摘要
【目的】针对北方森林,协同无人机激光雷达和无人机RGB影像进行单木地上生物量估测,以彰武县樟子松和杨树为研究对象,探究应用组合数据和单一数据对针、阔叶林单木地上生物量估测的影响,为彰武县防风固沙人工林单木地上生物量的精准预测提供技术参考。【方法】从LiDAR点云和基于RGB光学影像获取的数字正射影像图(DOM)中提取单木尺度高度、强度、密度、冠层结构、光谱、纹理和植被指数多特征,采用置换重要性(PI)和Boruta优选特征子集,结合地面实测单木地上生物量数据,使用随机森林(RF)、极端梯度提升树(XGBoost)和分类提升算法(CatBoost)3种典型机器学习方法构建樟子松和杨树地上生物量估测模型,对仅用LiDAR数据、仅用DOM数据以及联合二者的建模结果进行比较。【结果】1)点云高度和冠层结构是估测2个树种单木地上生物量的关键特征;纹理特征仅对樟子松地上生物量估测产生积极影响。2)对于樟子松,基于组合数据的单木地上生物量估测精度最高,优于单一LiDAR和单一RGB影像;3种数据集的最优模型分别为ALL-PI-XGBoost、LiDAR-PI-XGBoost和DOM-PI-RF,测试集R^(2)分别为0.77、0.69、0.67,RMSE分别为10.94、12.75、13.16 kg·plant^(−1)。对于杨树,基于组合数据和单一LiDAR数据的单木地上生物量估测精度相当,且优于单一RGB影像;3种数据集的最优模型分别为ALL-PI-XGBoost、LiDAR-Boruta-XGBoost和DOM-Boruta-CatBoost,测试集R2分别为0.85、0.85、0.59,RMSE分别为17.63、17.11、28.99 kg·plant^(−1)。【结论】基于低成本无人机遥感技术获取的高密度点云和高分辨率影像可以实现彰武县防风固沙人工林单木地上生物量高精度、快速且无损估测;应用组合数据和单一数据对针、阔叶林单木地上生物量估测有不同影响,组合数据可以显著提高樟子松单木地上生物量估测精度。
【Objective】The aim of this study is to estimate aboveground biomass(AGB)at individual tree scale in northern forests by synergistically utilizing UAV-LiDAR and UAV-RGB data.Pinus sylvestris var.mongolica and Populus.in Zhangwu County were used as the research object to investigate the influence of using combined data versus single data on AGB estimation in coniferous and broadleaf forests.The findings are to provide technical references for precise prediction of AGB at individual tree scale in windbreak and sand fixation plantations in Zhangwu County.【Method】Multiple features at individual tree scale,including height,intensity,density,crown structure,spectrum,texture and vegetation index,were extracted from LiDAR point clouds and digital orthophoto map(DOM)derived from RGB optical imagery.Permutation importance(PI)and Boruta methods were used to select feature subsets.Combining these features with the aboveground biomass(AGB)data of individual trees calculated from field-measured tree height and diameter at breast height,three typical machine learning methods,including random forest(RF),extreme gradient boosting(XGBoost),and categorical features gradient boosting(CatBoost),were adopted to construct biomass estimation models for the two tree species,P.sylvestris var.mongolica and Populus.The modeling results using only LiDAR data,only DOM data,and a combination of both methods were compared.【Result】1)Point cloud height and crown structure were identified as key features for AGB estimation at individual tree scale for both species,whereas texture features only positively influenced the estimation of AGB for P.sylvestris var.mongolica.2)For P.sylvestris var.mongolica,the estimation accuracy of AGB at individual tree scale based on combined data was highest,outperforming models based on single LiDAR and RGB imagery.The optimal models for the three datasets were ALL-PI-XGBoost,LiDAR-PI-XGBoost,and DOM-PI-RF,with R^(2) of 0.77,0.69,and 0.67,and RMSE of 10.94,12.75,and 13.16 kg·plant^(-1),respectively.For Populus,the estimation accuracy of AGB at individual tree scale was comparable when using combined and single LiDAR data,and both outperformed the model based on single RGB imagery.The optimal models for the three datasets were ALL-PI-XGBoost,LiDAR-Boruta-XGBoost,and 2 DOM-Boruta-CatBoost,with R of 0.85,0.85,and 0.59,and RMSE of 17.63,17.11,and 28.99 kg·plant^(-1),respectively.【Conclusion】The high-density point clouds and high-resolution images obtained from two types of low-cost UAV remote sensing technology can achieve high-precision,fast,and non-destructive estimation of individual tree aboveground biomass in the windbreak and sand fixation plantations in Zhangwu County.The use of combined data versus single data has different impacts on AGB estimation at individual tree scale in coniferous and broadleaf forests,with combined data significantly improving the accuracy of AGB estimation for P.sylvestris var.mongolica.
作者
张宇娇
赵恒谦
付含聪
刘哿
皇甫霞丹
刘轩绮
Zhang Yujiao;Zhao Hengqian;Fu Hancong;Liu Ge;Huangfu Xiadan;Liu Xuanqi(College of Geoscience and Surveying Engineering,China University of Mining and Technology(Beijing),Beijing 100083;Inner Mongolia Research Institute,China University of Minning and Technology(Beijing),Erdos 017010)
出处
《林业科学》
北大核心
2025年第8期129-141,共13页
Scientia Silvae Sinicae
基金
国家自然科学基金项目(41701488)
鄂尔多斯市科技重大专项(ZD20232304)
中央高校基本科研业务费专项资金项目(2025JCCXDC01)
中国矿业大学(北京)博士研究生拔尖创新人才培育基金项目(BBJ2023024)。
