融合无人机载激光雷达与多光谱遥感数据的冬小麦叶面积指数反演  

Inversion of Leaf Area Index in Winter Wheat by Merging UAV LiDAR with Multispectral Remote Sensing Data

作　　者：牛玉洁 李晓鹏[1] 张佳宝[1] 马东豪[1] 纪景纯 宣可凡 蒋一飞 汪春芬 邓皓东 刘建立[1] NIU Yujie;LI Xiaopeng;ZHANG Jiabao;MA Donghao;JI Jingchun;XUAN Kefan;JIANG Yifei;WANG Chunfen;DENG Haodong;LIU Jianli(Institute of Soil Science,Chinese Academy of Sciences,Nanjing 210008,China;University of Chinese Academy of Sciences,Beijing 100049,China;College of Hydrology and Water Resources,Hohai University,Nanjing 210024,China)

机构地区：[1]中国科学院南京土壤研究所,南京210008 [2]中国科学院大学,北京100049 [3]河海大学水文水资源学院,南京210024

出　　处：《土壤学报》2022年第1期161-171,共11页Acta Pedologica Sinica

基　　金：国家重点研发计划项目(2016YFD0300601);国家自然科学基金项目(41877021,41771265)资助。

摘　　要：为了进一步挖掘无人机载激光雷达(Light Detection and Ranging,Li DAR)在农作物长势监测方面的潜力,探究机载Li DAR与多光谱遥感数据融合反演冬小麦叶面积指数(Leaf Area Index,LAI)的效果,以无人机载Li DAR和可见光-近红外多光谱为研究手段,获取试验区冬小麦孕穗期的无人机载Li DAR点云和多光谱数据,从中提取并筛选合适的Li DAR点云结构参数和植被指数,借助多元线性回归法(Multivariable Linear Regression,MLR)和偏最小二乘回归法(Partial Least Squares Regression,PLSR),通过融合Li DAR点云结构参数与植被指数以及单独使用植被指数作为模型输入参数,分别与实测LAI构建了LAI反演模型。用决定系数(Coefficient of Determination,R^(2))和均方根误差(Root Mean Square Error,RMSE)来评价模型时,结果显示融合Li DAR点云与多光谱数据能够较好地反演冬小麦LAI。而且,无论是利用MLR还是PLSR法,融合Li DAR点云结构参数与植被指数的模型(MLR︰R^(2)=0.901,RMSE=0.480;PLSR︰R^(2)=0.909,RMSE=0.445(n=16))均优于仅使用植被指数的模型(MLR︰R^(2)=0.897,RMSE=0.492;PLSR︰R^(2)=0.892,RMSE=0.486(n=16))。因此,加入无人机载Li DAR数据可以一定程度上弥补光谱数据在作物垂直方向上信息提取不足的缺陷,提高冬小麦LAI的反演精度,为冬小麦LAI反演提供了更优的手段。【Objective】In order to further tap the potential of unmanned aerial vehicle(UAV) carried LiDAR to monitor crop growth and to explore effect of merging UAV LiDAR with multispectral data in inversing leaf area index(LAI) in winter wheat,this study was carried out.【Method】In this study, with the aid of UAV LiDAR scanners and visible-near infrared multispectral cameras, UAV LiDAR point cloud and multispectral data of the winter wheat at the booting stage in experiment zone were collected. From the data, four LiDAR point cloud structure parameters, i.e., three-dimensional volumetric parameters(BIOVP),mean plant height(Hmean), 75 percentile plant height(H75) and laser penetration index(LPI), and six vegetation indices, i.e.,NDVI, SAVI, MCARI, TVI, NDRE and RVI were extracted. Then correlation analysis was performed of these parameters for screening suitable modeling parameters. With the aid of the multiple linear regression(MLR) and the partial least squares regression(PLSR), a LAI inversion model was constructed through merging the LiDAR point cloud structure parameters with vegetation indices as input parameters of the model. In applying the MLR method, the two vegetation indices, NDVI and SAVI,that are the most closely correlated with the field-observed LAI and the two point cloud structure parameters, H75 and BIOVP,that are the most closely correlated with the field-observed LAI, were used as input parameters of the model. While in adopting the PLSR method, the number of principal components in modeling was determined in the light of the result of the cross-validation. Before modeling, the experimental dataset had been randomly divided into a modeling set(n=32) and a validation set(n=16) at a ratio of 7︰3 in all treatments. A LAI inversion model was built up based on the modeling dataset and then the validation dataset was used to evaluate effect of the model. Meanwhile, in order to determine whether the inversion with the LiDAR point cloud data merged with the multispectral data was better than that base

关 键 词：无人机 冬小麦 LiDAR点云结构参数 植被指数 叶面积指数 反演模型 

分 类 号：S158.3[农业科学—土壤学]