349
SPECTRAL VARIABILITY AND ANALYSIS PROCEDURES FOR HIGH
RESOLUTION REFLECTANCE DATA
JOHN C. PRICE
USDA Agricultural Research Service
Beltsville Agricultural Research Center
Beltsville, Maryland 20705
ABSTRACT:
Hyperspectral data (0.4 - 2.5 /¿m reflectance data at 0.01 fim resolution) are
considered for both surface and aircraft data sets representing soils,
vegetation, and other common surfaces. Both types of data are well described
by 20-30 spectral shapes, although minerals may require a larger number due to
sharp absorption features at the longer wavelengths. This suggests that, in
contrast with research studies, operational applications do not require 200 or
more spectral measurements at 0.01 //m resolution to obtain the useful
information in reflectance data. Comparison of surface and aircraft
observations shows that the types of shapes are similar, except for known
atmospheric water vapor features in the aircraft spectra. It thus should be
possible to estimate water vapor corrections in remotely sensed reflectance
data using relatively broad band (0.04/um) spectral observations.
KEYWORDS: Hyperspectral, Spectral Collections, AVIRIS, Basis functions
1. INTRODUCTION
Recent advances in instrumentation, combined with interest in global
environmental assessment, have prompted the development of high spectral
resolution sensors which provide imagery with large numbers of spectral bands.
The AVIRIS instrument (Vane, 1987) is a good example, obtaining data in 224
spectral bands in the range 0.4-2.5 ¿im for an image swath more than 600 pixels
wide. This discussion follows closely that of the previous meeting of this
symposium (Price, 1991), applying a newer methodology which has been developed
to address the expansion of spectra in basis functions (Price, 1993) when the
number of such functions becomes relatively large, i.e. >15. We first review
the description by spectral basis functions, then apply the formalism to
collections of surface/laboratory spectra, then to AVIRIS imagery, and then
describe the relationship between the two types of spectra, where the
systemmatic difference (atmospheric water vapor) is readily identified. 2
2. DESCRIPTION OF HYPERSPECTRAL DATA BY BASIS FUNCTIONS
Let x a (A) = (x^, x°, ...x^) represent a measured spectrum for the set of n
wavelength values A = (A^, A^, A^, ...A^), with superscript a denoting the
individual sample. Throughout we shall work with reflectance spectra, i. e.
the ratio of reflected to incident radiation, as this eliminates effects of
local illumination conditions and facilitates comparison of spectral
collections from many laboratories. For remote sensing applications the
illumination source is usually the sun. We shall describe visible to near
infrared spectra (0.4 to 2.5 ¿¿m) by a set of spectral basis functions:
