Rice Identification and Spatio-Temporal Changes Based on Sentinel-1 Time Series in Leizhou City, Guangdong Province, China
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe main problem about this study is the lack of innovative in the methodology. The generalization ability of the proposed rice label generation method was not sufficiently validated using the multiyear SAR data. Details should be supplement concerning the data processing strategy. In addition, the manuscript presented a lot of discussions about the results, but these contents lack persuasiveness because little details of the multiyear rice distribution results were provided that can match with the discussions.
1) The introduction part does not provide with enough reviews on previous researches. It does not reveal the novelty of the research.
2) Line 83-84 (and line 290-291). You mentioned 3 seasons of rice in previous sentence but only give 2 seasons’ phenology periods.
3) Please reconsider the contextual logics of the manuscript. For instance, the first paragraph of 2.3.1 is poorly organized and redundant.
4) The definitions of RICE and rice in Figure 3 was verbose and ambiguous. Besides, Line 104 mentioned that 98 ground samples were collected, how many of them are rice? And what are the differences of rice-1 and rice-2?
5) Line 154-170, is poorly organized. You didn’t clarify which polarization is demonstrated in Figure 3. If the zoom-in figures of Figure 3 are intending to demonstrate the inverted V shape of rice, it should be clearly stated in the text. Besides, previous studies have proposed the “V-shape” curve of rice in Sentinel-1 VH curve. What is the difference between your discoveries and the previous studies? Corresponding discussions should be briefly illustrated, and the reference paper should be added. Also, the differences of early rice and late rice are not demonstrated.
6) What does the “two vector range” refer to? Does “the intersection of the two vector ranges” mean that a sample should satisfy two equations simultaneously to be labeled as rice? Since previous contents still lack basic information about the samples utilized for analysis, Figure 3 does not provide enough support for the rules defined in equations (1) and (2). Besides, the “key phenological period” given in Table 1 lacks a clear definition or adequate descriptions. Corresponding rice sample curves should be given to make Table 1 convincing.
7) The manuscript lacks the details of building the rice recognition model. Exactly how much training / testing sample patches are utilized in the U-net model? What are the input features? Did you train one model and apply it to all years, or train an independent model for each year / each season?
8) The relative area errors for early rice were 2-3 times higher than late rice for all years. Maybe it indicated some systemic bias implied in the overall methodology. In addition, the statistics of early and late rice were similar in 2017-2018 and in 2020-2021. However, in figure 4 some distinct differences can be noticed. Please give some closer comparisons about the results (with some zoom-in images, maybe corresponding to the discussions in line 389-391), as well as some deeper analysis about the discrepancies and possible reasons.
9) Line 319-330, since section 2 lacks descriptions about the processing strategies of the training samples, these contents seem irrelevant to the paper.
10) The discussion part is too long. Many information are redundant in the two subsections, such as repeated statements of local climate, irrigation, cultivation patterns of watermelon, and so on.
11) Figure 5 is confusing. How did you calculate single and double rice areas? I suppose that double rice area should be the places that be labeled as early and late rice simultaneously, whereas single rice should be the areas that only be labeled as early rice or late rice. However, the values of Figure 5 does not seem to fulfill these conditions.
12) What is the data source of Figure 6?
13) The location of Figure 8-12 should be denoted in Figure 7. Besides, in Figures 8-11, plot polygons are in different colors, what are the meaning of each color? Rice extraction results should be overlayed on these patches to see whether the non-rice structures are accurately excluded by the U-net model.
Comments on the Quality of English Language
Check the writing errors of the whole paper. For instance, line 129, the capitalized writing of “double-bounce”. Line 145-147, the sentence is not finished (grammar error). Line 287, where is the result of SVM? Line 292-293 is also confusing “Because other crops are also in the growth period of early rice…”
Author Response
- The introduction part does not provide with enough reviews on previous researches. It does not reveal the novelty of the research.
A:Since our manuscript is about rice identification by using the method of rapid production of rice labels for key phenological periods of rice in South China, according to the opinions of reviewers, we add the current research progress of the phenology-based paddy rice mapping in the introduction and add some theoretical basis.
- Line 83-84 (and line 290-291). You mentioned 3 seasons of rice in previous sentence but only give 2 seasons’ phenology periods.
A:According to field investigation and official statistics, Leizhou Peninsula is the largest double-cropping rice growing area in Guangdong Province, mainly planting double-cropping rice, so only the planting distribution of double-cropping rice is analyzed.
- Please reconsider the contextual logics of the manuscript. For instance, the first paragraph of 2.3.1 is poorly organized and redundant.
