A New Multiple Phenological Spectral Feature for Mapping Winter Wheat

Round 1

Reviewer 1 Report (Previous Reviewer 1)

This manuscript has been well revised and there are no major problems on the whole.Small suggestions:

(1) Figure 4 can only show the effect of SG filtering, which is not significant to this article and can be removed;

(2) It is suggested that the author should adjust the parameters of SG filter. At present, the filtering effect is very poor;

(3) The analysis in section 2.3.2 is too general. It is suggested to analyze specifically in Fig. 5 to explain the characteristics of wheat different from other vegetation types and why the method you proposed can maximize the difference.

 

Author Response

Thanks again for your time and excellent comments to help us improve the manuscript sincerely. 

Please download the attachment file to view the detailed.

Author Response File: Author Response.docx

Reviewer 2 Report (Previous Reviewer 3)

Thanks for resubmitting your article to Remote Sensing Journal. Please see the comment below for your minor revision.

(1) Authors are using SVM classifcation adopted in Google Earth Engine, and Google Earth Engine is using LIBSVM for SVM classification. Therefore, it is more appropriate to cite the original paper as suggested by LIBSVM developers (https://www.csie.ntu.edu.tw/~cjlin/libsvm/faq.html#f203). The parameters used for one class SVM are well-described in the original LIBSVM implementation paper. I suggest authors figure out which parameters were used for your one-class SVM classification and their values. In the manuscript, authors mentioned they used default values for "other" parameters. Authors should clarify what is the other parameters and their values used, which could have been found from Google Earth Engine API.

(2) In scientific paper writing, it is generally more appropriate to cite a journal paper than a web page if applicable. Please cite the paper titled "Google Earth Engine: Planetary-scale geospatial analysis for everyone" published in Remote Sensing of Environment on 2017.

(3) I very much appreciate your effort you put into spectral separability evaluation section. Authors made a very good description of the classes that exists in their study area, and they clearly showed that garlic class has a very similar pattern of vegetation indices to winter wheat class. It makes sense to select the two classes (winter wheat and garlic) to test spectral separability, and the J-M distance clearly shows the separability was maximized when they used the features in the entire phenological periods of winter wheat. This result implies that feature(band) selection is not required for your research task. I think you can add your findings from the separability test to discussion section, maybe section 4.1.

(4) Authors showed the possibility of separating winter wheat and garlic with J-M distance. However, they did not demonstrated the performance of their classification method between the two classes. I suggest authors do one more experiment of classifying winter wheat and garlic classes. Since authors already achieved a decent overall performance, a low classification accuracy between winter wheat and garlic will not harm the legitimacy of the entire study at all. In case the classification accuracy turns out good, the result might be rather used to show off the performance their method in a challenging task.

 
(5) How about changing representation of the phenological period combinations in Table 3 to P₁(where 1 is subscribed), P₂, P_1,2, P_1,2,3 and so on? You can use your own way of representation if you have a better idea.

(6) The areas of ROIs were not found in the manuscript. Could you check if your uploaded version has the ROI areas?

Thank you again for your effort and keep up the good work!

Author Response

Thanks again for your time and excellent comments to help us improve the manuscript sincerely.

Please download the attachment file to view the detailed.

