Improving discrimination of savanna tree species through a multiple endmember spectral-angle-mapper (SAM) approach: canopy level analysis

Abstract

Differences in within-species phenology and structure are controlled by genetic variation, as well as topography, edaphic properties, and climatic variables across the landscape and present important challenges to species differentiation with remote sensing. The objectives of this paper were to (i) evaluate the classification performance of a multiple-endmember spectral angle mapper (SAM) classification approach (conventionally known as the nearest neighbour) in discriminating ten common African savanna tree species and (ii) compare the results with the traditional SAM classifier based on a single endmember per species. The canopy spectral reflectance of the tree species (Acacia nigrescens, Combretum apiculatum, Combretum Imberbe, Dichrostachys cinerea, Euclea natalensis, Gymnosporia buxifolia, Lonchocarpus capassa, Pterocarpus rotundifolius, Sclerocarya birrea and Terminalia sericea) were extracted from airborne hyperspectral imagery that was acquired using the Carnegie Airborne Observatory (CAO) system in the Kruger National Park, South Africa, in May 2008. This study highlights four important phenomena: (i) intra-species spectral variability affected the discrimination of savanna tree species with the SAM classifier, particularly the producer's accuracy, (ii) the effect of intra-species spectral variability was minimised by adopting a multiple endmember approach, (iii) the classification accuracy of the multiple endmember classifier was affected by the quality of the training endmembers, and (iv) targeted band selection improved be the classification of savanna tree species. The authors furthermore proposed bootstrapping as a method to obtain the best training subset for the classification.