期刊
PHOTOGRAMMETRIE FERNERKUNDUNG GEOINFORMATION
卷 -, 期 5, 页码 573-588出版社
E SCHWEIZERBARTSCHE VERLAGSBUCHHANDLUNG
DOI: 10.1127/1432-8364/2012/0140
关键词
tree species; classification; temperate forest; WorldView-2; linear discriminant analysis
Suitability of WorldView-2 data for tree species classification with special emphasis on the four new spectral bands. There is an increasing demand for information on tree species composition and spatial distribution. Detailed tree species maps are essential for addressing different ecological problems. Moreover, they are important in modern forest management as close to nature forest management is becoming more and more common. This study examines the potential of 8-band WorldView-2 data with 2 in ground sample distance (GSD) for identifying 10 tree species in a mid-European forest. We delineated well-illuminated tree crowns manually and did a supervised classification using linear discriminant analysis (LDA) implemented in a bootstrapping environment (1465 observations, 500 bootstrap samples). The overall accuracy for the 10 tree species was around 84 % (8 bands) with class-specific producer's accuracies ranging between 54 % (Carpinus betulus) and 96 % (Fagus sylvatica). In general, the accuracy was higher for the 5 deciduous tree species (89 %) than for the 5 coniferous tree species (79 %). The classification accuracies decreased significantly, if only the 4 conventional bands Blue, Green, Red, and Near Infrared I were used (78 % overall accuracy). However, when the classification focused only on the 4 main tree species in the investigation area, i.e. Picea abies, Pin us sylvestris, Fagus sylvatica, Quercus robur, the 4 conventional bands were sufficient to achieve high classification accuracies (95 % overall accuracy). Adding the 4 new bands (Coastal, Yellow, Red Edge, Near Infrared 2) to the input feature set did not further improve the overall classification accuracy for these 4 main tree species. Hence, the positive impact of the additional 4 new bands resulted from strongly increased classification accuracies of the 6 secondary tree species. Green and Near Infrared 1 (conventional bands) as well as the bands Red Edge and Near Infrared 2 (new bands) contributed most to class separability (according to Wilks' Lambda).
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据