High-Throughput Plot-Level Quantitative Phenotyping Using Convolutional Neural Networks on Very High-Resolution Satellite Images

Round 1

Reviewer 1 Report (Previous Reviewer 2)

Overall, the authors have addressed my major concerns. For this revised version, there are a few mirror concerns and suggestions.

-1, Though the authors have claimed that they have reviewed all tenses used within the document, "Using Past Tense for all references to the past; Using Present Simple Tense for references to the manuscript itself". But there are still some confusing expressions. For example, in L72-73, "We faced the same issue with our data, and therefore we predict traits in each image independently using spatial 2D CNNs." The mixing of tenses may not cause ambiguity, but it may greatly reduce interest in reading. I suggest authors approach this issue with caution and organize the entire text.

-2, L160-166, L182-200. The descriptions of data collection should be more detailed. For example, what is the Hiphen, and how did it measure canopy cover, greenness, and flowering? Adding at least some supplementary materials or necessary references could be suggested.

It is recommended to continue to approach the multiple tenses issue with caution.

Author Response

Please see attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report (Previous Reviewer 3)

The manuscript (remotesensing-2789029) has been consistently improved. Furthermore, it has become more interesting. The sections are organized, including introduction, materials and methods, results, and discussion. The discussion section still needs better referencing. Minor corrections regarding organization are needed.

Keywords in alphabetical order.

The table legend should be above the table, not below.

L462-496. References?

L499-527. References?

Check grammar and spelling.

Author Response

Please see attachment.

Author Response File: Author Response.pdf

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

The manuscript predicts plot-level quantitative phenotyping using Convolutional Neural Networks (CNNs) with very high-resolution satellite images. This is a compelling topic with the potential to enhance AI applications in quantitative remote sensing.

However, I have two major concerns that must be addressed before considering the publication of this manuscript:

The authors have not introduced the phenotyping parameters in this study. How were canopy cover, greenness, height, biomass, and NDVI measured? It's worth noting that biomass measurements can be obtained through various methods, each with varying levels of accuracy. How did you quantify the greenness of the plot? There are several ways to define greenness, each with different implications. Additionally, NDVI is a vegetation index that can be directly calculated from remote sensing images, so the need for CNN training in this context is not entirely clear. Ground data are crucial for AI training, and the authors should at least explain the significance of these parameters to the reader.

My primary concern is that the timing of satellite and ground data does not align. Crop growth is not typically linear; for instance, canopy cover, height, and biomass tend to grow rapidly during vegetative stages, stabilize during peak stages, and decrease rapidly during senescence stages. Therefore, I question the feasibility of interpolation for the extrapolation process. In other words, training with such data misalignment may introduce significant uncertainties.

Minor comment: Change "L76 phenology" to "phenotyping."

The sentences in the text are often lengthy and convoluted, making it challenging to follow the author's arguments. Breaking them into shorter, more concise sentences would improve readability.

Author Response

Thank you very much for taking the time to review this manuscript. Your comments and suggestions have helped us to improve the manuscript substantially.

We have highlighted our changes in blue. All reviewers noted necessary improvements to language, so there are many changes throughout.

Please find detailed responses to each comment in the attached pdf. Where appropriate we include the specific wording changes in each comment response.

Author Response File: Author Response.pdf

Reviewer 2 Report

The present study focuses on the utilization of satellite images and deep learning methodologies for the automated acquisition of plot-level phenotypic characteristics in plant breeding experiments. The authors present two methodologies for forecasting plot-level phenotypes. The first approach involves employing a classification model to predict the centered plot, while the second approach entails predicting per-pixel values and subsequently aggregating these forecasts to derive a value for each plot. The document additionally addresses pertinent literature on the topic and examines the difficulties associated with utilizing deep learning techniques to directly predict canopy features from satellite imagery. In summary, the current work is a really intriguing scientific endeavor. However, the shortcomings of this manuscript are also evident. On the one hand, the writing and grammar of this manuscript urgently need to be revised and improved. On the other hand, there are some shallow and conservative understandings of remote sensing. Therefore, I suggest a major revision.

major concerns:

-1, L45-76. In the introductory section, it is crucial to provide an objective description of various methodologies and elucidate their distinctions. The exposition of the methodologies employed in this work needs to be included in the methodology section. It is advisable to relocate these descriptions to the methods section. Meanwhile, it is not advisable to provide a detailed description of the results, conclusions, and contributions of your work in the introduction section.

-2 There is more than one tense in the active voice in this manuscript, such as:

L49 "we wished", L50 "We propose and compare", L56 "we use", L59 "we obtained", L67 "We describe", L154 "we used", L191 "we use", L192 "we purchased", L223 "we did not perform", L245 "we used", L256 "we propose", L467 "we have shown", L487 "we were limited", L525 "we have worked", L526 "we have proposed".... 

The utilization of the Present Tense, the Past Tense, and the Present Perfect Tense has been intermingled.

