Round 1

Reviewer 1 Report

I hope the paper will be useful for the scientific community. I wish the authors good luck. 

 

cheers,

Author Response

We appreciate the considerations and the revision performed in our manuscript. Your contribution was of high value to improve the quality of our work. Thank you.

Reviewer 2 Report

dear authors; formerly, I had problem with the size of samples to train the model; however, in this version you support with better, and solid references the experimental design. The results and discussion have a good presentation.

 

 

Author Response

We appreciate the considerations and the revision performed in our manuscript. Your contribution was of high value to improve the quality of our work. Thank you.

Reviewer 3 Report

The manuscript evaluates different machine learning methods on reflectance and 1rst derivative of reflectance to estimate macro und micro nutrients in citrus leaves. The novalty of this study is to apply the machine learning with 1rst derivative of reflectance. Hovewer, the given dependent calibration (n=256) and validation/test (n= 32 per each) data sets from one single orchard/field without validation with independent data from an other area or field demonstrates /results in an quite often published machine learning approach: the results are only valid for this field and can not be reproduced with the same methods on other fields. The reported results in this study are much better than in many other studies, however the application of those algorithms to another independent data set would complety fail  as usually reported. This study would have a higher value if these methods were validated with another citrus field, which is not located near by the calibration test site.

It is well known that hyperspectral data creates a high data amount of single wavelength which allows many combinations between single bands and a dependant nutrient. But the authors want to evaluate nutrients from in total 320 measured samples. The moist important point of criticism is the dependant data set what is even not mentioned and discussed!

It is unclear how many reflectance measurements were done with the ASD per each leaf/sample and if those were averaged. Or were just 320 measurements taken, so one single reflectance value per leaf (line 196)?

The given reference list shows that the authors did not sreen all publications related to micro- and macro nutrient estimation from hyperspectral reflectance measurements. Just four cited references have a direct link to macro and micro nutrients in field crops or citrus while many others have a link to just one single nutrient. This indicates that the focus here was more in machine learning than in understanding how macro and micro nutrients are linked to each other. The key word "precision agriculture" is here some how misleading because there are almost no links to the topic by citations and the term is even not explained. But it easily fits in this area.

Although the manscript was improved by adressing the reviewers' comments, it still need some significant modifications and explainations (see comments above). For the time beiing, the manuscript is more ready for a machine learning community (journal) than remote sensing with a link to precision agriculture.

 

Author Response

Thank you for the provided feedback. We have performed all the suggestions made. The contributions of this revision were vital to improve the manuscript quality. All of them were added to the new document. Please see the attachment below. We prepared a point-by-point response to the reviewer’s comments.

The manuscript evaluates different machine learning methods on reflectance and 1rst derivative of reflectance to estimate macro und micro nutrients in citrus leaves. The novalty of this study is to apply the machine learning with 1rst derivative of reflectance. Hovewer, the given dependent calibration (n=256) and validation/test (n= 32 per each) data sets from one single orchard/field without validation with independent data from an other area or field demonstrates /results in an quite often published machine learning approach: the results are only valid for this field and can not be reproduced with the same methods on other fields. The reported results in this study are much better than in many other studies, however the application of those algorithms to another independent data set would complety fail as usually reported. This study would have a higher value if these methods were validated with another citrus field, which is not located near by the calibration test site.

Thank you so much for this observation. We agreed with the observations pointed out. Because of that, we modified our manuscript description to a more limited range, and instead of applying this methodology to the “citrus” plants in a general sense, we restricted it to the Valencia-orange type, which was the plant evaluated in our study. We indeed used one single orchard/field to produce our data-set. Still, we made sure to collect and register different types of trees of different plantation fields during our field campaign. The differences in characteristics of these trees were demonstrated in Table 3 by the results of the chemical analysis, in which returned high coefficients of variation for all nutrients, a heterogeneous and a not-normal data-set. With this, we understood that, although not representative of the genre citrus (which we corrected it by modifying the title and every sentence regarding this theme, as previously explained), it should configure a data-set to represent different concentrations of leaf nutrients. In relation to the independent data-set, we believe that, since the goal was to predict different concentrations of nutrients, the testing data-set was appropriate, since neither of the algorithms was trained or had contact with none of the spectral wavelengths separated for the testing data-set (which, although from the same species and from the same orchard, were obtained from trees at different plantation fields and at different topographic and location conditions). Nonetheless, we understand the preoccupation regarding this theme and hope that this explanation provides a comprehensive idea of the main goal of the proposed approach. We sincerely thank you for this comment and hope that you find the alterations appropriate.

