Toward Improved Hyperspectral Analysis in Semiarid Systems

Idaho State University's Boise Center Aerospace Laboratory (BCAL) has processed and applied hyperspectral data for a variety of biophysical sciences in semiarid systems over the past 10 years. HyMap hyperspectral data have been used in most of these studies, along with AVIRIS, CASI, and PIKA-II data. Our studies began with the detection of individual weed species, such as leafy spurge, corroborated with extensive field analysis, including spectrometer data. Early contributions to the field of hyperspectral analysis included the use of: time-series datasets and classification threshold methods for target detection, and subpixel analysis for characterizing weed invasions and post-fire vegetation and soil conditions. Subsequent studies optimized subpixel unmixing performance using spectral subsetting and vegetation abundance investigations. More recent studies have extended the application of hyperspectral data from individual plant species detection to identification of biochemical constituents. We demonstrated field and airborne hyperspectral Nitrogen absorption in sagebrush using combinations of data reduction and spectral transformation techniques (i.e., continuum removal, derivative analysis, partial least squares regression). In spite of these and many other successful demonstrations, gaps still exist in effective species level discrimination due to the high complexity of soil and nonlinear mixing in semiarid shrubland. BCAL studies are currently focusing on complimenting narrowband vegetation indices with LiDAR (light detection and ranging, both airborne and ground-based) derivatives to improve vegetation cover predictions. Future combinations of LiDAR and hyperspectral data will involve exploring the full range spectral information and serve as an integral step in scaling shrub biomass estimates from plot to landscape and regional scales.