Estimating nitrogen in eucalypt foliage by automatically extracting tree spectra from HyMap™ data

Airborne HyMap™ data obtained from the crown reflectance of Eucalyptus melliodora were used to estimate nitrogen in the foliage. Estimating chemical concentrations in individual crowns by remote sensing is especially difficult for eucalypts because, first, there is marked variation between individual crowns and, secondly, separating leaf and background spectral information is difficult. We developed an automatic method to select relatively pure tree pixels for each tree. In this method, the background materials are modeled, and the pixels within a crown that do not resemble the background clusters are regarded as target pixels. A modified partial least squares gave an R² value for predicted versus determined nitrogen concentrations of 0.79, with an RMSE of 0.69 mg/g, less than half the standard deviation of the measured values. Automatically selecting tree pixels was more accurate than manual selection, while the study confirmed that using the maximum spectrum gives results that are as accurate as those from the mean spectrum.