It is well known that hyperspectral data creates a high data amount of single wavelength which allows many combinations between single bands and a dependant nutrient. But the authors want to evaluate nutrients from in total 320 measured samples. The moist important point of criticism is the dependant data set what is even not mentioned and discussed!

Thank you for pointing this out. We have inserted a paragraph in Section 2.2 (as well as some sentences in 2.1) describing the conditions of the dependent data-set. Most of which was already stated in the previous commentary. We thank you kindly for indicating it. We think that this suggestion has helped us to improve our manuscript quality. Thank you for this recommendation.

It is unclear how many reflectance measurements were done with the ASD per each leaf/sample and if those were averaged. Or were just 320 measurements taken, so one single reflectance value per leaf (line 196)?

Thank you for this question. We have added a sentence in lines 192-193 explaining that, for each leaf, we registered 10 measurements in order to produce a mean spectral signature. This was a standard procedure to generate a more confident reading of a single target (leaf) with less noise or errors related to the equipment, operator and local illumination conditions. Thank you for this recommendation.

The given reference list shows that the authors did not sreen all publications related to micro- and macro nutrient estimation from hyperspectral reflectance measurements. Just four cited references have a direct link to macro and micro nutrients in field crops or citrus while many others have a link to just one single nutrient. This indicates that the focus here was more in machine learning than in understanding how macro and micro nutrients are linked to each other. The key word "precision agriculture" is here some how misleading because there are almost no links to the topic by citations and the term is even not explained. But it easily fits in this area.

Thank you for this feedback. Although we had some trouble encountering references in remote sensing journals indicating predictions of macro and micronutrients with machine learning frameworks, we see how the keyword “precision agriculture” is misleading, as pointed out. Because of this reason, we have modified it to a more fitting word, “artificial intelligence”. We, again, thank you kindly for the commentary and appreciate this observation. This is something that we will consider in future research regarding this theme.

Although the manscript was improved by adressing the reviewers' comments, it still need some significant modifications and explainations (see comments above). For the time beiing, the manuscript is more ready for a machine learning community (journal) than remote sensing with a link to precision agriculture.

Thank you so much for your detailed revision of this manuscript. We have made all the necessary changes as pointed. We believe that the contribution provided by the reviewers helped to improve this paper. We hope that you find the alterations appropriate.

Our kind regards.

Author Response File: Author Response.docx

Reviewer 4 Report

 

Pag.1, line 33-34

You say: We measured their spectral data …

Suggestion: Improve writing. Do not refer in the first person (we). In a scientific article the wording must be in the third person.

Idem page 1, line 38, 41, 42, 46 and 47 (We, Our) Writing only in third person.

 

Pag. 2 line 56: known as proximal sensors [18–22]. Proximal sensors can measure…

Suggestion: Improve writing, Do not repeat the same words in the same sentence (proximal sensors).

 

Page 3, line 135 and 138

You say: “The contribution of our approach is two-fold. Firstly, we demonstrate a method to indicate the most suitable spectra (reflectance/first- derivative) to model the nutrient content according to the algorithms' performance. Secondly, we determine important wavelengths…

Suggestion: If these are the objectives of this work, you must say: “The aims of this work are a) show a method to indicate the most suitable spectra (reflectance/first- derivative) to model the nutrient content according to the algorithms' performance; b) determine important wavelengths…

 

Page 3 line 138-141

You Say: …” The rest of this paper is organized as follows: Section 2 details the method adopted to produce this framework; Section 3 presents the results obtained with our method; Section 4 discusses the results and; Section concludes this paper”.

