Cotton Growth Modelling Using UAS-Derived DSM and RGB Imagery

Round 1

Reviewer 1 Report

Dear Authors,

I have accurately reviewed the

Manuscript ID: remotesensing-2167169

Type of manuscript: Article

Title: Cotton growth modelling using UAS-derived DSM and RGB imagery

Authors: Vasilis Psiroukis *, George Papadopoulos, Aikaterini Kasimati, Nikos Tsoulias, Spyros Fountas Submitted to section: Remote Sensing in Agriculture and Vegetation

I recommend to Reconsider the article after major revision.

 

In the follow, my detailed remarks.

 

Best regard

 

Remarks

Line 18. Substitute “modeling” with “products”

Line 87.  Digital Photogrammetry thanks to the image matching procedures is the fundamental technique adopted in your research. In this section It is necessary highlight problems concerning the use of automatic algorithms and the difficulty to correctly estimate errors of the photogrammetric products Furthermore, I suggest to introduce the key publications.

Line 94. “ points (GPSs) located in the field is an essential practice” for several reason, for example to obtain metric products, to introduce the survey in a global or relative Datum, to minimizing model errors, to check the errors which characterize each product….

Line 109. “ii using image analysis techniques” applied to the orthophoto to extract …

Line 114. Which is the dimension of the surveyed area?  

Line 122. The Figure 1 is constituted by several images and maps with low resolution. It is necessary improve the Font dimension, enlarge each image and the Legend, insert abscissae and ordinate explanatory heading.

Line 124. With the goal to compare different metric products coming from photogrammetric surveys it is necessary to define a unique Datum. Some of the GCPs have to be permanently fixed on the ground and visible in the images collected by means of UAV. These points have to be used as a reference for frame each survey. On the contrary, have you adopted temporary GCPs?. Clarify. Furthermore, it is really fundamental to know the target dimension and the GCPs distribution on the surveyed area.

Line 128. Horizontal errors of 1 cm and vertical errors of 2 cm are only nominal values. However, there are several sources of error that could make the results inaccurate which have to be identified and described. Explain the procedure adopted to survey the GCPs by means RTK-GPS and to frame their coordinates in the same Datum.  Furthermore evaluate the errors concerning Datum transformation and coming from coordinate transformation. Using geiodal models, GNSS measured ellipsoidal heights can be converted to orthometric ones. These transformation add errors. Clarify.

Line 140. Describe the characteristics of each DEM obtained with the GPS survey: number of points, areal distribution, the interpolator used to define the surface to be subtracted from the Photogrammetric DSM.. These 3D coordinates surveyed by means of GPS have to be adopted to evaluate errors of each photogrammetric DSM obtained at the same time by automatic procedures.

Line 148. Improve the description of the procedure to elaborate DSMs and orthophotos.

Line 152. How many GCPs have you adopted? Show the errors concerning the orientation procedures. Have you adopted tie points detected manually? Have you adopted ellipsoidal or orthometric heights? Clarify. Introduce a description of each DSM (n of points, density, errors..) and orthophoto (geometric resolution).

Line 160-161. “..by interpolating the RTK GPS images ..” Clarify.

Line 168. With which software the estimation of indices was carried out? To better understand, it is possible to create a figure that shows each index overlapped to the orthophoto?

Line 170-172. Clarify.

Line 174. Figure 2. Enlarge the Legend. Insert abscissae and ordinate explanatory heading. Improve the description of the figure in the text.

Line 176. The photogrammetric data processing pipeline is excessively simplified and the scheme is not fully explained in the text.

Line 178. The sample adopted is not statistically significant

Line 186. “..   confirming that all generated DSM models were of reasonable accuracy.” The data obtained does not support the comment.

Line 204, 206, 209, 2013. Are all the charts fundamental? It is necessary improve the Font dimension, enlarge each image and improve the discussion concerning these distributions.

Line 219. Introduce in the manuscript comments concerning this table.

Line 223. It is necessary improve the Font dimension, enlarge each image and improve the discussion concerning these charts.

