Recognition of Eucalyptus forest species using hyperspectral reflectance data

Abstract

Leaf reflectance data (400-2500 nm) from six Eucalyptus species and one non-Eucalypt species was measured with a spectroradiometer over a period of ten months. The reflectance and first derivative spectra were assessed for species discrimination using linear discriminant analysis (LDA). Species recognition accuracies up to 93% were obtained with LDA. The recognition accuracies were better with the first derivative spectra and when leaves from different seasons and age classes were included in the training samples. Another important finding from this study was that species with concolorous leaves (same colour on both sides) were clearly separated from species with bicolorous leaves. Species with concolorous leaves have a steep leaf angle (vertically hanging leaves), allowing high light penetration through the canopy. Species with this leaf morphology mostly dominate the arid and semi-arid interior of Australia, but several such species also co-occur in the high rainfall coastal forests with species having shallow angled bicolorous leaves. The ability to discriminate between concolorous and bicolorous Eucalypt species looks encouraging for the study of forest ecology with hyperspectral imagery.