Suggestion: remove this paragraph

 

Page 4 line 162-163,

What is the topological arrangement?

 

Page 4, line 165 and 166

You Say: “The area is predominantly composed of red-yellow podzolic soil, situated in a Cwa

Köppen subtropical climate type unit…”

Suggestion: At least one reference is required.

 

Page 6,

Fig 2. If this is an example, I suggest eliminate this fig.

 

Page 8,

Fig 3. Structure of the machine learning architecture adopted in the proposed framework.

Suggestion: If Fig 3 is adopted in the proposed framework, it is necessary to include a reference.

 

Page 11

Improve the quality of the Figure 4 and eliminate the colours and words below the fig “Correlation test between the nutrients in the sampled citrus-leaves”.

 

Page 14

Figure 4 is repeated twice, it should be Fig. 5.

Fig. 4. Prediction comparison against laboratory measurements for the best algorithms’ results.

Suggestion: pay attention to the sequence of the figures and improve the quality of this figure.

 

Page 15

Figure 6 contains a lot of information and is not clear.

Suggestion: divide it in two Figs (Fig 6a and Fig 6b) and improve the wording of the title of the figure (Idem).

 

Page 18, line 480-483:

You say: “…In this paper, we have proposed a machine learning framework to predict nutrient content in citrus-leaves. Our approach uses leaf spectral data in the visible and near-infrared regions, and switches between reflectance and its first-derivative data to predict the amount of macro and micronutrient measured in the laboratory. Our method was…”

Suggestion: Remove “In this paper, we have proposed a machine learning framework to predict nutrient content in citrus-leaves” and improve writing it in the third person (IDEM)
(This approach ... This method). In general, improve the wording of the conclusions.

 

Finally, include suggested references.

Author Response

Thank you for the provided feedback. We have performed all the suggestions made. The contributions of this revision were vital to improve the manuscript quality. All of them were added to the new document. Please see the attachment below. We prepared a point-by-point response to the reviewer’s comments.

 

Pag.1, line 33-34

You say: We measured their spectral data …

Suggestion: Improve writing. Do not refer in the first person (we). In a scientific article the wording must be in the third person.

Idem page 1, line 38, 41, 42, 46 and 47 (We, Our) Writing only in third person.

We appreciate the reviewer’s suggestion. We have changed the writing style in the manuscript. Thank you for this recommendation.

 

Pag. 2 line 56: known as proximal sensors [18–22]. Proximal sensors can measure…

Suggestion: Improve writing, Do not repeat the same words in the same sentence (proximal sensors).

We appreciate the reviewer’s suggestion. We have changed the writing as pointed out. Thank you for this recommendation.

 

Page 3, line 135 and 138

You say: “The contribution of our approach is two-fold. Firstly, we demonstrate a method to indicate the most suitable spectra (reflectance/first- derivative) to model the nutrient content according to the algorithms' performance. Secondly, we determine important wavelengths…

Suggestion: If these are the objectives of this work, you must say: “The aims of this work are a) show a method to indicate the most suitable spectra (reflectance/first- derivative) to model the nutrient content according to the algorithms' performance; b) determine important wavelengths…

We appreciate the reviewer’s suggestion. We have changed the writing as pointed out. We believe that this has improved the presentation of our objective. Thank you for this recommendation.

 

Page 3 line 138-141

You Say: …” The rest of this paper is organized as follows: Section 2 details the method adopted to produce this framework; Section 3 presents the results obtained with our method; Section 4 discusses the results and; Section concludes this paper”.

Suggestion: remove this paragraph

We appreciate the reviewer’s suggestion. We have removed the paragraph. This has made the introduction of our manuscript clearer than before. Thank you for this recommendation.

 

Page 4 line 162-163,

What is the topological arrangement?

We appreciate the reviewer’s suggestion. We have added a phrase explaining the field arrangement in Section 2.1. Thank you for this recommendation.

