基于k-NN算法的叶面积指数遥感反演  

Study on remote sensing estimation of leaf area index based on k-NN algorithm

作　　者：孙华[1] 罗朝沁 林辉[1] 严恩萍[1] 罗喜华 罗孝云 

机构地区：[1]中南林业科技大学林业遥感信息工程研究中心,湖南长沙410004 [2]攸县黄丰桥国有林场,湖南攸县412300

出　　处：《中南林业科技大学学报》2016年第12期11-17,36,共8页Journal of Central South University of Forestry & Technology

基　　金：国家"十二五"863项目:"数字化森林资源监测关键技术研究"(2012AA102001);中国博士后科学基金项目:林分环境条件下的林木冠幅提取及冠形曲线参数化(2014M562147)

摘　　要：叶面积指数(Leaf Area Index,LAI)作为植被冠层结构的重要描述参数之一,能体现植被光合、蒸腾和呼吸作用的能力。借助GPS和LAI-2200冠层分析仪在攸县黄丰桥林场开展LAI测量。利用ENVI软件对Geo Eye-1数据进行了辐射定标,大气校正和正射校正。通过研究LAI与Geo Eye-1影像波段及其衍生指数的相关性,筛选出2组估算LAI的指数因子(6个指数因子和10个指数因子)。应用k-NN进行叶面积指数反演,同时将反演结果与多元线性回归模型结果进行比较。结果表明:利用2组指数因子进行多元线性回归模型反演LAI中,6个指数因子的模型决定系数R2为0.386,10个指数因子的模型决定系数R2为0.498。从回归模拟的角度分析,10个指数因子得到的模拟结果要优于6个指数因子的模拟结果。利用2组指数因子通过设置4个不同的k值(k=3,5,7,10)得到8个k-NN反演结果中,以10个指数因子得到的k-NN反演结果较好,其中在k=3时效果最好,其决定系数R2为0.733,精度为85.4%。建模精度分析表明选用10个指数因子进行LAI的反演优于选用6个指数因子,其中k-NN方法的反演结果优于多元线性回归模型,说明利用k-NN方法进行LAI的反演是可行的。As one of the most important description parameters for forest canopy structure, LAI （Leaf Area Index） has the ability to reflect the photosynthesis, transpiration and breathing for vegetation. Accurately mapping LAI often conducted by combining sample plots and remotely sensed images. The objective of this study was to employ k-Nearest Neighbor （k-NN） algorithm to estimate the LAI in Huangfengqiao forest Farm of You County, Hunan province of China using GeoEye-1 images and ground sample plots. In this study, ground measurement of LAI was conducted with the assistant of GPS and LAI-2200 canopy analyzer. The GeoEye-1 images were processed with the correction of radiation, atmospheric and orthographic. Moreover, through the correlation analysis of LAI and GeoEye-1 factors, 2 groups （6 variables and 10 variables respectively） of variables were selected for the estimation of LAI. Finally, k-NN algorithm method was used to simulation LAI, and compared with the multiple linear regression. Results show that multiple linear regression model of 10 independent variables was better than 6 variables.The R2 coefficient of the two models were 0.498 and 0.386. With the difference ofk value （k=-3, 5, 7, 10） and variable numbers （6 and 10）, 8 combinations and estimations for LAI were generated. The estimation derived from the combination of k=-3 and 10 variables had greatest accuracy, with the R2 of 0.733 and estimation accuracy of 85.4%. This implied that the algorithm of k-NN provided greater potential than multiple linear regression model to map LAI with the combination of sample plots and GeoEye-1 images.

关 键 词：林业遥感 叶面积指数 K-NN Geo Eye-1 黄丰桥林场 

分 类 号：S757[农业科学—森林经理学]