Line 242. The Section must be written after applying the suggested corrections. It is essential to correctly assess the accuracy of photogrammetric products and how this affects subsequent analysis.

Author Response

We are very grateful to the reviewer for the time they dedicated in both reading our manuscript and providing us with their invaluable comments, which most certainly helped us improve the quality of our work.

Below the changes that we incorporated:

Line 18. Substitute “modeling” with “products”

We have substituted the two words as requested.

Line 87.  Digital Photogrammetry thanks to the image matching procedures is the fundamental technique adopted in your research. In this section. It is necessary highlight problems concerning the use of automatic algorithms and the difficulty to correctly estimate errors of the photogrammetric products Furthermore, I suggest to introduce the key publications.

We thank the reviewer for this valuable comment. We have further elaborated on the requested parts in the respective Introduction paragraph.

Line 94. “ points (GPSs) located in the field is an essential practice” for several reason, for example to obtain metric products, to introduce the survey in a global or relative Datum, to minimizing model errors, to check the errors which characterize each product….

We thank the reviewer for this comment , and all requested points have been included alongside the additions from the previous comment.

Line 109. “ii using image analysis techniques” applied to the orthophoto to extract …

We have added the requested supporting information.

Line 114. Which is the dimension of the surveyed area? 

We are grateful to the reviewer for this comment, as we overlooked this important piece of information, that has now been added to the text.

Line 122. The Figure 1 is constituted by several images and maps with low resolution. It is necessary improve the Font dimension, enlarge each image and the Legend, insert abscissae and ordinate explanatory heading.

To address this comment, we have divided the initial map into two (2) separate Figures, which naturally increased the readability of the maps and accompanying elements.

Line 124. With the goal to compare different metric products coming from photogrammetric surveys it is necessary to define a unique Datum. Some of the GCPs have to be permanently fixed on the ground and visible in the images collected by means of UAV. These points have to be used as a reference for frame each survey. On the contrary, have you adopted temporary GCPs?. Clarify. Furthermore, it is really fundamental to know the target dimension and the GCPs distribution on the surveyed area.

We thank the reviewer for this valuable comments and observation. First of all, we have further elaborated on the datum used for the RTK point collection. Afterwards, regarding the GCPs, we indeed selected to deploy GCP targets each time we visited the field, as the area did not have any fixed GCPs that we could easily (and more importantly accurately) distinguish consistently across data collection visits. Finally, we have added the dimensions of the targets we used in the survey.

Line 140. Describe the characteristics of each DEM obtained with the GPS survey: number of points, areal distribution, the interpolator used to define the surface to be subtracted from the Photogrammetric DSM. These 3D coordinates surveyed by means of GPS have to be adopted to evaluate errors of each photogrammetric DSM obtained at the same time by automatic procedures.

We thank the reviewer for their valuable comment. We have elaborated on all aforementioned points, namely number of soil sampled points, spatial distribution and interpolation method in the “Data Acquisition” section of our manuscript, as well as the following “Photogrammetry” section.

Line 148. Improve the description of the procedure to elaborate DSMs and orthophotos.

We have improved the sentences mentioned above, and we have also included a few additional ones to make it easier for the readers to follow.

Line 152. How many GCPs have you adopted? Show the errors concerning the orientation procedures. Have you adopted tie points detected manually? Have you adopted ellipsoidal or orthometric heights? Clarify. Introduce a description of each DSM (n of points, density, errors..) and orthophoto (geometric resolution).

The number of GCPs we used was 6, scattered evenly across the field, as we have now also mentioned in the manuscript. Automatic detection was activated and in case a target was not detected, they were referenced manually. We have clarified on all aspects of the methodology that we followed in their respective sections.

Line 160-161. “..by interpolating the RTK GPS images ..” Clarify.

We have expanded this section to provide more information.

Line 168. With which software the estimation of indices was carried out? To better understand, it is possible to create a figure that shows each index overlapped to the orthophoto?

We thank the reviewer for noticing the missing piece of information. We have elaborated on the software used, and we have included the Figure 2 (currently Figure 3) for this exact reason as stated above.

