Wavelength Selection Method Based on Partial Least Square from Hyperspectral Unmanned Aerial Vehicle Orthomosaic of Irrigated Olive Orchards

Round 1

Reviewer 1 Report

Reviewer Blind Comments to Author

The paper presents a contribution to the field of classification methods from hyperspectral data in a very specific context (two types of irrigation techniques in olive orchards). In general, the article is well-organized and contains all expected components, from introduction to conclusions. In general, all sections are well-developed and clearly explained. English is not my native language but I feel that the article is well-written and easy to understand. The article does not supposes a major advance but a useful contribution to the knowledge base. Some minor changes are proposed, expecting to improve the clarity of the performed research ant their results.

Minor suggestions

(1) Title. Given that the framework if very specific (classification of only two types of irrigation) the title should also reflect this. For example: “Partial Least Square based wavelength selection for classification of two irrigation techniques of olive orchards from Unmanned Aerial Vehicle orthomosaick”, or “Partial Least Square based wavelength selection for olive orchards irrigation classification from Unmanned Aerial Vehicle orthomosaick”, etc.

(2) Keywords: I miss specific terms related with the core of the research, such as “PLS”, “wavelength selection”. Please revise.

(3) The amount of references if high (108). Although providing references is always positive, I feel that they could be condensed. Only in section 1 (Introduction, two pages, 89 lines) there are 72 references. I suggest authors to revise if all of them are relevant.

(4) The introduction is focused and it allows the reader to easily understand the scope in which the authors make their proposal. Nevertheless I have to make two comments for a improvement.

In line 102 it is said that “wavelength selection method is essential to apply”, and the reference 39 is cited. Given that this fact is the reason for the performed research, author should explain why wavelength selection is so important.

In lines 123-126 the contribution of the article and their objectives of the performed work is summarized. But is made so briefly that contrast sharply with the previous text in the introduction. I miss:

- A better presentation of the contribution. Only is said that “we report (…) the use of selection methods”, which tells nothing or very little.

- Which irrigation systems are that commonly used in olive orchards (to mention them at least).

- A better concretion of the objectives. “The evaluation of 16 methods” is not an objective but a mean of achieving them. I feel that several specific objectives could be stablished.

- It is a common practice to summarize the contents of the sections, which could help the authors to improve this part. This allows presenting the reader a comprehensive view of the whole paper and the performed research.

(5) In line 129 the location is cited as Marchena in the province of Seville. But in Figure 1.a the province of Seville is missed and the Andalusia region is included. Also the red dot in Figure 1.a is not placed in the correct location but in the approximate center of the Region. I suggest including a bigger figure to introduce details if necessary. I miss in Figure 1.b an image of the commercial orchard (7-ha) with an indication of the representative selected area.

(6) The work flow presented in lines 140-147 should be included in other section that “study area and UAV flights”. Also observe that section 2.1.1 is missing.

(7) In line 143 it is said that “different PLS- and non-PLS-based methods are used”, but in the flowchart in Figure 2 the corresponding step is “PLS methods”. Please correct this.

(8) In line 146 the classification methods are presented, which are LDA and KNN. There is a lack of an adequate justification of this selection. Why these methods? Why only two? Observe also that the introduction says nothing about them. Also observe that, while the authors include an high amount of references, here there is none which could justify why these only two classification methods are selected.

(9) In lines 146-147 it is said “evaluating the quality and efficiency of each method”. The word “method” could be ambiguous. It could made reference to “irrigation method” (line 145), PLS and non-PLS selection methods (line 144) or the classification methods. Given that the results depends on the combination of data+selection method+classification method, I would be more careful with the terms. For example “evaluating the quality and efficiency of the results”. In addition, I would reduce ambiguity if, for example, the term “irrigation techniques” is always used, like lines 31 and 40 (instead of “irrigation methods”). What is more, I suggest revising the whole text in order to disambiguate the term “method”, if needed.

(10) In line 198 it is indicated that SRTM is used. Since in Spain there are other DEMs available, with higher resolution and accuracy, a justification of use of SRTM is missed. Also I miss a reference when using third party data.

(11) First time it is introduce the term “calibration phase” is in line 290. It should be clearer what is made in this phase and in the “prediction phase”. Observe that Table 2 speaks about “calibration set” and “prediction set”, that is to say, sets and not phases. It is some confusing because in line 290 it is said that the results of the classification if applied in the calibration phase. Does not the classification need firstly the calibration? Please clear all this part of the research.

(12) In Section 2.5 the packages used for each PLS and non-PLS method are listed but we don’t know which package/s is/are needed for each method. It should be clearer if a table is included with this information.

(13) I consider that section 3.1 is essential to understand the utility of hyperspectral data in the context of this study. Figure 3 shows that both reflectance curves are very similar, therefore multispectral data are not enough for the pursued goal. In consequence, I suggest moving lines 315-321 to the end of the introduction section. This is related with my comment #3.

(14) I suggest moving text from lines 321-326 (including Table 2) to the Materials and methods section. This part is not related with results but with methodology.