A:Part of 2.3.1 has been deleted according to the research content and the opinions of reviewers.
- The definitions of RICE and rice in Figure 3 was verbose and ambiguous. Besides, Line 104 mentioned that 98 ground samples were collected, how many of them are rice? And what are the differences of rice-1 and rice-2?
A:We supplemented the sample information. According to the rationality of sample distribution, the total number of samples available was 72, including 21 samples of early rice, 12 samples of non-early rice, 20 samples of late rice and 19 samples of non-late rice. We have added information to the article to explain the meaning of rice, RICe-1, RICe-2.
- Line 154-170, is poorly organized. You didn’t clarify which polarization is demonstrated in Figure 3. If the zoom-in figures of Figure 3 are intending to demonstrate the inverted V shape of rice, it should be clearly stated in the text. Besides, previous studies have proposed the “V-shape” curve of rice in Sentinel-1 VH curve. What is the difference between your discoveries and the previous studies? Corresponding discussions should be briefly illustrated, and the reference paper should be added. Also, the differences of early rice and late rice are not demonstrated.
A:We supplemented the relevant information.
- What does the “two vector range” refer to? Does “the intersection of the two vector ranges” mean that a sample should satisfy two equations simultaneously to be labeled as rice? Since previous contents still lack basic information about the samples utilized for analysis, Figure 3 does not provide enough support for the rules defined in equations (1) and (2). Besides, the “key phenological period” given in Table 1 lacks a clear definition or adequate descriptions. Corresponding rice sample curves should be given to make Table 1 convincing.
A:We supplemented the relevant information.
- The manuscript lacks the details of building the rice recognition model. Exactly how much training / testing sample patches are utilized in the U-net model? What are the input features? Did you train one model and apply it to all years, or train an independent model for each year / each season?
A:We supplemented the relevant information in the manuscript.
- The relative area errors for early rice were 2-3 times higher than late rice for all years. Maybe it indicated some systemic bias implied in the overall methodology. In addition, the statistics of early and late rice were similar in 2017-2018 and in 2020-2021. However, in figure 4 some distinct differences can be noticed. Please give some closer comparisons about the results (with some zoom-in images, maybe corresponding to the discussions in line 389-391), as well as some deeper analysis about the discrepancies and possible reasons.
A:Affected by heat accumulation and other factors, the transplanting period of late rice is short, and the monitoring time tends to remain unchanged, while the transplanting and heading period of early rice are more affected by climatic factors and meteorological disasters, which is the possible reason why the identification accuracy of our analysis of late rice is higher than that of early rice. We also use Figure 7-12 for a more detailed analysis.
- Line 319-330, since section 2 lacks descriptions about the processing strategies of the training samples, these contents seem irrelevant to the paper.
A:The phenological information of rice is very important. Although the hydrothermal condition of Leizhou Peninsula is good, it is greatly affected by meteorological disasters, and Sentinel-1 image will be affected by uncontrollable noise in Leizhou Peninsula, so we believe that these factors should be taken into account.
- The discussion part is too long. Many information are redundant in the two subsections, such as repeated statements of local climate, irrigation, cultivation patterns of watermelon, and so on.
A:Leizhou Peninsula is the largest commercial grain planting base and double-cropping rice growing area in Guangdong Province. However, due to the influence of crop prices, characteristic agriculture, climate, agricultural planting structure and other factors, the planting area of rice and other grain crops has changed greatly. Therefore, analysis of the above factors is conducive to monitoring the "non-grain" tendency of cultivated land in traditional grain producing areas and the protection of cultivated land resources.
- Figure 5 is confusing. How did you calculate single and double rice areas? I suppose that double rice area should be the places that be labeled as early and late rice simultaneously, whereas single rice should be the areas that only be labeled as early rice or late rice.However, the values of Figure 5 does not seem to fulfill these conditions.
A:We have supplemented the relevant information in the manuscript.
- What is the data source of Figure 6? The location of Figure 8-12 should be denoted in Figure 7. Besides, in Figures 8-11, plot polygons are in different colors, what are the meaning of each color? Rice extraction results should be overlayed on these patches to see whether the non-rice structures are accurately excluded by the U-net model.
A:We have supplemented the relevant information in the manuscript.
Reviewer 2 Report
Comments and Suggestions for AuthorsOverall, the paper would benefit from an extensive review of the technical component employed in the study and a better description of the results obtained. I would not recommend the paper for publication until the above concerns are addressed. See more details in my attached report
Comments for author File: Comments.pdf
Extensive English editing is required
Author Response
- Line 98-100 – Please provide relevant literature supporting the pre-processing steps in handling the Sentinel-1 data.
