Isolating Individual Trees in a Savanna Woodland Using Small Footprint Lidar Data

This study presents a new method of detecting individual treetops from lidar data and applies marker-controlled watershed segmentation into isolating individual trees in savanna woodland. The treetops were detected by searching local maxima in a canopy maxima model (CMM) with variable window sizes. Different from previous methods, the variable windows sizes were determined by the lower-limit of the prediction intervals of the regression curve between crown size and tree height. The canopy maxima model was created to reduce the commission errors of treetop detection. Treetops were also detected based on the fact that they are typically located around the center of crowns. The tree delineation accuracy was evaluated by a five-fold, cross-validation method. Results showed that the absolute accuracy of tree isolation was 64.1 percent, which was much higher than the accuracy of the method, which only searched local maxima within window sizes determined by the regression curve (37.0 percent). Introduction Isolating individual trees and extracting relevant tree structure information from remotely sensed data have significant implications in a variety of applications. For example, detailed information at the individual tree level can be used for monitoring forest regeneration (Gougeon and Leckie, 1999; Clark et al., 2004a and 2004b), reducing fieldwork required for forest inventory (Gong et al., 1999) and assessing forest damage (Leckie et al., 1992; Levesque and King, 1999; Kelly et al., 2004). To study the interactions between vegetation and climate, we are applying an individual treebased model, called MAESTRA, over an eddy covariance tower site in Ione, California for quantifying the carbon fluxes. To parameterize the individual tree-based model, our research is ongoing to extract individual tree structure parameters such as tree height, crown height, crown size, leaf area index (LAI), and biomass using small-footprint lidar data over an area of 800 m by 800 m around the eddy Isolating Individual Trees in a Savanna Woodland Using Small Footprint Lidar Data Qi Chen, Dennis Baldocchi, Peng Gong, and Maggi Kelly covariance tower. However, to obtain such individual tree parameters, the initial process is to isolate individual trees and delineate tree crown boundaries. Intensive research has been done on isolating individual trees using remotely sensed data. However, previous data focuses on large-scale aerial photos or high-spatial resolution remotely sensed imagery. The methods for isolating individual trees from imagery or photos include: local maxima detection (Dralle and Rudemo, 1996), local maxima filtering with fixed or variable window sizes (Wulder et al., 2000; Pouliot et al., 2002), valley-following (Gougeon, 1995), edge detection using scale-space theory (Brandtberg and Walter, 1998), template-matching (Pollock, 1996; Larsen and Rudemo, 1998), local transect analysis (Pouliot et al., 2002), 3D modeling (Sheng et al., 2001; Gong et al., 2002), and watershed segmentation (Schardt et al., 2002; Wang et al., 2004). When isolating trees from a monocular image or photo, these methods are mostly based on the assumption that there are “peaks” of reflectance around the treetops and “valleys” along the canopy edges. However, the “peaks” and “valleys” are not always distinct since canopy reflectance is affected by various factors such as illumination conditions, canopy spectral properties, and complex canopy structure. Recently, researchers have begun to apply lidar data into individual tree isolation and canopy information extraction (Hyyppä et al., 2001; Persson et al., 2002; Brandtberg et al., 2003; Leckie et al., 2003; Popescu et al., 2003; Popescu and Wynne, 2004). Compared with passive imaging, lidar has the advantage of directly measuring the three-dimensional coordinates of canopies. Therefore, the geometric, rather than spectral, “peaks” and “valleys” can be detected. Several studies have extended methods developed for optical imagery and aerial photos into lidar data for tree detection (Brandtberg et al., 2003; Leckie et al., 2003; Popescu et al., 2003). Brandtberg et al. (2003) extended the scale-space theory to detecting crown segments. Leckie et al. (2003) applied the valley-following approach into both lidar and multi-spectral imagery and found that the lidar can easily eliminate most of the commission errors that occur in the open stands while the optical imagery performs better for isolating trees in Douglas-fir plots. This study attempts to use marker-controlled watershed segmentation in tree isolation. Watershed segmentation, first proposed by Beucher and Lantuejoul (1979), is a wellknown image segmentation method that incorporates the advantages of other segmentation methods such as regiongrowing and edge-detection (Soille, 2003). To avoid the oversegmentation problem, Meyer and Beucher (1990) introduced PHOTOGRAMMETRIC ENGINEER ING & REMOTE SENS ING Augu s t 2006 923 Qi Chen is with the Center for the Assessment and Monitoring of Forest and Environmental Resources (CAMFER), University of California at Berkeley, Berkeley, CA 94720 ([email protected]). Dennis Baldocchi is with the Department of Environmental Science, Policy, and Management, University of California at Berkeley, Berkeley, CA 94720 (baldocchi@nature. berkeley.edu). Peng Gong and Maggi Kelly are with the Center for the Assessment and Monitoring of Forest and Environmental Resources (CAMFER), University of California at Berkeley, Berkeley, CA 94720 ([email protected]; mkelly@ nature.berkeley.edu). Photogrammetric Engineering & Remote Sensing Vol. 72, No. 8, August 2006, pp. 923–932. 0099-1112/06/7208–0923/$3.00/0 © 2006 American Society for Photogrammetry and Remote Sensing 04-096 7/10/06 8:44 PM Page 923