34 
COMBINED SPATIAL FEATURES AND SPECTRAL AVERAGE FEATURES 
JULY 10 DATA/JULY 10 SIGNATURES 
TABLE IX. 
amiss ( 
1 ERRORS 
CATEGORY ACCURACY URBAN FARMII 
D IRRIGATI 
G FARMINt 
RANGELAND FORESTS 
Urban 96 
4 
Dryland 
Farming 100 
Irrigated 
Farming 93 2 
6 
Rangeland 99 
1 
Forests 97 
3 
WEIGHTED AVERAGE: 97 
CONCLUSIONS 
Figure 4 summarizes the average classification performance 
obtained using spatial features for the July and August imagery and the 
results obtained when July 10 spatial signatures are applied to August 15 
data. For comparison, the average classification performance attained with 
spectral signatures for the corresponding cases are plotted. Figure 5 sum 
marizes the average classification performances obtained for the various tests 
applied to the July 10 image. From Figure 4 it is evident that spatial sig 
natures discussed in the test do not extend in time for the land-use categories 
investigated at these two image dates. However, at both dates, classification 
performance using spatial features compares favorably with that obtained using 
spectral features of individual pixels. 
Several observations may be drawn from Figure 5. Classification 
using spatial features derived from only one spectral band (MSS 7) is greater 
than that obtained using average features derived from the same band. However, 
classification accuracy using spectral average feature vectors derived from 
all spectral bands is greater than that achieved from the spatial features 
derived from only one band. Appending the spectral average response from one 
band to the spatial features derived from another band increases classification 
performance to the level achieved by employing spectral average features 
using all bands. 
Absolute classification accuracies indicated in the two figures 
apply only to training data and may not extrapolate to test areas; however, the