(15) Related with #13, it should be justified, better with references, why 75% of trees are used in calibration and 25% in prediction. It seems to me that the capacity of extrapolation (prediction) of the applied methodology is not well-founded with such proportions. Furthermore, it should be better to have introduced different percentages in the research, analyzing the influence of the size of both sets (calibration and prediction). Also there is a lack of information of the spatial distribution of both sets. A figure from the QGIS digitized trees is suitable here, with an explanation of the spatial distribution criteria.

(16) Following suggestions #12 and #13, the section 3.2 could be renumbered to section 3.1

(17) In table 2, the term “summary” is bad used. I think the author are referring to “sum” of “total”.

(18) In table 2, also in line 322, the labels 1 and 2 do not contribute for a better understanding, since the columns are named as “sub-surface irrigation” and “surface irrigation”. Please remove.

(19) In line 329 it is said that depending on the selection method employed, the number of selected wavelengths vary considerably. This seems to say that the number of wavelengths depends only in the selection method. That is to say that each selection method offers a fix quantity of selected wavelengths. It does not also depends on parameters that control the selection method, ore any threshold for any measure? Please improve this part.

(20) I see a problem with the name of section 3.2 (line 327). It speaks about “performance”. What is performance? The term “performance” is used several times along the text but never defined and explained how to measure it. Only the term “efficiency” is defined in lines 294-297. In addition, I also see a problem with the name of the section since non-PLS method are also considered in the analysis. I wonder if a simple title of this section as “Wavelength selection results” is enough.

(21) In caption of Table 3 the term “effective wavelength” is introduced. But it is not used in the main text of the article. Please define what is an effective wavelength ore remove “effective” from the caption.

(22) In lines 340-344 an “All-together method” is stablished by de authors. Nevertheless is not clear the way it is applied. In lines 340-341 they say Figure 4 show the total number of times a wavelength has been selected by the total number of methods and they define it as “All-together method”. This is not a selection method but only a description of Figure 4. In lines 341-342 it is said that 18 wavelengths were selected by half of the methods, but this fact is not linked to the term “All-together method”. Furthermore, no justification or reference is given in relation to consider a relative frequency of 50% for this selection method. Why 50%? Have they tested other thresholds to justify the 50%? In addition, wavelengths with a frequency of 7 have been considered, which means a relative frequency lower than 50% (there are 15 selection methods).

For better clarity, I would first present and analyze Figure 4 (which is not a selection method). In this analysis is interesting to remark that frequency is low in visible wavelengths and relate it with Figure 3. But in Figure 3 it can be seen that there is a difference in the spectrum of SDI and DI in wavelengths around 600-650 nm. This deserves a comment.

Secondly, I would present the selection method based on the relative frequency from Figure 3. Perhaps the name “All-together method” is not appropriated because in Table 4 we have results from “All bands”.

Please revise caption in Figure 4. Is not the same “times a wavelength has been selected by a singular method” (which has no-sense) as “times a wavelength has been selected by the total number of methods” (line 340).

(23) I suggest change the name of section 3.3 to something similar to “quality and efficiency of classification with each selection method” (in line with line 147).

(24) From lines 363-366 it is clear that the classification method is essential to stablish any conclusion. Observe that if KNN is applied, the selection of wavelengths makes no sense. I have not found a comment like that in the analysis of Table 4. Authors should highlight the importance of the classification method and perhaps indicate as a future research, the exploration of other methods different from LDA and KNN.

(25) What is the utility of OA C in Table 4? If not necessary and analyzed, please remove. Also please change the caption in Table 4, because non-PLS methods are also applied, for example “…using different wavelength selection methods”. Please change the name of the first column to “method” (like Table 3).

(26) In addition, I miss an analysis of the confusions between both classes, SDI and DI. We don’t know if it is more frequent the confusion of DI with SDI or SDI with DI, or if confusions between both classes are similar. Furthermore, the performed research is not a case where confusion matrix applies, but QCCS (quality control column set) because the reference is accurate/true (the irrigation technique is known for each tree) (see doi: 10.3390/rs12050816). Therefore I suggest to provide in Table 4 both proportions of correctly classified items (SDI and DI) and include a citation to the suggested reference.

(27) I miss some further analysis on the classification results. For example, are there any spatial correlation in misclassifications? It has any influence the proximity between trees with different irrigation technique? Are there any edge effect with trees in the perimeter of the orchard? Etc…

A Figure with the orchard with the misclassified trees will be useful here. Since there are a high quantity of classifications (32), many figures should be performed. I suggest only perform the figure for the best cases (LDA: GA-PLS, Lasso).

(28) In relation with #24, I should not make an analysis that merges results in both classification methods. Results with KNN are poor and I would not look for selection techniques with adequate results in both LDA and KNN (lines 376-378). This makes that suitable results, like RC-PLS or FiPLS in LDA, go unnoticed. Please rewrite page 12 with this in mind.

(29) Please revise the conclusions section after #28. It should be clear that KNN does not offer good classification results and that wavelength selection does not makes an improvement. Therefore, focus should be maintained on LDA, where several wavelength selection techniques can be advised: RC-PLS, GA-PLS, Lasso or All-together, all of them with an efficiency over 15%.

Author Response

The paper presents a contribution to the field of classification methods from hyperspectral data in a very specific context (two types of irrigation techniques in olive orchards). In general, the article is well-organized and contains all expected components, from introduction to conclusions. In general, all sections are well-developed and clearly explained. English is not my native language but I feel that the article is well-written and easy to understand. The article does not supposes a major advance but a useful contribution to the knowledge base. Some minor changes are proposed, expecting to improve the clarity of the performed research ant their results.

