Early Crop Mapping Using Dynamic Ecoregion Clustering: A USA-Wide Study

Round 1

Reviewer 1 Report

The paper presents an innovative approach to enhancing early crop mapping accuracy using Dynamic Ecoregion Clustering. While the paper's ideas are novel and promising, there are several issues and questions that need to be addressed before it can meet the requirements for publication.

1.1. L52-71: More references are needed to support the statements made in this section.

1.2. Section 3.1: Please provide a detailed list of the metrics used and their data sources.

1.3. Basis for Metric Selection: What criteria were used to select these metrics? Were collinearity tests and filtering performed, as the choice and potential redundancy of metrics can directly impact clustering results?

1.4. L235-246: How were the training samples obtained, processed, and labeled? Please provide a detailed description.

1.5. L250: Conduct the necessary significance tests.

1.6. L329: In which part of the paper is it specifically mentioned, "Additionally, we prove that the mapping result with ten ecoregion clusters has the highest accuracy in the following experiments"?

2. Questions:

2.1. Section 3.2: Ecoregion clustering was trained using data from 2013 only. Is there a possibility of bias when using only 2013 data to predict ecoregions for the years 2013-2022? Would using multi-year data, such as 2013-2015, provide greater stability?

2.2. L235-240: The paper assumes that "soil, elevation, and slope conditions remain stable while climate conditions vary over time." Would it be more reasonable to use hierarchical clustering, with soil, elevation, and slope conditions as the first layer, and climate conditions built upon this foundation?

3. Minor Issues and Suggestions

3.1. L17-18: Is the use of the word "significantly" appropriate here?

3.2. Figure 4: It is recommended to include error bars.

Author Response

Dear Reviewer,

I would like to express my sincere gratitude for your valuable suggestions and the time you dedicated to reviewing my work. I have carefully considered and incorporated your feedback into my research. Below, I will address each of your comments and suggestions individually:

1.1 I put more references here (L52-62)

1.2 You could find the new metrics list between L206-207

1.3 This is a good one. In our work, we use PCA to reduce the dimensions of the data. But I missed this pre-processing in the paper. Now I write it in L259.

1.4 I add the details in L252-263.

1.5 Yes! I totally missed the significant test here. Now I add it in 288-304. And you can also find the MMD values table under L304.

2.1 Also a good one. In the initial stages of our research, we attempted to incorporate multi-year climate data into the training process. However, we encountered a challenge where, for instance, in one training year, only 498 or 499 ecoregions might exist out of the 500 originally designated. This situation posed a risk of the system being unable to establish correspondences between ecoregions across different years for the extraction of vegetation index (VIs) data. Therefore, in order to ensure the robustness and reliability of our system, we opted to utilize environment data from the year 2013 exclusively for training purposes. 

2.2 For this question, we add some discussion in L241-251. The hierarchical clustering should be a better choice! However, for our specific scenario, we prefer to adopt a simpler clustering method in order to let the system facilitate a smoother transition when more comprehensive soil and topography data becomes available.

3.1 We deleted the word.

3.2 Certainly! However, we believe that visualizing ten lines with error bars could potentially complicate the visualization for understanding. So we add the MMD based on your suggestion for the significant test to explain the VIs data distribution difference between different ecoregions.

I hope I answered all your questions. And thanks for the suggestions again!

Best regards,

Yiqun

Author Response File: Author Response.pdf

Reviewer 2 Report

Article ID : remotesensing-2616556-peer-review-v1

Article Name : Early Crop Mapping Using Dynamic Ecoregion Clustering: A USA-wide Study

This paper presented the following contributions

l It propose a novel approach for mapping target crops (soybean and corn) earlier than

the harvest period in the USA using time-series NDVI and EVI data with a dynamic

ecoregion clustering method.

l It  determines the number of ecoregions using the elbow method and demonstrate significantly higher mapping accuracy using the dynamic clustering method compared to the static clustering method.

Therefore, it is interesting and attractive. However, it should be major revised to enhance the quality, as follows:

1) In Section 1, authors should make three sub sections, motivation, contributions and organization of the paper

2) Related work is not upto the mark. Pl provide one table for overall idea about the current trends on the particular problem.

3) Contributions of the research paper is limited for this platform. Pl enhance it in to atleast 3 vital contributions which will reflect on your research paper

4) Figure 1 is not clearly derived about the proposed framework. Pl explain in details

5) Eq. 2,3 and eq 4 are not presented well. Pl elaborate .

6) Figure 3 is not properly derived. Pl represents in clear form.

7) Avoid too many subsections which is creating confusion the flow of the paper

8) In conclusion section, pl add future scope of the present research.

Author Response

Dear Reviewer,

I would like to express my sincere gratitude for your valuable suggestions and the time you dedicated to reviewing my work. I have carefully considered and incorporated your feedback into my research. Below, I will address each of your comments and suggestions individually:

1) I reconstructed the Introduction Section, in L87-100 I wrote the motivation. The contributions are located in L101-110. And the organization is in L111-114

2) Thanks for this suggestion! It makes the related works more clear. You can find the table under L115.

