Testing the Performance of LSTM and ARIMA Models for In-Season Forecasting of Canopy Cover (CC) in Cotton Crops

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The author constructed a CC estimation model using drones and the LSTM algorithm, but this study is very simplistic and roughly executed, offering no innovation. The writing of the manuscript is difficult to understand.

1. In the abstract, abbreviations must be clearly explained to enhance readability and comprehension for the reader.

2. The abstract's result section requires elaboration. It should offer a more detailed summary of the key findings, rather than just stating that the results are "satisfactory."

3. The introduction needs to be expanded. It should include a comprehensive overview of the current state of CC (presumably carbon capture or climate change, depending on the context) forecasting and explain why the author chose to use LSTM (Long Short-Term Memory) for this study.

4. In Section 2.1, a map illustrating the location and general view of the study area would greatly enhance the reader's understanding of the geographical context.

5. Section 3 should clarify how interpolation and time series variation analysis contribute to the construction of the CC estimation model. It's important to explain the role of these analytical tools in model development.

6. The description of the results section is unclear and needs to be reorganized. It should be divided into subsections for clarity, each focusing on a specific aspect of the results.

7. The discussion section requires significant enhancement. It should include a comparative analysis with relevant research and offer insightful scientific suggestions based on the study's findings.

Author Response

The answers to each of the queries have been attached.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

he authors conducted a comparison between LSTM-based models and ARESM in predicting cotton canopy cover (CC). The dataset was collected for 14 cultivation days, with additional data generated for 116 days using polynomial interpolation. The results indicate LSTM's strong performance in predicting cotton CC.

 

However, several drawbacks are noted:

 

1.    The paper lacks sufficient detail for comprehension. For instance, in section 2, clarification on individual cultivation clusters versus discrete cultivation clusters is needed. What distinctions exist between them?

2.    The authors stated, "Since canopy cover data was obtained for 3500 grids for the entire cultivation period of 116 days, it was not possible to train the LSTM models considering all the time series sequences, as it would have resulted in millions of training sequences." Why not reduce the number of grids from 3500? Is it necessary to use data from all 116 days?

3.  The clustering method groups each day's data into 14 clusters. Can it guarantee that the same plot results will be clustered similarly on different days? Could this inconsistency impact the results?

4.    The paper's organization needs improvement.

 

Comments on the Quality of English Language

Need sentence by sentence proofreading to make sure

Author Response

The answers to each of the queries have been attached.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

This paper tests the performance of LSTM and ARIMA models in predicting canopy coverage based on UAS data. Overall, the design of the article is innovative, and the comparison of results is very rich. However, there are some major issues in the article that need to be addressed and modified.

 

Main issues:

1.     Line 86-87, "A few things need to be added to crop management using the crop growth variables." I do not understand what this is specifically expressing.

2.     The last paragraph of the introduction needs to be revised. Currently, it does not highlight the innovative aspects, and the challenges your research aims to overcome are not clearly stated.

3.     Section 2.1, I believe a schematic diagram of the study area, including the actual range of UAS data collection, is necessary and should be added.

4.     Line 129-132, why is the growth cycle set to 116 days? This does not correspond with the information given in Table 1, which does not provide the emergence time.

5.     Section 2.1, from the description in the text, I cannot determine the specific timing of UAS data collection. However, it seems from subsequent results that data was collected over different 14-day periods. Please clarify the data collection in this section.

6.     Line 298-303, why choose 14 days as the forecasting time window?

7.     Line 311-316, the description is unclear; it seems to express that you calculate the average canopy coverage data for each cluster, then compute the DTW distances between pairs of average data and use the pair with the smallest distance as model input? But later, "The same process was repeated to gauge the testing efficiency of the trained model on the 2021 canopy data." I do not understand what this specifically refers to.

8.     Line 353-355, I think you need to reorganize the statement. From what is shown in the table, the Multiple input multi-step output LSTM model consistently performs the best, even 56 days later, and has significantly lower errors compared to other model variants. Even considering computational issues, the last model is undoubtedly the best performing.

9.     Line 356-358, it seems you mean to say "When data from 28 to 56 days after emergence is used as input, the multi-input multi-step output LSTM model predicts canopy coverage with the highest accuracy."

10.   Line 403, "most unique" why? A more detailed explanation is needed.

 

Other issues:

1.     The first occurrence of abbreviations is not standardized; full names need to be provided, such as Line 17 'ARIMA', Line 33 'DSSAT', 'APSIM'.

2.     Figure 2, ADF needs to include the full name.

3.     Figure 3, the composition of the box plot needs to be introduced, such as what the box body and the upper and lower axes correspond to.

 

4.     Table 3, two instances of the word 'model' are misspelled.

Comments on the Quality of English Language

Extensive editing of English language required

Author Response

The answers to each of the queries have been attached.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The concerns have been well addressed.

Comments on the Quality of English Language

 Moderate editing of English language required