Author Response File: Author Response.docx

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

In this paper, a feature image was built  by synthesizing the images of multiple phenological periods, and a one-class classifier was used to identify the winter wheat. On the whole, the method is good. But there are also some problems:
(1)"phenological spectral feature". This statement is not universal, so it is easy to create ambiguity.
(2)“The existing studies have mainly used a single phenological period to map winter wheat”. This statement is inappropriate. Crop identification using multiple phenological periods is common in time series remote sensing recognition. This paper belongs to the identification method using various characteristics in the key phenological period.
(3)“To our knowledge，”. This expression  should not appear in academic papers.
(4)Line69-70. The classification and statement of image resolution are inappropriate, and it is recommended to add references.
(5)In the introduction, the problems existing in the existing methods are summarized, which should be the problem to be solved in this paper. It is suggested to clarify the logical relationship and expression.
(6)“spectral temporal profiles”“spectral profiles”“Spectral indices temporal profiles”. There are multiple expressions in the description of Figure 4, and it is suggested to use the third expression.
(7) In Fig. 4, the profiles of winter wheat? The title is unclear.
(8)Line238-255. To identify winter wheat, it is necessary to analyze the differences between wheat and other vegetation, but figure 4 has only one type of profile. Without the profiles of other vegetation, can you explain the recognition ability of this method?
(9)Line260-269. Why using median synthesis? What is the reason? How do you use your three phenological periods in median processing? The second phenological period (growth period) is a wide range. What can be reflected by the median?
(10)The overall logic of the method part is poor, and it is suggested to reorganize.
(11)Line318-340. Line347-349. They arerepeatedly expressed with section 2.2.2.
(12)Fig. 6 is unclear. What is N? What is Y? Each diagram should be self-evident and each symbol should be explained clearly.
(13)The conclusion part should clearly puts forward the innovation of this article, and it should be specific.

Author Response

Thanks for your time and excellent comments to help us improve the manuscript.

Details can be seen in the attachment file.

Author Response File: Author Response.docx

Reviewer 2 Report

This study aims to mapping winter wheat area in Beijing, China using six vegetation indices (VI) through machine learning method (One‐Class Support Vector Machine, OCSVM). Overall, I think the manuscript lacks innovation, and there are many previous studies used varies Vis for different crop type mapping, also used deep learning and machine learnig methods. Therefore, one main concern is that what’s the main difference and innovation of the manuscript compared with other researches?

Besides, some specific comments are listed below.

1.     L92-94, I think this is not appropriate, because there are lots of studies have used varies vegetation indices in crop type mapping. So, what’s the innovation of this research?

2.     How do authors get the 1023 winter wheat samples? Only visual interpreting from google earth is unconvincing, because the crop type may change over time. Some farmer may plant different crops in different year. Therefore, mapping winter wheat of three years using one set of data is flawed, and at least should be discussed.

3.     Table 2, please add the references?

4.     Figure 4, the X axis is incorrect, there are two Oct. in figure 4(a) and (b)?

5.     It is confused in figure 4(b), why the highest NDVI value exist in September?

6.     The garlic have the similar growth period with winter wheat, author also mentioned, so why doesn’t author sample some garlic point for a more accurate winter wheat mapping?

7.     From Figure 7, we found that the “Mapping results in 30m product”, the author used, may not a good reference dataset for validation. Because the quality of this dataset is not good enough, and the higher accuracy of authors’ product than this reference dataset doesn’t mean much.

Author Response

Thanks for your time and excellent comments to help us improve the manuscript.

Details can be seen in the attachment file.

Author Response File: Author Response.docx

Reviewer 3 Report

The authors suggested using a combination of vegetation indices to detect winter wheat in Beijing area by remote sensing technology. The approach suggested by the authors, along with the use of GEE, is meaningful because it can be applied to detect other types of crops while overcoming disadvantages of previous approaches which relied on limited number of observations (RS images) and vegetation indices. However, the experiment design must be improved to support the excellency of the methodology.

The drawback of this study is that one class classification can easily misused to overestimate classification accuracy. It can be true especially when the counterpart of the object class is imbalanced with a class that is distinctively different than the object class. For example, the majority of the non-wheat class was urban area, the classifying winter wheat versus non-wheat becomes a very easy task while the classification accuracy being overestimated. Therefore, I suggest authors to (1) divide the non-wheat class into an adequate number of classes, e.g., forest, urban area, water, agricultural area (but not wheat) and clarify the areal proportion of the constituent classes,  and (2) perform either multi-class classification with new classification scheme or do one class classification between winter wheat versus other classes per se.

