Detection of Individual Tree Crowns in Airborne Lidar Data

Laser scanning provides a good means to collect information on forest stands. This paper presents an approach to delineate single trees automatically in small footprint light detection and ranging (lidar) data in deciduous and mixed temperate forests. In rasterized laser data possible tree tops are detected with a local maximum filter. Afterwards the crowns are delineated with a combination of a pouring algorithm, knowledge-based assumptions on the shape of trees, and a final detection of the crown-edges by searching vectors starting from the trees’ tops. The segmentation results are assessed by comparison with terrestrial measured crown projections and with photogrammetrically delineated trees. The segmentation algorithm works well for coniferous stands. However, the implemented method tends to merge crowns in dense stands of deciduous trees. Introduction For sustainable forest management, a great amount of information is required both for planning of future forest management and for documenting the activities of the last decade. Parameters, such as tree species and tree species distribution, timber volume, increment of timber volume, and mean tree height are usually needed. In a number of European countries these single tree-related parameters are the basis for a forest inventory, which is conducted every ten years. Currently, most of those variables are estimated by measuring sample plots manually in field surveys, thus, forest inventories are rather expensive. During the last several years more effort has been put into decreasing costs by developing inventory systems that are based on remote sensing. Airborne lidar (light detection and ranging) is becoming a promising technique for modeling the forest’s canopy and thus for completing several inventory tasks. Brandtberg (1999) and Hyyppä and Inkinen (1999) proved that single tree delineation in forest stands of Nordic countries can be detected by high-density laser data. Popescu et al. (2002) estimated plot level tree heights with lidar data based on local filtering with a canopy height-based variable window size with good success. Persson et al. (2002), Leckie et al. (2003), and Hyyppä et al. (2000) have demonstrated that tree heights can be measured with high accuracy from airborne lidar data. Yu et al. (2003) were the first to demonstrate the use of laser scanner data for change detection assessment of single trees in boreal forests. Several authors have shown mostly for coniferous forests that airborne lidar is also a good means for estimating other forest stand parameters (like volume or mean tree height) Detection of Individual Tree Crowns in Airborne Lidar Data Barbara Koch, Ursula Heyder, and Holger Weinacker with an averaging, stand-wise approach (Naesset, 1997; Magnussen and Boudewyn, 1998; Lefsky et al., 1999). However, under typical conditions in temperate forests a standwise approach comprises many difficulties. As several tree species with different growing behavior can occur in one stand, a-priori knowledge of stem number and tree-species distribution would be necessary for calculating stand parameters. Additionally, in diverse forests, a stand-wise result is usually not sufficient for forest management planning as established in a number of European countries. Especially for harvest management purposes, information on single trees is required. Therefore, single tree delineation and tree species classification seem to be a prerequisite to use remote sensing technologies for large-scale forest inventories under typical conditions in temperate forests especially in respect for calculation of timber volume, as well as harvesting and silvicultural treatment schemes. First approaches of single tree delineation showed promising results for conifer forests with multispectral imagery (Gougeon, 1995; Pollock, 1996), as well as with lidar data (Hyyppä and Hyyppä, 2001; Persson et al., 2002; Brandtberg et al. 2003) or by fusing both type of data (Popescu et al. 2004; Popescu and Wynne, 2004). The objective of this paper is to develop a robust algorithm to detect and delineate tree crowns and tree heights that is suitable for coniferous stands, as well as for deciduous and mixed, vertically-structured, temperate forests. The correct delineation of crowns is a prerequisite for other derived parameters like tree position, tree height, crown diameter, or crown volume. Even the extraction of tree species type from multispectral data needs the correct delineation of tree crowns to achieve good results (Koch et al. 2002). Due to the fact that in temperate forests the tree crown delineation is still a topic that needs to be improved, the presented study focuses on the refinement of algorithms to improve the tree crown delineation for dense and multilayered stands. The fusion of multispectral data with laser scanner data is also an important objective under investigation, but not the topic of this article. It also has to be taken into consideration that information extraction based on laser scanner data will provide more flexibility in the flight scheme. Therefore, the knowledge what may be exclusively based on laser scanner data has high priority for data provider and users. Study Areas The two study areas are located in the Southwest of Germany close to the City of Freiburg/Breisgau (Figure 1). They were PHOTOGRAMMETRIC ENGINEER ING & REMOTE SENS ING Ap r i l 2006 357 Department of Remote Sensing and Landscape Information Systems, Albert-Ludwigs University of Freiburg, Tennenbacherstr. 4, 79085 Freburg, Germany ([email protected]). Photogrammetric Engineering & Remote Sensing Vol. 72, No. 4, April 2006, pp. 357–363. 0099-1112/06/7204–0357/$3.00/0 © 2006 American Society for Photogrammetry and Remote Sensing 04-046 3/14/06 8:56 PM Page 357