A:Relevant information has been supplemented based on the research content and reviewer’s suggestion.
- Please include in Section 2.2 the source of your study's radar, optical, and DEM datasets.
A:Relevant information has been supplemented based on the research content and reviewer’s suggestion.
- The authors failed to discuss the sources and collection mode for data validation/training data to develop the deep learning U-Net model. Please provide more information.
A:Relevant information has been supplemented based on the research content and reviewer’s suggestion.
- In Figure 2, the authors state that simultaneous high-resolution optical imagery is used. Is this different from the Sentinel-2 imagery (which is medium resolution)?Please verify and ensure the exact datasets used in the study are reported in the manuscript and workflow.
A: Relevant information has been checked and supplemented in the manuscript based on the research content and reviewer’ s suggestion.
- Change the title of Figure 2 to a more descriptive sentence that tells readers what it represents.
A: Relevant information has been checked and supplemented in the manuscript based on the research content and reviewer’ s suggestion.
- Line 140 and Figure 3 – The authors provide change curves for identified crop samples. Can the authors please distinguish between the classes “RICE”, “rice-1”, and “rice-2”? This is not clearly stated and could be confusing to readers.
A: Relevant information has been checked and supplemented in the manuscript based on the research content and reviewer’ s suggestion.
- Lines 182 – 190—The authors refer to image segmentation as a step in creating the training labels used for the U-Net modelling. Please provide more details on this approach and cite more relevant studies demonstrating the employed method. Also, examples of the image segmentation outputs in the manuscript will be shown. This information could be provided in the main manuscript or supplementary information.
A: Relevant information has been checked and supplemented based on the research content and reviewer’ s suggestion.
- Lines 195-199 are not clear and need to be reviewed.
A: Relevant information has been checked and supplemented in the manuscript based on the research content and reviewer’ s suggestion.
- Table 1: For the key phenological period, please provide date ranges or exact dates, such as March to May 2017.
A: Relevant information has been checked and supplemented based on the research content and reviewer’ s suggestion.
- Lines 209-215: The authors provide the fine-tuning parameters of the U-Net model used in the study. In addition, please provide details of the training and validation labels used in the study. Additional information should include the number of label chips per class generated, dimension sizes, etc.
A: Relevant information has been checked and supplemented in the manuscript based on the research content and reviewer’ s suggestion.
- Section 2.3.3. discusses the spatio-temporal change analysis component of the study. This section must be elaborated and expanded to include more details on how the change detection analysis was performed. This is not clearly stated in the current version of the manuscript.Security Classification: Protected A
A: Relevant information has been checked and supplemented based on the research content and reviewer’ s suggestion.
- Line 287: The authors mention “SVM” – support vector machine for the first time in the paper. Was this machine learning algorithm employed in this study? If so, please include the results and description of the model development in the paper.
A:Relevant information has been deleted based on the research content and reviewer’ s suggestion.
- Most of the maps presented in the paper lack legends; please revise accordingly.
A:Relevant information has been checked based on the research content and reviewer’ s suggestion.
- In all the results presented, none of the maps shows the change detection analysis performed in the study.
A:In our study, we have shown the content of change detection analysis in the Figure7-12.
- The entire references section needs to be reviewed.
A: Relevant information has been supplemented based on the content and reviewer’ s suggestion.
- More relevant references need to be cited in the paper.
A: Relevant information has been supplemented based on the content and reviewer’ s suggestion.
Reviewer 3 Report
Comments and Suggestions for AuthorsThis study used the threshold segmentation method to identify rice labels in SAR intensive time series images and extracted the distribution range of early and late rice in Leizhou city from 2017 to 2021 using a U-Net model. Some details of the article can be further modified and supplemented. Please see the following comments:
1. The content arrangement is somewhat disorganized, particularly in the results and discussion sections. Given the article's title, the analysis of spatiotemporal changes is a significant part of the research, but this is not reflected in the results section; instead, it appears in the discussion. It is recommended to include the spatiotemporal change content in the results section, and the analysis of related change factors can be placed in the discussion section. Moreover, section 3.3 does not seem like results, but rather a discussion of the research methods used.
2. The quality of rice labels is a guarantee for the accuracy of deep learning classification. This study rapidly obtained labels for key growth stages using the threshold method. However, the quality of the labels is currently described qualitatively, without a quantitative evaluation or a spatial display. It is recommended to supplement and improve this part of the content.
