Extraction Of Three-Dimensional Crown Structure Of Larix Olgensis By UAV-borne LiDAR

The light detection and ranging(LiDAR)technique plays an increasingly important role in forest inventory,and various platforms are employed for LiDAR data acquisition to obtain forestry metrics at different scales.Herein,we evaluate the capabilities of a low-cost unmanned aerial vehicle laser scanning(UAVLS)system to extract fine-scale crown structural information of Larix olgensis,compare the extraction of crown characteristic attributes by UAVLS and airborne laser scanning(ALS).At the same time,we model the crown profile of Larix olgensis using UAVLS data.Firstly,we preprocess LiDAR point clouds,detect individual trees by local maxima filter and segment tree crown by marker-controlled watershed and improved K-means clustering,screened all detected Larix olgensis trees,which 1:1 matched with measured data.Then we extract tree crown characteristic attributes(tree height,crown base height,crown width,crown volume and surface area)using UAVLS data and ALS data,respectively.Finally,we choose the crown profile points from UAVLS individual tree points by width percentile method,and model the crown profile of Larix olgensis by four equations(parabola equation,Mitscherlich equation,power equation and modified Weibull equation).Using crown profile model to reconstruct the three-dimensional geometric structure of the tree crown,then compare the volume and surface area derived from crown geometric reconstruction and three-dimensional convex hull.The results showed that the detection rate of individual tree is increased by about 37%,the detection accuracy of individual tree position is improved by 0.19 m,the estimation of tree height is increased by about 1.29 m,and the estimation of crown width is increased by about 0.19 m.For individual tree crown segmentation,compared with improved K-means clustering method based on point cloud,grid-based marker-controlled watershed method can better segment Larix olgensis plantation based on UAVLS.Compared with the results of other UAVLS in estimating tree crown characteristic factors,the UAVLS in this study have a high accuracy in extracting tree crown characteristic attributes,especially for extracting tree height and crown width,RMSE is 0.58 m and 0.39 m,respectively.For the underestimation of tree height,it is proved that the degree of underestimation of tree height improves with the increase of point cloud density.UAVLS estimates the crown base height slightly inadequately,overestimating 0.6m,mainly due to the less energy echoes of LiDAR sensor,resulting in the uneven distribution of point clouds penetrating the canopy.The modeling results of crown profile show that the four profile models can well describe the crown profile constructed by UAVLS.The fitting R2 of the four models are all higher than 0.75.For the test of the model,the parabola model and the modified Weibull model have strong predictive ability.For the estimation of crown volume and surface area,the results of crown volume and surface area based on crown profile model are more accurate than those based on three-dimensional convex hull.Overall,this study provides an efficient procedure for extracting individual tree crown structure by UAV-borne LiDAR,which can replace part of field measurements without felling trees and thus improve the efficiency of forestry survey.