Next Article in Journal
Bale Ensiling Preserves Nutritional Composition and Phenolic Compounds of Red Grape Pomace
Next Article in Special Issue
ViT-RoT: Vision Transformer-Based Robust Framework for Tomato Leaf Disease Recognition
Previous Article in Journal
Application of Geotechnologies in the Characterization of Forage Palm Production Areas in the Brazilian Semiarid Region
Previous Article in Special Issue
Predicting Pineapple Quality from Hyperspectral Data of Plant Parts Applied to Machine Learning
 
 
Article
Peer-Review Record

Utilizing TabPFN Transformer with IoT Environmental Data for Early Prediction of Grapevine Diseases

AgriEngineering 2025, 7(6), 173; https://doi.org/10.3390/agriengineering7060173
by Nikolaos Arvanitis 1,2,*, Filippo Graziosi 3, Gina Athanasiou 1, Antonia Terpou 4, Olga Arvaniti 4 and Theodore Zahariadis 1,4
Reviewer 1: Anonymous
Reviewer 2: Anonymous
AgriEngineering 2025, 7(6), 173; https://doi.org/10.3390/agriengineering7060173
Submission received: 11 March 2025 / Revised: 1 April 2025 / Accepted: 9 April 2025 / Published: 3 June 2025

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This study aims to develop a machine learning-based framework for early prediction of grapevine diseases using Internet of Things sensors. We thank the authors for presenting this study, but it has serious shortcomings in several areas, including the novelty of the study, which distinguishes it from previous studies in the same field. Also, the study's linguistic formulation is weak and the inclusion of useless information. Furthermore, the results and discussion sections are very poor. Here are some detailed comments:
1. The abstract needs a complete rewrite (150-250 words, outlining the study's purpose, methods, key findings, and conclusions, while avoiding abbreviations, footnotes, and references).
2. In the abstract, what is this formulation? (Plant diseases caused by pathogens that affect vineyards can cause serious damage and lead to reduced quantity and quality of fruit.)
(Just a suggestion: Downy mildew and powdery mildew are among the most serious diseases affecting grapevines. They can cause severe damage such as yield loss, altered grape size, impaired sugar accumulation, and negatively affect the flavor and aroma of the fruit. Early prediction of these diseases is crucial for the timely application of treatments, which can help mitigate their impact on crop production. Therefore, this study aims to present a machine learning-based framework for early prediction of grapevine diseases using IoT sensors.)
After that, it is best to briefly explain the most important methods used and the most important results and conclusions.
3. Both Section 2. Impact of downy mildew, powdery mildew disease to grapevines, and Section 3. Effects of environmental conditions on enabling grapevine diseases), there is absolutely no need to make them separate from the introduction and include them as separate sections. It is better to include them in the introduction in the appropriate way.
4. The manuscript needs linguistic revision.
5. Despite the (useless) expansiveness in section 4.1.2. IoT Environmental Data Acquisition & Labeling, you did not clarify exactly which parameters you measured, what type of sensors were used, what type of IoT platform was used, and what measurements were applied to the vineyards to demonstrate the correlation between these parameters and diseases.
6. Figure 1. The workflow does not clearly outline the methodology used to predict plant diseases. Please reproduce this figure to explain the methodology in detail, from data monitoring to disease classification and model evaluation.
7. What measurements were taken on the vineyards that demonstrate disease infection and represent the (y-axis) in the model inputs? For example, was the percentage of downy mildew leaf area recorded, digital images of the disease taken, or disease spores collected? You may have included these, but I missed them, so please clarify.
8. The discussion section is very poor and needs to be reformulated and supported by citing relevant studies in the same field to highlight the novelty of the study.

Comments on the Quality of English Language

The English could be improved to more clearly express the research.

Author Response

Comments 1 & 2: [1. The abstract needs a complete rewrite (150-250 words, outlining the study's purpose, methods, key findings, and conclusions, while avoiding abbreviations, footnotes, and references).
2. In the abstract, what is this formulation? (Plant diseases caused by pathogens that affect vineyards can cause serious damage and lead to reduced quantity and quality of fruit.)
(Just a suggestion: Downy mildew and powdery mildew are among the most serious diseases affecting grapevines. They can cause severe damage such as yield loss, altered grape size, impaired sugar accumulation, and negatively affect the flavor and aroma of the fruit. Early prediction of these diseases is crucial for the timely application of treatments, which can help mitigate their impact on crop production. Therefore, this study aims to present a machine learning-based framework for early prediction of grapevine diseases using IoT sensors.)
After that, it is best to briefly explain the most important methods used and the most important results and conclusions.]

Response: The abstract has been modified to also inlcude (shortly) the methods and key-findings. The problematic formulation has been removed and the first part of the abstract has been changed considering your suggestion.

