I have a classified LiDAR point cloud for an area of 250*250 square meter, with 1 meter resolution. It has a 5 lakhs points with an average point density of 9 (means around 9 points within a pixel).

Now, I created a Canopy Height Model (CHM) model from the LiDAR point cloud and want to identify individual trees and segmenting their crowns. How can I do that?

  • 1
    @AndreSilva I'm voting to keep this open because it is specific to Lidar derived CHM's. The other question seems to focus on spectral data.
    – Aaron
    Commented Mar 15, 2016 at 12:11
  • 3
    I perceive some serious issues with your current methodology of "I classified the point clouds using hyperspectral". When you project 2D spectral data to each 3D return, the geometry does not work out. It seems more prudent to identify individual trees from the lidar point cloud, using one of the published methods, treat them as 2D image objects and then leverage the hyperspectral data for identifying species. In this way you can control the planar geometry. By attempting to assign spectra to the lidar your are over complicating things and precluding the advice provided. Commented Mar 15, 2016 at 17:07

4 Answers 4


I would encourage you to investigate the spatial wavelet analysis (SWA) method. This is an automated object oriented approach used to identifying individual tree canopies. The method has the potential to identify both tree height and canopy diameter from LiDAR derived canopy height models. The output is usually composed of a table with tree centroid coords, tree diameter and tree height (when coded for use with CHM's). The following paper goes into detail on the SWA method:

Falkowski, M. J., Smith, A. M., Hudak, A. T., Gessler, P. E., Vierling, L. A., & Crookston, N. L. (2006). Automated estimation of individual conifer tree height and crown diameter via two-dimensional spatial wavelet analysis of lidar data. Canadian Journal of Remote Sensing, 32(2), 153-161.

enter image description here

  • actually I found the individual trees and tree height etc, But I need to classify this trees. I am having a classified lidar point cloud, I'm trying to do classification of trees using these classified point cloud. Commented Mar 15, 2016 at 16:04
  • 6
    @bibinwilson, please do not be so quick to assume a dismissive tone. Participants on this forum bring notable experience and dedicate considerable time answering questions. This answer is on the mark as far as a place to start. You will need to identify individual trees, in the point cloud, before you can go about identifying species them via spectral data fusion. There are numerous issues with the geometry of projecting 2D into 3D data. So, it is not a trivial problem "classifying" a point cloud in the way you are inferring. If you start with something like SWE this is somewhat mitigated. Commented Mar 15, 2016 at 16:58
  • 1
    Also, recognize that this is somewhat considered the "holy grail" of forestry remote sensing. The environmental and sensor variables that come into play in this type of classification make it incredibly complex. If you're looking for a click-n-go solution with any kind of respectable accuracy, I'm afraid you might be disappointed. Commented Mar 15, 2016 at 17:33
  • @bibinwilson Are you after tree species rather than classifying presence/absence?
    – Aaron
    Commented Mar 15, 2016 at 17:54

The most often used method that I've encountered in the literature involves a "local maxima" identification and subsequent inverted watershed creation. This link gives one example using LiDAR data and the free USFS FUSION software

A simple Google scholar or other database search for "local maxima tree canopy" will yield many other peer-reviewed remote sensing articles that address this common use of LiDAR in canopy metrics.


A good place to ask what algorithms people use in practice to extract tree locations and tree heights and/or to segment tree crowns from a raster CHM or a raster DSM would be in the LAStools user forum. There seem to be a number of forestry people that are doing plot-scale analysis as well as single-tree analysis for actual production work.


There are two individual tree segmentation methods, CHM segmentation and point cloud segmentation in LiDAR360. enter image description here

CHM segmentation utilizes the watershed segmentation technique to identify and delineate individual trees, and therefore obtain individual tree information, such as tree location, tree height, crown diameter, crown area and tree boundaries.

Point Cloud Segmentation can directly segment LiDAR point cloud, which can avoid the loss of under canopy information of CHM segmentation method. Individual tree information, including tree location, tree height, crown diameter, crown area and crown volume can be obtained from the segmentation results.

LiDAR360 individual tree segmentation tutorial video.

  • 2
    These are just two of many methods for identifying individual tree crowns from lidar point clouds. Please do not imply that these are the only approaches. Commented Mar 24, 2018 at 13:30

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.