集成多源遥感与极限梯度提升的竹林地上生物量估测  

Above-ground biomass estimation of bamboo forests by integrating multi-source remote sensing and XGBoost machine learning

作　　者：杨可乐 谭艳 郭孝玉 余坤勇[1,3] YANG Kele;TAN Yan;GUO Xiaoyu;YU Kunyong(College of Forestry,Fujian Agriculture and Forestry University,Fuzhou,Fujian 350002,China;College of Resources and Cemical Engineering,Sanming University,Sanming,Fujian 365004,China;Fujian Provincial Key Laboratory of Resources and Environment Monitoring&Sustainable Management and Utilization,Sanming,Fujian 365004,China)

机构地区：[1]福建农林大学林学院,福建福州350002 [2]三明学院资源与化工学院,福建三明365004 [3]福建省资源环境监测与可持续经营利用重点实验室,福建三明365004

出　　处：《森林与环境学报》2024年第4期431-440,共10页Journal of Forest and Environment

基　　金：国家自然科学青年基金项目“顾及叶物候光谱特征的中亚热带山地毛竹林精细提取研究”(41801279);福建省资源环境监测与可持续经营利用重点实验室开放课题基金项目(ZD202104);竹资源开发利用福建省高等学校重点实验室开放课题“黄甜竹生物量分配规律及碳汇潜力评估”(KBJ2102)。

摘　　要：森林生物量遥感精准估测是评估森林碳汇潜力的重要依据,通过对比机载激光雷达(LiDAR)与机载多光谱数据在竹林地上生物量(AGB)估测中的差异,探索不同光学与激光点云数据协同估测森林AGB的潜力。以福建特色竹种黄甜竹(Acidosasa edulis)人工林为研究对象,同步采集无人机载多光谱影像和LiDAR数据,提取并优选单一LiDAR的优选特征集、单一多光谱影像数据优选特征集、联合机载LiDAR和多光谱影像的优选特征集,采用多元线性回归模型(MLR)和支持向量机(SVM)、随机森林(RF)、极限梯度提升(XGBoost)拟合竹林AGB估测模型。结果表明:(1)LiDAR数据的高度特征和密度特征是竹林AGB估测的重要特征,基于多源异构数据的特征集对竹林AGB估测的效果优于单一的遥感数据;(2)竹林AGB估测中,基于机器学习算法的非参数模型优于多元线性回归模型;集成XGBoost与多源特征集的黄甜竹AGB估测模型最优,R^(2)达0.64,Erms为9.90 t·hm^(-2)。竹林具有特殊冠层结构,应用LiDAR数据源能够有效获取林分垂直结构信息,联合多源异构数据能提高生物量估测模型精度。Accurate estimation forest biomass by remote sensing is an important basis for assessing the potential of forest carbon sinks.We explored the potential of synergistic estimation of above-ground biomass(AGB)in bamboo forests using different optical and LiDAR point cloud data,by comparing the differences between airborne multispectral and LiDAR data,respectively.In this study,the plantation forest of Acidosasa edulis,a characteristic bamboo species in Fujian Province,was used as the research object.Unmanned airborne multispectral imagery and LiDAR data were collected synchronously to extract and select the preferred feature set of a single LiDAR and multispectral imagery data each,the preferred feature set of the combined airborne LiDAR and multispectral imagery,and the combined feature set of the combination airborne LiDAR and multispectral imagery,using multivariate linear regression model(MLR)and support vector machine(SVM),random forest(RF),and eXtreme Gradient Boosting(XGBoost)to fit the bamboo forest AGB estimation model.Results showed that(1)the height and density features of LiDAR data are important for bamboo forest AGB estimation.The feature set based on multi-source heterogeneous remote sensing data was better than single data for bamboo forest AGB estimation.(2)The nonparametric model based on machine learning algorithms was better than the MLR model for bamboo forest AGB estimation;the AGB estimation model integrating XGBoost and multi-source feature set was the best fit for Acidosasa edulis bamboo,with R^(2) of 0.64 and Erms of 9.90 t·hm^(-2).Bamboo forests have a special canopy structure,and applying LiDAR data can effectively obtain information of the vertical structure of the forest stand.Moreover,the combination of multi-source isomorphic data can improve the accuracy of the biomass estimation model to provide a reference for the application of the biomass survey of the fixed sample plots of bamboo forests in subtropical areas and the estimation of carbon sinks.

关 键 词：竹林 地上生物量 多源遥感 激光雷达 多光谱 机器学习 

分 类 号：S757[农业科学—森林经理学]