Unsupervised algorithms to detect single trees in a mixed-species and multilayered Mediterranean forest using LiDAR data

• Published in: Canadian journal of forest research Vol. 51; no. 12; pp. 1766 - 1780
• Main Authors: Alvites, Cesar, Santopuoli, Giovanni, Maesano, Mauro, Chirici, Gherardo, Moresi, Federico Valerio, Tognetti, Roberto, Marchetti, Marco, Lasserre, Bruno
• Format: Journal Article
• Language: English
• Published: Ottawa Canadian Science Publishing 01.12.2021; NRC Research Press; Canadian Science Publishing NRC Research Press
• Subjects: Accuracy; airborne laser scanning (ALS); Airborne lasers; Algorithms; balayage laser aéroporté (BLA); Carbon; carbon sinks; carbon stock; climate-smart forestry; data collection; Density; détection des arbres; Forest management; forest structure; foresterie intelligente face au climat; Forestry; Forests; Heterogeneity; Land cover; Laser applications; Lidar; Measurement; Methods; Optical radar; prediction; stock de carbone; structure de la forêt; Sustainability management; Sustainable forestry; tree detection; tree height; Trees; Italy
• ISSN: 0045-5067; 1208-6037; 1208-6037
• DOI: 10.1139/cjfr-2020-0510

Accurate measurement of forest growing stock is a prerequisite for implementing climate-smart forestry strategies. This study deals with the use of airborne laser scanning data to assess carbon stock at the tree level. It aims to demonstrate that the combined use of two unsupervised techniques will improve the accuracy of estimation supporting sustainable forest management. Based on the heterogeneity of tree height and point cloud density, we classified 31 forest stands into four complexity categories. The point cloud of each stand was further divided into three horizontal layers, improving the accuracy of tree detection at tree level for which we calculated volume and carbon stock. The average accuracy of tree detection was 0.48. The accuracy was higher for forest stands with lower tree density and higher frequency of large trees, as well as a dense point cloud (0.65). The prediction of carbon stock was higher with a bias ranging from –0.3% to 1.5% and a root mean square error ranging from 0.14% to 1.48%.