Field Data Collection Methods Strongly Affect Satellite-Based Crop Yield Estimation

Round 1

Reviewer 1 Report

This manuscript is well written and organized. The author(s) should consider the following in improving this manuscript.

 

1. In the introduction, the importance of maize should be mentioned. Because it is the study object of this manuscript. Therefore, some relevant literatures need to be considered.
2. The conclusion section is missing in the current manuscript. It should not be missing.
3. Line 57, is OLS the abbreviation of “linear regression”? The complete spelling of OLS needs to be supplemented.
4. Line 117, what does "15 VI-(aggregation method) variables" mean? What does "in the univariate linear models" mean? Is it the “360 univariate ordinary least squares (OLS) linear models (LR-1)” in line 114? What does "LR" mean? Please clarify these issues.
5. Equation 3 and equation 2 have different styles, why? Replace N(ef) with Nef and replace A(efm) with Aefm.
6. In equation 1, some variables are not explained clearly. E.g., “Where y is crop yield for observation (field) i, field sampling method f, and modeling method m.” I didn’t find “y_ifm”. What is “n”? What does “Aefm” mean? What does “the predicted crop yield” mean? What is the role of “Aint”, “AEXT”, and “ATRA” in equation 1? Lack of relevant literatures about these equations. Other similar issues in Section 2.4 also need to be considered.
7. Line 161, what is “slop”? In the method section, it should be mentioned.
8. Section 3.1, the diagonal cuts method was considered to have better performance. This method includes three sub-plots. If the number of sub-plot is increased to 5, as shown in the figure below, what will be the result. Besides, for each sub-plot, it’s area was 4×4m. What impact will the area change of sub-plot have on the results?
9. Figure 2 needs to be further beautified. For example, the font size is inconsistent, the image is blurred, and so on.
10. Line 278, the words “vegetation index (VI)” could be replaced by “VI”. As for abbreviations in the whole manuscript, the authors need to check carefully them.

Comments for author File: Comments.pdf

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

This research by Tiedeman et al. investigates the impact of field data collection approaches when gathering training data for developing/calibrating crop yield predictive models. Overall, it was shown that collection of training sample data via the crop-cut approach had better predictive results.

I found the paper to be well written all figures and tables were clear. I would, thus, recommend publication of this article if the authors can address some relatively minor points. First, with regards to the field sampling method, I think the validity of the crop-cut approach will depend on whether the cut area location are sufficient to capture the variability of a given field. One approach would be to perhaps use satellite imagery (e.g., vegetation index map) in order to have a field view of the overall variability that could then inform where to capture the destructive measurements. Perhaps this points could be added to the Discussion section.

Second, with the random forest modelling method, could all of the Sentinel-2 bands have been included without the need to convert to the VIs? Please justify the use of a VI here.

Specific points:

L26-28: Modern agricultural machinery (i.e. combines) record the yield amounts and points within the field. Such technology may not be available for fields in Ethiopia, but this should be mentioned here.

L290: The validity of VI/Yield empirical relationships is also dependant on the crop growth stage.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 3 Report

Please see attached comments in manuscript.

Reviewer Comments

Field data collection methods strongly affect satellite-based crop yield estimation

Remote Sensing

remotesensing-1609939

 

I want to thank the authors and editors for the opportunity to review this work. The paper addresses an area of great interest to many of us who work with remote sensing of yield (or biomass) estimates from remotely sensed data: what direct sampling measures are best, and how many samples are required? Overall, the paper is very well organised and the figures are well presented. I have included my overall comments here, and refer the authors and editors to the attached copy of the manuscript with comments/questions within the text.

• Introduction and rationale of the work. I think that the introduction highlights the issues faced with yield prediction in many developing countries and/or, where the level of mechanisation is still low. To set the stage, I would suggest adding some context around this – the paper isn’t addressing production areas with large fields and available data streams such as yield monitor data.
• Discussion and results of sample method. Table 1 indicates an improvement with three sets of cuts (diagonal) over one. This isn’t surprising. However, later in the Discussion section, the sample methods are combined. Is there evidence presented to justify this? Line 233 states that a single random cut was best. That was not obvious to me looking at the analysis through the paper. Perhaps the authors could elaborate why they support this conclusion.
• Discussion and results of VI used. My impression as a reader was that the use of various VIs was more or less background (or black box) to the paper. I was surprised in the Discussion section that certain VIs (or VI-approaches) are being singled out as ‘best’. The authors need to support this with additional results presented, perhaps as an appendix?
• Models used. As with the VIs, the models used were not described in detail. However, I think that is not necessary, as the authors did present evidence that the sample method appeared to be more important than the VI or the prediction model used.

Overall, I would encourage the authors to read over the Discussion section again to verify that concluding remarks are supported by the results presented. My sense is that the authors have learned a lot from this study, and that some results have not been fully presented.

