The Prediction Model of Total Nitrogen Content in Leaves of Korla Fragrant Pear Was Established Based on Near Infrared Spectroscopy

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Introduction presents the topic of the paper: the use of near-infrared spectroscopy (NIRS) to establish an efficient, accurate and non-destructive detection method able to predict the nitrogen deficiency of pear trees.

Related work in terms of NIRS use to determine nutrients in plants have been presented. Some used NIRS to predict content of protein, nitrogen of phosphorus deficiency. However, there is no mention on the methods used (statistical regression or machine learning), the evaluation metrics and procedure to validate this and the results obtained. Please complete this section with specific related work and more details as mentioned, at least top 5-10 most relevant papers strictly related to the work in this paper (machine learning!). What datasets are they using? Results, methods, etc.

The collection of the dataset sample were from Korla Fragrant Pear in the fifteenth company of the ninth regiment 80 of Alar City, Xinjiang Production and Construction Corps, from different experimental treatments. The process of obtaining each spectrum is described in detail, as well as the process of determining the total nitrogen, and the interpretation of the spectrums. Anomalies were removes from the dataset using the Mahalanobis distance method. Wavelength preprocessing involved applying different derivatives. Please explain the chosen preprocessing methods and their necessity. What are other researchers in the field using, and why? How do they perform?

Section 3.5.2 Please explain what CARS algorithm is. Please explain what the RMSEC value refers to.

The method is very vaguely described. Section 3.6.1 what implementation of the random forest algorithm model, the genetic optimization random forest algorithm model, the radial neural network model, and 240 the extreme learning machine model were used and how was the data split? Any parameters to set? Validation/test data? Cross validation? Repetitions?

Please explain acronyms in the Figure 5 (ELM? GA-FR? RBF? )

Please describe what are the algorithms involved doing, and which version of implementation you are using.

Figure 6 is wrong. There is no a , b, c in the subgraphs, so it’s impossible to say which is which, especially as they are on separate pages. A figure should fit one page, otherwise, split in 2 figures on two pages.

“The fitting results of the measured values and predicted values of all samples;” – for which algorithm? How was this calculated? It is really unclear.

The discussion section presents again the summary of the paper and some related work, which some are unrelated (classification of apple leaves to recognize green apples ? it’s unclear why this study is here). The discussion should be something different – if should discuss the results, limitations of the study, how these results compare with related literature (compare some similar metrics), what novelty does this paper bring to the state of the art, advantages to other approaches, disadvantages, what is there still to do etc. Please rewrite this section based on these lines.

Please consider making the dataset publicly available.

Please consider clarifying the dataset characteristics and target attribute used in prediction.

Algorithm vs model - two different things, please clarify this to yourself and use appropriate terms in the paper. Also terms like training, testing, evaluation, validation, attributes, tuning, parameters, and prediction target are part of any paper that desires to perform machine learning regression. They describe important parts of the experiments, which are not present here.

The paper is an interesting research, but fails to clearly describe the methods and experiments, as well as the results obtained. I strongly recommend clarifying these points and resubmitting after a careful restructuring and editing of the paper and it’s figures.

-       

Comments on the Quality of English Language

Avoid using uncommon abbreviations in the abstract (SVN, SD, CARS, RBF), explain them first

-       Please rephrase abstract using common language (less use of + sign – as it makes little sense at this point) and summarizing the relevant aspects of the paper to be clear for someone reading it the first time

-       Algorithms vs model? 

-       Language is a bit hard to read, as some phrases start with a mixed order of words or use work terms (or definite article instead of indefinite). Pay attention to ‘the’ and ‘a’ (e.g. “In order to realize the rapid and non-destructive detection of total nitrogen content 17 in Korla fragrant pear leaves, the detection model was established by near infrared spectroscopy”)

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript presents an interesting investigation. The manuscript, in its current form, needs significant improvement. The methods and presentation of the obtained results should be improved.

Note to the manuscript:

Line 70 - Some abbreviations are not explained, for example, CARS, SPXY, etc.

