Predicting Crop Growth Patterns with Spatial–Temporal Deep Feature Exploration for Early Mapping

Round 1

Reviewer 1 Report

The manuscript titled "A Spatial-Temporal Deep Feature Exploration Approach for Early Crop Mapping" presents a novel method that utilizes spatial-temporal satellite data and Deep Learning techniques to accurately map crop areas. Overall, the paper demonstrates a logical and well-structured approach. The content is highly detailed, presenting the methodology clearly, and the results are appropriately aligned with the methodology. The authors also acknowledge the limitations of their model and suggest future research directions.

However, several limitations need to be addressed before the paper is suitable for publication.

-  L130-131: Second purpose somehow is questionable. Authors mentioned that they proposed early crop mapping strategy but what I have seen is the longer observation yielded better results. They haven’t focused on the “strategy” as they claimed here.

-    Insufficient Explanation of Section 2.1. Section 2.1 should provide a more comprehensive explanation of each subsection depicted in Figure 1, rather than just listing them.

-     Misplacement of Content (Lines 154-165): The content found in Lines 154-165 should be appropriately placed within Section 2.2.

-  Methodology Content in Figure 2 Caption. The caption of Figure 2 contains methodology-related content ("After processing, input samples…"). Please revise the caption to describe the figure accurately.

-     The title of Section 2.3 appears to be identical to the title of Section 2.2. Please review and correct any inconsistencies.

-  L315-316: Please provide reference documentation for Google Earth Engine and S2 Level-2A data mentioned in Lines 315-316.

-  L321: Authors should provide an explanation for the "10-day median synthesis method" mentioned in Line 321.

-   The rasters used in Figure 6 should be labelled as NDVI data instead of RGB composited images. Please rectify this error.

-  L338-339: Please provide details on the number of points used, the percentage in comparison to total pixels, the random point sampling method employed, and a reference source for this information.

-   Figure 8: Without information related to the scenes, it is challenging to determine which scenes the model performs best on and which ones yield poor results. The variability in the model predictions is evident, with R-squared values ranging from 0.49 to 0.99. Please provide relevant details about the scenes

-     Appearance: The distinction between figure captions and main text should be enhanced to improve readability.

Therefore, significant revisions are necessary for this manuscript to be considered for publication in this journal.

English Language Revision. This paper requires revision for English language usage. Numerous instances of awkward English and occasional difficulty in understanding were observed. For example:

o   Figure 7’s caption: “The prediction NDVI scenes”

o   L521-522: “In the above section, we presented the proposed spatial-temporal prediction model based on remote sensing data…”. The above section (4.1) primarily presents the evaluation results obtained from the model, rather than describing the model itself. Alternatively, do the authors intend to mention the results ("prediction") at this point?

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Mapping crops at an early stage is crucial, and the authors propose the STPM model, which utilizes crop growth patterns for early mapping. The topic is both novel and intriguing.

However, there are several issues that need to be addressed:

1. The article has significant problems with citing literature. Citations are omitted in many places, such as GEE, study area overview, CDL, PyTorch, etc. It is recommended to standardize citations to distinguish between original and non-original content more effectively.

2. Many references contain incomplete information, including missing page numbers and other issues. For example, references [10], [20], [21], and [38] need to be revised.

3. On Page 1, Line 24, please provide specific values for "achieving reasonable results."

4. On Page 1, Line 26, the abstract information is incomplete. For instance, when mentioning "is analyzed in this paper," please provide a brief summary of the analysis results.

5. On Page 4, Line 149, modify "(1), (2), (3)" to "(a), (b), (c)."

6. On Page 4, Line 163, delete the text "are shown in this figure" in the title of Figure 2.

7. On Page 5, Line 174, add a reference after "Compared with commonly..."

8. On Page 5, Line 213, write "x" in italics.

9. On Page 6, Line 251, ensure that sections 2.2 and 2.3 have distinct titles.

10. On Page 6, Line 235, write "K" in italics.

11. On Page 6, Line 254, add a reference to "existing early crop mapping methods."

12. On Page 12, Line 443, avoid using "x" if it does not refer to an input sample.

13. On Page 13, Line 450, write "R" in italics and "2" in superscript. Check for this issue throughout the manuscript.

14. On Page 20, Line 636, consider changing the right axis color to black, as it is not only "Region" that refers to it.

15. On Page 22, Line 700, please provide the exact value of the improvement.

The English in the text requires modification to improve clarity and academic writing style. For example, on Page 23, Line 707, replace "such as" with "for example," as "such as" is typically followed by several nouns rather than a complete sentence. Additionally, the usage of "this" is inappropriate in some instances. On Page 14, Line 466, replace "this" with "this model." Furthermore, there is confusion regarding singular and plural forms and tenses throughout the text.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

Excellent work. Predicting future images is very challenging and the research is innovative. I recommend the revision for publication. I have the following comments:

 

Major Comments:

Both the title and the text of the manuscript emphasize early crop mapping, which confused me because the manuscript does not provide a calendar of the various crops. I cannot determine if all crops were in the early part of the growing season at the time of crop mapping. The authors need to explain the definition of early crop mapping in this study. I think early mapping in this paper refers to crop mapping using imagery from earlier in the year. But in fact, it is possible that some of the crops were harvested earlier in the year (I am not very sure). Therefore, I would also suggest that the authors provide crop calendar information in the supplementary material.

 

Minor Comments:

The authors used CDL data as the reference data for crop classification. As far as I know, the resolution of CDL is 30 m, and the resolution of S2 Level-2A is up to 10 m. How did the authors avoid the problem of spatial scale differences in sampling?

 

S2 Level-1C has more images than S2 Level-2A in many areas of the world, so why not choose S2 Level-1C?

 

What is the algorithm to remove the S2 Level-2A cloud?

 

What is the window size for SG filtering?

 

What is the sample size of the training set used in crop classification? Not the ratio

 

Please adjust the font size in Figure 8 to make the figure clearer.

 

Why do the scatter plots in Figure 8 and Figure 11 use randomly sampled points instead of all pixels?

 

What is the actual crop category of the field with lower accuracy in Figure 10? Please explain

 

The table style in the manuscript is not standardized, please use the standard three-line table.

 

What is the value of acceptable accuracy of the classification model? What is satisfactory accuracy?

 

Predicting future images is a big challenge, and sudden natural disasters will affect crop growth. I think the authors' current model does not have this prediction ability, which may affect the prediction accuracy of the model. Please add this shortcoming of the model in section 5.3.

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

I have reviewed the issues I mentioned in my previous review, and I am pleased to note that the authors have addressed all of them. However, upon re-examining the paper, I came across a minor style problem on L440-443. It appears that Figure 7 and its caption are mixed together. To ensure clarity and consistency, the caption should be placed entirely below the figure.

Apart from this observation, I am confident that the paper is now prepared for publication.

Author Response

Question:

I have reviewed the issues I mentioned in my previous review, and I am pleased to note that the authors have addressed all of them. However, upon re-examining the paper, I came across a minor style problem on L440-443. It appears that Figure 7 and its caption are mixed together. To ensure clarity and consistency, the caption should be placed entirely below the figure.

Response:

Thanks for your kind reminder. Regarding the mixed placement of Figure 7 and its caption on lines 440-443, it is indeed important to maintain clarity and consistency in the document. To rectify this, we have adjusted the position of the Figure 7 placement.