Table 3. Variables Required to Meet Varying Accuracies 
354 
Accuracy 
(100-E) 
# variables 
Moffet field 
(1 scene) 
# variables 
1% sample 
(28 scenes) 
# variables 
1% sample +1% bad 
(28 scenes) 
90% 1 1 
99% 3 3 
99.9% 9 8 
99.95% - 15. 
99.96% - 20 
1 
4 
12 
19 
23 
The recommended spectral bands are presented in table 4. 
Table 4. Recommended bands for the 28 AVIRIS scenes (¿tm) 
0.40-0.44 
0.48-0.55 
0.60-0.68 
0.69-0.70 
0.72-0.74 
0.75-0.85 
0.90-0.93 
0.97-0.99 
0.99-1.08 
1.11-1.16 
1.17-1.22 
1.23-1.28 
1.29-1.31 
1.43-1.48 
1.50-1.56 
1.57-1.71 
1.97-1.99 
2.07-2.08 
2.09-2.16 
2.21-2.24 
2.26-2.31 
2.41-2.47 
2.48-2.50 
Finally, each of the coefficient images (S^) for each of the 28 scenes was 
studied visually. Small variance images (i>23) still showed signals at a 
level of a few tenths of a percent. Possible explanations include instrument 
noise, spectral misregistration, broad atmospheric variations between images 
(aerosols), and true scene to scene variability. Little or no evidence was 
found for isolated surface types with extraordinary spectral features such as 
minerals. 
5. COMPARISON OF SURFACE AND AVIRIS SPECTRA: THE EFFECT OF THE ATMOSPHERE 
There exists a systematic difference between the shapes described by the basis 
functions for surface/laboratory spectra, and those derived from AVIRIS data. 
This difference is due to water vapor absorption features in the AVIRIS 
spectra. The sample sizes for the data sets are disproportionate, consisting 
of a few thousands of surface spectra, and almost 10 million AVIRIS spectra. 
Therefore we have used all the surface spectra plus a sample of the AVIRIS 
spectra, with a larger number of AVIRIS spectra because these have not been 
corrected for atmospheric absorption or solar zenith angle, and thus have 
lower "effective" reflectances. By combining these data sets, with the 
regions of strong water absorption (defined previously) omitted, we find that 
the weak absorption features in the AVIRIS spectrum represent the third most 
important shape variable (basis function) in the ensemble. However since our 
goal is to invert AVIRIS spectra to obtain spectra having the same shapes as 
surface spectra, we select spectral intervals which are suitable for 
describing both types, then identify a spectral band which specifies the 
atmospheric effect. Thus using 11 bands at (0.40-0.49, 0.50-0.70, 0.71-0.73, 
0.74-0.83, 0.97-1.11, 1.23-1.34, 1.48-1.54, 1.55-1.77, 2.03-2.14, 2.15-2.36, 
and 2.41-2.50/im), we can describe both surface and AVIRIS spectra very well, 
except in the regions of strong water vapor absorption. Are these 11 broad 
band intervals sufficient to provide a reasonable estimate of reflectance in 
regions where atmospheric water vapor causes a difference from ground values, 
as is required for estimation of atmospheric water vapor, and for possible 
identification of surface types? By calculation on the surface ensemble we 
may compare this approach with the Continuum Interpolated Band Ratio (CIBR) 
and the Narrow/Wide (N/W) algorithms, as discussed in Carriere and Conel 
(1993). The CIBR method averages spectral values at 0.88 ¿¿m and 1.10 Atm to