| The main objective this research was to assess two types of emerging remote sensing technology, hyperspectral and lidar sensors, for the automated discrimination of tropical rain forest tree (TRF) species. The hyperspectral data contain information on the biochemical and structural properties of crowns, while the lidar data contain structural information. I hypothesized that these two datasets combined would permit greater species classification accuracy than either dataset alone.; Working in an old-growth TRF in Costa Rica, canopy-emergent individual tree crowns (ITCs) for seven target species were manually digitized with reference to high spatial resolution hyperspectral and lidar datasets that were acquired from airborne sensors. Multispectral and hyperspectral classification was performed using pixel- and crown-scale spectra and spectral angle mapper (SAM), maximum likelihood (ML), and linear discriminant analysis (LDA) classifiers. Pixel-majority and crown-scale ITC classifications were significantly more accurate with hyperspectral data relative to multispectral data, revealing the importance of the spectral detail offered by hyperspectral imagery. Additional techniques were explored to best harness this spectral information. These included incorporating hyperspectral metrics into decision trees (DTs) and multiple endmember spectral mixture analysis (MESMA). The best spectral-based classification accuracy was with crown-scale spectra and a relatively simple LDA procedure. These results suggested that hyperspectral imagery need not be acquired at a very high spatial resolution or analyzed with sophisticated techniques to provide adequate discrimination of species. Leaf phenology was important in mapping TRF tree species. Leaf-off trees had distinct volume-scattering and spectral mixing properties that influenced classifier variable selection as well as final classification accuracy.; Crown-scale hyperspectral data were combined with structural data from the lidar sensor in LDA and DT classifiers. There were significant differences in the majority of lidar-derived structural metrics among the study tree species; however, the addition of this information to the classifiers did not improve classification accuracies. Although lidar data was not useful for species discrimination, it did provide an unprecedented view of canopy topography and sub-canopy elevation that is difficult to measure using traditional techniques. |