Thank you for the opportunity to resubmit our manuscript. Our manuscript has benefited greatly from your review and we appreciate the favorable comments from reviewer.  We have revised it to clarify your suggestions. We believe our revised manuscript aligns with your comments.

Those added texts have been marked in yellow while the moved texts have been colored in blue.

Minor suggestions

(1) Title. Given that the framework if very specific (classification of only two types of irrigation) the title should also reflect this. For example: “Partial Least Square based wavelength selection for classification of two irrigation techniques of olive orchards from Unmanned Aerial Vehicle orthomosaick”, or “Partial Least Square based wavelength selection for olive orchards irrigation classification from Unmanned Aerial Vehicle orthomosaick”, etc.

We appreciate your comment and title has been rewritten. Your suggestion has been taken into account, see Line 1.

“Wavelength selection based on Partial Least Square from hyperspectral Unmanned Aerial Vehicle orthomosaic of irrigated olive orchards”

(2) Keywords: I miss specific terms related with the core of the research, such as “PLS”, “wavelength selection”. Please revise.

Your suggestion has been taken into account and new keywords have been included. See line 37.

(3) The amount of references if high (108). Although providing references is always positive, I feel that they could be condensed. Only in section 1 (Introduction, two pages, 89 lines) there are 72 references. I suggest authors to revise if all of them are relevant.

Regarding your suggestion, we have tried to reduce as much as possible the number of references (107), but finally we could not reduce them consistently due to the changes suggested. However, in section 1 (Introduction, three pages, 144 lines) the number have been proportionally reduced (there are 82 references).

(4) The introduction is focused and it allows the reader to easily understand the scope in which the authors make their proposal. Nevertheless I have to make two comments for a improvement.

• In line 102 it is said that “wavelength selection method is essential to apply”, and the reference 39 is cited. Given that this fact is the reason for the performed research, author should explain why wavelength selection is so important.
• In lines 123-126 the contribution of the article and their objectives of the performed work is summarized. But is made so briefly that contrast sharply with the previous text in the introduction. I miss:
• A better presentation of the contribution. Only is said that “we report (…) the use of selection methods”, which tells nothing or very little.
• Which irrigation systems are that commonly used in olive orchards (to mention them at least).
• A better concretion of the objectives. “The evaluation of 16 methods” is not an objective but a mean of achieving them. I feel that several specific objectives could be stablished.
• It is a common practice to summarize the contents of the sections, which could help the authors to improve this part. This allows presenting the reader a comprehensive view of the whole paper and the performed research.

Your suggestion has been taken into account and we have clarified it.

See line 105:

“As a result of the high number of spectral bands, many of them are highly correlated and therefore, a dimension reduction or wavelength selection method is essential to apply in pre-processing of the hyperspectral image to improve its usability [39].”

See lines 184-189:

“As per the above discussion, this article focuses on the use of wavelength selection methods in UAV hyperspectral images to compare two irrigation systems commonly used in olive orchards, SDI and DI. For this purpose, 16 methods (13 based on PLS) were evaluated, and the quality of the results were assessed by two linear and nonlinear classification techniques. This paper is organized as follows: Section 2 describe the materials and methods used, Section 3 shows the results and Section 4 includes the final concluding remarks.”

(5) In line 129 the location is cited as Marchena in the province of Seville. But in Figure 1.a the province of Seville is missed and the Andalusia region is included. Also the red dot in Figure 1.a is not placed in the correct location but in the approximate center of the Region. I suggest including a bigger figure to introduce details if necessary. I miss in Figure 1.b an image of the commercial orchard (7-ha) with an indication of the representative selected area.

Thank you so much for your comment. We have redesigned Figure 1 taken into account the limited area and therefore the scale limitation. We have verified location of study area. We have included road and rivers. Figure 1.b include now limits of both irrigation techniques, but it was not possible to include a general view of the plantation.

See Figure 2, line 214.

(6) The work flow presented in lines 140-147 should be included in other section that “study area and UAV flights”. Also observe that section 2.1.1 is missing.

Lines 140-147 and Figure 2 have been moved to the beginning of section 2 to describe in general terms the research.

Moreover, there was a numbering error that has been corrected.

(7) In line 143 it is said that “different PLS- and non-PLS-based methods are used”, but in the flowchart in Figure 2 the corresponding step is “PLS methods”. Please correct this.

Your suggestion has been taken into account and Figure 1 has been updated.

(8) In line 146 the classification methods are presented, which are LDA and KNN. There is a lack of an adequate justification of this selection. Why these methods? Why only two? Observe also that the introduction says nothing about them. Also observe that, while the authors include an high amount of references, here there is none which could justify why these only two classification methods are selected.

In regards with your suggestion, a paragraph justifying the use of LDA and KNN was included in the introduction section. After a revision of the most effectives machine learning algorithms for the classification of hyperspectral imagery, these two models were selected.

