Distinguishing Buildings from Vegetation in an Urban-Chaparral Mosaic Landscape with LiDAR-Informed Discriminant Analysis
and Jason V. Lombardi 1,†Round 1
Reviewer 1 Report
Identification of buildings from remotely sensed imagery in urban and suburban areas is a challenging task. In this paper, the authors present a technique using discriminant analysis for distinguishing building from non-building planar surfaces. Using discriminant analysis, the authors grouped potential building polygons into building and non-building classes using the point densities of ground, unclassified, and building points. This should be an extension to software-based building-classification that increases accuracy and feasibility in areas with complex vegetation, and the accurate was up to 95% at distinguishing buildings from non-buildings.
However, there are several key problems in the paper.
1. As the main content of the paper, the description about the discriminant analysis is rather poor.
2. In the method section, the structure of the content is in the format of the report or processing procedure but not a paper.
3. The authors mentioned several problems in the current studies, but why the new method can solve or avoid the existed problems is still not clear in this paper.
4. The discussion section should be improved.
5. The structure of the paper is bad.
Other comments:
1. The title is not corresponding to the content of the paper.
2. What is “discriminant analysis”, it is too simple in abstract.
3. Line 27: it is hard to understand the expression “analyses based on these classified images are only as useful as the classification itself” and “Therefore, it is important to accurately 28 classify the image.”?
4. Lines 89-91,” The dense internal structure of thornscrub prevents a LiDAR beam from penetrating causing several last-return points to occur at the same height (~ 3 m), appearing as a one-story building.” The expression is problematic.
Author Response
Please see the attachment
Author Response File: 
Author Response.docx
Reviewer 2 Report
The authors present a method using discriminant analysis for distinguishing building from non-building planar surfaces. The manuscript is easy to follow with a good structure. The authors calculated proportions of points for each class as input of QDA to distinguish building polygons. As can be seen from Table 2, building polygons can be easily identified from non-building polygons using point densities. So my major concern is that why the authors use QDA method instead of distinguishing building polygons directly using the point densities which I think is more simple to finish the same task. Another concern is that how the authors set the prior probability of a polygon belonging to building type in QDA. Please give more details.
Author Response
Please see the attachment
Author Response File: Author Response.docx
Reviewer 3 Report
The paper is not completely self-explanatory. Therefore it requires details on
(a) Ground extraction process
(b) Building identification process
(c) Formulation of discriminant analysis, suited to the problem.
(d) It is not clear whether the experiments / work is based on raster (created from the pt cloud) or the raw point cloud itself.
(e) There are some papers that have taken a density-based clustering approach on LiDAR point clouds. The authors should prove the advantage of the proposed procedure over the clustering-based approaches.
The authors should detail these points in order to make it a complete paper.
Author Response
Please see the attachment
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
The authors gave a good response to my comments. The revised paper is more rational compared to the previous edition, and the approach is simpler and more user-friendly for a broader group of applied researchers. I have no new comments.
Author Response
We thank the reviewer for their time reviewing our manuscript.
Reviewer 3 Report
The paper is currently clearly written. However, there are similar papers that execute the separation between buildings and trees using the RANSAC algorithm. That paper must be cited as a comparative effort.
Author Response
We thank the reviewer for their comments. We have added reference to the RANSAC method for identifying buildings to the Introduction (line 47) and Discussion (line 261) sections.
