不同龄组木麻黄地上生物量估测模型构建  

Construction of aboveground biomass models for different age groups of Casuarina equisetifolia

作　　者：古丽再排尔·安外尔 尤龙辉 叶功富[3] 聂森[3] 胥喆 陈凤娇 ANWAIER Gulizaipaier;YOU Longhui;YE Gongfu;NIE Sen;XU Zhe;CHEN Fengjiao(College of Forestry,Fujian Agriculture and Forestry University,Fuzhou,Fujian 350002,China;Fuzhou Forestry Administration,Fuzhou,Fujian 350007,China;Fujian Academy of Forestry,Fuzhou,Fujian 350012,China;Fujian Forestry Prospect and Design Institute,Fuzhou,Fujian 350001,China;Luoyuan State-owned Forest Farm,Fuzhou,Fujian 350603,China)

机构地区：[1]福建农林大学林学院,福建福州350002 [2]福州市林业局,福建福州350007 [3]福建省林业科学研究院,福建福州350012 [4]福建省林业勘察设计院,福建福州350001 [5]福建省罗源国有林场,福建福州350603

出　　处：《森林与环境学报》2025年第1期62-72,共11页Journal of Forest and Environment

基　　金：福建省林业科技项目“基于无人机激光雷达的木麻黄防护林生物量及碳储量反演研究”(2022FKJ19);中央引导地方科技发展专项“功能导向型沿海防护林结构定向调控技术研究”(2021L3017)。

摘　　要：为提取木麻黄高精度的单木结构参数,并建立多样化的木麻黄地上生物量估测模型,快速、高效地调查和监测木麻黄林分的生长趋势,以福建省平潭岛不同龄组木麻黄为研究对象,利用无人机激光雷达(UAV-LiDAR)点云、无人机-地基激光雷达融合激光雷达(Fusion-LiDAR)点云,快速、精确地获取单木尺度下木麻黄的树高、冠幅、胸径等关键结构参数,并结合实地调查数据,运用偏最小二乘法、随机森林以及反向传播神经网络(BPNN)等算法构建木麻黄地上生物量估测模型。结果表明:基于Fusion-LiDAR点云构建的冠层高度模型的单木分割精度明显优于UAV-LiDAR,尤其是在幼龄林中差异较大;与UAV-LiDAR点云提取的结果相比,Fusion-LiDAR点云提取的树高和冠幅的决定系数(R^(2))普遍更高,特别是在过熟林中,其树高、冠幅决定系数分别比UAV-LiDAR点云增大了11.41%、16.73%;在3种算法模型中,不同龄组BPNN模型的R^(2)均大于0.75,相对分析误差均大于1.40,展现出了优越的性能;随着林龄的增长,木麻黄单木分割精度、单木结构参数提取精度及模型预测精度均会逐步下降。无人机与地基激光雷达的融合显著提高了木麻黄单木分割的准确度和单木结构参数的提取精度,BPNN模型在预测不同林龄木麻黄地上生物量方面表现出了较优异的性能,进一步提升了建模的效率和预测的准确性。We used Casuarina equisetifolia of different age groups on Pingtan Island,Fujian Province as the research objects to extract the high-precision structural parameters of individual C.equisetifolia trees,establish diversified aboveground biomass estimation models for C.equisetifolia,and quickly and efficiently investigate and monitor the growth trends of C.equisetifolia stands.We utilized point clouds data acquired from unmanned aerial vehicles(UAV)equipped with LiDAR(UAV-LiDAR)and fusion laser point clouds(Fusion-LiDAR)formed by combining data from both UAV and terrestrial LiDAR scanners(TLS)to rapidly and accurately obtain the key structural parameters of C.equisetifolia at the individual tree scale,including tree height,crown width,and diameter at breast height.Combining field survey data,we used algorithms such as partial least squares(PLS),random forest(RF),and back propagation neural network(BPNN)to construct an aboveground biomass estimation model for C.equisetifolia.The results indicated the following:the canopy height model constructed based on Fusion-LiDAR point clouds significantly outperformed that constructed solely based on UAV-LiDAR in terms of individual tree segmentation accuracy,particularly in young forests;the determination coefficients(R^(2))for tree height and crown width derived from Fusion-LiDAR point clouds were generally higher than that extracted from UAV-LiDAR point clouds.Specifically,the R^(2)values for tree height and crown width increased by 11.41%and 16.73%,respectively,in overmature forests compared with those obtained from UAV-LiDAR point clouds;among the three models,i.e.,BPNN,PLS,and RF,the BPNN model exhibited superior performance in predicting the biomass of C.equisetifolia across various age classes,with R^(2)values exceeding 0.75 and relative prediction deviation values surpassing 1.40 for all age groups;and as the age class increased,the accuracy of individual tree segmentation,precision of stand parameter extraction,and predictive performance of the models tended to declin

关 键 词：木麻黄 无人机 地基激光雷达 偏最小二乘法 随机森林 反向传播神经网络 单木分割 地上生物量 

分 类 号：S757.1[农业科学—森林经理学]