See line 170-181:

“On the other hand, machine learning algorithms, including Linear Discriminant Analysis (LDA) and K-Nearest Neighbors (KNN), are powerful tools for analyzing hyperspectral information since they can process a large number of variables efficiently [79,80]. LDA is a subspace technique that seeks to find the maximum Fisher’s ratio [81] while KNN is a non-parametric learning algorithm since there is no assumption for the underlying data distribution. KNN also utilizes lazy learning, meaning that it does not need any training data point for the generation of the model [82]. Specifically, these machine learning algorithms have been widely used in the remote sensing field for agricultural applications. For instance, Suarez et al. [81] estimated phenoxy herbicide dosage in cotton crops through the analysis of hyperspectral data with LDA. As well, Bohnenkamp et al. [35] utilized KNN to detect yellow rust in wheat with the application of hyperspectral imaging technology. In general, these models have shown to be effectives for investigating agricultural features using hyperspectral imagery.”

(9) In lines 146-147 it is said “evaluating the quality and efficiency of each method”. The word “method” could be ambiguous. It could made reference to “irrigation method” (line 145), PLS and non-PLS selection methods (line 144) or the classification methods. Given that the results depends on the combination of data+selection method+classification method, I would be more careful with the terms. For example “evaluating the quality and efficiency of the results”. In addition, I would reduce ambiguity if, for example, the term “irrigation techniques” is always used, like lines 31 and 40 (instead of “irrigation methods”). What is more, I suggest revising the whole text in order to disambiguate the term “method”, if needed.

Your suggestion has been taken into account. Lines 136-140 has been rewritten

See lines 194-198:

“Then, different PLS- and non-PLS-based methods were used to select the most significant wavelengths to classify the two irrigation techniques methods of the study area. Finally, with the selected wavelengths, two classifications were performed using a Linear Discriminant Analysis (LDA) and K-Nearest Neighbors (KNN), evaluating the quality and efficiency of the results.”

In addition, the whole text has been revised and the term “method” has been replaced by “technique” is those sentences where irrigation technique is named.

(10) In line 198 it is indicated that SRTM is used. Since in Spain there are other DEMs available, with higher resolution and accuracy, a justification of use of SRTM is missed. Also I miss a reference when using third party data.

Thank you very much for your appreciation. Indeed, there are other DEMs available. We used SRTM model because it is the Headwall recommended by default. In addition, the topography of the area is smooth, and the quality of the DEM does not have a significant effect on the orthorectification process.

Your suggestion has been taken into account. We have clarified it, sentence has been rewritten and a new reference has been added linked to SRTM.

See lines 262-263.

“The accuracy and spatial resolution of the SRTM model was adequate due to the smooth relief of the study area.”

In addition, a new reference has been added, see reference 88:

• van Zyl, J.J. The Shuttle Radar Topography Mission (SRTM): a breakthrough in remote sensing of topography. Acta Astronaut. 2001, 48, 559–565, doi:10.1016/S0094-5765(01)00020-0.

(11) First time it is introduce the term “calibration phase” is in line 290. It should be clearer what is made in this phase and in the “prediction phase”. Observe that Table 2 speaks about “calibration set” and “prediction set”, that is to say, sets and not phases. It is some confusing because in line 290 it is said that the results of the classification if applied in the calibration phase. Does not the classification need firstly the calibration? Please clear all this part of the research.

We have clarified it. Your suggestion has been taken into account. See Lines 298-302:

“Before applying LDA or KNN classification, the data set was divided into two smaller data sets that were used for calibration and prediction purposes. The calibration subset of data was used to estimate the parameters of the classifier model and the prediction subset of data was used to check the results of the model. Calibration and prediction were performed following an iterative process where subsets of data changed per iteration.”

(12) In Section 2.5 the packages used for each PLS and non-PLS method are listed but we don’t know which package/s is/are needed for each method. It should be clearer if a table is included with this information.

According your suggestion, Table 3 has been included, indicating the packages used for each PLS and non-PLS method was included.

(13) I consider that section 3.1 is essential to understand the utility of hyperspectral data in the context of this study. Figure 3 shows that both reflectance curves are very similar, therefore multispectral data are not enough for the pursued goal. In consequence, I suggest moving lines 315-321 to the end of the introduction section. This is related with my comment #3.

We appreciate your suggestions, and it is true that both reflectance curves are very similar, therefore multispectral data are not enough for the pursued goal. However, we can detect visually differences in Figure 3 and as indicated in the text: “different magnitudes of spectral reflectance were found in the 550∼880 nm range. Therefore, within that range, there is a total of 165 possible wavelengths to differentiate between both irrigation techniques”, so we consider that section 3.1 should be placed in the results section for a better understanding of the of the work for readers. 

(14) I suggest moving text from lines 321-326 (including Table 2) to the Materials and methods section. This part is not related with results but with methodology.

The suggestion was considered and text from lines 321-326, including Table 1, were moved to the Materials and methods section.

(15) Related with #13, it should be justified, better with references, why 75% of trees are used in calibration and 25% in prediction. It seems to me that the capacity of extrapolation (prediction) of the applied methodology is not well-founded with such proportions. Furthermore, it should be better to have introduced different percentages in the research, analyzing the influence of the size of both sets (calibration and prediction). Also there is a lack of information of the spatial distribution of both sets. A figure from the QGIS digitized trees is suitable here, with an explanation of the spatial distribution criteria

The percentages used for calibration and prediction are those generally used in other works. The creation of the model follows an iterative process. In each iteration the algorithm selects 75% of the sample for calibration and 25% for prediction, changing the selected sample in the following iterations. It is not possible to present this spatial distribution in each iteration.