Line 170-172. Clarify.

We have clarified on the process of calculating the Coverage metric.

Line 174. Figure 2. Enlarge the Legend. Insert abscissae and ordinate explanatory heading. Improve the description of the figure in the text.

We have enlarged the legend of this Figure, and as mentioned in the reply of the previous comment, we have made changes on the description of the image in the corresponding paragraph.

Line 176. The photogrammetric data processing pipeline is excessively simplified and the scheme is not fully explained in the text.

We thank the reviewer for their comment. As a matter of fact, we did not want to bloat this section with unnecessary information, which have already been explained in the earlier section “Photogrammetry”. Nevertheless, we have tried to include a few more details in this paragraph prior to Figure 3 (currently Figure 4) better explaining the process.

Line 178. The sample adopted is not statistically significant

We thank the reviewer for this comment and we agree that this section of the experiment can be further improved. The sample strategy that we used (4 plants per GCP, resulting in a total of 24 sample points per orthomosaic) could definitely be improved, but considering the limitations and challenges we faced during the field visits (difficulty to access all points of the field in order to achieve an evenly distributed layer of GCPs) was the best to our abilities. In a similar experiment that we plan to initiate this summer, among other things that we want to improve, the number of both GCPs and sampled crops will be significantly higher, and following the instructions of the reviewer, a new experimental location with easily accessible and distinguishable fixed GCPs will be selected.

Line 186. “..   confirming that all generated DSM models were of reasonable accuracy.” The data obtained does not support the comment.

As mentioned within the paragraph, the arbitrary value threshold of 5% that we adopted was referring to the vertical development of the plants. The generated errors are of course existent, but it is very challenging to eliminate after a certain threshold in complex systems such as agricultural row crops. Additionally, other researchers have encountered similar performance values (e.g. Wu et al. 2022, which is now referenced in the Discussions section). In future research, we will follow the instruction of the reviewer and lower this acceptable threshold to a lower value, in order to achieve DSM values of higher accuracy.

Line 204, 206, 209, 2013. Are all the charts fundamental? It is necessary improve the Font dimension, enlarge each image and improve the discussion concerning these distributions.

We thank the reviewer for their comment. We have tried to enlarge the resolution of these charts, and we have provided the individual images in case the editor requests that these Figures are broken down in smaller (potentially separate) charts to make the values more distinguishable. Finally, we have improved the discussion prior to the charts, and we have further elaborated on the captions of each individual Figure.

Line 219. Introduce in the manuscript comments concerning this table.

We have introduced the Table in the first line of the previous paragraph as following “The average VI data and calculated horizontal cotton growth curve (DSM values) for each date of record are shown in Table 2 and Figure 11”.

Line 223. It is necessary improve the Font dimension, enlarge each image and improve the discussion concerning these charts.

We thank the reviewer for their comment. Similarly to the comment regarding the font on a previous chart, we have increased the resolution and provided the separate charts in case the editor also suggests that we break them down to individual ones.

Line 242. The Section must be written after applying the suggested corrections. It is essential to correctly assess the accuracy of photogrammetric products and how this affects subsequent analysis.

We are grateful to the reviewer for their invaluable comments. We have integrated several new parts in the Discussions section of our manuscript, and we have both included metrics from other studies but also explained in a greater detail our results and limitations, which we plan to solve in the next iteration of the experiment.

Reviewer 2 Report

Overall comments

The subject of this work falls within the general scope of Remote sensing Journal and, specifically, within the Special Issue: 3D Modelling and Mapping for Precision Agriculture. This article aims to investigate the potential of a data processing pipeline based on UAS-derived RGB vegetation indices and photogrammetric modeling to estimate cotton plant height. The study is original and the results derived from it can be considered of interest for the scientific community. The data processing pipeline seems to be very robust and reliable. Furthermore, the manuscript is overall quite well written, and the conclusions and interpretations are sound and consistent with the objectives.

However, I have some serious concerns issues that need to be addressed in order for this article to be published.

