Yield Estimation of Paddy Rice Based on Satellite Imagery: Comparison of Global and Local Regression Models

Round 1

Reviewer 1 Report

The authors present work on estimating rice yield from processed multispectral imagery from a satellite. I find the work interesting and useful. However, the description of methods and results needs improvement.

What is the resolution of the satellite imagery?

The vegetation indices and texture indices in tables 1 and 2 are introduced without references as far as I can see. Please add references, particularly for the GLCM indices which some readers may not be familiar with.

Has GLCM been applied to rice yield prediction before? Please add references if so. If not, please expand on how you came up with the descriptions in table 2 (eg how do you know GLCM variance is lower for higher yield etc).

For the yield data (dependent variable), it is hard to understand exactly what this data is. Is it from hand sampling? Or overall per-field data? For the hand samples in Figure 2 - what is the collection area for each Datum Mark? Particularly on the paragraph under Figure 3, it is hard to know what you mean by "yield data for each paddy rice land parcel is not available. Therefore, we used the total yield data..." What is total yield data - is this for the whole region? How does this relate to the ground survey data? Is the model trying to predict ground survey data or end of season yield data? 

The distinction between yield and ground survey data is also unclear in Figure 4 - are Training Data from Ground Survey and Testing Data both from mid-season samples?  I assume training and testing is split from the same dataset, including both image data (independent variables) and ground sample data (dependent predicted variables). Again, how do ground samples relate to "Yield Production"? 

What is "First Cultivation"? This term is used a number of times but not defined as far as I could tell.

In section 2.1, please explain clearly what the combinations 2 and 3 are. How do you "use Pearson correlation to select variables from past theories and experiments"? What variables did you end up selecting? How many variables were included in combinations 2 and 3? Please report this process clearly. It would be interesting to the reader to know whether raw bands, vegetation indices, or texture indices were more important, and which ones. What do you mean by "unexpected positive and negative" in the paragraph above Table 3?

There is a paragraph on page 8 that maybe shouldn't be there ..."where wij is the weight value..."

Again, I think this paper is interesting and useful but you need to explain your methods and results more clearly. 

Author Response

Dear Editors and Reviewers:

Thank you for your letter and for the reviewers’ comments concerning our manuscript entitled “Yield Estimation of Paddy Rice Based on Satellite Imagery: Comparison of Global and Local Regression Models” (Manuscript ID: remotesensing-390713). Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our researches. We have studied comments carefully and have made correction which we hope meet with approval. Revised portion were highlighted with Track Changes in the manuscript. The main corrections in the paper and the responds to the reviewer’s comments are as following:

Author Response File: Author Response.docx

Reviewer 2 Report

Major comments:

·         Spatial variability is a common problem in many data analysis techniques. Since season and geographical location is influencing the model performance, you should try with factorial regression combining different locations and years.

·         The authors mentioned different machine learning algorithms. However, it is necessary to discuss the limitations of those techniques and why GWR is the best option for yield estimation of rice.

·         Comparing just only three regression models will not benefit the research or grower community.

I hope this will help the author to improve the quality of paper and add the innovation as well.

Author Response

Dear Editors and Reviewers:

Thank you for your letter and for the reviewers’ comments concerning our manuscript entitled “Yield Estimation of Paddy Rice Based on Satellite Imagery: Comparison of Global and Local Regression Models” (Manuscript ID: remotesensing-390713). Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our researches. We have studied comments carefully and have made correction which we hope meet with approval. Revised portion were highlighted with Track Changes in the manuscript. The main corrections in the paper and the responds to the reviewer’s comments are as following:

Author Response File: Author Response.docx

Reviewer 3 Report

The authors presents the usefulness of GWR model for the estimation of paddy rice yield based on satellite imagery. The work is interesting and deserves a constructive discussion, but it still needs a considerable revision to be acceptable for Remote Sensing. Especially, ground survey data (yield) in this study site is not clear in this manuscript.

Major comments:

1. Yield data in this study site is not clear. Please add the yield data of maximum, minimum, average and standard error in this study site in Table 5.

2. I think Table 3, 4 and 6 should be omitted because the information in the Tables was too little. Please combine such information with another Figures or Tables.

3. You said that GWR model was superior for yield estimation to other two models. I think the reason and/or the result for your conclusion is not clear. You need to rethink your whole your result and conclusion.

4. There is much difference of the measuring dates of Spot-7 images between years and study sites (Table 6). I think the comparison of the difference of error rates between years and study sites is difficult.

Minor comments

1. I think you need to add the background in this study in abstract.

2. P1L11 You said rice yield can be estimated in multiple ways using ground survey data. Please include enough information.

3. P2L29 You explained about rice growth model ORYZA. I think explanation of another rice growth model related to remote sensing data (e.g. SIMRIW-RS, Homma et al., 2017, Journal of Agricultural meteorology, 73, 9-15) should be provided.

