each index in the article (see the blue text in Lines 243–245).

 

Comments 10: [It is unclear what tools the researchers used to apply machine learning methods such as RF, CART Decision Tree, and SVM for classification. Specifically, what software was used in the research? In some studies, researchers use the Google Earth Engine platform and code within it. It is essential for the researchers to specify the type of software they used in the Materials and Methods section of their paper. Did they write their own code in a different environment?]

Response 10: In this study, eCognition software was used to classify the three algorithms (random forest, CART, and SVM). We have identified the various pieces of software used in this study in the text (see the blue text in Lines 156–158).

 

Comments 11: [Line 324: Overall Accuracy (OA)]

Response 11: Thank you for your suggestion. We have corrected this in the article (see the blue text in Line 333).

 

Comments 12: [Section 4.1 and Figure 5: Understanding this section and Figure 5 was somewhat challenging for readers. Is there a reference for this analysis? The authors state, 'After conducting comparison experiments at various segmentation scales, a scale parameter of 70 was selected for use in this study (Fig. 5).' Does this mean they chose 70 because it represents the intersection of the local variance and rate of change curves? This section requires more explanation to clarify its relevance to the research objectives.]

Response 12: Thank you for your valuable comments on our article. In response to your questions, we provide the following answers. ① In Section 3.1 Multi-scale and optimal segmentation scale selections, part of the object-oriented multi-scale segmentation is utilized to make the corresponding explanation. In addition, I have added two references. ② The peak inflection point of the rate of change (ROC) indicates that the value at that point is a better segmentation scale (see Fig. 5) (now labeled Fig. 4). The peaks of the ROC occur at four points, 70, 123, 127, and 136, the four values are tested for segmentation, and the optimal segmentation scale of the peak at 70 is obtained at the end. ③ We have made changes to Section 4.1 to explain its relevance to the purpose of the study (see the blue text in Lines 341–345).

 

Comments 13: [Line 346: needs explanation and reference for Savitzky–Golay filter]

Response 13: Thank you for the comments. We have explained the Savitzky-Golay filter in Lines 316–321 and have added two references to supplement it (See references 31 and 32; Zhang et al., 2023 and Wu et al., 2023)).

 

Comments 14: [In Figure 6, it is not clear what the dates are? Also, all Figures can be combined in one framework.]

Response 14: In response to your question, we provide the following answer. ① The meaning of the date in Fig. 6 (now labeled Fig. 5) is the date of the remote sensing image used for 2023, which is also explained in the article (see the blue text in Lines 362–367). ② Considering the clarity of the images and the harmony of the overall layout, we decided to temporarily separate the images that we had planned to show together. This is a slight adjustment, but it is intended to ensure that each image is presented in the best way possible, both to ensure a detailed presentation of the information and to avoid the inconvenience of reading a densely-packed layout. We sincerely hope that this adjustment is satisfactory.

 

Comments 15: [In Figure 7, it is not clear what the dates are? Also, all Figures can be combined in one framework.]

Response 15: Since this question is similar to that in comment 14, we answer it as follows. ① The meaning of the date in Fig. 7 (no labeled Fig. 6) is the date of the remote sensing image used for 2023. ② Considering the clarity of the images and the harmony of the overall layout, we decided to temporarily separate the images that were originally planned to be displayed together. This is a slight adjustment, but it is intended to ensure that each image is presented in the best way possible, which ensures the detailed presentation of information and avoids the reading inconvenience caused by too dense a layout. We sincerely hope that this adjustment is sufficient.

 

Comments 16: [Line 400: repeated sentence. It was written before.]

Response 16: Thank you very much for your detailed feedback. We have removed the duplicate sentence.

 

Comments 17: [Figure 8: Legend in the Figure is not clear. As well, it is better to show the layers in the period of 2019 -2023 from left to right.]

Response 17: Thank you for your suggestion. To address the problems you have pointed out, we have made the following changes: ① Regarding the issue that the legend is not clear in Figure 7, we have readjusted the font of the legend. ② Regarding the order of the year, we have adjusted the order to 2019–2023 from left to right (see Figure 7).