 

Page 4, line 165 and 166

You Say: “The area is predominantly composed of red-yellow podzolic soil, situated in a Cwa

Köppen subtropical climate type unit…”

Suggestion: At least one reference is required.

We appreciate the reviewer’s suggestion. We have added a reference (see [50]) indicating this citation. Thank you for this recommendation.

 

Page 6,

Fig 2. If this is an example, I suggest eliminate this fig.

We appreciate the reviewer’s suggestion. It was a mistake on our behalf to use the word “example” in this sentence. We removed this word and maintained the figure since these are in fact the 32 curves used in the testing set. Thank you for this recommendation.

 

Page 8,

Fig 3. Structure of the machine learning architecture adopted in the proposed framework.

Suggestion: If Fig 3 is adopted in the proposed framework, it is necessary to include a reference.

We appreciate the reviewer’s suggestion. It was a mistake on our behalf to use the word “adopted” in this sentence. We remove this word. Thank you for this recommendation.

 

Page 11

Improve the quality of the Figure 4 and eliminate the colours and words below the fig “Correlation test between the nutrients in the sampled citrus-leaves”.

We appreciate the reviewer’s suggestion. We believe we have modified the figure accordingly and think its quality has been improved. Thank you for this recommendation.

 

Page 14

Figure 4 is repeated twice, it should be Fig. 5.

Fig. 4. Prediction comparison against laboratory measurements for the best algorithms’ results.

Suggestion: pay attention to the sequence of the figures and improve the quality of this figure.

We appreciate the reviewer’s suggestion. As further suggested in Fig 6, we have also divided this Figure into two Figs. We believe that this suggestion was also of value for this Fig 5 since it helped us to improve its presentation. Thank you for this recommendation.

 

Page 15

Figure 6 contains a lot of information and is not clear.

Suggestion: divide it in two Figs (Fig 6a and Fig 6b) and improve the wording of the title of the figure (Idem).

We appreciate the reviewer’s suggestion. We divided the figure into tow Figs and modified its title. We believe that this has helped us to improve its quality. Thank you for this recommendation.

 

Page 18, line 480-483:

You say: “…In this paper, we have proposed a machine learning framework to predict nutrient content in citrus-leaves. Our approach uses leaf spectral data in the visible and near-infrared regions, and switches between reflectance and its first-derivative data to predict the amount of macro and micronutrient measured in the laboratory. Our method was…”

Suggestion: Remove “In this paper, we have proposed a machine learning framework to predict nutrient content in citrus-leaves” and improve writing it in the third person (IDEM)

(This approach ... This method). In general, improve the wording of the conclusions.

Finally, include suggested references.

We appreciate the reviewer’s suggestion. We modified text accordingly and also performed an evaluation of the subsequent phrases following the reviewer’s recommendations. Thank you for this recommendation.

 

Thank you so much for your detailed revision of this manuscript. We have made all the necessary changes as pointed. We believe that the contribution provided by the reviewers helped to improve this paper. We hope that you find the alterations appropriate.

Our kind regards.

Author Response File: Author Response.docx

Round 2

Reviewer 3 Report

The authors have siginificatly improved the manuscript. Attached some relevant references for your study:

Barker and Pileam (2007). Handbook of plant nutrition

Obreza and Morgan. Nutrition of Florida citrus trees

Jull et al. (2018) Nutrient quantification in fresh and dried mixtures of ryegrass and clover leaves using laser-induced breakdown spectroscopy. Precision Agriculture, 19, 823-839.

Galvez Sola et al (2015). Rapid estimation of nutritional elements on citrus leaves by near infrared reflectance spectroscopy. Frontiers of Plant Science

Johnson et al. (2013).Identification of Water Stress in Citrus Leaves Using Sensing Technologie. Agronomy, 3, 747-756

 

 

Author Response

Thank you very much for this feedback. We appreciate all the help offered to improve our manuscript. We have read and inserted all the suggested references. Please see the attachment below.

 

Best regards,

Lucas.

Author Response File: Author Response.docx