The rationale for automatic processing of multispectral scanner data 
are summarized as follows: 
1. Automatic processing can be done in quasi-real time, that is, 
before the information content of the data can significantly decay in 
worth. 
2. Automatic processing of large volumes of data can be done more 
cost effectively (not necessarily more cheaply) than manual processing 
and interpretation. 
3. Although the information volume of the desired kind may be 
small, the data volume is exceedingly high. Automating this data volume 
reduction to information frees people for creative tasks less well 
suited for machines. 
4. Automatic processing gives a potential for a greater consistency 
of results with objective classification standards. Some would also 
mention higher accuracy here for automatic over manual processing. 
5. Derived information from automatic interpretation is in a form 
for quick and easy integration with other data bases in automatic infor- 
mation systems such as automatic mapping and compilation of statistical 
records or summaries. 
6. Multispectral scanner data is recorded in electronic form 
intended for automatic processing in contrast to photography in which 
film is the recording medium. 
A multistage sampling technique is envisaged for some operational 
applications whereby an earth resource or land use inventory would be 
performed using three data-collecting systems: satellites, aircraft, 
and ground observers, in that sequence. Each of these systems would 
provide progressively closer looks at progressively smaller areas, and 
would provide more detailed information about those areas. Then, the 
more detailed information would be applied to a much larger area for 
which the limited sample appeared to be representative, as evidenced 
by the similarity of the small areas to certain surrounding areas as 
recognized in the space and aircraft data. The first tests of this 
three stage approach for multispectral scanners will come from Earth 
Resources Technology Satellite (ERTS) experiments. 
But to achieve a practical tool, the sensors and automatic processors 
must replace a sufficient number of ground observers for the costs to 
be reasonable (a few U.S. dollars per square kilometer is the goal). 
2. Basis of Spectral Discrimination 
The most important keys in photographic interpretation are generally 
spatial in nature, e.g., the shape, size, texture of an object class if 
given sufficient contrast, although color may also be important. However, 
the multispectral scanner (see Fig. 1) is a device which emphasizes 
spectral information for recognition, although temporal, spatial, and