Practical limits on hyperspectral vegetation discrimination in arid and semiarid environments

Hyperspectral remote sensing is a promising tool for the analysis of vegeta tion and soils in remote sensing imagery. The purpose of this study is to a scertain how well hyperspectral remote sensing data can retrieve vegetation cover, vegetation type, and soil type in areas of low vegetation cover. We use multiple endmember spectral mixture analysis (MESMA), high-quality fie ld spectra, and AVIRIS data to determine how well full-range spectral mixtu re analysis (SMA) techniques can retrieve vegetation and soil information. Using simulated AVIRIS-derived reflectance spectra, we find that, in areas of low vegetation cover. MESMA is not able to provide reliable retrievals o f vegetation type when covers are less than at least 30%. Overestimations o f vegetation are likely, but vegetation cover in many circumstances can be estimated reliably. Soil type retrievals are more than 90% reliable in disc riminating dark-armored desert soils from blown sands. This simulation comp rises a best-case scenario in which many typical problems with remote sensi ng in areas of low cover or desert areas are minimized. Our results have br oad implications for the applicability of full-range SMA techniques in anal ysis of data from current and planned hyperspectral sensors. Several phenom ena contribute to the unreliability of vegetation retrievals. Spectrally in determinate vegetation types, characterized by low spectral contrast, are d ifficult to model correctly even at relatively high covers. Combinations of soil and vegetation spectra have the potential of generating mixtures that resemble an unmixed spectrum from different material, further confounding vegetation cover and soil type retrievals. Intraspecies spectral variabilit y and nonlinear mixing produce uncertainties in spectral endmembers much la rger than that only due to instrumental noise modeled here. Having establis hed limits on linear spectral unmixing in areas of low cover through spectr al simulations. we evaluate AVIRIS-derived reflectance data from the Mojave Desert, California. We show that MESMA is capable of mapping soil surface types even when vegetation type cannot be reasonable retrieved. (C) 2001 El sevier Science Inc. All rights reserved.