Analysis of Hyperspectral and High-resolution Data for Tree Species Classification

Current tree species classification algorithms often use high-resolution satellite data and are in many cases based on forest stands. The spectral bands of the sensors used for data acquisition are given and cannot be chosen regarding the needs of tree species classification. Furthermore distinction is often limited to deciduous trees, coniferous trees and other land use classes. Single tree based tree species recognition needs very high resolution data. Classification of several distinct species based only on spectral data requires an analysis of the available data to extract information on the spectral signatures of tree species and to find the bands which contain the most relevant information. Only limited information regarding the signatures of tree species in airborne images can be found in the literature, especially regarding those sorts common in central Europe. In this paper, 235 bands in the near infrared and short wavelength infrared regions are evaluated and compared to RGB and color infrared images (CIR) as well as SPOT4 images. The spectral signatures and the overlaps in the spectral signatures of different tree species are analyzed. Using this information, suitable algorithms can be developed and existing algorithms can be evaluated. Furthermore the decision making process for a specific data source can be supported and the gained information can be used in the developments of new sensors to fulfill the requirements of tree species classification.