353 
4. APPLICATION TO AVIRIS IMAGERY 
Twenty eight AVIRIS scenes were selected using the quick look data at the Jet 
Propulsion Laboratory. They were transformed to apparent reflectance by 
dividing by the solar constant. The description of spectra in terms of 
reflectance as opposed to radiance greatly equalizes the wavelength 
dependence. However it increases the effect of noise. 
Table 2 lists the 
data sets used, and the scene type. 
The selection largely 
spans the natural 
. variability 
represented in the AVIRIS data sets, with the 
exception of clouds. 
The two 
spruce forest scenes (920616B and 920615B) are 
for overlapping areas, 
as was 
discovered only during 
image processing. 
Table 2. AVIRIS 
Data 
Scenes and Surface Types 
Flight 
Run 
Scene 
Catalog name 
type of scene 
920602A 
9 
8 
Moffett Field 
suburb, shallow water 
920826B 
3 
2 
Maricopa farm 
agriculture 
920828B 
13 
1 
Los Alamos 
geology, town 
920827B 
2 
5 
Rodgers Dry Lake 
geology 
921119B 
9 
8 
Tampa Bay 
city, water 
920603B 
2 
3 
Cuprite 
geology 
920826B 
5 
1 
Camp Pendleton 
water, military base 
920615B 
2 
3 
Harvard Forest 
forest 
920612B 
2 
5 
Indian Pines 
agriculture, forest 
920531C 
6 
1 
Death Valley 
geology 
920603B 
14 
3 
Cima volcanic field 
geology 
920708B 
1 
2 
Gainesville, FL 
lake, vegetation 
920819B 
2 
1 
Denver 
suburb, agriculture 
920828B 
2 
5 
San Juan Mtns 
snow, geology 
920602A 
6 
2 
Jasper Ridge 
suburb, vegetation 
920826B 
4 
1 
Fort Huachuca 
a.f.base, geology 
920616B 
2 
1 
Spruce forest 
forest, clear cuts 
920820B 
6 
1 
Pleasant Grove 
agriculture 
921117D 
2 
30 
Jackson, TN 
agriculture, forest 
921119B 
5 
4 
Tampa Bay 
island, shallow water 
920826B 
6 
2 
San Joaquin 
agriculture 
920531C 
2 
1 
Owens Valley 
geology 
920820B 
7 
1 
Dunnigan, CA 
agriculture 
920819B 
10 
14 
San Berdardino 
agriculture 
920708B 
5 
1 
Gainesville, FL 
town, vegetation 
920621C 
2 
1 
Blackhawk Island 
ag, forest, water 
920615B 
8 
1 
Spruce forest 
forest, clear cuts 
920820B 
8 
3 
Davis, Webster 
agriculture, town 
Since 28 AVIRIS images represent 4 gigabytes of data an efficient analysis 
strategy was needed. First a scene showing considerable variability in 
surface types (Moffet Field) was analyzed. It was found that 9 variables 
(spectral intervals) described E to within 0.1%. Then a 1% sample of the 28 
scenes (every 100th pixel) was analyzed. Finally, the basis function 
expansion to level 20 was subtracted out from each scene to select "bad" 
pixels, with the worst 1% in terms of residuals being saved from each. Then 
these were added to the original 1% sample, so that the significance of poorly 
described spectra was multiplied by a factor of 50. Even this requires only 
approximately 20 basis functions to describe the AVIRIS signal very well, as 
illustrated in table 3.