Intra-Annual Variabilities of Rubus caesius L. Discrimination on Hyperspectral and LiDAR Data
Round 1
Reviewer 1 Report
- The time period of the variabilities of Rubus caesius L. should be clearfied in the manuscirpt.
- I suggest to merge the two sections: "2. Aim of the study " and "3. Object and area of research "
- Lines168-171, “during the first campaign, 50 patches in which this species dominated…, and each polygon contained a compact and homogeneous dewberry patch”, and line 178 “During the three campaigns, we acquired a total of fifty polygons with vegetation…”. Is this a contradiction? Do the other two campaigns have no polygons?
- Lines 179-183, Do these polygons (50) not contain R.caesius? Figure 3 is not show polygons of Bidentetea tripartiti, Filipendulion ulmaria and Cirsium arvense, why?
- What is bands resolution of hyperspectral image?
- Lines 232-234, form table 2, I think that other classes (non-Rubus) were divied based on dominant species and vegetation classes. Please check it carefully.
- Lines 236, Do reference polygons refer to the surveyed polygons described above?
- The meaning of the histogram (for Rubus 40-60%, 70-80%, 90-100%) is not clear in Figure 6.
- Figure 8 should be more beautiful, such as the abscissa scale line (C1-3) and the other two borders.
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
General Comments
The authors have undertaken a detailed study of the utility of hyperspectral imagery and Lidar for the purpose of mapping Rubus plants in a mixed-vegetation landscape. The methods of the study are generally reasonable and well described. The results are well presented too. One general suggestion is to make the manuscript more concise by about 300 words (especially the Introduction, last part of Results, and the Discussion section), by keeping only the most important information. Doing this will make it easier to read. The quality of English in this manuscript is impressive.
This reviewer finds a key weakness in the accuracy of the objectives statement (lines 112-113). This study does not address the functional traits of the plant species (which relate to water use, competition, response to disturbance, nutrient use, etc.). It seems that the aims of this study were:
- to evaluate a wide variety of hyperspectral vegetation indices and structural metrics derived from ALS for their utility in discriminating Rubus at different cover densities in spring, summer, and autumn
- to determine which data are most discriminant of Rubus under different growth and pigmentation phases
The manuscript should be reviewed carefully by the authors to improve any statements that could be unclear. Below are some specific recommendations to achieve this (but authors should not limit revision to these comments).
19: recommend removing unneeded “which is an example of” portion of sentence
21: recommend ‘monitoring plant distributions over large areas requires mapping that is fast,…’
21-24: the purpose of these two sentences is not clear, and they could be removed, or made clearer
41-44: this first sentence could be more concise
51: unclear what is meant by “brambles limit their populations”. “Their” could refer to the brambles or the other plants, so this sentence could be improved by stating that ‘brambles limit the populations of other species by competition…”
164: 30-31 of May is not summer, but rather late spring. Please correct this here and throughout the manuscript. Summer begins June 21.
173: please remove “/”, and use either “and” or “or”
181: More accurate to say that ‘the following species were present within the non-Rubus reference plots:…’, unless the polygons were actually smaller in scale than the plant species patches.
185: how can the plots “move” or “graze”? Please revise.
204: Formulae of the spectral indices should be described, but I do not see a Table S1 in the manuscript. Please include.
225: Do you mean twelve layers derived from Lidar that relate vegetation structure? It is not clear whether you estimated vegetation height based on difference between first and last return of the laser pulses (the common approach). Please clarify in the text about the structural metrics you derived.
248: Please give citation for Linear Discriminate Analysis so readers can seek further information on this technique.
259-261: It is unclear how the single pixel samples are related to the polygon samples you obtained. Please clarify whether you sampled only one pixel from each polygon, or several from each polygon.
286: It seems you have eliminated the ALS data layers simply because they were less correlated to Rubus cover than the spectral data (85 percent correctness is very good). This makes sense, but it seems that you did not explore how the ALS data might improve the classification accuracy IN COMBINATION with the spectral data using multiple predictor variables or criteria.
Figure 6: the signatures in the bottom two graphs are squeezed at the bottom. It would be best to change the max values on the Y axes in these graphs to 6000 rather than 10000.
Table 4: in this table or nearby, it would be nice to include wavelengths the key spectral bands used in each of these indices that performed well.
400: suggest ‘the ALS data were less useful in this application’ rather than ‘worse’.
404: unclear what is meant by ‘different vegetation condition’. Do you mean phenology or pigmentation?
438: Please specify the cause for the shifts in the spectrum. Changes in incoming radiation?
451: It is unclear how noise resulted from errors in data acquisition. Either there was a problem with the sensor (if so, please specify exactly what it was), or there was atmospheric interference (which seems unlikely, but if so should have been corrected for). Please clarify.
515: somewhere in this paragraph, you should comment on the spatial scale of the Rubus patches relative to Sentinel 2 resolution. If the patches are typically smaller than 20 m in dimension, than linear spectral unmixing might be necessary with Sentinel.
Comments for author File: 
Comments.pdf
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
This manuscript reports the result of using hyperspectral and LiDAR data for Rubus caesius classification. Authors collected their images at three different stages during summer and analysed their hyperspectral images through different techniques. In general, the topic is interesting while the collection of the ground data for species percentage coverages was not reported through the whole manuscript. Authors must add the ground data collection methodology and how they gave the coverage percentages for each sampling point. However, I have the following specific comments:
L23: I suggest removing this sentence as it is not adding to your abstract.
