UAV-Based Characterization of Tree-Attributes and Multispectral Indices in an Uneven-Aged Mixed Conifer-Broadleaf Forest

Round 1

Reviewer 1 Report

Review of the manuscript ID: remotesensing-1715662 entitled ‘UAV-based characterization of tree-attributes and multispectral indices in an uneven-aged mixed conifer-broadleaf forest’

The assessment of resources and biodiversity of forested areas is one of the main currents of environmental research using remote sensing data and methods. The Authors join this trend with a study showing the possibilities given by the application of unmanned aerial vehicles (UAVs) for the multi-parametric evaluation of forest canopy and the ecological processes occurring in it. In order to characterize the natural forest stand of El Cordoncito forest located in Mesa de Pawiranachi, in the municipality of Guachochi, in the Sierra Madre Occidental Mountain range of northern Mexico. Using photogrammetric data obtained with a DJI 4 UAV, the Authors proposed a multi-parameter framework allowing the calculation of nine individual tree metrics, including several multispectral indices. Analyses conducted in the free software environment OpenDronMap and QGIS, allowed to calculate, among others. The analyses conducted in the free software OpenDronMap and QGIS allowed to calculate among others: tree height, canopy area, number of trees and multispectral indices (NDVI -normalized difference vegetation index; GNDVI -green NDVI, LCI -leaf chlorophyll index; NDRE -normalized difference red edge index; OSAVI -optimized soil adjusted vegetation index; RVI -ratio vegetation index; TVI -transformed vegetation index, NDGI -normalized difference greenness index NIR -near infrared band; RED - red band; RedEdge -red edge band; GREEN -green band). Although the area analyzed is characterized by a significant dispersion of trees, the authors indirectly confirm the weakness of this study. An important limitation is that automated tree detection and tree quantification were disrupted by the presence of overlapping crowns. Admittedly, the Authors proposed a new way of mapping stand density and estimating crown attributes to prevent this, but the study lacks an example from an area with high crown density to confirm effectiveness. However, assuming that areas with high tree dispersion are to be the subject of analysis, the framework proposed by the authors is a useful tool for estimating stand parameters that can be an effective supplement to forest inventories. As demonstrated by the authors, the obtained metrics are reliable and can serve as input to models in modern forest area inventories. The biggest weakness of this study is the lack of analysis of the accuracy of tree crown autodelimitation. In the discussion, the authors acknowledge this fact while emphasizing that better results were clearly seen in deciduous species, but these statements were not sufficiently supported by evidence. The topic of the paper is interesting and fits well with the remote sensing and special volume topics. The layout of the paper is correct and the proposed framework has utilitarian potential. I believe that the paper should be published in spite of some weaknesses of this work, and these weaknesses should give impetus to the authors to develop the scope of the study, deepen it in the direction of making the process of self-modeling more plausible, and conduct comparative effectiveness studies comparing the proposed framework with the effectiveness of LiDAR-based studies.

Author Response

Please find PDF attached.

Author Response File: Author Response.pdf

Reviewer 2 Report

Vivar-Vivar et al present a UAV photogrammetric and multispectral study of a complex forest site. The goal of the study has merit, and I also found the study interesting to read.  I was happy to see the authors using open-source SFM-MVS / segmentation software to achieve reasonable results in their height comparison for a complex site. The main negative was the multispectral data presentation, and especially the interpretation and discussion of multispectral results. Before this is published, please address the issues with the multispectral data outlined below.  

To be clear, NDVI is a very indirect measurement of plant photosynthetic activity or “productivity”, as are most indices.  Red edge indices can be used to get a handle on chlorophyll which relates tangentially to photosynthetic rate. However even this relationship is not straightforward. The reason I am emphasizing this is because some statements regarding multispectral data in the discussion are not only inaccurate but wrong e.g. “Unlike NDVI, the TVI is sensitive to crown structure” [L344]. The NIR plateau, and by consequence the NDVI*, is mostly a function of canopy structural properties such as leaf area or leaf angular distribution, up to a saturation point. Further, the “success” of some indices is over emphasized. Without validation or calibration data (i.e. pigment measurements), the authors are on thin ice when using reflectance spectrometry to infer relative pigment contents, as a proxy of vigor, across multiple species.  To fix this the authors need to dial down the inferences/conclusions drawn from their data. Canopy reflectance spectrometry is affected by multiple factors which need to be carefully considered; these include: LAI/LAD, view angle/sun directionality, atmos scattering, sunlit-shaded fractions, specular leaf reflectance, the list goes on. Section 4.2 needs a complete re-write, based more thoroughly in the classic (non-drone) literature.  

*Gamon, J.A., Field, C.B., Goulden, M.L., Griffin, K.L., Hartley, A.E., Joel, G., Penuelas, J. and Valentini, R., 1995. Relationships between NDVI, canopy structure, and photosynthesis in three Californian vegetation types. Ecological Applications, 5(1), pp.28-41. 

A couple of separate points on the Multispectral data. It would be useful to see some basic plots of multispectral reflectance, i.e. band values as a function of wavelength, for a couple of crowns. The NDVI values look a little low to me, though you do have juvenile trees. Finally there is a only a very cursory description of data collection. What is the instrument specs (band locations, FWHM etc)? What were the sky conditions? How were you normalizing from DN to reflectance (panel, irrad. sensor?)?   

A few of the graphs are also low quality e.g. Fig 5, lower panels of Fig 1. 

Minor issues with line numbers below: 

L15 should you define UAV here? 

L21 OpenDroneMap typo (missing e) 

L71 refer to “non-visible” here but use a sensor with presumably visible to NIR bands? Perhaps remove the term visible. 