3) The new on is in L101-110

4) I added more details in L220-236, L252-263, and L332-345.

5) I added the definitions of TP, TN, FP and FN in L362-365.

6) Now you may find the correct forms.

7) I have removed five subsections and replaced them with bold text in L332,343,358,375,and 394

8) I added the future work in L540-546.

Once again, I want to express my gratitude for your valuable suggestions. They have significantly enhanced the clarity and comprehensibility of our paper.

Best regards,

Yiqun

Author Response File: Author Response.pdf

Reviewer 3 Report

The article is well written and describes an attempted classification of soybeans and corn across the entire US.
To do this the authors first divided the area into 10 ecoregions with an unsupervised classification algorithm using environmental and climate data. The ecoregions changed over the years as the climate changed. Once the annual ecoregions were defined, a classification of corn and soybeans was made using MODIS satellite data and a supervised classification algorithm. The latter was trained and validated thanks to a crop map of the entire USA.

The article has some minor issues and major issues

MINOR  ISSUES

In the "Introduction", close to line 82 you must explain or mention what CONUS is

Since your work is focused in eraly mapping (mapping before harvesting) you must explain when corn and soybean are harvested in USA

From "3.4.1 Model training ...." I didn't understand what "training image" is.

MAJOR ISSUES

The NDVI and EVI are very similar indeces, it is confirmed also by the time trends in figure 4. It would be great to repeat the classification with RF with only one index (with only NDVI and then with only EVI) and comparing the results with those in table 1. I think you would obtain very good results using only NDVI

Please use also other methods for the determination of the optimal number of cluster (i.e. silohuette and GAP methods) and compare it, since the elbow method alone is quiet misleading. From figure 7 I would rather say that the ideal number of clusters is 20 and not 10

Please describe what are the benefits of having a US corn and soybean crop map at 250m versus having the same (similar) map at 30m later (I mean the CDL)? How far in advance is your map produced compared to USDA's CDL supply? What are the benefits of your map compared to CDL's?

Table 1 also shows the accuracy obtained without dividing the territory into ecoregions

 

Author Response

Dear Reviewer,

I would like to express my sincere gratitude for your valuable suggestions and the time you dedicated to reviewing my work. I have carefully considered and incorporated your feedback into my research. Below, I will address each of your comments and suggestions individually:

MINOR  ISSUES:

1) You could find the definition of the CONUS in L3 in the abstract.

2) That's a great suggestion! I added this information in L230-233.

3) Sorry for the confusion. I explained it in more detail in L332-345.

MAJOR ISSUES:

1) That's an interesting one. So we do an experiment only using NDVI. You can find the experiment metrics table above L410. And we add some discussion in L428-439. It makes our paper more comprehensive.

2）So important! Thanks for your suggestion. We add the silhouette method in L379-385. You can find Figure 8 about the silhouette scores above L376.

3) We added it in L510-522.

4) We conducted experiments using a single cluster (the normal case without the ecoregion method), and the results are presented in table under Line 409.

Once again, I want to express my gratitude for your valuable suggestions. They have significantly enhanced the clarity and comprehensibility of our paper.

Best regards,

Yiqun

Author Response File: Author Response.pdf

Reviewer 4 Report

This manuscript proposed a dynamic ecoregion clustering approach to map crop types (soybean and corn) in the entire conterminous USA land area. In general, this manuscript is logically organized and would be of interest to the community and the readership of Remote Sensing. There are, however, some issues that need to be resolved before this manuscript is acceptable for publication. Additionally, some revisions should be made to improve the readability of this paper. Please, find my detailed comments below and revise them accordingly.

Comments:  

1. The satellite data used in this study have different spatial resolutions, e.g., land cover dataset at 30 m resolution, soil data at 250 m resolution, climate data at 11132 m resolution, elevation data at 231.92 m resolution and MODIS EVI and NDVI at 250 m resolution. How do you process those data sources with different spatial resolutions? Please clarify it.

2. How do you define the early crop mapping in this study? I only read that you retrieved the VIs data from April 1^st to mid-July, why this period was selected for crop type classification? What are the phenophases of soybean and corn in the study area?

3. The sub-title (a) and (b) in Figure 4 should be presented.

4. The sub-title (b) in Figure 6 should be corrected. Besides, what do the x-axis and y-axis mean? I could not understand it.

5. Why do you validate soybean and corn separately? I think it would be better to combine them for analysis.

6. Figure 9. The overall accuracy in Figure 9 (a) lower than 0.5 should be explained. Does the green color in Figure 9 (b) highlight the regions where the pixel-wise overall accuracy from the dynamic method outperforms the static method? It seem that the pixels in white color are more than that in green color.

7. The discussion part need to be extended based on your results and previous studies. Besides, the limitations of this study should be pointed out.

None.