Line 77-78 - “The Fourier near-infrared spectrometer (Antaris-II, USA) was used to scan the spectrum” – scan the spectrum or scan the samples?

Line 88-89 - How 28 different experimental treatments were collected? In the material and methods section, 7 fertilization periods and three nitrogen application levels were mentioned.

The part “Data processing” does not describe the methods used to create models for nitrogen content determination and the software used.

Line 145-146 – “The total nitrogen content of pear leaves at different growth stages can be calculated by the formula” - Which formula? Is the one presented on line 121 or another?

Line 180-181 – “In the experiment, a total of 180 pear leaf samples were removed by the Mahalanobis distance method” - What does this mean? How many samples did they work with then? The same is in lines 197-178.

Fig. 3.  - The quality of the figures is poor. The selected wavelengths are not clearly visible. The scale of the X-axis of the figures in the second column is reversed, but only for MSC pretreatment the graph is correct. There are no comments on the selected spectral information and what chemical bonds it is associated with.

Line 214-215 – fig 3 or fig 4?

In general, there is no explanation of what spectral information is used in the models. The calibration equations appear to be overdetermined, resulting in much larger errors of determination for the samples from the validation set.

Comments on the Quality of English Language

There are some unclear sentences.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Relate work section in introduction has been extended to include relevant works.

Achronyms description has been more attentively performed in the manuscript. However some make no sense - why GA-RF is Random Forest for Genetic Optimization?

Action 3 is titled results and analysis, but it presents the methodology. Please adjust title as it is misleading. It seems to present data analysis first - the sample data which are preprocessed and analysed (wavelengths etc).

The methodology section is not complete. The experiments are not described in what concers the training of the machine learning methods, the implementation used for the algorithms, hyperparameters values and validation methodology (cross validation ? train-test split? anything else relevant here?).  [Please describe what are the algorithms involved doing, and which version of implementation you are using - in the paper.]

Figures 6 and 7 are unclear because of PDF rendering, please consider a better rendered pdf for review, with simple highlight of changes, rather than complete markup which is hard to follow. Thank you!

The paper is an interesting research, but it still fails to clearly describe the methods and experiments, like the previous version, thus the scientific soundness is hard to evaluate. Please clarify if the focus of this paper is machine learning. Please use only English in your reply.

Author Response

Dear Editors and Reviewers:

Thank you for your letter and for the reviewers’ comments concerning our manuscript entitled “The Prediction Model of Total Nitrogen Content in Leaves of Korla Fragrant Pear was Established Based on Near Infrared Spectroscopy” (ID: agronomy-3008146).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 are marked in yellow in the paper. The main corrections in the paper and the responds to the reviewer’s comments are as flowing:

1. Achronyms description has been more attentively performed in the manuscript. However some make no sense - why GA-RF is Random Forest for Genetic Optimization?

The translation error of GA-RF was identified as stemming from inaccuracies in the translation process. In the revised manuscript, we have rectified 'Random Forest for Genetic Optimization' to 'genetic algorithm-based random forest', as indicated on line 388 of the paper and updated in the Abbreviations section.

2. Action 3 is titled results and analysis, but it presents the methodology. Please adjust title as it is misleading. It seems to present data analysis first - the sample data which are preprocessed and analysed (wavelengths etc).

The article introduces the method in Action 3 for presenting preprocessed and analyzed sample data, which includes details such as wavelength. Consequently, the title for Action 3 is identified as 'Data Analysis and Results' on line 291.

3. The methodology section is not complete. The experiments are not described in what concers the training of the machine learning methods, the implementation used for the algorithms, hyperparameters values and validation methodology (cross validation ? train-test split? anything else relevant here?).  [Please describe what are the algorithms involved doing, and which version of implementation you are using - in the paper.]

In order to address the incomplete method section lacking a description of the machine learning method, a Modeling methodology section has been added from lines 225 to 283. This new section details the training process, algorithm implementation, hyperparameters, and validation method used in this article.

4. Figures 6 and 7 are unclear because of PDF rendering, please consider a better rendered pdf for review, with simple highlight of changes, rather than complete markup which is hard to follow. Thank you!

