Predicting Leaf Phenology in Forest Tree Species Using UAVs and Satellite Images: A Case Study for European Beech (Fagus sylvatica L.)

Round 1

Reviewer 1 Report

Overall an interesting work, here some points for clarification and further improvement, particularly regarding the applied statistics: 

* Fig. 1: include site labels and a legend (I assume this is supposed to be a topographic map, in accordance with the elevational gradient mentioned in the study design?)

* Table 3: some sentences on the applied vegetation indices should be included in the Introduction, to introduce them somewhat more broadly: how do they work, what do they show and how have they been used in previous studies? How do they differ among each other (i.e., why not just use the most common ones like NDVI? what added value do you expect by including further indices?)

* Statistical analysis: some general knowledge sections (e.g. on correlations, regression) could be shortened substantially

* Statistical analysis: it is often recommended not to use location variables such as latitude and longitude in machine learning algorithms ("ID-effect"); please discuss this in the context of your work (see e.g. results Fig. 8), including references to other studies

* Table 5: it is often recommended to have 15-20 observations for each predictor or interaction term in multiple (linear and non-linear) models; as the models presented in Table 5 partially have a very large amount of parameters to be estimated, this needs to be justified and discussed further in Methods and Discussion in the context of current literature on model overfitting; I would further suggest to present more model performance metrics (such as R2) for the compared models.

* also Table 5 / Statistical analysis: in (linear and non-linear) multiple modelling approaches, care should be taken with multicollinearity. it is recommended to only include independent variables as predictors, i.e. variables that are not closely correlated (-0.7 < R < 0.7) among each other. As can be seen from Fig. 3, many of the predictors used in the models in Table 5 are closely correlated (e.g. GLI and NGBDI, R = 0.83), which is critical from a statistical point of view and needs to be addressed. 

* Fig 7. : there is a typo ('fenology')

* in general: the presented results are quite lengthy and somewhat fail to put the focus on the key outcomes by presenting several display items. I suggest to present some of these results as supplementary information rather than in the main text (e.g. the correlation matrices). 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

The manuscript is well structured and well written: According to the present reviewer some revisions have to be performed before going ahead. If authors well follow the suggestion given, I will certainly recommend this manuscript for pubblication on Remote Sensing.

Firstly I suggest you to discuss a little bit more in the introduction section the role of climate change in phenology in also different areas of the globe before focusing as you correctly done on the area of study. Therefore, I deeply suggest you to consider to include these works in the introduction in order to improve the quality of the work and better depict the role of remote sensing and phenology. Here it is the work to be considered:

-  https://doi.org/10.3390/rs12213542

- https://doi.org/10.1016/j.rse.2021.112568

Then, I suggest you to include in teh manuscript the UAV flight parameters in a supplementary material section or directly inside the material and methods. They are crucial and at this time are completely missing. I suggest also to include the calibration parameters. (i know you perform many filght but I guess you have standardized all the acqusition with same parameters in terms of flight speed, camera angle etc) 

Moreover, I sugget you to include the software adopted to performe the entire processing and workflow. I saw Pix4D but not R or other to retreive and compute phenological metrics. I suggest you to include the settings.

Finally, please include in each map reported into the manuscript in the caption sections or directly into the maps the datum, reference system, nominal and representation scale in a geomatics and remote sensing journal is highly appriciated.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

The use of UAV as a surrogate of field observations has been proved to be satisfying regarding the track of phenology of fagus sylvatica trees at the crown level. The use of satellite data time series has shown promissing results in tracking the phenology at the stand level.

The research topics are very interesting and the data sets (ground, drone, satellite) are appropriated. Nevertheless, the quality of the presentation is weak and needs to be considerably improved before publication. Among other things, I think that the quality of the written English needs to be improved in order to make the research more understandable. The figures requires also certain improvements.