 

Comments 18: [Figure 9: all figures should be shown in one frame. They took two and a half pages, too long.]

Response 18: Considering the clarity of the images and the harmony of the overall layout, we have decided to temporarily separate the images that were originally planned to be displayed together. Although this is a slight adjustment, it aims to ensure that each picture is presented in the best way possible, which not only ensures the detailed presentation of information, but also avoids the inconvenience of reading due to the dense layout. We sincerely hope that this adjustment is sufficient.

 

Comments 19: [Figure 11: all figures should be shown in one frame. They took two and a half pages, too long.]

Response 19: Thank you for your suggestion. Since this issue is the same as that in Comment 18, we have made the same decision, and we hope that this decision will receive your understanding and support.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The authors present an interesting study which is well presented. However, the main issue with this work is how the authors fail to state the novelty and aim of the study explicitly. The hypothesis under review is also not being expressed by the authors. In addition, the methods section misses critical information which makes the work impossible to reproduce. The authors need to improve on that. I have included some sections that need attention. 

1. Figure 1 should be re-adjusted and improved. The current figure is not ordered properly to show the study area. Look at more recent publications and adjust your figure accordingly. 

2. Figure 3 could be changed to a table. I think this could do more justice. The points can be shown on the study area map

3. Under section 2.2.4. Statistics kindly include a summary of how the statistics were derived. This improves clarity. 

4. Section 3.1 needs more information. In its current state, it has more literature than the actual method that was used to derive the segments. Also, include software etc. 

5. The authors should consider testing the significant difference in the classification outputs beyond the classification outputs. 

 

 

 

 

 

 

 

 

 

 

 

Comments on the Quality of English Language

N/A

Author Response

Comments 1: [Figure 1 should be re-adjusted and improved. The current figure is not ordered properly to show the study area. Look at more recent publications and adjust your figure accordingly.]

Response 1: Thank you for your valuable comments. We have revised the overview map of the study area according to your request and have laid it out according to the country, province, city, watershed, and irrigation district (see Figure 1).

 

Comments 2: [Figure 3 could be changed to a table. I think this could do more justice. The points can be shown on the study area map]

Response 2: Thank you very much for your valuable feedback. We have fully considered your comment and made the following adjustments. ① After in-depth discussions and careful planning by the team, we have decided to use the picture display, aiming to present the reader with more vivid and easy to understand information through the intuitive comparison from the three-dimensional perspective. This decision was made to improve the effectiveness of information delivery and the readers' understanding, so the traditional table format was not used. We sincerely hope that this change will meet with your approval and believe that it will bring you a better reading experience. ② We have carefully revised and optimized Figure 3(a) (changed to Figure 2). Now, the distribution of each category is clearly labeled on the sample point map, making the categorized information clear at a glance and making it easy for readers to quickly and accurately grasp the core points of the data. We hope that this improvement will better meet your needs (See Figure 2).

 

Comments 3: [Under section 2.2.4. Statistics kindly include a summary of how the statistics were derived. This improves clarity.]

Response 3: Thank you very much for your detailed feedback. In response to the issues you raised, we have revised the content of Section 2.2.4 (see the blue font in Lines 195–206).

 

Comments 4:[Section 3.1 needs more information. In its current state, it has more literature than the actual method that was used to derive the segments. Also, include software etc.]

Response 4: Thank you for your valuable comments. We provide the following answers to your questions and have made the suggested revisions. ① In Section 3.1, we have added the specified reference (Zhao et al., 2023). ② In the results section corresponding to Section 3.1, we have added the specified content and references (Fan et al., 2021) (see Lines 339–343). ③ In the Materials and Methods section, we have added the software used (see Lines 156–158).

 

Comments 5: [The authors should consider testing the significant difference in the classification outputs beyond the classification outputs.]

