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

基于MODIS混合像元分解的湖南省森林碳密度反演

严恩萍1,林辉1**,王广兴1,2,陈振雄3   

  1. 1中南林业科技大学林业遥感信息工程研究中心, 长沙 410004;  2南伊利诺伊大学地理系, 美国卡本代尔,  62901;  3国家林业局中南林业调查规划设计院, 长沙 410004)
  • 出版日期:2015-11-18 发布日期:2015-11-18

Estimation of Hunan forest carbon density based on spectral mixture analysis of MODIS data.

YAN En-ping1, LIN Hui1, WANG Guang-xing1,2, CHEN Zhen-xiong3   

  1. (1Research Center of Forest Remote Sensing & Information Engineering, Central South University of Forestry & Technology, Changsha 410004, China; 2Department of Geography, Southern Illinois University, Carbondale IL 62901, USA; 3Central South Forest Inventory and Planning Institute, State Forestry Administration, Changsha 410004, China)
  • Online:2015-11-18 Published:2015-11-18

摘要: 随着遥感技术的快速发展,基于遥感影像和地面样地的方法成为目前森林碳密度估算的常用手段.然而由于混合像元的存在严重制约了区域森林碳密度反演精度的提高,特别是MODIS这种低空间分辨率影像.本研究以MODIS影像和固定样地为数据源,开展森林碳密度的反演研究.首先利用不带约束、带约束的线性分解和非线性分解3种方法进行混合像元分解,导出不同土地利用/覆盖类型的丰度图;然后采用结合和未结合丰度图的序列高斯协同模拟算法对湖南省森林碳密度进行反演.结果表明: 3种混合像元分解模型中,带约束线性分解估计的地物丰度精度最高(平均均方根误差0.002),明显优于不带约束线性分解和非线性分解模型;通过将混合像元分解模型和序列高斯协同模拟算法结合,森林碳密度反演精度从74.1%提高到81.5%,均方根误差从7.26减小到5.18;2009年湖南省森林碳密度的平均值为30.06 t·hm-2,变化范围介于0.00~67.35 t·hm-2之间.这表明混合像元分解在提高区域和全球尺度森林碳密度反演精度方面显示出巨大的潜力.

Abstract: With the fast development of remote sensing technology, combining forest inventory sample plot data and remotely sensed images has become a widely used method to map forest carbon density. However, the existence of mixed pixels often impedes the improvement of forest carbon density mapping, especially when low spatial resolution images such as MODIS are used. In this study, MODIS images and national forest inventory sample plot data were used to conduct the study of estimation for forest carbon density. Linear spectral mixture analysis with and without constraint, and nonlinear spectral mixture analysis were compared to derive the fractions of different land use and land cover (LULC) types. Then sequential Gaussian co-simulation algorithm with and without the fraction images from spectral mixture analyses were employed to estimate forest carbon density of Hunan Province. Results showed that 1) Linear spectral mixture analysis with constraint, leading to a mean RMSE of 0.002, more accurately estimated the fractions of LULC types than linear spectral and nonlinear spectral mixture analyses; 2) Integrating spectral mixture analysis model and sequential Gaussian cosimulation algorithm increased the estimation accuracy of forest carbon density to 81.5% from 74.1%, and decreased the RMSE to 5.18 from 7.26; and 3) The mean value of forest carbon density for the province was 30.06 t·hm-2, ranging from 0.00 to 67.35 t·hm-2. This implied that the spectral mixture analysis provided a great potential to increase the estimation accuracy of forest carbon density on regional and global level.