Mapping and Spatial Characterization of Nonnative Grasses in the Big Island, Hawaii Using Hyperspectral Imagery

A cross-sensor (hyperspectral and high-resolution data sets) hybrid approach was used to map grass species in the coastal lowland area of the Hawaii Volcanoes National Park. AVIRIS imagery was selected for hyperspectral data and its 20-meter resolution was compensated with IKONOS 1-meter resolution data. Three main native and nonnative grass species were focused in the study, including broomsedge (Andropogon virginicus), natal redtop (Melinis repens), and pili grass (Heteropogon contortus). A 3-step, hybrid approach, combining an unsupervised and a supervised classification schemes, was applied to grass mapping. First, the IKONOS 1-m high-resolution data were classified to create a binary image (vegetated vs. non-vegetated) and converted to 20-meter resolution percent cover vegetation data to match AVIRIS data pixels. Second, the minimum noise fraction (MNF) transformation was used to extract a coherent dimensionality from the original AVIRIS data. Since the grasses were sparsely distributed and most image pixels were intermingled with lava surfaces, the reflectance component of lava was filtered out with a binary fractional cover analysis assuming that the total reflectance of a pixel was a linear combination of the reflectance spectra of vegetation and the lava surface. Finally, a supervised approach was used to classify the grass species based on the maximum likelihood algorithm. The classification result showed that there was much confusion between the grasses, especially between broomsedge natal redtop. Knowing that there was co-occurrence of one or more grass species, more accurate sampling schemes and additional phenology characteristics of the species would be needed to better define training sites.