A Machine Learning Model for Early Prediction of Crop Yield, Nested in a Web Application in the Cloud: A Case Study in an Olive Grove in Southern Spain

Round 1

Reviewer 1 Report

General comments:

The manuscript presents an interesting description of an expert system for olive producers in which a yield prediction model based on meteorological data from a single measuring station is implemented.

 

The authors set themselves two objectives, one concerning the yield prediction algorithm and the other concerning the development of an expert system. A major problem for the authors is deciding which objective is more important, and there is an imbalance between the approaches and descriptions in the following chapters. I would suggest that the authors focus more on the prediction algorithm than on the expert system. It is also argued that the expert system has not been tested on other sites, by different users. A description of the expert system with test results would be the subject of a separate manuscript.

 

The methodology of the study requires additions. Details are given below.

 

The major shortcomings are evident in the discussion chapter. The content presented is not a critical scientific discussion of one's own results in the context of the results of other authors.

 

Specific comments and suggestions:

In the introduction, the authors describe the applications of AI in many places. In the methodology, results and discussion this element is omitted.

 

L100: “It should be borne in mind that the harvest is normally collected between the months of December-January, with January beinging the month in which pruning, tillage...works begins." The sentence needs to be corrected. The use of an ellipsis seems unjustified.

 

L147" The methodology followed in this work (...)" The whole paragraph up to L195 is methodology. Should be moved to methodology

 

L203" The olive orchard studied is 9.5 ha with 600 olive trees of similar properties, they are all about 25 years old and their mean fruit harvest is 60 kg/year."

600 trees x 60 kg/tree = 36,000 kg / 9.5 ha = 3,789 kg/ha; These values seem to contradict the data in Table 1. Perhaps it is a typo error.

 

Orchard description. A description of the soil conditions is missing. There is no description or reference to the cultivation technologies and agrotechnical treatments used. A reference is sufficient.

 

L207: Figure 2 (a) Please indicate the source of the data on the olive crops shown in the region.

Figure 2 (b). Please provide the source of the image.

 

Table 1. In the literature, crop yield is expressed as 'yield data' not as 'harvest data' Applicable to all paper .

Yield is usually expressed in kg /ha. Here it is yield per 9.5 ha. I propose to change.

If the harvest is done December-January then the yield value probably refers to the growing season 2013, 2014, not the years.

How was the harvest done? By hand? By machine? By combine? How was the yield measured?

Methodology for measuring yield, its errors are important for the evaluation of the study.

 

Metorogical data: It is worth informing what is the standard of measurement of meteorological data in public net . Is it a national standard? International? Are the sensors accredited?  In general, the reader does not know what the standard of measurement of meto data is and whether it has changed over the years in these 8 years. 

 

At what distance was the meteostation from the farm?

The paper uses a single source of data from a single meteo station. So therefore why does it mention "multi-source web services"?

 

Table 2. 30mm >> 30mm ; Km/h >> km/h

 

L255 "Another important phase in the process of generating the predictive model is the determination of the level of influence of the variables used on the target attribute. In this case, the Minimum Description Length (MDL) algorithm [39] is used.". This sentence gives references to the MDL algorithm hence lines L256-263 seem redundant.

 

 Three regression models were used in the model. That is, no AI or ML methods were used. So what is the reason for the description of ML and AI in the introduction and background?

 

The methodology used for the use of regression algorithms is very modestly described and has shortcomings.

 

L306: This action is performed by the user through the application by a user-friendly interface, Figure 4." Figure 4. shows a diagram of the expert system, not the user interface.

The issue of 'detection of anomalous data' in terms of yield data needs clarification. The description provided in L410-424 is ambiguous as to what the model for detection of anomalous data in yield data ultimately looks like.

 

Chapters 3.4.1 to 3.4.3 need to be rewritten and adapted to the style of the whole manuscript. The omission/deletion of common information is worth considering.

 

L426: Figure 8. Please express yield in kg/ha

 

L523: Table 5. The fifth column shows relative error rather than absolute error.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

Dear Authors,

I revised the manuscript "Web application in the cloud for early prediction of crop yield. A case study in an olive grove in southern Spain." submitted to Agriculture journal. The manuscript is interesting. In my opinion, the authors concentrated too much on describing the web application instead of describing the research material and methods used more widely. In addition, some of the information in the section "4. Results and Analysis" should have been included earlier in the section "3. Materials and Methods". Unfortunately, the weakest part of the manuscript is the discussion of results. In addition, I have some concerns which need to be addressed.

 

Minor comments:

1. Line 1. Use only one word "Article".

2. Abstract. In the abstract, it is worth adding the most important results of the study.

3. Keywords. It is worth adding words regarding the prediction method used.

4. Section "2. Background". In my opinion, section 2 can be included in section 3.

5. Figure 1 adds nothing to the work. It can be deleted.

6. At the end of section 2, I suggest adding a figure showing the data flow - see the example at https://doi.org/10.3390/agronomy11050885 - Figure 2. In the next section, you present fig. 3, which is similar to the one I suggest.

7. Section " 3. Materials and Methods". Add information about the data analysis software used (except Oracle).

8. Line 204. How does the value of 60 kg/year relate to the yields in Table 1. Check this value because 600 x 60 kg/year = 36000 kg.

9. Table 1. I suggest changing the word "data" to "yield" in this table.

10. Table 2. In this table, add a column with a data range.

11. In the manuscript, there is no subsection 2.2. There is only section 2, and within it, item 2.

12. Figure 4 adds nothing to the work. It can be deleted.

13. Table 4. What unit of measurement is used for the column "Level of influence on target"?

14. Figure 10. Instead of the figure, a table should be inserted.

15. Table 5. Is "Absolute error" a MAPE?

16. Figure 12. Use a bar graph to compare actual and predicted yields.

17. Section "References". The style of the references is not in accordance with the requirements of Agriculture. I suggest using bibliography software package like Mendeley, Zotero, EndNote etc. Check out the instructions for authors at https://www.mdpi.com/files/word-templates/agriculture-template.dot

 

Major comments:

1. Section "3. Materials and Methods". This section lacks detailed descriptions of the algorithms and methods used (statistical and ML). Equations could also be added. In addition, information on yield prediction validation methods and determination of prediction errors (e.g. MAPE, MAE, RAE, RMS ...) should be added.

2. Section "3. Materials and Methods". In this section, you must describe in detail the materials and research methods you used. Unfortunately, often information about methods is presented in the next section.

3. Subsection "3.2. Dataset". In this subsection, add a table with all the variables and data ranges used to build the models. It is not clear what data you used to build the models. Was the soil data also used to build the models?

4. Section "5. Discussion". This section must be greatly expanded. You must also compare your results with other papers. Scientific work needs a real discussion of results! Additional references are also required.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Dear Authors,

Round 2 comments and suggestions are presented in the attached file.

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Dear Authors,

Most of my earlier comments were addressed correctly.

I still have doubts about two aspects of the paper:

1. There is still no specific information about determination of prediction errors (e.g. MAPE, MAE, RAE, RMS ...).

2. Section "5. Discussion". This section should still be expanded. Additional references are also required.

 

In summary - as it stands, the paper is most suitable for publication as "Communication". If the authors want to publish the paper as "Article", my earlier comments should be properly addressed.

Author Response

Please see the attachment

Author Response File: Author Response.docx