A biogeophysical approach for automated SWIR unmixing of soils and vegetation

Arid and semiarid ecosystems endure strong spatial and temporal variation of climate and land use that results in uniquely dynamic vegetation phenology, cover, and leaf area characteristics Previous remote sensing efforts have not fully captured the spatial heterogeneity of vegetation properties required for functional analyses of these ecosystems, or have done so only with manually intensive algorithms of spectral mixture analysis that have limited operational use. Those limitations motivated the development of an automated spectral unmixing approach based on, a comprehensive analysis of vegetation and soil spectral variability resulting from biogeophysical variation in arid and semiarid regions. A field spectroscopic database of bare soils, green canopies, and litter canopies was compiled for 17 arid and semiarid sites in North and South America, representing a wide array of plant growth forms and species, vegetation conditions, and soil mineralogical-hydrological properties. Spectral reflectance of dominant cover types (green vegetation, litter, and bare soil) varied widely within and between sites, but the reflectance derivatives in the shortwave-infrared (SWIR2: 2,100-2,400 nm) were similar within and separable between each cover type. Using this result, art automated SWIR2 spectral unmixing algorithm was developed that includes a Monte Carlo approach for estimating errors in derived subpixel cover fractions resulting from endmember variability. The algorithm was applied to SWIR2 spectral data collected by the Airborne Visible and infrared Imaging Spectrometer instrument over the Sevilleta and Jornada Long-Term Ecological Re-search sites. Subsequent comparisons to field data and geographical information system (GIS) maps were deemed successful. The SWIR2 region of the reflected solar spectrum provides a robust means to estimate the extent of bare soil and vegetation covers in arid and semiarid regions. The computationally efficient method developed here could be extended globally using SWIR2 spectrometer data to be collected from platforms such as the NASA Earth Observing-1 satellite. (C) Elsevier Science Inc., 2000.