Estimation of stand parameters and tree species classi̇fi̇cati̇on airborne lidar data

Developing remote sensing technologies have made the monitoring and management of forest ecosystems more sensitive and efficient. In this context, LiDAR technology offers the opportunity to reliably determine the three-dimensional position and structural properties of trees. However, it is known that LiDAR data is limited in tree species determination and generally needs to be supported by additional data or methods. In this study, stand parameters (tree height, trunk diameter, crown diameter, crown area, trunk volume and crown volume) were determined using airborne LiDAR data and tree species classification was performed without using optical data. Focusing on spruce (Picea orientalis), beech (Fagus orientalis) and dry spruce individuals; PointNet, PointNet++ and DGCNN algorithms performed species classification with 86%, 94% and 89% accuracy, respectively. In addition, trunk diameters of individuals in other closure classes were estimated with a 3–5 cm error margin using Random Forest (R²=0.95) and Extreme Gradient Boosting (R²=0.92) models established with reference data obtained from the first closure level. While the segmentation success was high in spruce individuals, lower accuracy was obtained in beech trees with a broad crown structure. Although individual separation becomes difficult in areas with high crown closure, it has been shown that reliable stand parameters and tree species estimation can be made with data obtained from different closure classes.