 

Comments for author File: Comments.pdf

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 4 Report

The article, "Field data collection methods strongly affect satellite-based crop yield estimation", uses eight field data collection methods to demonstrate that high-quality measurements is crucial to build yield estimation models with satellite data. The results are solid and useful for building remote sensing models to estimate crop yield. Generally, the manuscript is well organized and quite suitable for the scope of the journal. Whereas there are still some issues need to be solved before the acceptance, I would thus recommend major revisions.

1. Cob density is used in the transect methods in Table 1, but how to obtain cob density does not seem to be described in detail? It may be the reason why transect methods is overestimated.
2. Line 106, why use the median VI to represent each field for each date instead of the mean?
3. It is recommended to add RMSE to the panels in Figure 1 and Table 3, because R² sometimes does not represent the estimation accuracy.
4. The quality of figures is poor, the title of Figure 1 is too close to the upper frame line and the y axis format of the final panel is also inconsistent with others. The format of different figures should also be consistent as far as possible, otherwise they will be difficult to use.
5. Line 191-194, if the figures in the appendix are quoted in the text, they should also be clearly marked. It seems that the data or figure of two most important variables in the RF model cannot be found both in the main text and appendix.
6. In section 3.3, when calculating the extrapolation accuracy, how to determine the modeling set and verification set, and what is the proportion? Please specify.
7. Line 329, Figure S1 should be A1?
8. The figure notes of Figure A2 & A3 are not clear enough. Figure A3 should be the result of LR-2? And where is the RF?

 

Author Response

Please see the attachment

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

The author didn't take all the suggestions and comments seriously. I regret for that. There are still some major problems that have not been solved.

1. In the introduction, the author only explained that they wanted to evaluate the impact of sampling methods on yield estimation. However, other more important literature reviews are missing. For example, what is the current research progress on field sampling methods? What is the accepted sampling method in academic circles? Are the sampling methods applicable to different crop types consistent? Can the sampling method proposed by the author form an industry standard? If the conclusion obtained by the author can only be applied to the experiments in this manuscript, it has no scientific value and application value.
2. The author devotes a lot of space to introducing the defects of asking farmers about crop yield. I think this defect is common sense. No scientist will use this yield data as the training data of a model. The current literature review of this manuscript is unscientific. In addition, I can't find the source of some references.
3. Line 48-55, What is the purpose of this paragraph? “balancing the costs and benefits”, it is a problem to be considered in application, not what scientific papers should focus on. This manuscript should address which parameters in the sampling scheme will affect the yield estimation of the model to what extent. Relevant scientific issues should be studied in a scientific and quantitative research paradigm.
4. Line 57-65, maize is the subject of this manuscript, why? Why not other crops? Is this a random decision? Are the conclusions drawn in this manuscript applicable to other crops?
5. In table 1, For each sampling method, what scientific paradigm does the author get the parameters of the sampling method based on? What are the effects of different parameters on the results? Shouldn't these effects be the scientific questions to be answered in this manuscript? For example, “At each point, three cobs were harvested.” Why is this parameter three? If it was 30, would the result be better? “recorded the number of cob-bearing plants within c of the area surrounding the sampling points.” Why is this parameter 1 m²? How did the author consider the influence of plant spacing and row spacing on this parameter? All sampling schemes in Table 1 have similar defects. This will affect the correctness and scientificity of the results and conclusions in this manuscript. Therefore, I think the experimental scheme of this manuscript is unscientific and incomplete.
6. Line 108, “removed very low or removed very low or high values (outliers) that were missed by the cloud filter”, what are these values? Why removed?
7. Satellite data section, how does the author consider and deal with the difference of image spatial resolution?
8. Conclusion section, this conclusion has almost no scientific value.
9. What will be the result of directly extrapolating the overall true yield through the sample yield? For example, sample yield multiplied by total area. Instead of obtaining the estimated yield of the model based on the sample yield, and then comparing it with the overall true yield.
10. According to my understanding, the ultimate purpose of sampling is to estimate the yield of the entire This estimate does not refer to remote sensing estimation. In order to achieve this goal, evenly arranging sampling points should be the most scientific and commonly used scheme. But the author did not adopt it. What impact does the collaborative change of the area and number of sampling points have on the yield estimation of the entire field? This is the scientific problem studied.
11. For section 2.4, one or two classical maize yield estimation models are enough. Because the focus of this manuscript is not the estimation model.

Based on the above comments, I think this manuscript does not clearly study the relationship between sampling scheme and yield estimation, as title "field data collection methods affect yield estimation". There are major defects in the research scheme. The scientific value and academic contribution of this manuscript are insufficient.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 4 Report

The authors provided relative good answers in terms of the reviewers' comments and the manuscript improved a lot compared to the previous version. I have no further comments.

Author Response

Thank you very much for your time reviewing this manuscript. We appreciate your valuable feedback.