Use of a hybrid supervised and unsupervised classification model to determine nitrogen concentration of eucalypt tree foliage using HyMap data

In this study we assess the potential for remote measurement of leaf nitrogen concentration in eucalypts, through the use of high spectral resolution remote sensing. A hybrid supervised and unsupervised classification model for sub-pixel analysis is applied to HyMap (Hyperspectral Mapper) data to identify pixels containing relatively pure Eucalyptus melliodora foliage spectra. Two types of threshold used by the model are assessed, a fixed threshold that selects a variable number of pixels per tree canopy, and a relative threshold that selects a constant number of pixels per tree canopy. The wavelength positions of most of the log (1/R) correlations with foliar nitrogen concentration coincide with reported absorption locations of nitrogen-containing compounds, suggesting a physical basis for the canopy-level predictions of nitrogen obtained in this study