A critical mistake was found while calculating classification accuracy in Table 3. The PA of winter wheat should be 47% and UA of non-wheat should be 82% for overall Beijing area. Authors should have also clarified the classification accuracy in all eleven districts. The rest eight districts should have a lower classification accuracy than the three districts presented, and the reason for the lower accuracy should be addressed appropriately.

Please also see the comments below.

Line 34. Please change "the spread of COVID-19" to "the supply chain disruptions due to COVID-19"

Lines 69-70. Change the order of items in the list. It seems high, medium-high, and medium seems more legitimate.

Line 88. Does the continuous extreme weather include snow or frost? Please clarify it or list its examples.

Lines 91-101. Authors should include literature review of crop detection research with multiple vegetation indices, for crops other than winter wheat.

Lines 116-117. Use en dash instead of tilde. Inconsistent use of space before minute symbol.

Table 1. Add area of ROIs. Please include data of other eight districts.

Line 216. Add what is the word for the acronym BSI.

Line 254. Does the land signal mean BSI?

Figure 4. Please clarify the source the data in Figure 4 was derived from. Was that all winter wheat ROIs, ROIs of training data, or else? It would be nice to highlight planting season, vegetative and matured season as shaded area in the background.

Line 260. Please clarify why median filter was used. You could have used other measures of central tendency such as mean.

Lines 267-269. Please clarify why use chose specific vegetation indices for each period. Authors can use figures or tables to show the difference of feature values in various land cover classes (multiple classes).

Line 284. Is SPAD a gold standard to measure of chlorophyll content or a measure generally used to roughly estimate plant growth or health? Please clarify.

Lines 307-315. Use general parameter names instead of gamma and nu. Authors did not mentioned other parameters used for SVM such as regularization parameters. Authors should also clarify scaling or standardization was done for the input variables since the average and standard deviation values are different according to vegetation indices used.

Lines 366-378. Move to discussion.

Lines 474-475. The data presented in this paper cannot support the statement. Authors should prove that the temporal changes of vegetation indices in Figure 4 is different than those of other land cover classes. Authors could also compare it with corn, garlic, etc.

Lines 486-487. This is considered a justification for conducting one-class classification.

Thank you.

Author Response

Thanks for your time and excellent comments to help us improve the manuscript.

Details can be seen in the attachment file.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

The manuscript has been well revised, but there is still one problem that I am not convinced (The 8th question).
How can you prove that the spectral separability has been maximized without comparison with other vegetation types. I still think it is not enough to just learn the spectral characteristics of winter wheat.  I hope you can explain and describe clearly in the manuscript.

Author Response

Thanks for your time and excellent comments to help us improve the manuscript again.

Details can be seen in the attachment file.

Author Response File: Author Response.docx

Reviewer 2 Report

Authors has answered all my concerns, I have no more comments by now.

Author Response

Thanks for your time and excellent comments to help us improve the manuscript again.

Reviewer 3 Report

Thanks for revising the manuscript based on the reviewers' comments. However, I have to reject your paper because of some serious flaws described below.

 

Authors were asked to clarify the general terms of the SVM and RBF kernel parameters to help the readers understand the study method and make the study reproducible. Authors should have at least clarify the value of regularization parameter. Authors should also have clarified whether a slack variable was used and its value if used. Authors could have at least add a mathematical expression of RBF kernel including "gamma" and "nu" parameter to help readers know what the "gamma" and "nu" parameters are.

Authors claims that they found the selected vegetation indices in winter wheat area showed significantly different response than other classes. However, there is no data, graph, or description that supports their claim.

Lines 404-414 should be moved to discussion section (minor).

Authors were asked to include areas of each class, but they were not able to do so (minor).

 

Author Response

Thanks for your time and excellent comments to help us improve the manuscript again.

Details can be seen in the attachment file.

Author Response File: Author Response.docx