Imaging spectroscopy for desertification studies: comparing AVIRIS and EO-1 Hyperion in Argentina drylands

Arid and semiarid ("dryland") regions are complex mosaics of vegetation cover, structure, and phenology. Few multispectral remote sensing approaches have quantitatively resolved the spatial complexity of photosynthetic vegetation (PV), nonphotosynthetic vegetation (NPV), and bare soil in drylands. In combination, these surface properties provide insight to land degradation and desertification known to be occurring in many drylands worldwide. Given sufficient spatial resolution and sensor performance, imaging spectroscopy provides this information using reflectance measurements over the 0.4-2.5-μm region. We tested and compared the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) and the spaceborne Earth Observing 1 Hyperion imaging spectrometer for measuring PV, NPV, and bare soil fractional cover in the Monte Desert biome of Central Argentina. A probabilistic spectral mixture model was used to decompose image pixels into subpixel surface constituents using a spectral endmember bundling approach. AVIRIS (4.5-m pixels) and Hyperion (30-m pixels) data were collected over a 763-km² region containing the Nacunan Man-and-Biosphere Reserve and surrounding unprotected areas. The AVIRIS data, combined with the mixture modeling, provided highly accurate estimates of PV, NPV, and bare soil in comparison with field measurements (0.832<0.86). Hyperion provided accurate estimates of PV (r²=0.68) due to good red-edge (0.69-0.71 μm) performance, but NPV and bare soil estimates were accurate only when cover values exceeded ∼30% (r²=0.39-0.40). Lower NPV and soil cover accuracies from Hyperion were due to lower S/N performance in the shortwave-infrared (2.0-2.4 μm) region. Convolving the AVIRIS reflectance data to 30-m pixel size showed that apparent accuracies decrease by 20% due to field/remote sensing colocation errors. AVIRIS 30-m results for PV were statistically similar to Hyperion 30-m results.