3. Suggest making the layout of the figures more aesthetically pleasing and rational. For instance, multiple separate small images, as shown in Figure 4, can be integrated using Geographic Information System software like ArcGIS, or graphic design software such as Adobe Photoshop or Illustrator.
4. Line 25 Change the initial letter of Rice to lower case.
Author Response
- The content arrangement is somewhat disorganized, particularly in the results and discussion sections. Given the article's title, the analysis of spatiotemporal changes is a significant part of the research, but this is not reflected in the results section; instead, it appears in the discussion. It is recommended to include the spatiotemporal change content in the results section, and the analysis of related change factors can be placed in the discussion section. Moreover, section 3.3 does not seem like results, but rather a discussion of the research methods used.
A:Relevant information has been supplemented based on the research content and reviewer’s suggestion. Figure7-12 were shown the spatiotemporal changes details in Leizhou.
- The quality of rice labels is a guarantee for the accuracy of deep learning classification. This study rapidly obtained labels for key growth stages using the threshold method. However, the quality of the labels is currently described qualitatively, without a quantitative evaluation or a spatial display. It is recommended to supplement and improve this part of the content.
A:According to the reviewers’ suggestions, the process of making rice labels has been described in detail. Due to the fine fragmentation of rice labels, it is necessary to optimize and merge the label vectors through Sentinel-2 images at the same time to improve the accuracy of rice labels. It is true that there is no quantitative analysis of label quality in the manuscript. I am so sorry that I need to learn more to solve this problem.
- Suggest making the layout of the figures more aesthetically pleasing and rational. For instance, multiple separate small images, as shown in Figure 4, can be integrated using Geographic Information System software like ArcGIS, or graphic design software such as Adobe Photoshop or Illustrator.
A: Because the classification results are relatively fragmented and require a lot of post-classification processing to display, the results are presented in detail in the manuscript mainly through the cities and towns attached to the main rice producing area of Leizhou city, as shown in Figure 7-12.
- Line 25 Change the initial letter of Rice to lower case.
A:I am so sorry that I couldn’t find this mistake.
Author Response File: Author Response.docx
Reviewer 4 Report
Comments and Suggestions for AuthorsIn this paper, the identification and spatio-temporal changes of rice in Leizhou City, Guangdong Province, China is studied based on Sentinel-1 time series. The topic is interesting and within the scope of RS. There are some issues should be addressed.
1. According to the paper's introduction, the main innovation and objective are the creation of the samples. However, this aspect occupies only a small portion of the paper and lacks a detailed description. The presentation of results in this area should be strengthened.
2. The input of the UNet model, as well as implementation details, are not provided in the description, they should be given.
3. Some parts of the article are difficult to understand, such as the terms "RICE," "rice-1," and "rice-2" in Figure 3. Similarly, terms like "9-10 seasons" in Figure 8 are also unclear.
4. How do you get the polygons in Figures 8-12.
Author Response
- According to the paper's introduction, the main innovation and objective are the creation of the samples. However, this aspect occupies only a small portion of the paper and lacks a detailed description. The presentation of results in this area should be strengthened.
A:According to the opinions of the reviewers, the introduction of the phenological method of rice mapping is added to the related literature. In our manuscript, we supplemented the process of rapid production of rice labels based on threshold method, and optimized the rice labels (with reference to Sentinel-2 images of the same period) to improve the accuracy of rice labels.
- The input of the U-Net model, as well as implementation details, are not provided in the description, they should be given.
A: Relevant information has been supplemented based on the research content and reviewer’s suggestion.
- Some parts of the article are difficult to understand, such as the terms "RICE," "rice-1," and "rice-2" in Figure 3. Similarly, terms like "9-10 seasons" in Figure 8 are also unclear.
A: Relevant information has been supplemented based on the research content and reviewer’s suggestion.
4) How do you get the polygons in Figures 8-12.
A:In our manuscript, we mapped the paddy rice distribution in Leizhou from 2017 to 2021, so the distribution results of a total of 10 seasons of rice were obtained. The distribution of rice in the attached towns of Leizhou was calculated by summing up, and Figure 7 was finally obtained. Then, according to the planting times, typical rice fields were selected for specific display, and Figure 8-12 was finally obtained.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
Comments and Suggestions for Authors1. Line 60-62: “To date, the majority of rice mapping approaches based on SAR images have relied on prior knowledge (such as planting dates) and empirical thresholds specific to certain regions, which has restricted their application over large spatial scales. ”
I can agree with this statement. Plenty of references can be found in recent years regarding SAR-based rice mapping with machine learning or deep learning models. The authors should investigate more previous researches.
