A Fusion Approach for Tree Crown Delineation from Lidar Data

Abstract

This study proposes a multi-step method (the COTH method) to delineate individual tree crowns in dense forest conditions using lidar data, with the intent for the final delineation results to be used in a biomass estimation study. The study was conducted for an even-aged Norway Spruce (Picea abies) plantation containing 188 trees located in Tully, New York, and owned by the State University of New York College of Environmental Science and Forestry (SUNY ESF). Lidar data with a point density of 12.7 points/m(2) was collected in August 2010, and field data were collected to measure tree height and species in August 2010 as well. Field data containing tree height and crown width, an important component of treetop detection, were collected in summer 2006. By combining heuristically (genetic algorithm) optimized object recognition to detect tree crown objects, local maxima filtering with variable window size to detect treetops, and a modified hill climbing algorithm to segment crown objects; treetops were identified with 86.2 percent accuracy and 23.9 percent commission error. The overall areal accuracy of the delineation was 72.5 percent. The automated COTH method represents an improvement in crown delineation accuracy for lidar data.