 

Introduction

Overall, the Introduction is quite complete, with a very good review of the state of the art. Perhaps, some deeper ideas LiDAR tech and RGB spectral indices could be provided in this section. Please see the specific comments below.

Line 35: scientific name should be in italic.

Line 56-58: this sentence is very difficult to understand. Please, re-write it in a clearer way.

Line 63 and 64: Remove “the” from “the many crops elements”. Remove the period after citation [30] as well.

Line 72: “to spatio-temporally characterize large populations …” sounds better.

Line 80: remove double-space just before UAS.

Line 85: “UAS” instead “UAV”. Revised all along the manuscript.

Line 87-95: This paragraph is very interesting but I reckon it would be more valuable if a comparison with LiDAR technology would have been performed. I strongly recommend the authors to establish this comparison and justify the subsequent use of SfM.

Line 96-104: Perhaps this paragraph lacks of a deeper insight on the indices that have been used in this study. If the authors can just mention these indices and expose some other applications of them, the Introduction will be truly complete.

Material and Methods

Line 114: Please, state the dimension of the experimental cotton field.

Line 116-120: This part should be moved to the next section  “Data acquisition”. Furthermore, in the “Study area” section, authors should include a brief comment of the meteorological conditions of the study area.

Line 138: I suggest the authors to use UTC instead of local time.

Line 139: Can you also mention the flight speed?

Line 140: The points taken to elaborate the DSM are not strictly GCP. Please change the name. In this line, you need to state the number of points you take to build the DSM, which should be in accordance with the size of the experimental plot.

Line 160: Which interpolation algorithm did you use? Kriggin polygons? Nearest neighbor?

Line 163-166: It is necessary to explain the meaning of each of the VI studied. I mean, what does each index represent? The higher the index, the higher the vegetation? Or vice versa? Is it instead bare soil? Please, explain this.

Results

Line 201: “Figures 5 – 9”.

Table 2: Does the DSM values are also normalized between 0 and 1, as are the indices values?

Figure 10: In the caption you mention DSM values but in figure b) the DEM values are shown? I believe this is not correct.

Table 3: I have a serious concern about these results. Pearson's correlation coefficient is correctly interpretable in those situations in which the variables have a linear relationship with each other, a relationship that is barely observable in Figure 5 and non-existent in Figures 6 - 9. For those cases in which there is no linear relationship between the variables, it is necessary to use other correlation indexes, such as those provided by quadratic regressions.

Discussion

 

Overall, the discussion lacks of a deeper review of the current scientific literature. There is a lot of redundant information from the previous Result section, but scarce discussion about it (only one single citation). The authors should compare their results with the work of other researchers that use those same indices, maybe not in cotton but in other crops. I can assure you that there is a vast scientific literature about the use of vegetation indices to calculate vegetation height. This section needs to be significantly improved.

Author Response

Overall, the Introduction is quite complete, with a very good review of the state of the art. Perhaps, some deeper ideas LiDAR tech and RGB spectral indices could be provided in this section. Please see the specific comments below.

We are very grateful to the reviewer for the time they dedicated in both reading our manuscript and providing us with their invaluable comments, which most certainly helped us improve the quality of our work. Below the changes that we have implemented based on their comments:

 

Line 35: scientific name should be in italic.

We thank the reviewer for their comment. We have switched the scientific name in Italics.

Line 56-58: this sentence is very difficult to understand. Please, re-write it in a clearer way.

We agree with the reviewer that indeed, this sentence was difficult to understand. We have rephrased and simplified it.

Line 63 and 64: Remove “the” from “the many crops elements”. Remove the period after citation [30] as well.

It has been removed.

Line 72: “to spatio-temporally characterize large populations …” sounds better.

We thank the reviewer for their suggestion. This sentence has been changed accordingly.

Line 80: remove double-space just before UAS.

We thank the reviewer for noticing this error. We have deleted the extra space.

 

Line 85: “UAS” instead “UAV”. Revised all along the manuscript.

We thank the reviewer for noticing this inconsistency. We have changed all UAV mentions to UAS.

Line 87-95: This paragraph is very interesting but I reckon it would be more valuable if a comparison with LiDAR technology would have been performed. I strongly recommend the authors to establish this comparison and justify the subsequent use of SfM.

