基于Hyperion数据的香格里拉森林郁闭度遥感估测研究  

Estimating Methods of Forest Canopy Closure Based on Hyperion Data of Shangri-La

作　　者：胡振华[1] 王丽媛[1] 岳彩荣[1] 王宗梅[1] 

机构地区：[1]西南林业大学西南地区生物多样性保育国家林业局重点实验室,云南昆明650224

出　　处：《西南林业大学学报（自然科学）》2017年第3期159-164,共6页Journal of Southwest Forestry University：Natural Sciences

基　　金：国家自然科学基金项目(31260156)资助;云南省林学一流学科建设项目(51600625)资助;西南林业大学林学一级学科研究生创新基金项目资助;西南林业大学科技创新基金项目(C15086)资助

摘　　要：基于Hyperion高光谱数据,采用2种方法进行波段选取,将选择的波段数据进行特征提取变量,采用偏最小二乘法分别对2种方法选择的特征变量建立香格里拉主要树种郁闭度遥感估测模型,并进行精度检验评价。结果表明:基于实测样地郁闭度差异特征分析选择的Hyperion特征波段建立的模型R^2为0.837、估测精度为82.09%,基于遥感影像进行分段主成分分析选择的Hyperion特征波段建立的模型R^2为0.764、估测精度为78.4%,基于样地数据郁闭度变化敏感性分析模型优于基于Hyperion影像的分段主成分分析模型;分段主成分分析法所选出的特征波段虽然包含了较多的波段信息,但是很多为连续波段或者波长较近的波段,波段之间的相关性较高,导致建模精度不如预期。Use 2 methods to band selection and choose the band data feature extraction variables of the hyper- spectral data. Using partial least squares to the Shangri-La main tree crown density remote sensing estimation model building, and its accuracy were presented and checked in the study. The results showed that R2 of the model based on Hyperion characteristic band sensitive to forest crown closure was 0. 837, the estimation precision was 82. 09% ; R2 of the model based on Hyperion characteristic band via selection of segmented principal component was 0. 764, the estimation precision was 78. 4%. The accuracy and fitting effects of based on inventory data model were better than based on hyperspectral data model. Although the selected band characteristics from segmented principal com- ponent analysis contained more information, but many for continuous band or wavelength band. The highly correla- tive of bands which lead to modeling precision accuracy was lower than expected.

关 键 词：HYPERION数据 二类调查 森林郁闭度 主成分分析法 偏最小二乘法 香格里拉 

分 类 号：S757.2[农业科学—森林经理学]