(16) Following suggestions #12 and #13, the section 3.2 could be renumbered to section 3.1

Number section have not changed because we have maintained section 3.1.

(17) In table 2, the term “summary” is bad used. I think the author are referring to “sum” of “total”.

Your suggestion has been considered. See Table 2 at line 306.

(18) In table 2, also in line 322, the labels 1 and 2 do not contribute for a better understanding, since the columns are named as “sub-surface irrigation” and “surface irrigation”. Please remove.

Your suggestion has been taken into account and label-row has been removed. See Table 2.

(19) In line 329 it is said that depending on the selection method employed, the number of selected wavelengths vary considerably. This seems to say that the number of wavelengths depends only in the selection method. That is to say that each selection method offers a fix quantity of selected wavelengths. It does not also depends on parameters that control the selection method, ore any threshold for any measure? Please improve this part.

It is true that each selection method also depends on parameters that control the selection method. However, we optimize threshold arguments before select variables (e.g., by using the ithresh function) and in general, when compare other arguments, no changes were obtained in the number of wavelengths selected.

(20) I see a problem with the name of section 3.2 (line 327). It speaks about “performance”. What is performance? The term “performance” is used several times along the text but never defined and explained how to measure it. Only the term “efficiency” is defined in lines 294-297. In addition, I also see a problem with the name of the section since non-PLS method are also considered in the analysis. I wonder if a simple title of this section as “Wavelength selection results” is enough.

Your suggestion has been taken into account. Section 3.2 has been renamed as “Wavelength selection results”, see line 340.

(21) In caption of Table 3 the term “effective wavelength” is introduced. But it is not used in the main text of the article. Please define what is an effective wavelength ore remove “effective” from the caption.

We have clarified it and “effective” term has been removed in caption. See line 351.

“Table 4. Wavelength selected by different methods.”

(22) In lines 340-344 an “All-together method” is stablished by de authors. Nevertheless is not clear the way it is applied. In lines 340-341 they say Figure 4 show the total number of times a wavelength has been selected by the total number of methods and they define it as “All-together method”. This is not a selection method but only a description of Figure 4. In lines 341-342 it is said that 18 wavelengths were selected by half of the methods, but this fact is not linked to the term “All-together method”. Furthermore, no justification or reference is given in relation to consider a relative frequency of 50% for this selection method. Why 50%? Have they tested other thresholds to justify the 50%? In addition, wavelengths with a frequency of 7 have been considered, which means a relative frequency lower than 50% (there are 15 selection methods).

For better clarity, I would first present and analyze Figure 4 (which is not a selection method). In this analysis is interesting to remark that frequency is low in visible wavelengths and relate it with Figure 3. But in Figure 3 it can be seen that there is a difference in the spectrum of SDI and DI in wavelengths around 600-650 nm. This deserves a comment.

Secondly, I would present the selection method based on the relative frequency from Figure 3. Perhaps the name “All-together method” is not appropriated because in Table 4 we have results from “All bands”.

The suggestion was considered (see below) and we firstly present and analyze Figure 5 (Figure 4 in the first version) in lines 353-356 by remarking that frequency is low in visible wavelengths but in Figure 4 (Figure 3 in the first version) it can be seen that there is a difference in the spectrum of SDI and DI in wavelengths around 600-650 nm. Then, we present All-together method in lines 356-359.

“The total number of times a wavelength has been selected by the total number of methods simultaneously employed is shown in Figure 5. Although there was a difference in the spectrum of SDI and DI in the wavelengths of around 600 to 650 nm (Figure 4), the frequency of selection by the selection methods employed in this range was low (Figure 5). Of all the methods applied, 18 wavelengths were selected by almost half the methods (All-together method; Table 4). All within the infrared region and for a total of 9 times the wavelength 920 nm was selected the most to predict the irrigation method used in this study. The utility of the All-together method was demonstrated for investigating agricultural features using hyperspectral imagery.”

Please revise caption in Figure 4. Is not the same “times a wavelength has been selected by a singular method” (which has no-sense) as “times a wavelength has been selected by the total number of methods” (line 340).

We have rewritten Figure 4 caption. See line 362.

“Figure 4. Number of times a wavelength has been selected when performing the All-together method.”

(23) I suggest change the name of section 3.3 to something similar to “quality and efficiency of classification with each selection method” (in line with line 147).

Your suggestion has been taken into account and section 3.3 has been renamed. See line 363.

(24) From lines 363-366 it is clear that the classification method is essential to stablish any conclusion. Observe that if KNN is applied, the selection of wavelengths makes no sense. I have not found a comment like that in the analysis of Table 4. Authors should highlight the importance of the classification method and perhaps indicate as a future research, the exploration of other methods different from LDA and KNN.

Your suggestion has been taken into account and we have included to explore other classifiers (Line 377-378) and we have highlight LDA results (Line 419-420).

“In addition, while LDA showed similar accuracy classifying both irrigation techniques, KNN offered worse results at SDI technique.”

“In addition, other features such as soil properties or olive cultivars along with other classifiers.”

