Title: Tree Parameter Detection from UAV LiDAR Point Cloud in Latvia
Precise forest resource information is important for sustainable and cost-effective forest management. Laser scanning technologies can give significant support for regional and stand level forest inventories reducing labor participation. This study analyzed high density UAV LiDAR point cloud usage for tree parameter detection in Latvia. Laser scanning with density of 900 points/m2 was made in central part of Latvia in total are of 0.6 ha. Two selected polygons were even-aged mixed species boreo-nemoral stands. Laser scanning was used for tree detection and stem DBH model generation which was based on tree height and crown features. Tree detection and point cloud segmentation was developed with layer stacking approach were point cloud was sliced in 1 m height intervals and trees were classified in each layer using local maxima algorithm. For each tree stem and crown points were classified. Crown points were used for crown feature calculation. Results revealed that tree detection accuracy was 94.4%, but tree height accuracy was 36.7 cm. Modelled DBH compared to reference measurements were estimated with accuracy of 2 cm. Author claims that there is still room for developments in tree detection and segmentation methodology. More precise variables could give better modelled DBH predictions. Inside canopy data is recommended for better overtopped tree detection rate. It would be worthy to analyze UAV LiDAR parameter detection accuracy in leaf-off conditions when laser point cloud give detailed stem information. That would allow to detect tree DBH directly from point cloud. Modelled DBH results are comparable to MLS system accuracy. Tree height and DBH detection accuracy fulfilled Latvia’s forest inventory accuracy standards and methodology could be useful and implementable in everyday operations. High density LiDAR inventory results could be useful as calibration data sets to improve national level forest inventory.
Keywords: UAV, LIDAR, tree DBH, tree segmentation