欢迎访问《应用生态学报》官方网站,今天是 分享到:

应用生态学报

• 研究报告 • 上一篇    下一篇

天然草地植被覆盖度的高光谱遥感估算模型

刘占宇1;黄敬峰1;吴新宏2;董永平2;王福民1;刘朋涛3   

  1. 1浙江大学农业遥感与信息技术应用研究所,杭州 310029; 2中国农业科学研究院草原研究所,呼和浩特 010010; 3内蒙古大学生态与环境科学系,呼和浩特 010021
  • 收稿日期:2005-05-23 修回日期:2006-02-28 出版日期:2006-06-18 发布日期:2006-06-18

Hyperspectral remote sensing estimation models on vegetation coverage of natural grassland

LIU Zhanyu1; HUANG Jingfeng1; WU Xinhong2; DONG Yongping2; WANG Fumin1; LIU Pengtao3   

  1. 1Institute of Agriculture Remote Sensing Information System Application, Zhejiang University, Hangzhou 310029, China;2Grassland Research Institute, Chinese Academy of Agricultural Science, Huhhot 010010, China;3Department of Ecology and Environment Science, Inner Mongolia University, Huhhot 010021, China
  • Received:2005-05-23 Revised:2006-02-28 Online:2006-06-18 Published:2006-06-18

摘要: 利用ASD FieldSpec Pro FRTM光谱仪,对内蒙古自治区锡林郭勒盟的天然草地进行高光谱遥感地面观测,并计算天然草地植被覆盖度;选择25个高光谱特征变量与天然草地植被覆盖度进行相关性分析.结果表明,共有17个变量通过极显著性检验,尤以红边波长范围内一阶微分波段值总和(SDr)的相关系数0.781为最高在此基础上将观测数据分成两组:一组观测数据作为训练样本(n=49),运用单变量线性、非线性和逐步回归方法,建立植被覆盖度高光谱遥感估算模型;另一组观测数据作为检验样本(n=32),进行精度检验分析结果显示,采用逐步回归分析方法,运用冠层原始反射率数据估算草地植被覆盖度的效果并不理想;而以红边波长范围内一阶微分波段值的总和(SDr)为变量的线性回归方程是最佳估算模型,模型标准差为10.4%,估算精度为83.99%.

关键词: 生态系统, 管理, 地理信息系统, 多目标最佳空间规划

Abstract: By using ASD FieldSpec Pro FRTM spectroradiometer, the spectral measurement of natural grassland in Xilingole Leaguer of Inner Mongolia was performed, with the vegetation coverage of natural grassland calculated, and the correlation of 25 hyperspectral feature variables with the vegetation coverage of natural grassland was analyzed. The results showed that there were 17 variables correlated significantly with the vegetation coverage of natural grassland, among which, the correlation coefficient between vegetation coverage and the area of red edge peak calculated as the sum of the amplitudes between 680 nm and 780 nm (∑dr 680~780 nm) was the highest, with the value of 0.781. The basic experimental data including the vegetation coverage and canopy reflectance of natural grassland were classified into two groups. One group was used as the training sample to build the regression models with onesample linear method, nonlinear method, and stepwise analysis method, while the other was used as the testing sample to test the precision of regression models. It was suggested that the variable of the area of red edge peak calculated as the sum of amplitudes between 680 nm and 780 nm (∑dr 680~780 nm) was the best one to univariate general linear model, with a standard deviation of 10.4% and an estimation precision of 83.99%, while the stepwise regression technique was not effective to estimate the grassland coverage with raw hyperspectral canopy reflectance.

Key words: Ecosystem, Management, GIS, Multi objects optimal spatial plan