Due to poor rendering in the PDFs of Figures 6 and 7, we have re-drawn the images. In Figure 7-c, 45 samples were not visible in the image due to a drawing error. This issue has been addressed in the revised manuscript.

5. The paper is an interesting research, but it still fails to clearly describe the methods and experiments, like the previous version, thus the scientific soundness is hard to evaluate. Please clarify if the focus of this paper is machine learning. Please use only English in your reply.

In the Modeling methodology section (2.5) of the new manuscript, the author clarifies that the algorithms discussed are all related to machine learning. Specifically, RF is categorized as an ensemble algorithm in machine learning, utilizing the bagging idea. ELM is a neural network algorithm. GA-RF is an optimization algorithm within machine learning. RBF is considered a deep learning algorithm for model building.

The manuscript has been carefully reviewed and revised to enhance clarity and coherence. Thank you for your valuable suggestions. If you have any further inquiries, please do not hesitate to contact us. We eagerly await your response.

Yours sincerely,

Mingyang Yu

Corresponding author:

Name: Jinhu Zhi

E-mail: [email protected]

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript has been improved but still needs some clarification.

Note to the manuscript:

Line 374-375 - The indices in the formulas should be corrected.

After the RMSE and RMSEP values, there must be units of measurement.

Fig.4 - The quality of the figures is poor. The selected wavelengths are not clearly visible.

Comments on the Quality of English Language

Moderate editing of English language required

Author Response

Dear Editors and Reviewers:

Thank you for your letter and for the reviewers’ comments concerning our manuscript entitled “The Prediction Model of Total Nitrogen Content in Leaves of Korla Fragrant Pear was Established Based on Near Infrared Spectroscopy” (ID: agronomy-3008146).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 are marked in yellow in the paper. The main corrections in the paper and the responds to the reviewer’s comments are as flowing:

1. Line 374-375 - The indices in the formulas should be corrected.

A correction has been made in the new manuscript at lines 422-423, and we sincerely apologize for any inconvenience caused by our previous oversight.

2. After the RMSE and RMSEP values, there must be units of measurement.

In this experiment, it is common practice to express the total nitrogen content of leaves by adding a percentage symbol (%) after the total nitrogen value, and by adding (%) after the RMSE and RMSEP values. This convention is followed in chapters 30, 58-60, 406, Lines 491, and 509.

3. 4 - The quality of the figures is poor. The selected wavelengths are not clearly visible.

In order to prevent the situation described in the question from occurring, this experiment emphasized the position of the selected wavelength in the image（Figure 4 of the“Spectral characteristic wavenumber extraction”）.

The manuscript has been carefully reviewed and revised to enhance clarity and coherence. Thank you for your valuable suggestions. If you have any further inquiries, please do not hesitate to contact us. We eagerly await your response.

Yours sincerely,

Mingyang Yu

Corresponding author:

Name: Jinhu Zhi

E-mail: [email protected]

Round 3

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript has been improved from the previous version in terms of providing relevant abbreviations and descriptions for machine learning algorithms. The experimental set-up has also been described, including algorithms parameters, train:test data split, and discussion of the results.

However, for the prediction result values to be relevant a cross validation would be preferable, with several repetitions. So from this perspective the scientific soundness is low.  While there is room for improvement in this regard, I can appreciate the magnitude and multidisciplinary nature of this research and the paper provides the minimum sufficient information for the purpose, considering all aspects.

The paper still does not motivate the choice of algorithms, nor does it provide some future work directions. The conclusion is very careless written with no clear conclusion for the purpose of the research, just focusing on experiments, but not their meaning for the research community.

I recommend further improvements before paper publication.

Author Response

Dear Editors and Reviewers:

Thank you for your letter and for the reviewers’ comments concerning our manuscript entitled “The Prediction Model of Total Nitrogen Content in Leaves of Korla Fragrant Pear was Established Based on Near Infrared Spectroscopy” (ID: agronomy-3008146).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 are marked in yellow in the paper. The main corrections in the paper and the responds to the reviewer’s comments are as flowing:

However, for the prediction result values to be relevant a cross validation would be preferable, with several repetitions. So from this perspective the scientific soundness is low.  While there is room for improvement in this regard, I can appreciate the magnitude and multidisciplinary nature of this research and the paper provides the minimum sufficient information for the purpose, considering all aspects.