4. There is no statement concerning SVR model in Introduction part.

5. P3L9 What does it mean “a more compact structure” ??

Author Response

Dear Editors and Reviewers:

Thank you for your letter and for the reviewers’ comments concerning our manuscript entitled “Yield Estimation of Paddy Rice Based on Satellite Imagery: Comparison of Global and Local Regression Models” (Manuscript ID: remotesensing-390713). Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our researches. We have studied comments carefully and have made correction which we hope meet with approval. Revised portion were highlighted with Track Changes in the manuscript. The main corrections in the paper and the responds to the reviewer’s comments are as following:

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

The authors have provided helpful replies in their response document and have made some additions to the paper. However, some of their responses to reviewers are not yet reflected in their paper. For example, in response 3, they describe their sampling method more fully (1.2m x 1.2m etc), but this has not been added to the paper. Another example is response 6, where the authors explain a little how they arrived at variable combinations 2 and 3. I think more detail is needed here, describing the variable selection process more rigorously, and again, this needs to be added to the paper, not just to the response to reviewers document.

They have added references showing previous work that have used GLCM and VIs for rice yield prediction.

Reviewer 2 Report

Dear Authors

Thanks for improving the manuscript as suggested in the previous version. However, I would like to suggest some more revisions as given below:

·         The abstract should start with the overarching goal or main objective(s) as well as the purpose/rationale of the study, which was not articulated properly in the study. Please try to improve first three sentences. Why your approach is important? Why yield estimation of paddy rice is important? etc.

·         Page 2, paragraph 2, line 1, please rewrite the sentence as ‘Numerous studies used satellite imagery combined with …..’ and complete the sentence with ‘estimation of agricultural crops’. Please add relevant references.

·         Page 2, paragraph 2, last line, please consider adding the limitations of different regression models developed using VIs.

·         Page 2, paragraph 3, sentence 1, please add relevant references.

·         Page 2, paragraph 3, last sentence, please explain the limitations of ORYZA and SIMRIW models. Why you have not used those models in your study?

·         Page 2, paragraph 4, line 4, please use ‘SVM’ after ‘support vector machines’ and ‘support vector regression’ before SVR.

·         Page 2, paragraph 4, last sentence, please rewrite the sentence ‘The advantages of …’. Now it does not have any meaning. Please delete next sentence ‘With remote ….’, as that is not relevant to your study. Please rewrite the last sentence of this paragraph ‘ Some were….’

·         Page 2, paragraph 4 and page 3, paragraph 2, I would suggest to rewrite those paragraphs and add explanation given in the response for the previous version of manuscript. If you explain the limitations of OLS and SVR in such way, why you have used and compared those models with GWR? Better you explain the models in introduction and describe the limitations in the discussion.

·         Section 2.1, The VIs were generated from 4 spectral bands and ground survey data. Please split the sentence into two.

·         Table 1, Please use the original references of the VIs. How did you get the parameter values for NIR(Soil)? Please explain why you have chosen these VIs. What is the R value in TSAVI, PVI and GESAVI?

·         Section 2.2, first line, please delete ‘where’ and add the latitude, longitude of the study area.

·         Section 2.2, please add the source of climate data.

·         Section 2.2, please add ‘respectively’ in the last sentence.

·         Figure 1, please follow the standard format of the journal. Remote sensing journal has specific format, which you can find in the instructions for authors section in their website.

·         Section 2.3, please include the website address of AFA.

·         Section 2.3, paragraph 3, last sentence, why you changed the sentence from previous version? It is now incorrect.

·         Figure 3, please follow the standard format of the journal.

·         Section 2.3, last paragraph, line 3, please use ‘acquired’ instead of ‘captured’.

·         Section 3, Results and Discussion, please move first three paragraphs (up to equation 5) to Methodology section.

·         Figure 5, please follow the standard format of the journal. Please add a, b, c, d for the subfigures. Same for the caption as well.

·          Page 13, citation 77 was given twice.

·         Section 3, paragraph 12, line 1, please consider using ‘In order to increase’ not ‘increasing’.

·         Section 3, paragraph 12, line 6, please use ‘Dapi are also reduced’ instead of ‘reduction’.

·         Section 3, paragraph 12, last line, please delete ‘The above discussion…’. Don’t need to describe same thing twice.

·         Please separate the Results and Discussion chapter in the manuscript.

·         Conclusions, first paragraph, please use the previous version.

Reviewer 3 Report

I have annotated the manuscript with several minor corrections, which I believe will improve the readability of the paper.

The text becomes well organized but there are quite a few repetitions and thus the paper should be shortened to become more concise.

Table 4

The unit of Yield is normally presented by t ha^-1

Introduction

Homma→Homma et al.