Representation Learning with a Variational Autoencoder for Predicting Nitrogen Requirement in Rice

Round 1

Reviewer 1 Report

In the work: "Representation Learning with a Variational Autoencoder for Predicting Nitrogen Requirement in Rice," a prediction rice yield model was constructed based on soil and climatic data, remote sensing indices, and management practices using Explainable Artificial Intelligence techniques.

Interesting conclusions are drawn, with possible applications in agricultural modelling, although they suggest some changes and some questions arise for this reviewer:

1) In the abstract, acronyms are used without defining the acronym: ML. An abbreviations list could be added to show all abbreviations used.

2) The statement (line 75): "Note that in case of high temperatures, the fertilization rates should be higher than normal and vice versa" should be referenced and explained for the case of rice, please.

3) Please provide a figure showing the experiment's location in the region (for example, study area, rice crop, and local administration limits).

4) In line 195, it is indicated that the images from 2017 that have been used are TOA. Why not use BOA (Surface Reflectance) images that are available (or they can be corrected) and are more accurate? Comment and justify the choice, please.

5) In table 1 learning control parameter: min_data_per_group is not shown.

6) Adding a block diagram showing the entire data treatment and analysis process is recommended.

7) There are numerous errors of reference to figures, for example, in lines 388, 404,417, 421, 423, 426, 462, 466, 473, and others. This makes it difficult to read the article.

8) Typographical errors are detected in the lines: 69, 134, 143, 268, 341, 560...

Comments for author File: Comments.pdf

Author Response

Please, check the attached file.

Author Response File: Author Response.pdf

Reviewer 2 Report

The authors note that meteorological conditions have an important impact on rice yield. The manuscript developed a new prediction model to improve nitrogen management by incorporating the predicted climate data. The paper was well written, and much of it was well described, especially for long-time and large-scale field data collection. Because of this, the current study is on a topic of relevance and general interest to research the fusion of remote sensing and crop models. Therefore, a Minor revision is warranted.

 

1、 The manuscript submitted for review has obvious typesetting errors, including the citation index. Please improve.

2、 This manuscript has a few suggestions for the author's reference. In field production, the achievable crop yield varies due to the differences in climate conditions between years. The achievable yield required for N fertilizer also varies. Therefore, combine RS growth monitoring, crop models and predicted meteorological conditions to predict achievable yields more accurately and then determine N management recommendations. It is indispensable.

Author Response

Please, check the attached file.

Author Response File: Author Response.pdf

Reviewer 3 Report

Dear authors

Thank you for giving me the opportunity to read your work. The manuscript “Representation Learning with a Variational Autoencoder for Predicting Nitrogen Requirement in Rice” is very interesting and denotes a lot of work. The study of food production (and food security), especially for studying nitrogen needs is a relevant topic in research. I really like the theme. Congrats on that. References are generally adequate and up to date.

However, I have to say that in the end of the introduction is not clear what kind of novelty your work brings. As main opinions I have to say that, despite this is an article about remote sensing and submitted to a high impact factor remote sensing journal, there is almost no work done in remote sensing (just two indexes), this is clearly a paper about modelling either a remote sensing research. Within the modelling procedure it is not clear how authors deal with data with such different spatial resolutions.

As less general comments I would like to point that:

Line 11 - Authors use Explainable Artificial Intelligence (AI) in abstract and eXplainable Artificial Intelligence (XAI) in keywords. I prefer as it is in the keywords.

Line 21 - I understand that ML stands for machine learning?

Line 127 - Personally I found the sentence "The dataset was organized and managed in term of a geodatabase in ArcGIS Pro" unnecessary.

Line 157 - Please indicate how the resolution 0.1° x 0.1° translates in meters at the equator. Its also important to explains how this cell size relates to a study area of s 3.18 ha.

Line 195 - "Top of atmosphere for 2017". This data has any use?

Lines 221-222 - grid of 30x30m was ... A 30-m grid interval. Authors are repeating the same information.

Line 253 - "The subset of the initial variables that remained after removing the low importance features

were checked for collinearity using Spearman’s rank correlation". Some authors claim that one not need to do this procedure in ML. But if we do it should be in the beginning. Low importance variables can be due to multicollinearity. In this cases VIF will be preferable to spearmen.

Line 336 - From this line forward all references are "Error! Reference source not found..". This turns the scientific reading very difficult.

Line 366 - Figure 1 has very poor quality.

Line 444 and 451 - Figure 3 and 4, the names of the variables should be written properly, or else give them codes and explain them in the figure caption.

I recommend you read "Evaluation of the factors explaining the use of agricultural land: a machine learning and model-agnostic approach. Ecological Indicators. 131, 108200" which can be useful.

And that’s all, sorry, for being so picky and continue the good work

Best regards

Author Response

Please, check the attached file.

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

Dear authors

First of all sorry for taking more time than expected in the second review. However, this is because I have paid careful attention to your extensive changes in the text and complete and clear answers. It looks like I was the annoying reviewer: sorry about that.  

I must congratulate you for the extremely careful job and for the great number of changes. I would take the chance to called improvements, because they in my opinion are and if you agree to make such changes is because you somehow agree. This to say that is not usual for me to pass from a major revision directly to an acceptance. However, in this case in conscience I have to do it.

I just have two observations. First, regarding line 253, I am sorry but I do not agree with you. In my modest opinion Spearman rank correlation by itself is not  appropriate  for removing multicollinearity. However, as you well said, ML methods deal very well with this issue so I will not stress this issue. Second, is just a matter of appearance, in line 444 and 451, is just a matter of opinion. For me it not look good to have a name in a graph or map like we have it in a spreadsheet (e.g., Temperature_August) but has I said is a all about personal taste.

I hope you do not find my comments to much meticulous. Instead, I like you to regard them as a way to give some suggestions and raise some questions intending to improve the article.

Continue the good work and I hope you would come to publish the paper.

Best regards