We agree with the reviewer that a comparison with LiDAR would be a good fit to this section, which is something that we considered during the writing of this manuscript. Therefore, we initially decided to mainly focus on vision based systems and approaches, as we did not wish to include information about something that we did not use on the survey. To this end, however, a number of LiDAR-based studies have been included, and additional ones have been included in the Discussion section, to provide a comparison between achieved metrics and performances in similar cases.

Line 96-104: Perhaps this paragraph lacks of a deeper insight on the indices that have been used in this study. If the authors can just mention these indices and expose some other applications of them, the Introduction will be truly complete.

We thank the reviewer for their comment regarding this paragraph, which we agree that was missing some useful information about the selected indices. We have expanded this paragraph and improved the description, uses and expected trends of each selected VI.

 

Material and Methods

Line 114: Please, state the dimension of the experimental cotton field.

We thank the reviewer for noticing that this information was missing. We have included he total area of the experimental field.

Line 116-120: This part should be moved to the next section “Data acquisition”. Furthermore, in the “Study area” section, authors should include a brief comment of the meteorological conditions of the study area.

We thank the reviewer for their comment. These sentences have been moved to the indicated section, and a few new sentences have been included to enrich the section that describes the experimental location.

Line 138: I suggest the authors to use UTC instead of local time.

The reason we included local time is due to the importance of the solar angle during the flights. In any case, we have also included the time in UTC.

Line 139: Can you also mention the flight speed?

Flight speed is calculated automatically based on the provided overlap values (specifically frontal overlap) and the altitude along with the camera parameters, resulting in the GSD parameter that we mentioned in the following section.

Line 140: The points taken to elaborate the DSM are not strictly GCP. Please change the name. In this line, you need to state the number of points you take to build the DSM, which should be in accordance with the size of the experimental plot.

We are grateful to the reviewer for their comment, which is of course correct. We have changed the misused term GCPs with a more appropriate “ground elevation points”.

Line 160: Which interpolation algorithm did you use? Kriggin polygons? Nearest neighbor?

Yes used Kriging to interpolate the ground elevation points, which is the standard method for generating DSMs as the terrain of the field is spatially correlated and naturally has directional bias. We have included this piece of information to the respective sentence

Line 163-166: It is necessary to explain the meaning of each of the VI studied. I mean, what does each index represent? The higher the index, the higher the vegetation? Or vice versa? Is it instead bare soil? Please, explain this.

We thank the reviewer for their comment. We have further elaborated on the indices we used, along with some additional information such as expected trends

Results

Line 201: “Figures 5 – 9”.

We have changed the word with “–“ as requested.

Table 2: Does the DSM values are also normalized between 0 and 1, as are the indices values?

No, the DSM values are in meters (m), as shown in the following charts (of which we have also increased the resolution in an attempt to make more easily distinguishable).

Figure 10: In the caption you mention DSM values but in figure b) the DEM values are shown? I believe this is not correct.

We are grateful to the reviewer for noticing this mistake. Indeed, the correct caption is DSM, and we have changed the chart title and legend accordingly.

Table 3: I have a serious concern about these results. Pearson's correlation coefficient is correctly interpretable in those situations in which the variables have a linear relationship with each other, a relationship that is barely observable in Figure 5 and non-existent in Figures 6 - 9. For those cases in which there is no linear relationship between the variables, it is necessary to use other correlation indexes, such as those provided by quadratic regressions.

We agree with the reviewer that the Pearson correlation is not suitable for the use in our data. We mainly used this metric as it is the most widely used one, however, but as the data are not linear (as the reviewer stated correctly), this is an error that we should fix. According to your suggestion, a different type of correlation (e.g. spearman correlation) would be better, or should we move to a completely different approach? In any case, we will be able to re-submit a newer version of the manuscript very shortly after your valuable feedback.

