Next Article in Journal
An Appraisal of Dynamic Bayesian Model Averaging-based Merged Multi-Satellite Precipitation Datasets Over Complex Topography and the Diverse Climate of Pakistan
Previous Article in Journal
Three Decades of Coastal Changes in Sindh, Pakistan (1989–2018): A Geospatial Assessment
Open AccessArticle
Peer-Review Record

Identifying and Quantifying the Abundance of Economically Important Palms in Tropical Moist Forest Using UAV Imagery

Remote Sens. 2020, 12(1), 9; https://doi.org/10.3390/rs12010009
Reviewer 1: Anonymous
Reviewer 2: Bahareh Kalantar
Remote Sens. 2020, 12(1), 9; https://doi.org/10.3390/rs12010009
Received: 20 November 2019 / Revised: 11 December 2019 / Accepted: 16 December 2019 / Published: 18 December 2019
(This article belongs to the Section Forest Remote Sensing)

Round 1

Reviewer 1 Report

 

The authors developed an object-based classification workflow for RGB UAV imagery to identify and delineate palm tree crowns in the tropical rainforest. The technical route is reasonable and the results obtained are acceptable. There is no great innovation in the whole method, but it can provide application guidance for palm tree crown recognition in tropical rainforest area.

Some specific problems are as follows:

 

In line 113,” their identification on large scales could help to support sustainable management of these resources.” However, the use of small UAVs is also not suitable for large-scale research.

 

How to determine the optimal combinations of threshold value and minimum segment size? Whether optimization algorithm is used.

 

In line 237-238, “In order to constrain the classification for the identification of palm tree species, some ground data of soil, water and other trees were included as extra classes to avoid that the classifiers would identify them as palm trees.” However, accuracy assessment about the soil and water was not shown in table 5. Water and tree shadows are likely to be confused.

 

The author calculated 13 texture features and selected 5 to participate in the final classification. The author does not describe how to choose the five best parameters, but directly gives the conclusion.

 

In line 273-274, “The splitting process consisted of intersecting the 5 height classes from the K-means unsupervised classification of the Canopy Height model, with the classification crown mask.” According to this description, the segmentation is only based on the canopy height. Is it impossible to separate trees that are highly similar and close to each other?

 

In line 277-279, “The accuracy of the palm tree quantification for each plot was determined by comparing palm trees that were recorded during the ground data survey and also visible on the orthomosaics with the palm trees detected by the classification algorithm.” According to this description, the reference data for verification is the intersection of field investigation and visible palm trees on the image. Considering that the image can only reflect the situation above the canopy, it may not be able to distinguish the palm trees close to each other visually. Considering that the image can only reflect the situation above the canopy, it may not be able to distinguish the palm trees close to each other visually. Therefore, the so-called intersection mainly depends on visual results. The accuracy of the palm tree quantification for each plot was determined mainly by comparing palm trees that visible on the orthomosaics with the palm trees detected by the classification algorithm. Therefore, the accuracy of the evaluation (the number of palms by classification divided by the number of palms by vision) is actually higher than the real accuracy (the number of palms by classification divided by the number of palms by ground survey).

 

In line 280, suggest revising” Results “to “results and Analysis” since there were many discussion and analyses in the section 3. Of course, the best way is to move the discussion text in section 3 to section 4. For example, In Section 3.2, we should only introduce what is the best feature selected, what is the result precision based on the best feature classification, and compare it with the classification precision of other features not selected to show that the selected feature is indeed the best. The description of why the selected features can improve the classification accuracy should be put into the discussion part.

 

In section 3.3, Figure 6 can only see the trend consistency between the number of palm trees identified by classification and the reference value, and cannot see the quantitative error or accuracy information. Why don't use such quantitative indexes as absolute error, relative error, or absolute precision, relative precision to make evaluation.

 

Section 3.4 is not very related to the theme of the article and can be deleted.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

This paper is about palms tree species identification using object-based classification and random forest, support vector machine, kernel nearest, and recursive partitioning. The authors used unmanned aerial vehicle (UAV) images as data set. In my idea this paper will be interested to reader especially in the field of agricultural and remote sensing although there is not significant novelty in the model. I recommend a major correction for this paper as the method part needs to improve.

There is general literature on UAV and OBIA but it is better to add some literature about classification models as well. The authors need to draw a flowchart of their methodology. In the Materials and Methods, please put number as 2.1 for study area and 2.2 Data collection and so on. The explanation about methodology is not complete. For example, there are not any theory about the models (Kernal Nearest, Recursive Partitioning, Random forest and support vector machine). Moreover, the validation part is missed. The authors need to add a section for validation and put the suitable formula related to the producer, user, and overall accuracy, etc. The explanation about segmentation is not clear. For example, which algorithm are used for segmentation, what is the scale? Line 328 “species like Oenocarpus spp..” please remove extra “.”

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

All my concerns have been solved. Agree to publish

Author Response

Thank you for your time and input for the manuscript.

Reviewer 2 Report

In the figure 2, Orthomosiac should be connected to Mosaicking not DSM. So Mosaicked images will have two products which are DSM and Orthomosaic. In figure 2, in the classification box, please mention the name of classification algorithms.

Author Response

We agree with the reviewer and we have edited the flowchart (Figure 2).

Back to TopTop