2. Line 130-132:It is recommended that the authors include photos of different types of crop samples.
3. Line 212, “second stage” and “first stage” are not properly defined. “…and the second stage was determined respectively, in which the first stage of rice backscattering coefficient continued to decline and continue to rise” The meaning of this sentence is not clear.
4. Line 260 : “while table 2 displays the distribution of the validation samples for multi year cotton cropping pattern classification.” Is “Cotton” correct here?
5. The authors did not add enough information about the training / testing sample, specifically, regarding its volume, size, the applied data enhancement strategies, et al. The modifications in line 245-247 just didn’t answer my previous questions.
6. The enlarged image displayed in Figure 9 seems odd. Please check the file. Besides, from figure 8 to figure 12, the authors still did not present any actual rice extraction result.
Comments on the Quality of English Language
The manuscript has many writing problems. For instance, Line 71, “ndvi, evi and lswi”, these professional items should be capitalized. Line 76, PPM has not shown in previous context, so it should be spelled out. Line 65-70 describes how SAR is used to extract rice, whereas following contents are more about other sensors. The context is not smoothly connected or transitioned. Line167-168:”In order to reflect the difference of different types of crops, the smooth curve of all kinds of crops. ”The sentence is not complete. Besides, in figure 3, you should label clearly which ones are the smoothed curve. Line 342-343, “However, the resolution of SAR data after pretreatment is about 17.96m*17.96m, which is relatively lower than most of optical images.” Usually the word “preprocess” is used instead of “pretreatment”.
Author Response
1)Line 60-62: “To date, the majority of rice mapping approaches based on SAR images have relied on prior knowledge (such as planting dates) and empirical thresholds specific to certain regions, which has restricted their application over large spatial scales. ”
I can agree with this statement. Plenty of references can be found in recent years regarding SAR-based rice mapping with machine learning or deep learning models. The authors should investigate more previous researches.
A: Thank you for your constructive comments. We added detailed information at the end of the second paragraph.
- Line 130-132:It is recommended that the authors include photos of different types of crop samples.
A: Thank you for your constructive comments. We supplemented the Figure 3 in our manuscript.
- Line 212, “second stage” and “first stage” are not properly defined. “…and the second stage was determined respectively, in which the first stage of rice backscattering coefficient continued to decline and continue to rise” The meaning of this sentence is not clear.
A: Thank you for your constructive comments. We added detailed information in Line 219-227.
- Line 260 : “while table 2 displays the distribution of the validation samples for multi year cotton cropping pattern classification.” Is “Cotton” correct here?
A:Thank you for your constructive comments.“Cotton”is not correct here and we have delete this word.
- The authors did not add enough information about the training / testing sample, specifically, regarding its volume, size, the applied data enhancement strategies, et al. The modifications in line 245-247 just didn’t answer my previous questions.
A: Thank you for your constructive comments. We added detailed information in Line 258-262.
- The enlarged image displayed in Figure 9 seems odd. Please check the file. Besides, from figure 8 to figure 12, the authors still did not present any actual rice extraction result.
A: Thank you for your constructive comments. The vector outlined by the red line represents the extent of the area where double cropping rice was planted from 2017 to 2021. Combined with the rice identification results of Fucheng from 2017 to 2021, the results of rice planting in 5 years were superimposed to obtain the current Figure 9 (original Figure 8),therefore, the enlarged images(Figure 9-13)displayed the detail information about actual rice extraction result. Meanwhile, due to the large number of images, we are also considering whether it is necessary to display the recognition results for a total of five years from 2017 to 2021.
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThe authors have addressed my concerns. I recommend this paper for publication.
Author Response
Thank you very much.
Reviewer 3 Report
Comments and Suggestions for AuthorsThe author gave a good response to the review comments and recommended accepting this article
Author Response
Thank you very much.
Reviewer 4 Report
Comments and Suggestions for AuthorsThe authors have addressed my questions.
Author Response
Thank you very much.
Round 3
Reviewer 1 Report
Comments and Suggestions for AuthorsThe manuscript has been improved with details. I have no further comments.
Comments on the Quality of English LanguageHowever, writing problems are still frequent in the contents. For instance, 'ndvi, evi and lswi', these items are usually capitalized, whereas 'Double-bounce' (line 67) should not be capitalized. Clearly, the authors didn't correct all the writing problems I pointed out last time. The whole paper should be carefully checked.