Discussion

Overall, the discussion lacks of a deeper review of the current scientific literature. There is a lot of redundant information from the previous Result section, but scarce discussion about it (only one single citation). The authors should compare their results with the work of other researchers that use those same indices, maybe not in cotton but in other crops. I can assure you that there is a vast scientific literature about the use of vegetation indices to calculate vegetation height. This section needs to be significantly improved.

We agree with the reviewer that the initial Discussions section was very limited. To this end, we have both expanded the entire section and enriched it with similar studies, while also we have improved the commentary and insights regarding the results that we achieved in this study.

Reviewer 3 Report

The manuscript is well written and very well structured.

The text needs to be restructured in places. I have made several comments,  recommendations and corrections which are highligted in the pdf attached with my review.

appart form the comments and corrections, I would suggest the authors to carefully read through the manuscript for corrections in english language.

Comments for author File: Comments.pdf

Author Response

We are very grateful to the reviewer for the time they dedicated in both reading our manuscript and providing us with their invaluable comments, which most certainly helped us improve the quality of our work. We have addressed all comments accordingly, which can be found on the most recent version of the manuscript.

Round 2

Reviewer 1 Report

Remarks

The article is interesting but some problems are not solved yet. There are different topographic procedures available to frame multitemporal surveys in an unique Datum.

The authors don’t adopt fixed GCPs to carry out RTK GNSS surveys. The DSMs comparison can be carried out only if the 3D surfaces are in the same Datum. The accuracy of the XYZ coordinates of each Master Station define a first uncertainty of the procedure.

The authors have to be asses this uncertainty and clarify in the manuscript.

 

Line 159 -> The errors concerning the GNSS survey of 1 cm and 2 cm, respectively are not real.

Line 532 – Leave the “UAV” correct term in the titles of the articles listed.

It is strongly recommend to prepare references following the “Instructions for Authors”.

Insert the doi number.

Author Response

We would like to once again express our gratitude for their valuable comments and time they dedicated to review our work. Regarding the changes in the References, we indeed by mistake changed all UAV abbreviations to UAS across the text, without thinking about the changes on the references. We are thankful to the reviewer for noticing this error, which we have naturally addressed. Finally regarding the errors and all improvements/shortcoming of certain aspects of the geomatic procedure of our experimental pipeline, we definitely agree with the comments of the reviewer. However, for this specific application, as the influence of such error component can be considered neglectable, ergo not significantly affecting the obtained results, we have decided to prioritise these optimizations (and accompanying information that are requested) in our upcoming publication on the same topic (after this season’s data collection), utilising and fully covering all concern made by the reviewer.

Reviewer 2 Report

I would like to thank the authors for the conscientious work they have put into this review. Regarding the doubt about the correlations, since the data do not present an apparent linear relationship, the so-called Continuous Analysis of Variance (CANOVA) should be used. You can find more information in this article (Wang, Y., Li, Y., Cao, H. et al. Efficient test for nonlinear dependence of two continuous variables. BMC Bioinformatics 16, 260 (2015). https://doi.org/10.1186/s12859-015-0697-7). However, I have rechecked the Pearson correlation values and they are values that could be taken as representative of the relationship between the variables. Therefore, the authors could decide to keep this analysis, as long as they clearly state in the text that "despite the absence of a clear linear relationship, in this particular case the Pearson's linear correlation coefficient was sufficiently robust to detect a correlation between the variables studied".
The authors have definitely addressed all my concerns and, therefore, I consider that the article should be published in its present form.

Author Response

We are glad that we have successfully addressed all of the reviewer’s comments, and we would like to once again express our gratitude for their invaluable comments and time they dedicated to review our work. Regarding the statistical analysis, we attempted to use CANOVA in Python, but unfortunately we could not make it work as the documentation was fragmented and we did not have enough experience to produce results that we would feel confident in publishing. To this end, we tested several other correlation indices, such as Spearman’s Correlation index (as mentioned during the revision process), and the results were very similar to the ones obtain using Pearson (Spearman for instance, yielded slightly lower values overall). Therefore, we have decided, according to the reviewer’s suggestion, to keep the Pearson results, and naturally, we have included the proposed sentence in the respective Section (2.6 Statistical Analysis).