Individual Tree Crown Delineation Based On Marker-controlled Region Growing And Airborne Laser Scanner Data

Based on LiDAR, orthoimagery and forest inventory data acquired in 2009, this study delineated individual tree crown in coniferous and deciduous forests with high canopy density in Liangshui National Nature Reserve. Local maximum method with variable window size was applied to detect individual treetop locations and influences of different canopy height models and window sizes on detection results were discussed; marker-controlled region growing method was used to delineate individual tree crown boundary based on the detected treetops. Six stop conditions were designed in the region growing method, including two conditions that control canopy shape (rectangularity and ratio of length and width) and four conditions that control property of inner crown (neighborhood, threshold of variability, threshold of crown area and threshold of height difference). The accuracy of treetop detection and tree crown boundary delineation were assessed on both plot and individual tree levels.The results indicated that the local maximum method using canopy maximum model (CMM) and 95% lower predicting limit of height-crown nonlinear regression as variable window size provided the highest accuracy of treetop detection:detection percentage reached above 84.8%, up to 98.6%; user’s accuracy (UA), producer’s accuracy (PA) and the number of 1:1 matched tree all met the application requirements. For tree crown boundary delineation, relative error of crown area (RECA) of coniferous and deciduous trees were 8.74% and-8.24%, respectively. PA and UA for conifers were 62.2%～77.3% and 71.5%～83.9%; while PA and UA for deciduous trees were 76.1%～91.2% and 78.5%～92.5%. The deciduous forests obtained obvious higher accuracy than conifers due to the lower canopy density of deciduous plots in the study. The case of "match but over growing" was the main reason why the accuracy was low in coniferous forests. This study provides theoretical basis and technology support for automatic individual tree crown delineation and precise forestry.