Enhancing Maize Yield Simulations in Regional China Using Machine Learning and Multi-Data Resources

Round 1

Reviewer 1 Report

The present study used random forest (RF) to improve the capability of global gridded crop models in simulating China’s county-level maize yield. With the RF model, it also identified potential yield gaps by using the difference between simulated maize yields between the Harmnon and the Fullharm scenarios. Overall, the methods and results are generally sound. The paper had good understanding on the GGCMs, well organized the experiments, and the manuscript was clear to follow. I have some suggestions to further strengthen the study:

1、 In the introduction, provide some more background on why improving crop yield simulations at finer spatial scales like county-level is important, beyond just data mismatch issues. What kinds of applications would benefit from better county-level yield estimates?

2、 Elaborate more on why the specific machine learning algorithm (random forest) was chosen and its advantages for this purpose compared to other ML methods.

3、 In the discussion, comment on any limitations of the methodology, datasets used, or assumptions made that could be addressed in future work. For example, effects of management factors beyond what's implicit in the crop model inputs.

4、 Figure 7: results indicate that the highest yield gap was in southeast and southwest China. Sown area of maize has been quite small in southeast China. What is the major constraint of maize production in southeast China, and what factors contribute the most to the yield gap? Does the result mean China have large potential to grow more maize in the south? More explanation would benefit the study.

5、 The writing could be tightened up in places to improve clarity and flow. Carefully proofread to fix minor errors. For example, page 7, “GGCMI defines three distinct model configurations based on different crop management practices”, “defines” should be “definded”.

Overall, the study makes a valuable contribution, and with minor revisions to address the points above, I think the manuscript would be appropriate for publication after fully revisions. Let me know if you would like any clarification or have additional questions!

The writing could be tightened up in places to improve clarity and flow. Carefully proofread to fix minor errors. For example, page 7, “GGCMI defines three distinct model configurations based on different crop management practices”, “defines” should be “definded”.

Author Response

Dear Reviewer,

We deeply appreciate your valuable suggestions. Below please find responses in a point-by-point basis. Please see the attachment.

Best regards,

Yangfeng Zou.

Author Response File: Author Response.pdf

Reviewer 2 Report

The authors skillfully integrated the random forest algorithm with multiple data sources, training it on low -resolution maize yield simulations from GGCMs. This amalgamation was subsequently applied to a finer spatial resolution at the regional scale in China. The original GGCMs faced challenges in accurately simulating county-level maize yield, primarily attributed to their limited spatial resolution. Nevertheless, the enhanced GGCMs exhibited a significant improvement in the model performance. The paper introduces an innovative methodology that effectively accomplishes its stated goal, as outlined in the title. Below are some of my general suggestions:

1.      In the introductory section, the GGCMs were initially developed with the primary purpose of assessing the impact of climate change on crop production, incorporating the ability to retrieve historical yield data. The accuracy has not been high for several reasons. For instance, models were not capable of reproducing the damage of excessive wetness and other climate extremes. Models were not so sensitive to physical damages such as wind or hail. As models were using global datasets of crop calendar, soil and fertilizer use, there would be considerable errors induced. The authors tried to claim that spatial resolution is one major challenge of GGCMs in reproducing China’s county-level yield data, however, in the introduction, this part of the description needs to be strengthened.

2.      The downscaling model bears resemblance to the statistical emulators of GGCMs. Notably, there are existing studies, such as Franke et al., 2020, which delve into emulator development and enable researchers to employ lightweight simulations for various purposes. Consequently, the discussion sections would benefit from a more comprehensive exploration and comparison with these existing studies.

3.      Furthermore, the paper’s figures require enhancement in terms of clarity. For instance, in Fig. 2, it is advisable to label the dashed lines in subfigures a and b to enhance the self-explanatory nature of the image.

Moderate editing of English language required

Author Response

Dear Reviewer,

We sincerely appreciate your kind suggestion. In response to your feedback, we improved our manuscript substantially and addressed below on a point-by-point basis. Please see the attachment.

Best regards,

Yangfeng Zou.

Author Response File: Author Response.pdf

Reviewer 3 Report

The paper introduces an interesting approach to improve the accuracy of county-level maize yield simulations in China. This is achieved by the integration of global gridded crop models (GGCMs) with machine learning techniques. The clarity and completeness of the introduction, materials, and methods contribute to the paper’s overall coherence. The results are presented clearly, and the figures and graphs accompanying the paper facilitate a comprehensive understanding of the conducted work. My main concerns are as follows:

1.     Figure 3 was showing that all land in China were covered with maize. Why not using the maize distribution mask? The authors used the MIRCA crop mask data, right?

2.     The discussion of the paper should provide a more detailed explanation of how the findings currently discovered can be further utilized by peers and the broader implications of such improvement. This will help enhance the reliability of results.

3.     Remove "China" from the following keywords.

4.     In Fig. 1, the term "Shanxi" appears twice; please verify the spelling. Additionally, remove the "legend" and ensure there is a space between the colon and the letter.

5.     In Fig.2, the unit of legend is unclear and please add. Also please add space for the (a) ff and the others.

6.     Similarly, please add space for the number and unit for other parts in your article.

7.     Transform all tables into the official three-line table format.

8.     Review lines 163, 168, and others; the website format does not align with the others.

9.     The formatting of all formulas appears disorganized.

10.   Ensure both numbers and units are isolated with space. For example, Fig.5 and so on.

11.   Reduce the size of (a) (b) (c) in Fig. 7.

12.   Some grammatical errors and irregular English writing can be found. Please revise them.

13.   The type of the reference should be revised according to the RS journal.

14.   I suggested that authors should re-organized the discussion section to highlight the practical significance of the research.

15.   Please add some description on why use the RF algorithm in this study. Why not select other machine learning algorithms.

Once these issues are addressed the paper is worthy of publication after a carefully major revision.

Minor editing of English language required

Author Response

Dear Reviewer,

In accordance with your suggestions, we have made substantial improvements to our manuscript. We have also checked through the entire manuscript to avoid minor mistakes, thanks again for your useful feedback. Please see the attachment.

Best regards,

Yangfeng Zou.

Author Response File: Author Response.pdf

Reviewer 4 Report

Dear Authors,

I've found your paper interesting and well written. The topics were well explained and the reading was easy. The results were quite interesting and I hope you will continue to study in this research field.

Author Response

Dear Reviewer,

Thank you for your positive feedback on our paper. We appreciate your interest and acknowledgment of the clarity and interest in the topics discussed. Your encouraging words inspire us to continue our efforts in this research field.

We are pleased to hear that you found the results intriguing, and we are committed to furthering our studies in this area to contribute more meaningful findings.

Once again, we sincerely appreciate your recognition and encouragement.

Best regards,

Yangfeng Zou.

Round 2

Reviewer 2 Report

The authors have carefully revised their manuscript according to the reviewers' comments. I do not have additional comments and recommend it to publish on Remote Sensing in present form.

Minor editing of English language required

Reviewer 3 Report

The author have revised all the questions and the article can be accepted.