Thank you for your feedback. As you mentioned, cross-validation is indeed crucial for validating the relevance of predicted results. In this study, we aimed to assess prediction error size and model explained variables' variation using root mean square error and coefficient of determination. We analyzed the variation degree by comparing the linear fitting of predicted and actual values, hence did not conduct cross-validation. Future research will focus on implementing cross-validation to enhance and validate the model prediction results.

The paper still does not motivate the choice of algorithms, nor does it provide some future work directions. The conclusion is very careless written with no clear conclusion for the purpose of the research, just focusing on experiments, but not their meaning for the research community.

The motivation for choosing the algorithm is explained in lines 231-241 of the new manuscript：Given the extensive amount of data in this experiment, despite preprocessing the spectral information and removing outliers, there is still a possibility of noise and outliers impacting the model's establishment. The Random Forest (RF) algorithm is known for its robustness in handling noisy data and outliers, although it may face challenges such as overfitting and local optimal solutions. To address these issues, the Genetic Algorithm-based Random Forest（GA-RF ）algorithm and the Extreme Learning Machine (ELM) algorithm are employed in this study. The focus of this experiment is on detecting the total nitrogen content of plant leaves, a task that requires considering the growth and development of plants. Due to the limitations of ordinary linear functions in this context, the Radial Basis Function (RBF) algorithm, known for its non-linear function approximation capabilities, is utilized.

The conclusion section of the new manuscript has been extensively revised to offer a more coherent explanation of the future research directions and goals.

The manuscript has been carefully reviewed and revised to enhance clarity and coherence. Thank you for your valuable suggestions. If you have any further inquiries, please do not hesitate to contact us. We eagerly await your response.

Yours sincerely,

Mingyang Yu

Corresponding author:

Name: Jinhu Zhi

E-mail: [email protected]

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript has been improved but still needs some clarification.

Note to the manuscript:

Line 135 – “The spectrum was analyzed using a Fourier near-infrared spectrometer (Antaris-II, USA)” – unclear. One spectrum? The spectra were analyzed or obtained using a Fourier near-infrared spectrometer.

Figure 7 c – In the title was written “The relative error between the measured value and the predicted value of the validation set sample”, but in the Y axes in the figure was mentioned absolute deviation.

Comments on the Quality of English Language

Moderate editing of English language required

Author Response

Dear Editors and Reviewers:

Thank you for your letter and for the reviewers’ comments concerning our manuscript entitled “The Prediction Model of Total Nitrogen Content in Leaves of Korla Fragrant Pear was Established Based on Near Infrared Spectroscopy” (ID: agronomy-3008146).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 are marked in yellow in the paper. The main corrections in the paper and the responds to the reviewer’s comments are as flowing:

Line 135 – “The spectrum was analyzed using a Fourier near-infrared spectrometer (Antaris-II, USA)” – unclear. One spectrum? The spectra were analyzed or obtained using a Fourier near-infrared spectrometer.

A Fourier transform near-infrared spectrometer (Antaris-II, USA) was utilized in the experiment to scan and collect spectral data from leaves. This update has been included in the revised manuscript on lines 139-141.

Figure 7 c – In the title was written “The relative error between the measured value and the predicted value of the validation set sample”, but in the Y axes in the figure was mentioned absolute deviation.

This issue has been addressed in the revised manuscript, and the title of section Figure 7-c has been changed to "The absolute deviation between the measured value and the predicted value of the validation set sample", in lines 444-446 of the new manuscript.

The manuscript has been carefully reviewed and revised to enhance clarity and coherence. Thank you for your valuable suggestions. If you have any further inquiries, please do not hesitate to contact us. We eagerly await your response.

Yours sincerely,

Mingyang Yu

Corresponding author:

Name: Jinhu Zhi

E-mail: [email protected]