Response 5: Thank you very much for your valuable suggestion to ensure that the evaluations based on significant differences in the classification outputs may not be sufficient to fully understand the model performance and its applicability in different contexts. In this article, the following measures have been taken to extend the scope of the testing. ① Diversification of performance evaluation metrics: in this study, multiple metrics, including the Producer's Accuracy (PA), User's Accuracy (UA), Overall Accuracy (OA), and kappa coefficient, were used to conduct the evaluation. ② Independent testing set: in this study, tests were conducted separately for each year from 2019 to 2023, and an independent testing set was used to verify the generalization ability of the model. ③ Comparative testing: we compared our model with other relevant models in the existing literature to assess its performance advantages (see Section 5).

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Comments about the Paper: “The identification and monitoring of irrigation areas in arid regions based on Sentinel-2”.

 The paper aims to monitor irrigation areas the arid environment based on Sentinel-2 time-series imagery vegetation indices and classification algorithms such as Random Forest, CART Decision Tree, and Support Vector Machine. The results are promising regarding the extracted irrigated areas from 2019 to 2023 by applying Random Forest algorithm. However, the Methodology needs some improvements (see in the attached file).  Please provide a clear flowchart.

All in all, my recommendation is to accept the paper for publication, subject to minor revisions. Please find in the attached file the amendments which I believe are required prior to accepting the paper.

 

Comments for author File: Comments.pdf

Author Response

Comments 1: [Line 136, Please provide guide map about the location of Xinjiang in China.]

Response 1: Thank you for your valuable comments. We have revised the overview map of the study area according to your request and have laid it out according to the country, province, city, watershed, and irrigation district (see Figure 1).

 

Comments 2: [Line 140, Please provide more details. How many S2 images did you process per year? (2019, 2020, 2021, 2023). For what time period each year (i.e. May to November)?]

Response 2: Thank you very much for your detailed feedback. We have made the appropriate additions based on your valuable comments. We have added Table 1, which lists the specific dates of the remote sensing images used in this experiment, providing a clear record of the timeline of the study. We believe that this addition will help readers to better understand the source and acquisition times of the experimental data, which will in turn enable them to assess the timeliness and accuracy of the research results in depth. Thank you again for your attention to detail and guidance.

 

Comments 3: [Line 149, I don't understand what do you mean by......acquired over 48 days?]

Response 3: Thank you very much for your valuable comments on our article. Corrections have been made to address the issues you raised (see the blue text in Lines 148–150).

 

Comments 4:[Line 158, How many sample points did you acquired from May to November 2023?]

Response 4: Thank you very much for your detailed feedback. A total of 723 sample points in 2023 were used (see Figure 2(b)).

 

Comments 5:[Line 164, The samples were from 2019 to 2022 or and 2023? Did you acquire them the same period (May to November)? If yes, how? Please clarify.]

Response 5: Thank you very much for your detailed feedback. We answer your questions in several ways. ① First, the 2023 sample points are field sample points (see Lines 162–171). ② The 2019–2022 sample points were determined to be irrigated by analyzing the spectral profiles and index characteristics of the 2023 field sample points for each year during 2019–2022, after which we added or removed visually deciphered points, which were screened as sample additions based on spectral characteristics (see Lines 172–176).

 

Comments 6:[Line 201, for each year separately?]

Response 6: Yes, this study was tested separately each year from 2019–2023.

 

Comments 7:[Line 245, Somewhere in the methodology (maybe materials section) you have to mention what kind of software have been used.]

Response 7: Thank you very much for your detailed feedback. We have made changes to address the issues you have raised (see the blue text in Lines 156–158).

 

Comments 8:[Line 279, It's not clear to me how did you identify the irrigated areas. By performing the classification procedure (I don't think so because the applied classification scheme doesn't contain this type of land cover) or afterwards?Please clarify.]

Response 8: Thank you very much for your detailed feedback. The land use classification is conducted first in order to distinguish the cropland from other land use types and then to distinguish the agricultural wasteland and abandoned land from areas that require irrigation through differences in crop spectral characteristics and vegetation indices due to irrigation (see Section 3.4).

 

Comments 9:[Line 355, What does the red and blue line indicate? Please explain.]

Response 9: Thank you very much for your detailed feedback. In response to your questions, we provide the following answers. ① The blue line represents the homogeneous local variance (LV) of the segmented object. ② The red line represents the rate of change (ROC) of the segmented object (see Lines 220–238 in Section 3.1).