L78: Remove or move this sentence to your methodology.
L116: What do you mean by remote sensing layers? Do you mean spectral bands?
L118: I do not see any need to have this separate section. I suggest moving the part from L119 to L145 to your introduction section and the rest to the methodology as the study area.
L169: How did you measure these percentages of coverage?
L215: What do you mean by salt and pepper effect??
L231: You have to give more details on how did you measure these species percentage coverage? This part is not clear in your methodology.
Figure 6: Add X and Y-axis units.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 4 Report
The paper ‘Inter-annual variabilities of Rubus caesius L. functional traits using hyperspectral and LiDAR data’ deals with the use of hyperspectral and Light Detection and Ranging (LiDAR) in differentiating and classifying the European dewberry (Rubus caesius L.), different the vegetation patches. A case study in southern Poland was provided also taking into account the seasonal variability (early summer, summer, and autumn) and different percentages of R. caesius within the other vegetation types.
General comments
About the paper, in terms of proposed methods, it is clear what the content is. Moreover, I appreciate the coupling use of Lidar and hyperspectral data in analyzing natural vegetation components.
What is not understandable for a potential reader is why differentiating R. caesius should be worthy of interest. On the one hand, they highlighted the role of R. caesius as an ecologically significant component of natural or semi-natural ecosystems while, on the other hand, it represents a weed in maize or winter wheat crops. This study seems to fall in the first case but there is the need to specify why it is crucial to differentiate just this shrub species. In addition, this explanation should be provided in ecological and planning terms of Natura 2000 site management. In other words: in a Natura 2000 natural or semi-natural site, obtaining this knowledge of R. caesius presence and distribution in which terms could be of interest? Moreover, the authors must clearly state why they propose to differentiate R. caesius instead of proposing a classification of all study area vegetation classes.
Having said that, I want to be clear. I consider this paper worthy of interest to be published in this journal. What is needed is to clarify the aim of this piece of research clearly.
Going to the manuscript content, the final part of section 1(lines from 98 to 107) refers to the study aim and the proposed datasets and methodology. In section 2, the aim of the study is reported. I think these two mentioned parts should be merged. This also avoids having a section (2) with jus ten rows.
About the reference polygons, it is not clear if they are circles or not. In this case, I suggest changing accordingly along with the manuscript.
About the methodology section, the authors must specify how they ensured the geometrical co-registration of the ALS and Hyperspectral datasets. Moreover, while the authors clarified they use a GNSS to geolocated reference polygons, they do not refer to ground control points (GCPs). My concerns are about the source of Z dimension. It is referred to a specific geoid (in this case, there is a need for GCPs with this information or the application of PPK or RTK techniques) or just referring to the ellipsoid? Obviously, in this case, the Z information is less precise.
Moreover, I ask the authors to provide additional information on surveying reference polygons with the GNSS mobile mapper 120. Which survey technique has been used? As for example, single band absolute positioning Real-time DGPS, Real-time SBAS, Post-processing.
About the LDA and NPMANOVA, I ask the authors to specify if they performed these analyses just for the first campaign or for the three datasets. I guess they performed these analyses for all the three datasets, but I suggest clearly explaining this.
About the results and discussion, I really consider interesting what the authors stated about the potentiality of satellite information such as the Sentinel-2 platform. Also very interesting satellite data are those surveyed by the World-View3 sensor that couple a high spectral and geometrical resolution. In this direction, interesting applications investigated the use of WV-3 also to derive vegetation indices at tree crown detail in the framework of geographic object-based image classification (GEOBIA) techniques.
I understand this paper is showing a part of a more comprehensive and ongoing research that is developing new insights. By the way, in the discussion, I do not agree to consider as separate information the spectral (i.e., HS dataset) and the Lidar data layers. Indeed, as the authors suggested, there are several studies coupling topographic and spectral information used in machine learning algorithms that improve the classification also using lower spectral resolution and free and open data. In this direction, I suggest the authors to consider the following research papers: doi 10.1016/j.jag.2018.10.009; doi 10.1016/j.compag.2020.105500; 10.1016/j.jag.2018.05.003; doi: 10.3390/rs11101238).
Technical comments
Keywords: the use of only acronyms does not make sense (e.g., ALS, LDA, etc.).
Line 193 : please specify what INS data means.
Lines 193 - 195: I ask the authors to refer to the websites and version of the used software (ReSe Applications in this case). This helps a potential reader to understand the proposed workflow and used tools. Moreover, this information if also needed in case of the OrthoVista software, now part of the Inpho suite of Trimble.
Lines 200 – 204 : reporting all the 49 acronyms of the calculated remote sensing vegetation indices does not make sense.
Line 248: add the acronym LDA.
Line 423: MNF stands for?
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 3 Report
Authors modified their manuscript and I do not have further comments.
Author Response
Thank you for the information.
Reviewer 4 Report
The authors responded in a quite satisfactory shape to all the reviewer's comments.
Indeed some minor changes are required. For example, about the WV-3 data. I understand that these satellite data are not free of charge. By the way, in the proposed research the authors used airborne hyperspectral and LiDAR data that are very expensive. Therefore, I suggest to reshape the sentence about WV-3 data, hopefully highlighting some research papers on using WV-3 and GEOBIA that I consider very promising in the further steps of this interesting topic.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
