SNAMP Publication #6: Finding the trees in the forest using lidar
Title: A new method for segmenting individual trees from the lidar point cloud
Authors: Wenkai Li, Qinghua Guo, Marek K. Jakubowski, and Maggi Kelly
- Lidar (defined below) data has been widely used to characterize the 3-dimensional structure of forests;
- Information about individual tree location and size is useful for many areas of forest science, and thus isolating individual trees from the raw lidar data is an important process for forest mapping;
- In this study we develop a new method to delineate individual trees from lidar data;
- We applied this new method to our lidar data in the Sierra Nevada Mountains in California, USA;
- This method was very accurate: the overall accuracy was 90% as confirmed by field surveys on 30 plots.
Airborne lidar data is increasingly used in forest sciences. A benefit of airborne lidar technology is that in addition to mapping the forest ground in great detail, it can also capture information about the height (and sometimes the size) of trees and shrubs in the forest. A challenge is converting the raw data, which is just a collection (or “cloud”) of points (indicating x, y location and height above the ground), into meaningful information about individual trees. Information about individual trees would be useful for wildlife studies, carbon estimates, and forest planning for example.
Many researchers have developed methods to delineate individual trees from lidar data, but most of these methods do not use the raw data – rather they use a transformed version of the data. The new method discussed in this paper uses the raw lidar data “cloud”, and thus is able to work with more detailed data. The new method starts with the highest point in an area, and “grows” individual trees by adding points within a certain distance of the original point. It works from top to bottom and isolates trees individually and sequentially from the tallest to the shortest. We compared our results to field data across dense and sparse forests.
We compared the number of existing trees (from field surveys) and the number of lidar-derived trees within 30 plots. In general, our method underestimated the number of trees. There were 380 trees in total in our 30 test plots, but only 347 trees were segmented. The algorithm missed 53 trees, and falsely detected 20 trees. Overall, the accuracy was about 90%.
- We developed a new method to extract individual trees from the lidar point cloud;
- The method performed well at mapping individual trees from the lidar point cloud in complex mixed conifer forests on rugged terrain;
- The accuracy is relatively high, indicating that the new algorithm has good potential for use in other forested areas, and across broader areas than is possible with field work alone.
Full Reference: Li, W., Q. Guo, M. Jakubowski and M. Kelly. 2012. A new method for segmenting individual trees from the lidar point cloud. Photogrammetric Engineering and Remote Sensing 78(1): 75-84.
The full paper is available here.
To learn more about lidar data, check out our lidar FAQs sheet. For more information on lidar data in SNAMP, see the spatial team website: http://snamp.cnr.berkeley.edu/teams/spatial, and our spatial team newsletters that focus on lidar: Vol. 2, No. 3, and Vol. 5, No. 1.