Author Response

Dear Reviewer,

I would like to express my sincere gratitude for your valuable suggestions and the time you dedicated to reviewing my work. I have carefully considered and incorporated your feedback into my research. Below, I will address each of your comments and suggestions individually:

1. We added the pre-process details in L252-263
2. This is a really important one. We added the information in L209-213.
3. Yes! However, it is one figure with two subfigures. So we add the information above L285.
4. Sorry for the confusion. We added details above 312. The x- and y-axes represent the features following dimension reduction using PCA.
5. Thank you for your valuable suggestion. In our study, we trained our ecoregion clustering models specifically tailored to the target crop area. This means that distinct ecoregion clustering models were developed for each target crop. It is an interesting point you've raised about the potential benefits of jointly training these models. This aspect could indeed become a noteworthy focus for our future research. We will explore whether combining the ecoregion clustering models improves the accuracy of early crop mapping results, which presents an intriguing avenue for further investigation.
6. (a) Yes! This is an interesting point. We wrote the discussion about this question in L480-490. (b) Thanks for the question. The white color on the chart shows also the region where the static method works just as well as the dynamic one. This happens because the white part represents all the farmland in the USA, and a big part of it isn't corn or soybeans. Both methods do equally well in this area. However, we want to point out that our main focus in this study is to prove that the dynamic method is better than the static one in places where the types of regions change from year to year.
7.  We added more discussions in L480-490 and L510-522.

Once again, I want to express my gratitude for your valuable suggestions. They have significantly enhanced the clarity and comprehensibility of our paper.

Best regards,

Yiqun

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Authors are addressed all the queries related to the present manuscript. Now , it may considered for the publications in this journal.

Author Response

Dear reviewer,

Thanks a lot for your suggestions and time again.

Best regards,

Yiqun

Reviewer 4 Report

This manuscript proposed a dynamic ecoregion clustering approach to map crop types (soybean and corn) in the entire conterminous USA land area. In general, this manuscript is logically organized and would be of interest to the community and the readership of Remote Sensing. However, some revisions have not been done in the first round because of my late feadback. There are, however, some issues that need to be resolved before this manuscript is acceptable for publication. Additionally, some revisions should be made to improve the readability of this paper. Please, find my detailed comments below and revise them accordingly.

Comments:  

1. The satellite data used in this study have different spatial resolutions, e.g., land cover dataset at 30 m resolution, soil data at 250 m resolution, climate data at 11132 m resolution, elevation data at 231.92 m resolution and MODIS EVI and NDVI at 250 m resolution. How do you process those data sources with different spatial resolutions? Please clarify it.

2. How do you define the early crop mapping in this study? I only read that you retrieved the VIs data from April 1^st to mid-July, why this period was selected for crop type classification? What are the phenophases of soybean and corn in the study area?

3. The sub-title (a) and (b) in Figure 4 should be presented.

4. The sub-title (b) in Figure 6 should be corrected. Besides, what do the x-axis and y-axis mean? I could not understand it.

5. Why do you validate soybean and corn separately? I think it would be better to combine them for analysis.

6. Figure 9. The overall accuracy in Figure 9 (a) lower than 0.5 should be explained. Does the green color in Figure 9 (b) highlight the regions where the pixel-wise overall accuracy from the dynamic method outperforms the static method? It seem that the pixels in white color are more than that in green color.

None.

Author Response

Dear Reviewer,

I think I have already carefully considered and incorporated your feedback into my research during the first review. Below, I will address each of your comments and suggestions individually again:

1. We added the pre-process details in L252-263
2. This is a really important one. We added the information in L209-213.
3. Yes! However, it is one figure with two subfigures. So we add the information above L285.
4. Sorry for the confusion. We added details above 312. The x- and y-axes represent the features following dimension reduction using PCA.
5. Thank you for your valuable suggestion. In our study, we trained our ecoregion clustering models specifically tailored to the target crop area. This means that distinct ecoregion clustering models were developed for each target crop. It is an interesting point you've raised about the potential benefits of jointly training these models. This aspect could indeed become a noteworthy focus for our future research. We will explore whether combining the ecoregion clustering models improves the accuracy of early crop mapping results, which presents an intriguing avenue for further investigation.
6. (a) Yes! This is an interesting point. We wrote the discussion about this question in L480-490. (b) Thanks for the question. The white color on the chart shows also the region where the static method works just as well as the dynamic one. This happens because the white part represents all the farmland in the USA, and a big part of it isn't corn or soybeans. Both methods do equally well in this area. However, we want to point out that our main focus in this study is to prove that the dynamic method is better than the static one in places where the types of regions change from year to year.

Once again, I want to express my gratitude for your valuable suggestions. They have significantly enhanced the clarity and comprehensibility of our paper.

Best regards,

Yiqun

Author Response File: Author Response.pdf

Round 3

Reviewer 4 Report

Authors have responded all my comments and relieved most concerns, which contribute to the advancement of the science in the field of expertise.