(25) What is the utility of OA C in Table 4? If not necessary and analyzed, please remove. Also please change the caption in Table 4, because non-PLS methods are also applied, for example “…using different wavelength selection methods”. Please change the name of the first column to “method” (like Table 3).

Your suggestion has been taken into account. OA C column has been removed. Caption table and name of first column have been rewritten. See Table 5.

(26) In addition, I miss an analysis of the confusions between both classes, SDI and DI. We don’t know if it is more frequent the confusion of DI with SDI or SDI with DI, or if confusions between both classes are similar. Furthermore, the performed research is not a case where confusion matrix applies, but QCCS (quality control column set) because the reference is accurate/true (the irrigation technique is known for each tree) (see doi: 10.3390/rs12050816). Therefore I suggest to provide in Table 4 both proportions of correctly classified items (SDI and DI) and include a citation to the suggested reference.

Your suggestion has been taken into account.

See line 377-378 and Table 5. In addition, suggested reference has been included.

“In addition, while LDA showed similar accuracy classifying both irrigation techniques, KNN offered worse results with the SDI technique.”

A new reference has been added, see 94:

• Alba-Fernández, M. V.; Ariza-López, F.J.; Rodríguez-Avi, J.; García-Balboa, J.L. Statistical methods for thematic-accuracy quality control based on an accurate reference sample. Remote Sens. 2020, 12, 1–16, doi:10.3390/rs12050816.

(27) I miss some further analysis on the classification results. For example, are there any spatial correlation in misclassifications? It has any influence the proximity between trees with different irrigation technique? Are there any edge effect with trees in the perimeter of the orchard? Etc…

Your suggestion has been taken into account, see Line 382-385.

“In addition, the irrigation classification maps of some of the selection methods assessed are shown in Figure 7. From a visual analysis, no spatial correlation was detected in the errors obtained as well as the presence of a higher concentration of errors in the perimeter of each type of irrigation.”

A Figure with the orchard with the misclassified trees will be useful here. Since there are a high quantity of classifications (32), many figures should be performed. I suggest only perform the figure for the best cases (LDA: GA-PLS, Lasso).

Your suggestion has been taken into account, see Figure 7.

(28) In relation with #24, I should not make an analysis that merges results in both classification methods. Results with KNN are poor and I would not look for selection techniques with adequate results in both LDA and KNN (lines 376-378). This makes that suitable results, like RC-PLS or FiPLS in LDA, go unnoticed. Please rewrite page 12 with this in mind.

We have clarified it. A new sentence has been added, see line 403-404:

“In addition, RC-PLS and FiPLS methods showed a high overall accuracy using LDA.”

(29) Please revise the conclusions section after #28. It should be clear that KNN does not offer good classification results and that wavelength selection does not makes an improvement. Therefore, focus should be maintained on LDA, where several wavelength selection techniques can be advised: RC-PLS, GA-PLS, Lasso or All-together, all of them with an efficiency over 15%.

Your suggestion has been taken into account. See line 430-433:

“In addition, LDA offered more accurate results than KNN. In our study, GA-PLS, RC-PLS, Lasso, FiPLS, Boruta and All-together showed an overall accuracy of 75% or higher. They were all highly efficient methods and resulted in an improved classification.”

Reviewer 2 Report

Title:

• is it not supposed to read: ...wavelength selection method for....
• orthomosaic not orthomosaick – check also on line 188

Revise the sentence on line 22

Line 71: Earth Observation Programs – remove caps. Check for consistency in the document, eg in some instances it’s written as Principal Component Analysis and in some its principal component analysis.  Check line 105 to 106 and line 213

Line 71: Modis must be MODIS and must be written in full the first time since its an abbreviation

Line 123 and Line 124: Are you reporting in first person? Why use we and our

Line 128 to line 137. Can you include climatic characteristics of the area e.g. average temperature, seasons, rainfall

Line 138 and 139: The map must include base data e.g. roads, place names and rivers

Line 140 – you cannot start a paragraph or sentence with a cross reference. First give your readers the context of the issue the refer to the table or diagram later. Same on Line 315, 328, 340. 349,

Line 148 to 149 – its orthomosaic not ortomosaic can you change on the flow chart

Line 189 – there is a missing word on the sentence starting with at line 189

Line 191 to 198 - There is repetition here. Remove the first part where your named the stages or the second part which starts with Firstly.... There is no need to mention Bareto et al, just reference the author in your methodology

Line 200 - why digitizing when you could classify them using spectral signatures? Where is the digitized map?

Line 201 - Why didn’t you use thee spectral reflectance to map the olive trees??

Line 210 – the sentence starting in line 210 is not clear

Check your methodology - what did you do in this study? Most of the information here must be part of the introduction or literature review not methodology. You are just giving us literature and you are not telling us what you did.

Line 210 – More information…. is this statement really necessary or you just need to reference the author

Section 2.2, 2.2.1, 2.2.2, 2.2.3, 2.3 – if you check properly you will find that the whole paragraph will be describing the method but without mentioning what you did? Some of the information is not referenced (section 2.2.1) and you must add more.

Line 296 – you mentioned defined by these authors – why not cite them?

Line 309 to 311 – you didn’t mention the software used

Line 363 – revise the sentence starting with:  Except

Line 368 – make the footnotes

Results: where is the digitised map of olive trees? Where is the classified irrigation map showing the SDI and DI?