 

Comment 3: Both Section 2. Impact of downy mildew, powdery mildew disease to grapevines, and Section 3. Effects of environmental conditions on enabling grapevine diseases), there is absolutely no need to make them separate from the introduction and include them as separate sections. It is better to include them in the introduction in the appropriate way

Response: Sections 2 & 3 were written separately from the introduction to emphasize on the importance of early predicting the two diseases using environmental parameters that enable the diseases. However, the two sections have been moved to the introduction section.

 

Comment 4: The manuscript needs linguistic revision.

Response: Indeed, there were some sentences in the document that were not meaningful. This has been changed.

 

Comment 5:  Despite the (useless) expansiveness in section 4.1.2. IoT Environmental Data Acquisition & Labeling, you did not clarify exactly which parameters you measured, what type of sensors were used, what type of IoT platform was used, and what measurements were applied to the vineyards to demonstrate the correlation between these parameters and diseases.

Response: We now include all available information we had about the IoT system (sensors and platform) and the monitored parameters. The correlation between these parameters and the two diseases stand due to the fact that the annotation of a disease presence refers to when (fungicide) treatment was applied to the grapevines and this is well noted now in the document.

 

Comment 6: Figure 1. The workflow does not clearly outline the methodology used to predict plant diseases. Please reproduce this figure to explain the methodology in detail, from data monitoring to disease classification and model evaluation.

Response: Figure 1 has been improved. It describes all the steps (monitoring, data augmentation, preprocessing, training of the ML model). I could not include more information to a figure, because it is a figure. Also the caption now is more informative.

 

Comment 7: 7. What measurements were taken on the vineyards that demonstrate disease infection and represent the (y-axis) in the model inputs? For example, was the percentage of downy mildew leaf area recorded, digital images of the disease taken, or disease spores collected? You may have included these, but I missed them, so please clarify.

The y-axis of whether there is or not high risk of a disease presence (binary labeling) comes from the historical records of when fungicide treatment was applied. The annotators also labeled some days before the actual treatment as highly possible disease days given the respective environmental parameters. This has been clarified now.

 

Comment 8: 8. The discussion section is very poor and needs to be reformulated and supported by citing relevant studies in the same field to highlight the novelty of the study.

Response: I do not think the discussion was poor; I had included two paragraphs about limitations and how the suggested pipeline could be improved considering the results. Some similar works mentioning how our work distinguishes from these are included as related work at the introduction.

Reviewer 2 Report

Comments and Suggestions for Authors

The paper focuses on the use of iot environmental data and machine learning methods for early prediction of vine diseases, an important agricultural issue, which has positive implications for improving grape yield and quality and reducing pesticide use. Before the paper is published, I think the authors still need to solve some problems. Here are my comments:


1. The Internet of Things sensor is mentioned and used in the paper to collect temperature, humidity and rainfall information, but the sensor model and related measurement accuracy parameters are not clearly defined in the paper. It is recommended to complete this part of information.
2. It is mentioned in the paper that Gaussian Copula and Additive Gaussian Noise are used for data enhancement, but the problem of missing values in the original data is not dealt with in the paper. It is suggested to complete the missing data and the data filling method adopted.
3. In the data standardization part of the paper, it is simply mentioned that each feature is converted into a distribution with zero mean and one standard deviation by using standardized methods. It is suggested to supplement the standardized specific operation process.
4. The model performance in this paper is not deep enough. It is suggested to supplement the model architecture and principle.

Author Response

Comment 1: 1. The Internet of Things sensor is mentioned and used in the paper to collect temperature, humidity and rainfall information, but the sensor model and related measurement accuracy parameters are not clearly defined in the paper. It is recommended to complete this part of information.

Response: The sensor model and any related information we had available are now included in the respective subsection.

 

Comment 2: 2. It is mentioned in the paper that Gaussian Copula and Additive Gaussian Noise are used for data enhancement, but the problem of missing values in the original data is not dealt with in the paper. It is suggested to complete the missing data and the data filling method adopted.

Response: No missing or NaN values were included at the dataset, so there was no need for such a preprocessing. However, I wrote a sentence clarifying this, because it is something common to have missing values when you work with real-world tabular data.

 

Comment 3: 3. In the data standardization part of the paper, it is simply mentioned that each feature is converted into a distribution with zero mean and one standard deviation by using standardized methods. It is suggested to supplement the standardized specific operation process.

Response: I have included the formula of the standardization process. Initially I did not write it because it is a common normalization step at machine learning works.

 

Comment 4: 4. The model performance in this paper is not deep enough. It is suggested to supplement the model architecture and principle.

Response: All ML models used in the experiments are known and there was no need to analyze them further. But after your comment, I elaborated more on TabPFN-Transformer which is a recent method and was the one that performed better among the other models in our work.

 

 

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The esteemed authors have addressed all the concerns I raised previously.

Back to TopTop