Seeing Trees from Drones: The Role of Leaf Phenology Transition in Mapping Species Distribution in Species-Rich Montane Forests
Round 1
Reviewer 1 Report
Dear authors,
The paper presents the results of applying different object-based classification methods in order to distinguish tree species in a mountainous subtropical forest of Zhejiang Province, China. The considered species in the study are: Cryptomeria japonica (L. f.) D. Don, Liquidambar acalycina Chang, Ginkgo biloba L., Pseudolarix amabilis (J. Nelson) Rehder, Litsea auriculata Chien et Cheng, and Pinus taiwanensis Hayata as well as standing dead trees and canopy gaps. For the classification, the authors used a mosaic image acquired with a UAV on a single date in October 2017 (leaf senescence period). They tested different sets of features (spectral, geometrical and textural) as the main drivers for the classification. Moreover, to test for the hypothesized role of topographical heterogeneity in influencing the classification of tree species, they performed two analyses: one considering the whole area and a second one distinguishing three types of habitats based on elevation and slope characteristics.
In my view, the objective of the paper be of great interest to be published in the journal Forests. However, I recommend major revisions and clarifications under many aspects prior to publication, especially in the methodological part.
General comments:
I advise a revision by an English native speaker. The overall manuscript is understandable, but many sentences are difficult to interpret or not entirely clear in their meaning.
All figures presented in the supplementary material should be included in the main body of the article. The authors refer to them multiple times in the manuscript and they are needed to follow the reasoning and presentation of the results and methodology. Some of them should be revised (see specific comments).
From a methodological point of view, it is not clear how the comparison between the classification based on the whole area and the classification based on the individuated three different habitats was performed. Moreover, it is stated in the text that additional ground reference data was collected for the second one. If the comparison happened between two classifications based on a different training set of data, it would constitute a major methodological flaw of the work. Finally, these three habitats were used to take into account topographical heterogeneity in the classification, but it is not provided a clear explanation on how such heterogeneity might have influenced the classification in the reviewed study.
Specific comments
Introduction
- Line 102: you refer to the study are size in hectares (ha), while later in the methodology you use km2. Please be consistent.
- Lines 107 – 114: separate the different research question (RS). You list 3 RS but they are actually 5.
- Line 108: you write ‘to improve the accuracy…’, what is the reference term of this improvement? Moreover, here you talk about ‘spectral and texture features’ while in the next line you add ‘spatial geometry features’. Please be consistent.
Methods
- Lines 142-143: add some more information about the target species of your study. Which are deciduous and which are evergreen? You give this information scattered through the manuscript. It is a rather important information for the understanding of your methodology and results.
- Line 153: Please be more specific, which of your target species?
- Line 170: Provide more explanation of why you decided to reduce the extent of your study area. Were these distortions related to terrain topography?
- Line 187: it is not clear whether you considered sample trees only above certain dimensions. E.g., you considered a DBH threshold when collecting ground data? If so, please specify.
- Line 194: same as line 187.
- Line 196: the sampling scheme is a bit unbalanced, both in the number of samples per species and in the spatial distribution of tree samples within the study area (fig. 2a). I suggest that you provide some justifications in this sense.
- Line 278: The figure you refer to in the supplementary material (figure S1) should contain in the caption the reference for the sample size at which Cryptomeria japonica reaches 70% Producer accuracy, as it is not shown in the graph (supposedly for visualization purposes).
- Line 279: please provide a reference for considering the threshold of 70% producer accuracy as ‘high accuracy’.
- Line 295: you refer to figure S1 but this figure does not refer to the number of samples for each tree species in each habitat type, please clarify.
- Line 296: you refer to ‘low classification accuracy’, can you please be more specific? Did you run a classification prior to add the extra ground reference points? If so, what results did you obtained and how much did the accuracy changed by adding the extra reference points? Moreover, it is not methodologically sound to say that you added extra reference sample just to improve classification accuracy.
- Lines 298-299: It is not clear what figure S3 represents. Are those all the points used in the classification for each habitat? Moreover, figure S3 is not easily interpretable to distinguish how many samples for each tree species are present in each habitat. I suggest adding a table with the N° of samples for each tree species in each habitat.
- Line 299: How many additional reference points were added? Please clarify.
- Line 301: please provide a justification of why you did not consider the CART classification algorithm for the classification within each habitat.
Results
- Line 353-354: Please specify what you mean by ‘inconsistent classification accuracies’.
- Line 371-378: The whole paragraph is very unclear. What did you consider as a reference for a ‘realistic distribution’ of the tree species? How was the comparison made? Was it a comparison between the spatial distributions of the tree species, a comparison between the number of individuals per species in each classification or a comparison between accuracies?
- Line 376: clarify what you mean by ‘patches were distributed abnormally in the combined map’.
- Figures S4 and S5: consider providing some more information about the data contained in the maps (e.g., cover %, number of detected trees etc…). It is very difficult to make comparisons just by a visual inspection.
Discussion
- Line 402: the example you provide is contradictory, since the cited study actually found textural features to play a role in classifying tree species.
- Line 418: provide explanation of why Metrosideros Excelsa is a comparable species to the ones used in you study. You mention flowers’ spectral characteristics of the species while in your study you focus on leaves.
- Lines 424-425: The differences in the phenological phases of plants from the same species might be due to altitudinal differences. In this study, the considered altitudinal gradient might have been too ‘short’ (ca. 350 m) to show time shifts in phenological phases of plants, but that is something that should always be taken into account when working in mountainous areas.
- Line 450: consider using another term than ‘invisible wavelength’, e.g., ‘wavelength outside the visible spectrum’.
- Lines 481-484: the results you presented refers to the spatial distribution of tree species only.
Author Response
Dear reviewer,
Please find our point-to-point responses in the attached files. Thanks.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The manuscript is well written, most of the methods are clearly explained, the results are well described and discussed. I encourage the Authors to develop the section “2.4.1. Image segmentation”, namely, how the segmentation parameters, shape and compactness, were determined (lines 222-223).
Author Response
Dear reviewer,
Please find our responses in the attached files. Thanks.
Author Response File: Author Response.pdf
Reviewer 3 Report
The paper presents an interesting study of the ability of unmanned vehicles to map tree species in topogrphically complex areas in the mountains of China.
I think that the paper is good and well written; only the english is sometimes poor, in particular in the discussison and should be imporved; I provided some suggestion in the notes in the t ect
Abstract
The bstract is ok but the aim of your paper is aminly to test the classification potentiality of UAV and remote sensing in montaner areas and this is not very evident from your abstract; you should better focus the abstract to show the main meaning of the paper
Inotrduction
The introduction is good but you shopuld connect the last part the deals with the topographic complexity to the rest of the introduction
Methods, resutls
Ok
Discussion
The discussion si very itneresting but the english is poor and a revision is needed
Comments for author File: Comments.pdf
Author Response
Dear reviewer,
Please find our point-to-point responses in the attached files. Thanks.
Author Response File: Author Response.pdf