I find it difficult to keep up with the abbreviations (they are too many and can be confusing). Hope you are also not confusing them in your writing (just verify that)

Author Response

Thank you for the opportunity to resubmit our manuscript. Our manuscript has benefited greatly from your review and we appreciate the favorable comments from reviewer.  We have revised it to clarify your suggestions. We believe our revised manuscript aligns with your comments.

Those added texts have been marked in yellow while the moved texts have been colored in blue.

Title:

is it not supposed to read: ...wavelength selection method for....

orthomosaic not orthomosaick – check also on line 188

We appreciate your comment. Title has been rewritten, see Lines 2-4

“Wavelength selection based on Partial Least Square from hyperspectral Unmanned Aerial Vehicle orthomosaic of irrigated olive orchards”

See Line 253:

“Finally, individual images were rectified and mosaicked to generate an RGB UAV orthomosaic of the study area”

Revise the sentence on line 22

Sentences has been revised, see line 26.

“…Competitive Adaptive Reweighted Sampling-PLS; and an Embedded method: Sparse-PLS.”

Line 71: Earth Observation Programs – remove caps. Check for consistency in the document, eg in some instances it’s written as Principal Component Analysis and in some its principal component analysis. Check line 105 to 106 and line 213

Your suggestion has been taken into account. See lines 72, 108 and 112.

We have added acronyms PCA in line 108. It has been used in line 112.

Line 71: Modis must be MODIS and must be written in full the first time since its an abbreviation

Your suggestion has been taken into account. See line 73:

“…programs provide free low spatial-resolution datasets, like Moderate Resolution Imaging Spectroradiometer (MODIS),”

Line 123 and Line 124: Are you reporting in first person? Why use we and our

Your suggestion has been taken into account and sentences have been rewritten. See lines 184-189.

this article focuses on the use of wavelength selection methods in UAV hyperspectral images to compare two irrigation systems commonly used in olive orchards, SDI and DI. For this purpose, 16 methods (13 based on PLS) were evaluated, and the quality of the results were assessed by two linear and nonlinear classification techniques. This paper is organized as follows: Section 2 describe the materials and methods used, Section 3 shows the results and Section 4 includes the final concluding remarks.

Line 128 to line 137. Can you include climatic characteristics of the area e.g. average temperature, seasons, rainfall

Your suggestion has been taken into account. See lines 208-210.

“The climate is Mediterranean with an average annual temperature and precipitation (concentrated mostly in late fall and winter) of 18.8 ºC and 544 mm, respectively.”.

Line 138 and 139: The map must include base data e.g. roads, place names and rivers

The map was re-designed as shown in line 213-214.

We have taken into account your suggestion and roads and rivers have been included. Because the scale limitation we have just include countries’ name.

Line 140 – you cannot start a paragraph or sentence with a cross reference. First give your readers the context of the issue the refer to the table or diagram later. Same on Line 315, 328, 340. 349,

All paragraphs starting with a cross reference were rewritten in lines 140, 315, 328, 340 and 349.

Lines 191-192

“The workflow for classifying olive tree crowns according to the irrigation technique used is summarized in Figure 1.”

Lines 331-332

“The mean spectral reflectance curves of olive trees, being irrigated with SDI or DI systems, were similar to the results obtained in related studies [105] (Figure 4).”

Lines 341-342

“The results of the application of the methods used, including the number of the selected wavelengths as well as their wavelengths, are shown in Table 4.”

Lines 353-354

The total number of times a wavelength has been selected by the total number of methods simultaneously employed is shown in Figure 5

Lines 364-365

The accuracy and efficiency results of the irrigation technique rating using LDA and KNN methods and their efficiency are shown in Table 5 while Figure 6 shows their efficiency.

Line 148 to 149 – its orthomosaic not ortomosaic can you change on the flow chart

Your suggestion has been taken into account. See Figure 1 at line 199.

Line 189 – there is a missing word on the sentence starting with at line 189

Your suggestion has been taken into account and the sentence has been rewritten, see Lines 254-256

“This methodology has been validated in previous research projects [19,85,86] and was performed using Pix4Dmapper (Pix4D S.A., Prilly, Switzerland).”

Line 191 to 198 - There is repetition here. Remove the first part where your named the stages or the second part which starts with Firstly.... There is no need to mention Bareto et al, just reference the author in your methodology

Your suggestion has been taken into account and sentence has been rewritten. See lines 256-257

“The hyperspectral data processing was divided into three stages [87].”

Line 200 - why digitizing when you could classify them using spectral signatures? Where is the digitized map?

In a hedgerow orchard it is difficult to segment individual trees, more so if the base cartography is from a hyperspectral sensor. This is why RGB orthomosaic was used as a reference to digitize the crowns.

Also, because the scale of the map in the manuscript is reduced digitized crowns are not properly identified. However, we have included at the beginning of the result section a partial view of the olive trees in an RGB and hyperspectral orthomosaic.

We have clarified it at lines 265-268, in addition a new Figure (Figure 3) has been added to the manuscript, see line 272.