L75 goal 2 was to investigate the “applicability”, but the reason (e.g management, carbon estimation etc) for the application is missing.   

Figure 1. Climograms hard to read. Increase font size/quality. 

L91. Latin names of plants please. 

L99. Why measure in October? What were the day(s), sky conditions, temperatures on the day?  

L103. Incomplete sentence “which were ….” 

L112 cite the ForestTrees package here, likewise ODM. 

L113 Missing “E” from ODM 

L113. How long did your ODM run take? On what sort of computer? Did you set quality to maximum? You can look at your ODM log file to see how long it took. 

L120 did you use Ground Control Points? If not, then perhaps note this in the discussion and the possible impacts on accuracy. 

L130 Spectral reflectance indices are not a panacea. They are senstivie to many factors, pigment content being just one. So their question of adequacy is debatable.  

L135 centre wavebands and FWHM would be more useful than names of bands 

L142 How much does reflectance tell us about productivity? This is a very complicated question 

L149 Video is a good idea, but Vimeo requires a login in to view your video. Any reason why you did not use youtube? 

Figure 5. This is low quality, improve this graph to make it publication quality. Add some units to labels. Draw 1:1 line. There is clearly an offset from the 1:1 line, what is that average prediction error in meters please?   

L196. These are residuals of regression model. However the error from the 1:1 line is also of interest. 

L215 “attributable to different levels of stress”. This conclusion cannot be drawn without independent measurements of “stress” on the ground.  

L258-262. This is over-exaggeration of the impact of the current study. Drones can have an important role in management, but progress is incremental. The authors also did not directly quantify/qualify how this study improves forest management. Remove these sentences or re-write.  

L316 Quantify this bias as prediction error Vs 1:1 line. 

Section 4.2 needs a re-write. Link more to the literature, and do not just go through each reference one by one.  

Author Response

Please find PDF attached.

Author Response File: Author Response.pdf

Reviewer 3 Report

The article is analyzing UAV images and workflow for obtaining information about tree parameters in "El Cordoncito" in Mesa de Pawiranachi and comparing it with in situ measurement. The article is well structured and concise, but there is some thing should be further explained:

Line 103 - missing sentence?!?

Figure 2 - Correct english text in figure, ie. multiespectral

Line 115 - What methods were used - you just mentioned that you used SfM - which SfM since there are several others (with explanation how it works).
The same for obtaining orthomosaic, DSM, DTM

Line 225 - There is no figure 8.

Figure 7 - Which value was used for vegetation indices, max? if so, why did you used max, and not median or arithmetic mean?

Line 600 - References?!?!

Author Response

Please find PDF attached.

Author Response File: Author Response.pdf

Reviewer 4 Report

Dear Authors,

I think that your manuscript entitled  “ UAV-based characterization of tree-attributes and multispectral indices in an uneven-aged mixed conifer-broadleaf forest” is systematic and scientifically interesting, as well as brings new and valuable ideas and results to forest remote sensing.

This work is interesting from a both scientific and practical point of view, it is also tidy, hence it is suitable for publication in Remote Sensing in the present state.

                                                                                                                                                                                                                                          Sincerely yours

Reviewer

 

 

Author Response

Please find PDF attached.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

The authors did address most of my points, so I commend them for that. However, the discussion still falls short and the authors are extending too far with their interpretation of multispectral data. In simple terms airborne reflectance spectroscopy is really not the same as photosynthetic rate and should not be interpreted as such.  

The following phrase sums up the issue with the discussion (L433): 

“The differential reflectance rates denoted differentiated ecophysiological processes” 

The authors did not measure photosynthetic rate, pigments, or any other ecophysiological process. What they have is multispectral reflectance measurements. At my request, some examples of data are now shown in Fig 6. These spectra look ballpark correct, albeit with a positive bias of 15% or so suggesting the calibration/referencing method needs work in future flights. The point remains that leaf reflectance can be used to estimate pigments (e.g. chlorophyll or sometimes minor pigments), and that pigments are related to photosynthesis via light capture. But this is at least two steps from canopy/shoot UAV data, which are shown here. Put simply, UAV reflectance  does not directly indicate photosynthesis. It is my opinion that the authors need to edit section 4.2 and show that they understand this critical point. Particular places for attention include: 

 L379 “vigor”. What do you mean by this? 

L384 “it presents medium to weak conditions of photosynthetic rates across all of the genera”    

You cannot say this from reflectance data presented in this study. NDVI (or other indices) has sometimes been used as a proxy of fAPAR (abosrbed light), to subsequently model photosynthesis (GPP) where GPP = fAPAR . LUE. But you see there is still an LUE term? Reflectance does not equal photosynthesis even in the LUE model.  

L396 “water” By what mechanism? See above. 

L407 “abrupt changes in photosynthetic rate” Completely disagree. This index may relate to chlorophyll or canopy structure. That is not photosynthetic rate.  

L409/409 I do not understand this sentence.  

There are also a few minor grammatical issues, especially in the new text; a further check could be helpful if possible.

Good luck!

Author Response

See Pdf file attached

Author Response File: Author Response.pdf

Reviewer 3 Report

 

The article is much better, but there are still some corrections that should be made.

Lines 131-136 - You still hasn’t answered and explained used SfM, DSM and DTM – how were DSM and DTM generated (which method). I understand that you used OpenDroneMap, but what are the methods used in this. For example, for SfM you have multiple tools for point cloud generation, for example VisualSfM, OpenSfM, etc., which generates points in slightly different way. So, I think that the reader, how is not completely informed about OpenDroneMap software, should know which methods were used for SfM, DSM and DTM generation.

Line 451 – reference 71 should be in square brackets not round

Author Response

See Pdf file attached

Author Response File: Author Response.pdf