Next Article in Journal
Partitioning Global Surface Energy and Their Controlling Factors Based on Machine Learning
Next Article in Special Issue
Classification of Nemoral Forests with Fusion of Multi-Temporal Sentinel-1 and 2 Data
Previous Article in Journal
Sensitivities of Quantitative Precipitation Forecasts for Typhoon Soudelor (2015) near Landfall to Polarimetric Radar Data Assimilation
Previous Article in Special Issue
The Best of Both Worlds? Integrating Sentinel-2 Images and airborne LiDAR to Characterize Forest Regeneration
 
 
Article
Peer-Review Record

Mapping Species at an Individual-Tree Scale in a Temperate Forest, Using Sentinel-2 Images, Airborne Laser Scanning Data, and Random Forest Classification

Remote Sens. 2020, 12(22), 3710; https://doi.org/10.3390/rs12223710
by Veerle Plakman 1,*, Thomas Janssen 1, Nienke Brouwer 2 and Sander Veraverbeke 1
Reviewer 1: Anonymous
Reviewer 2: Anonymous
Reviewer 3: Anonymous
Remote Sens. 2020, 12(22), 3710; https://doi.org/10.3390/rs12223710
Submission received: 7 October 2020 / Revised: 9 November 2020 / Accepted: 10 November 2020 / Published: 12 November 2020
(This article belongs to the Special Issue Forest Monitoring in a Multi-Sensor Approach)

Round 1

Reviewer 1 Report

Review of manuscript “Mapping Individual Trees to Tree Species in a 2 Temperate Forest, Using Sentinel-2 Images, LiDAR 3 Data and Random Forest Classification”

This paper deals with an important topic in remote sensing-based forest inventory, and the experiment described here utilizes up-to-date classification methods and remote sensing material.

Structurally the paper is good, although some important pieces of information seem to be missing (see my further comments).

The list of references is extensive (an there are a multitude of scientific papers that could be cited in this context, so naturally not all key papers can be included. However, in some aspects the list of references could be extended (see further comment).

The general quality of presentation is good, as well as the linguistic form of the paper.

I have listed some (mostly minor) comments that should be considered before publication.

Otherwise I think that the paper is quite ready for publication.

 

  1. 63 “ Species classification at individual level…” …individual tree level… ?

 

  1. 66 “the combination of reflectance from the tree crown, tree crown shadows and from the understory, e.g. soil, herbaceous vegetation and litter”

Also the reflectances of illuminated and shadowed parts of the crowns are an important factor.

 

  1. 122-123. Random stratification is a somewhat strange concept to me.

Stratification (usually in the context of stratified sampling) in forest inventory is usually based on using apriori data, often auxiliary data such as remote sensing data, for dividing the population into mutually exclusive subsets, i.e. strata. Often the aim of stratification is to have smaller variation within subsets compared to the entire population. The sample can be picked from the strata based on random or systematic sampling.

What benefit would be achieved by carrying out the stratification in a random way would is not clear to me.

  1. 134-136 & 237-240: It seems that the smaller crowns are considerably smaller than the size of sentinel pixel. In that case the spectral properties represent not the reflectance of a crown but several crowns (or reflectance from crown and understory around it). How was this taken into account?

Figures 3 &5. It was quite difficult to find out, how the original population presented in Table 1 shrunk to the 218 gymnosperms and 343 angiosperms presented confusion matrices.

 

Chapter 4.3. The authors should include comparison to studies with high species diversity as well as utilizing different multi- hyperspectral data sources. The cited articles [14,16,30,66] are not very good references (except [30]) since they either are not tree level classifications or have relatively few species examined.

Author Response

"Please see the attachment"

Author Response File: Author Response.pdf

Reviewer 2 Report

The authors tried to map individual trees to tree species in a temperate forest by using Sentinel-2 Images, LiDAR Data, and random forest classification. However, it’s not fully clear that how Sentinel-2 (10 m) resolution data was combined with airborne laser scanning (ALS) LiDAR-derived canopy height model (CHM) using random forest (RF) for trees species classification. 10 m resolution can cover multiple small trees as compared to 0.5-meter resolution CHM. Please make it a bit more clear how all this information were combined in the prediction of the trees species classification. Please clarify more novelty of this paper as much worthy’s literature available regarding the combination of spectral + structure information for trees species classification. Further, I would say that the manuscript is interesting for possible consideration, and would like to suggest the following detailed comments.

  • Please always used airborne laser scanning (ALS) instead of LiDAR
  • Line 74, you stated that many parametric methods used for tree species classification. Please give examples?
  • How the accuracy was calculated in the methods. It’s between the ground inventory points and classified? Whether the confusion metrics used?

Author Response

"Please see the attachment."

Author Response File: Author Response.pdf

Reviewer 3 Report

 

The manuscript shows the use of spectral and structural parameters from Sentinel-2 and LiDAR data for individual tree species mapping using RF classification.

The authors have conducted an intensive field data collection as well as satellite processing to map tree species in the Veluwe forest of the Netherland and this study provides excellent information on this topic. So I suggest it for publication.  

I have the following queries and suggestion:

Authors should include some pictures of various tree species in the appendix

Authors should include a table of satellite data used in this study

Authors should revise the figure 2 flowchart and include LiDAR data

The authors should include a table that shows the summary of field data (inventory data).

As the authors mentioned about spectral mixing using Sentinel-2, 10m resolution data. So I would appreciate if the authors can clarify what is the threshold of crown diameter to have better accuracy in the identification of tree species. For example: if a tree with a crown diameter more than 100m2 can be detected better as compared to 80m2.

Authors have mentioned that Red-Edge causes low accuracy so which bands of sentinel-2 data were effective to get high accuracy?

Could you please provide more details on which parameters (spectral or structural) has a major role in classification to achieve high accuracy?

What about the effects of a multi-canopy tree structure in the study area?

What about the application of this method in other types of forests like tropical evergreen forests?

 

Author Response

"Please see the attachment."

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Please change the title, and rephrased it from a native speaker. For instance, in the title, the words "in a" repeats twice, and at the end, the word "in a Random Forest Classification" looks awkward

Author Response

Thank you for your comment, we have changed our title to:

"Mapping Individual Trees to Tree Species in a Temperate Forest, Using Airborne Laser Scanning Data and Sentinel-2 Imagery"

Back to TopTop