 

Comments 10:[Line 477, Maybe here a short description about how could your method be transferred elsewhere.]

Response 10: Thank you for your valuable comments. In this section, we first analyze the applicability of the three different algorithms, the impacts of the different vegetation indices on the extraction of the irrigated area, and the uncertainty of the process. We provide a short description in the article (see the blue text in Lines 549–553).

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Dear Authors

Please find my comments in the attached file.

Comments for author File: Comments.pdf

Comments on the Quality of English Language

My comments are in the attached file.

Author Response

Comments 1: [Round 1 : At the end of the introduction section, it is customary for authors to clearly present the research gap and the innovation of their study. However, in this article, I did not find such a point. The authors have provided a comprehensive literature review of other studies but have not specified what the research gap is and what innovation they are pursuing in this study. Instead, at the end of the introduction, they have defined their methodology. Round 2 : In my opinion, the answer given regarding the innovation of the article and filling the research gap was not convincing and this explanation cannot show that this article was innovative. My expectation was that the authors would clearly explain what was the innovation of this research in relation to the previous research. Currently, the use of machine learning methods in remote sensing research, such as RF and SVM, has become a routine practice, and it can be done very easily using smart software.]

Response 1: [Thank you again for your suggestions. In response to the questions about the innovation of the article and filling the research gaps, we have organized and summarized the article in the following three directions: i) innovation in the specific research perspectives of irrigated areas in arid zones, ii) innovation in the application of combining remotely sensed data with advanced machine learning algorithms, and iii) innovation in the application of the Soil Adjusted Vegetation Index(SAVI) in arid zones, see Lines103-127, Blue Font section.]

 

Comments 2: [Round 1 : In Figure3, better colors should be used in part(a) to clearly distinguish the shapes. Additionally, the colors in(a) should match the colors in(b) and(c) for the years under study. Figures(b) and(c) can all be placed in a single frame to reduce the length of the article. Figures(b) and(c) also need further explanation. Round 2 : Figure2(a) is good now, however, I asked to put the figure b and c in one frame like figure2a. I didn't understand why you separated in two figures and take one page for these two figures.]

Response 2: [Thanks again for your suggestions, and in response to your suggestions, I have put Figures2(b) and2(c) in the same frame.]

 

Comments 3: [Round 1 : need to write these indices completely Round 2 : partly accepted. However I suggest at the end of the article, before the references, the authors put a list of abbreviation with their complete names. there are too much abbreviations in the article without their complete name.]

Response 3: [Thank you again for your suggestions, and in response to your suggestions, I have provided additional descriptions of each index in Table 4 of the article. See the blue-font section of Table 4.]

 

Comments 4: [Round 1 : It is unclear what tools the researchers used to apply machine learning methods such as RF, CART Decision Tree, and SVM for classification. Specifically, what software was used in the research? In some studies, researchers use the Google Earth Engine platform and code within it. It is essential for the researchers to specify the type of software they used in the Materials and Methods section of their paper. Did they write their own code in a different environment? Round 2 : I suggest the author dedicates a section in the method and Material section, to explain these software, because these software are their tools.]

Response 4: [Thank you for your suggestions. According to your suggestions, I have added2.3Software used in this study, which gives a brief introduction to the software used in the study. See Lines215-222, Blue Font section.]

 

Comments 5: [Round 1 : In Figure 6, it is not clear what the dates are? Also, all Figures can be combined in one framework. Round 2 : Not accepted the reason. I think these figures can be in one frame like the figure7.]

Response 5: [Thanks again for your suggestion, and in response to your suggestion, I have put the images in the article in the same frame.]

 

Comments 6: [Round 1 : Figure11: all figures should be shown in one frame. They took two and a half pages, too long. Round 2 : Not accepted the reason. I think these figures can be in one frame like the figure7.]

Response 6: [Thanks again for your suggestion, and in response to your suggestion, I have put the images in the article in the same frame.]

Author Response File: Author Response.pdf