文献类型：期刊文献

中文题名：无人机激光雷达人工林林分高估测模型分析

英文题名：Analysis of estimation models of plantation stand heights using UAV LiDAR

作者：李梅[1,2] 刘清旺[1,2] 冯益明[3] 李增元[1,2]

第一作者：李梅

机构：[1]中国林业科学研究院资源信息研究所,北京100091;[2]国家林业和草原局林业遥感与信息技术重点实验室,北京100091;[3]中国林业科学研究院荒漠化研究所,北京100091

年份：2022

卷号：26

期号：12

起止页码：2665-2678

中文期刊名：遥感学报

外文期刊名：NATIONAL REMOTE SENSING BULLETIN

收录：CSTPCD;;EI(收录号：20230113344691);Scopus;北大核心:【北大核心2020】；CSCD:【CSCD2021_2022】；

基金：国家重点研发计划(编号:2017YFD0600904)。

语种：中文

中文关键词：遥感;无人机;激光雷达;点云;人工林;林分高

外文关键词：remote sensing;UAV;LiDAR;point clouds;plantation;stand height

分类号：P2

摘要：中国人工林面积居世界第一,精确地对人工林结构进行监测具有重要意义。本研究以内蒙古自治区赤峰市旺业甸林场内的落叶松和油松人工林为研究对象,利用无人机激光雷达LiDAR(Light Detection And Ranging)离散点云数据和地面样地调查数据对人工林林分高进行建模,通过点云特征变量与地面测量的6种林分高(包括:Lorey’s高、算术平均高、最大高、优势树高、中位数高和树冠面积加权高)间的Pearson’s相关性筛选自变量,然后利用全子集回归构建不同林分高估测模型,并采用交叉检验法进行精度评价。结果表明:激光雷达点云高度百分位数与不同林分高相关性均较高,通过一元线性回归构建的不同林分高结果最优,且估测模型的自变量均为高度特征变量。Lorey’s高(R^(2)=0.91—0.97,rRMSE=2.75%—3.96%)、优势树高(R^(2)=0.86—0.97,rRMSE=3.72%—3.83%)和树冠面积加权高(R^(2)=0.86—0.96,rRMSE=3.81%—4.73%)估测精度最高,算术平均高(R^(2)=0.85—0.94,rRMSE=4.52%—6.07%)和中位数高(R^(2)=0.80—0.95,rRMSE=5.37%—7.34%)次之,最大高(R^(2)=0.69—0.87,rRMSE=6.19%—8.09%)最低。针对不同森林类型,落叶松人工林林分高估测精度最优,优于不区分森林类型模型的估测精度(ΔR^(2)=0—0.05,ΔrRMSE=-0.69%—1.97%),优于油松林林分高模型的估测精度(ΔR^(2)=0.06—0.18,ΔrRMSE=-1.90%—1.13%)。无人机激光雷达可以用于估测北方温带针叶林的林分高,能够满足人工林资源调查快速、精确的要求。
The plantation area of China is the largest in the world.It is very important to precisely monitor plantation structure.The study area is located at Wangyedian forest farm,Chifeng,Inner Mongolia.The dominated tree species include Larix principis-rupprechtii and Pinus tabuliformis.The stand height models of plantation were established using the UAV(Unmanned Aerial Vehicle) LiDAR(Light Detection and Ranging) data and in situ sample plots measurements.The significant independent variables were selected based on the Pearson’s correlations between the six stand heights(Arithmetic mean height,Lorey’s height,Dominated height,Maximum height,Median height and Crown area weighted height) and the statistical metrics of discrete point cloud.The branch-and-bound search of best subset was conducted to fit the estimation models of stand height.The model accuracy was assessed by the cross validation.The results showed that the correlations between the height metrics of LiDAR point cloud and the different stand heights were high.The linear regression obtained the best result for different stand heights.The independent variables of the estimation model were all height metrics.For the six stand heights,the Lorey’s height(R^(2) = 0.91—0.97,rRMSE = 2.75%—3.96%),dominated height(R^(2) = 0.86—0.97,rRMSE = 3.72%—3.83%) and Crown area weighted height(R^(2) =0.86—0.96,rRMSE = 3.81%—4.73%) had the highest accuracy,while arithmetic mean height(R^(2) =0.85—0.94,rRMSE = 4.52%—6.07%) and median height(R^(2) =0.80—0.95,rRMSE =5.37%—7.34%) had a lower accuracy,maximum height(R^(2) =0.69—0.87,RMSE = 1.30—1.40 m) was the lowest.Considering the forest types,the estimation accuracies of larch plantation stands were better the estimation accuracies of all forest types(ΔR^(2) = 0—0.05,ΔrRMSE =-0.69%—1.97%),which were better than the estimation accuracy of the stand height models of pine stands(ΔR^(2) = 0.06—0.18,ΔrRMSE =-1.90%—1.13%).The UAV LIDAR can be used to estimate the stand height of the northern temperate coniferous forest,and applied for the rapid and accurate investigation of plantation resources.