Abstract
A major problem in classification of hyperspectral datasets is the reliable selection of reference spectra to classify heterogeneous vegetated areas. In this work, different approaches of end-member collection for spectral unmixing of hyperspectral data are presented, evaluated and compared to each other. The selected methods include high resolution ground spectroradiometric measurements, a scene based approach and modelling of the end-member spectra using a SVAT (soil-vegetation-atmosphere-transfer) approach. The presented methods are discussed and verified with an extensive ground truth collected in the observed area.