The authors have decided to works with two differents remote sensing products ; UAS imagery and satellite derived observation. I think that the paper focus more on the UAS imagery. They should give more place to the analysis of satellite biophysical parameters. For example, in the introduction, the number of sentences that introducte the use of satelitte images for phenology tracking is limited to L48-51 ,L67-71 and L89-90.  At least some additionnal description of previous scientific work on the same topic should be proviced, as well as the opportunities of using phenology information at the stand level (health status monitoring, forest species discrimination, etc). In the Material and Method section, further description of the satellite imagery used should be added ; temporal and spatial resolution of the biophysical parameters and from which sensor it has been generated.

 specific remarks

L11-14 ; sentence too long, review

title and intro : UAVs and satellite images are presented together, it should  be explained clearly that they are used separately.

L32 increasing trend of temperature : review sentence formulation, the temperature increase, not the trend

L57-61 sentence too long, review

L62 it should be added that UAS and satellite will probably be used complementary in the future. UAS is not really an alternative to satellite imagery, exept in Country where cloud are almost permanent.

L83 site index refers usually in forestry to the productivity potential of the forest site. I suggest that you  change site by stand or something more appropriated.

L86 leaf phenalolgy is introducted but not really defined anywhere in the document.

Fig 1: extend of the main map look totally inappropriated, as its does not shows the altitudinal gradient. I suggests that you zoom in

L110-114 paragraph is about the methodology of phenology tracking. This is not very well written and should maybe be moved after L120

L132; "for a bud we..." do you means for a branch?

L151 : Why a 70° camera angle and not a nadir one?  Specify the average ground sample distance, and the one use for computing the orthophotos

Fig 2 label of images are too small.

L161-162 : ground control point are usually used both for georeferencing and for camera (self-)calibration. They thus are used during the photogrammetric process, during the camera orientation computation by bundle block adjustement.  Why haven't you used GCP in Open Drone Map?

L163 : image was noramlized--> how

pixels classified as shadows --> how are they classified?

section 2.4.1 : expend the description of biophysical parameters;  temporal and spatial resolution

formulas 3, 4, 5, 6 ; very simple and generalistics formulas, I do not think you should present them in your manuscript

L253 the variables data series were normalized --> how

location was added as predictors ; it is X and Y coordinates, or site number? explain

L287 model 3 ; harmonize model 3 and model three in the document, the same for model 10

Fig 7 label fenology --> phenology

Fig 8 change label %incMSE and IncNodePurity

Fig 10 caps for accronym, FCover instead of fcover, etc

Discussion : you should discuss the fact that the camera you used is low grade camera which is not good in term of radiometric accuracy. It is even more important than any consideration about the choice of the proper VI or the proper ground sample distance

L410 : reformulate, the resolution is important and best VI changes with resolution, not with fligth altitude.

L443 reformulate

L469-470 reformulate

L485 phenology

L491 harmonize fcover as FCover

 

 

 

 

 

 

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

All my previous comments were addressed, I recommend to accept the manuscript in its present form. 

Reviewer 3 Report

Dear authors,

Thanks you for your review. I just want to point out that your SfM workflow seems to be poorly understood.

"Due to the rough terrain in the two
sites (Appendix A. Figure A 1), we used double grid missions to capture pictures at a 70°
camera angle and 80%/ 60% overlap for a high-quality result of the Digital Surface Model,
which increased the quality of the resulted ortho-photo."

 

--> On the one hand, increasing the quality of DSM do not improve the quality of orthophoto, even it is counter-intuitive. In particular when photogrammetric is applied to forest canopy. Traditionnally, aerial ortho are rectified on digital terrain model instead of digital surface model in forested area, because it provide a better ortho result.

-->On the other hand, you performed a rather complicated fligth acquisition (double gridded flight plan with oblique camera), which is totally fine and wich has a positive impact on reducing non-linear distorsion of the 3D model, but you failed to describe your camera calibration procedure (use of embedded GPS during the alignement? no use of GCP, although it has most of the time a positive impact on camera calibration and thus on model accuracy).

So what I suggest is that for the next time you either improve you image processing workflow, either you choose a less complex flight plan, especially if you have no use of the canopy surface model.

regards