An approach for analysis of reflectance spectra

A band selection procedure is applied to 45 AVIRIS hyperspectral images with wavelength range 0.4-2.5 mu m, in order to establish the physical significance of individual spectral bands. Five bands (0.46-0.54 mu m, 0.61-0.69 mu m, 0.99-1.09 mu m, 1.52-1.61 mu m and 2.08-2.17 mu m), which describe 98% of the mean square spectral signal, can discriminate water, snow, fire, vegetation, and a residual class which is principally soil. For the five classes, 20 bands represent all spectra with a mean square residual of 0.1%, which approaches the noise level of the data. Atmospheric water vapor is a major source of variability for soil and vegetation spectra. Spectral bands of 0.99-1.09 mu m, 1.12-1.16 mu m, and 1.20-1.31 mu m are processed in conjunction with field and laboratory spectra and the Lowtran radiative transfer code in order to quantify this effect. Further study is required to characterize properties associated with additional bands. (C)Elsevier Science Inc., 1998.