“Due to the difficulty in segmenting individual trees an RGB orthomosaic was used to manually digitize polygons of all the olive trees, using the QGIS desktop Geographic Information System. As an example, a partial view of the hedgerow in the RGB and hyperspectral orthomosaic is shown in Figure 3”

Line 201 - Why didn’t you use thee spectral reflectance to map the olive trees??

Thank you for your comment. In that case it would be necessary later to assign only one type of irrigation to each tree. Therefore, the average value has been used as the most representative of each tree.

Line 210 – the sentence starting in line 210 is not clear

Your suggestion has been taken into account and the sentence has been rewritten, see Lines 275-276.

As such, the variables analysed in this project were the wavelengths registered by the hyperspectral sensor.

Check your methodology - what did you do in this study? Most of the information here must be part of the introduction or literature review not methodology. You are just giving us literature and you are not telling us what you did.

The methodology section was organized according to your comments, please see next points. Most of the information from paragraphs were deleted and included in the introduction section

Line 210 – More information…. is this statement really necessary or you just need to reference the author

Your suggestion was considered, and the information included in this line has been deleted from methodology section and included in the introduction section. See Lines 111-118.

Section 2.2, 2.2.1, 2.2.2, 2.2.3, 2.3 – if you check properly you will find that the whole paragraph will be describing the method but without mentioning what you did? Some of the information is not referenced (section 2.2.1) and you must add more.

As indicated above, most of the information included in these sections have been deleted from methodology and included in the introduction section. See Lines 125-169.

Line 296 – you mentioned defined by these authors – why not cite them?

Your suggestion has been taken into account and sentence has been rewritten. See lines 311-312.

“According to Xia et al. [98], the efficiency of a wavelength selection method is based on the prediction rate and the number of variables, being calculated as follows:”

Line 309 to 311 – you didn’t mention the software used

Your suggestion has been taken into account, clarifying we used R Studio and Python. See line 323 and 326.

Line 363 – revise the sentence starting with:  Except

Your suggestion has been taken into account and the sentence has been rewritten, see Lines 379-380.

“While LDA offered better results than KNN, all the methods used were highly efficient except for LW-PLS.”

Line 368 – make the footnotes

Thank you for your comment. We have rewritten table caption, including all acronyms. see Lines 386-388.

Table 5. Overall accuracy (OA), Accuracy of Sub-Surface irrigation (A SDI), Accuracy of Surface irrigation (A DI) and Efficiency (E) results of Linear Discriminant Analysis and K-Nearest Neighbors using different selection methods.

Results: where is the digitised map of olive trees? Where is the classified irrigation map showing the SDI and DI?

We have tried to include the digitized area as explained previously but scale of the Figure does not allow to visualize them properly. However, we have included a Figure with the results of the classification of some methods used, representing each tree by means of a punctual symbol. See Figure 7 at line 393.

I find it difficult to keep up with the abbreviations (they are too many and can be confusing). Hope you are also not confusing them in your writing (just verify that)

We appreciate your comments. All abbreviations have been checked.

Round 2

Reviewer 2 Report

Hi Author

Thank you for entrusting me to review your paper. From my own assessment you must attend to the comments below. 

Title must read: Wavelength selection method based on Partial Least Square from….

Line 42-44, 51 -54, 64-65, 68, 70-74, 84-88, 95-97, 126-132, 134-135, 139-144, 175-177, 179-181, 182-183  – there are no references, the sentences are just left hanging (make sure all the information you are getting from literature is referenced with recent literature). Check on your reference list the majority of the reference were published before 2010 are you saying there is no recent literature?

Line 197 – remove a

Line 213 – hydrography not hidrography

The author must revisit their literature, some of the literature is too old. 

Author Response

Thank you for entrusting me to review your paper. From my own assessment you must attend to the comments below. 

Thank you for the opportunity to resubmit our manuscript. We have revised it to clarify your suggestions. We believe our revised manuscript aligns with your comments.

Added texts have been marked in yellow.

Title must read: Wavelength selection method based on Partial Least Square from….

We have taken into account your comment and title has been rewritten:

‘Wavelength selection method based on Partial Least Square from hyperspectral Unmanned Aerial Vehicle orthomosaic of irrigated olive orchards’

Line 42-44, 51 -54, 64-65, 68, 70-74, 84-88, 95-97, 126-132, 134-135, 139-144, 175-177, 179-181, 182-183  – there are no references, the sentences are just left hanging (make sure all the information you are getting from literature is referenced with recent literature).

As suggested, references, as recent as possible, were included in lines 44, 54, 65, 69, 71, 74, 85, 87, 97, 129, 132, 136, 141, 144, 178, 182 and 184.

Check on your reference list the majority of the reference were published before 2010 are you saying there is no recent literature?

We have taken into account your comment and we have checked the reference list and we have sought for recent literature. However, although we have found more recent references, they are not as relevant as the ones included in the original version.

Line 197 – remove a

As suggested, “a” was removed from line 198.

Line 213 – hidrography not hydrography

We have taken into account your comment and the word “hidrography” was corrected with “hydrography” in Figure 2 (line 214)

The author must revisit their literature, some of the literature is too old. 

We appreciate your suggestion. It is true that 29 of the 110 references from the reference list were published before 2010. However, based on the authors´ knowledge, we have not found similar references as relevant as the ones included in the original version.

Author Response File: Author Response.pdf