-3 L121-122. "To our knowledge, no other work has attempted to predict canopy traits directly from satellite images using deep learning." I can't agree with this view. This is a very shallow assertion about satellite remote sensing monitoring of plant canopy traits. A number of studies have been performed in recent years. Please refer to the following literature, including but not limited to:

[1] J. Cao, Z. Zhang, Y. Luo, L. Zhang, J. Zhang, Z. Li, F. Tao, Wheat yield predictions at a county and field scale with deep learning, machine learning, and google earth engine, European Journal of Agronomy, 123 (2021) 126204.

[2] T. Chang, B. Rasmussen, B. Dickson, L. Zachmann, Chimera: A Multi-Task Recurrent Convolutional Neural Network for Forest Classification and Structural Estimation, Remote Sensing, 11 (2019).

[3] S. Jeong, J. Ko, J.M. Yeom, Predicting rice yield at pixel scale through synthetic use of crop and deep learning models with satellite data in South and North Korea, Sci Total Environ, 802 (2022) 149726.

[4] Y. Xie, J. Huang, Integration of a Crop Growth Model and Deep Learning Methods to Improve Satellite-Based Yield Estimation of Winter Wheat in Henan Province, China, Remote Sensing, 13 (2021).

-4 I would strongly suggest the authors either 1) go over the manuscript as a whole and prepare to edit from the abstract to the discussion, not just sections where reviewers have left comments 2) get an additional co-author who can edit the manuscript, or 3) look into using an editing service.

minor concerns:

-1, L8, 112. NDVI. The full name should be displayed the first time it is used.

-2, L25-30. Please pay attention to the grammar and punctuation issues in this paragraph.

It is necessary to undertake a thorough English editing process for better understanding.

Author Response

Thank you very much for taking the time to review this manuscript, and we are glad you find it to be a really intriguing scientific endeavour. Your comments and suggestions have helped us to improve the manuscript substantially.

We have highlighted our changes in blue. All reviewers noted necessary improvements to language, so there are many changes throughout.

Please find detailed responses to each comment in the attached PDF. Where appropriate we include the specific wording changes in each comment response.

Author Response File: Author Response.pdf

Reviewer 3 Report

In this manuscript (remotesensing-2644611), the authors aim to optimise the process of crop breeding trials to contribute to global food security. They utilised satellite imagery and deep learning to automatically gather data on various crop phenotypes from trials in South Australia and Sonora, Mexico. Two computer vision methods were tested: one that focused on the centre of each crop plot and another that analysed data per pixel and aggregated it to provide a plot-level perspective. The first method, which uses a modified ResNet18 model, proved to be the most effective. This study illustrates the potential of combining remote sensing and machine learning to enhance the efficiency and outcome of crop trials.

After carefully evaluating this manuscript, I am not in favour of its publication in its current form.

The manuscript clearly refers to a dissertation or thesis that was simply “placed” into a model of remote sensing to receive feedback in the pursuit of publication.

The “manuscript” did not follow the guidelines present in Remote Sensing or the state of the art for a scientific publication, with a clearly defined introduction, hypotheses, and objectives. An introduction section and another review section are addressed, but without a clear and defined perspective for readers.

Consider adding more keywords and changing “satellite imagery”, which is very similar to that presented in the title.

The Materials and Methods section is confusing and does not cite pertinent bibliographies. Furthermore, the Data and methods sections were not adequate. There is a need for a clear and substantiated material and methods section. The maps also presented in the form of figures are of low presentation quality, and there are errors and a lack of other clear elements, such as adequate scales and legends. By line 297, 84 references have already been cited, and the manuscript has not yet reached the results or discussion.

Figure 6 – Material and methods clearly describe the results; however, it has some image alteration/manipulation/lighting, with RGB colours apparently identified along the image in a gradient.

Training details? The materials and methods, results, and discussion are unclear. Lines 339-347. References???

The legends of Tables 1 and 2 and Tables 3 and 4 are not adequate, nor are the formatting, data, and description of the presented tables. Is this the result section? I do not think so; as the results section starts at 6, does not it? The Results section is not clear, with a clear presentation of the discussion.

The Discussion section was also inadequate. For example, what references were used? What advances have been made in this area? How is the proposed method suitable for theoretical and practical applications? The data have not been discussed. Furthermore, among all the lines 450-523, only references 5 and 69 were presented, inclusive with the presentation of the same researcher “Chapman, S.C.” For example, what were the advances in Machine Learning in your work?

The conclusion section “8. The Conclusion section,” with the exception of lines 536-539, is interesting; however, it was not what I observed throughout the manuscript of the authors. The idea was interesting, but the execution was inadequate. The references need to be updated. For example, 1980?

Of the 85 references, all were cited in the upper half of the manuscript.

There is a need to check adjectives that are used constantly and to adapt the language to a scientific style. Additionally, some sentences present redundancy and verbosity and can be simplified to express the same idea.

Author Response

Thank you very much for taking the time to review this manuscript. Your comments and suggestions have helped us to improve the manuscript substantially.

We have highlighted our changes in blue. All reviewers noted necessary improvements to language, so there are many changes throughout.

Please find detailed responses to each comment in the attached PDF. Where appropriate we include the specific wording changes in each comment response.

Author Response File: Author Response.pdf