775 
N is dimension of spectral vector G. . (in our case G. . was 4- 
dimensional in single-date classification and 12-dimensional 
in multidate classification). 
G^ j is spectral vector corresponding to pixel ^ 
G is estimated mean spectral vector of Class C 
is estimated covariance matrix of Class C 
The classification procedure used in this study was the maximum likelihood 
classification rule which is based on the above equation. Pixels were 
classified according to their location in spectral space, that is, according 
to the probability of occurrence of their respective spectral vectors in each 
class. If the probability M. . for class yielding the highest value was 
1 i,j J 
under a threshold value of 0.01, the pixel W as classified as 
"unknown". * 
DISCUSSION OF RESULTS AND ACCURACY ESTIMATES 
Computerized classification of ERTS digital images resulted in four 
thematic maps of the Larose Forest test area: the first three are based on 
single-date ERTS scenes recorded at different dates while the fourth one is 
based on multidate combination of the above three scenes. Two display media 
were selected for presentation of the ground classes: line printer and 
digital film printer. The output of line printer were alphabetically coded 
computer printouts (Figs. 2-5) which might be hand colored to enhance class 
separation. The main advantage of this output is that it does not require 
specialized hardware. However, the number of pages becomes cumbersome when 
larger areas are to be displayed. While the relatively small Larose Forest 
test area required 72 pages of line printouts, the whole ERTS scene containing 
close to 7.5 million pixels would be displayed on about 2,500 pages. The film 
image on the other hand represents an ideal data compression and display 
medium for thematic mapping of natural resources. The problem is that digital 
film printers are not readily available in most computer facilities though 
this situation will certainly change by the time the earth resources 
satellites become operational. The thematic maps in Figures 6-9 were printed 
by the ELECTRON BEAM IMAGE RECORDER (EBIR). Besides using the EBIR, 
successful film images were printed from digital tapes by the less expensive 
DICOMED and OPTRONICS digital image recorders. 
Investigation of thematic maps reveals some interesting features. 
The multidate composites (Figs. 5 and 9) provide the best display of ground 
classes because they are free of temporal changes of information which varied 
from one scene to the other as a function of the date of ERTS pass. Thus the 
changing patterns of field cultivation, surface water and image noise are 
greatly reduced on the multidate composite. It can be readily distinguished