Evaluation of discrimination measures to characterize spectrally similar leaves of African Savannah trees

Abstract

Deterministic and stochastic measures of spectral similarity are essential for determining both the geometric characteristics of spectra and variability, respectively. Deterministic measures such as Spectral Angle Mapper (SAM) are useful in establishing similarities between spectra and are also functional in identification of vegetation types. The stochastic spectral similarity measures such as spectral information divergence (SID) describe the spectral prosperities essential for discriminating between the species observed through remote sensing technologies, by modeling spectra as probability distributions. These methods have been extensively utilized in geological studies and are now becoming very useful in vegetation spectroscopy. Little information is, however, available on the performance of these methods. The main objective of this study is to examine the statistical measures essential in distinguishing between the seven major savannah trees located in the largest game reserve (the Kruger National Park) situated in South Africa, at leaf level. The leaf measurements were obtained through the Analytical Spectral Radiometer (ASD) spectrometer of a hyperspectral sensor. The study also assesses the statistical significance of variations (the within and between spectral variations) in leaf spectra with respect to discriminating between these major tree species.