58 
Not only does the total number of possible classes remain unknown, but On 
the nature of the classification schemes most appropriate to machine process- environ 
ing has yet to be determined. Evidence from work accomplished in this re- illustr 
search suggests that since the scanner/computer combination records the land Distric 
use of each resolution element rather than generalizing by land-use polygon, a clust 
possibilities exist for mapping variations within a land-use type which might semi-ci 
be generalized by the air photo interpreter. Units of open space, such as signatu 
vacant lots within type areas -- for example, planned industrial parks -- similar 
are discernible. Also, thanks to spectral differences, parking lots may be ing den 
separated from the buildings they surround in such generalized area types dominan 
as shopping centers. 
Th 
Still undetermined is whether or not the attainment of a high enough in arid 
level of accuracy has been reached to make the computer-aided land-use map identif 
valuable to planners. For some of the land uses, results are consistently within 
high while others are too low; an average of about 85 to 90 percent accuracy numerou 
generally prevails. For purposes of relating uses to particular ownership contim 
parcels, anything short of 100 percent is inadequate, while for general ground 
monitoring of urban growth, a lower figure is acceptable, especially con 
sidering the inherent ability to repeat the process with great frequency. Or 
A full evaluation by potential users needs to be made to know the potential preter 
utility of the system. natural 
in whic 
To be of value to planners in making policy decisions, land-use data low der 
must be aggregated by jurisdictional and other statistical units. In this, they ar 
success has been reached in aggregating land uses by counties and by census the exp 
tracts. Boundaries of other units -- corporate cities, transportation zones, in area 
etc. -- could be placed on line-printer maps and the data tabulated following forest 
the procedures now developed. Easily done also is the ordering of the classi- deciduc 
fication results into numerous forms and levels of discreteness. desert 
from lc 
A common question raised during the early development of computer-aided sponse 
mapping of multi spectral data has been whether or not the system functions 
equally well in all types of physical environments. Even working within the 
United States, significant differences in environment are encountered. 
Broadly, the problems caused by the differences between the humid East and 
the arid West have been faced in this study; similar experience was gained in T\ 
our parallel work in California, the District of Columbia, and Connecticut. Survey 
is autf 
As might be expected, the principal variable between East and West affect 
ing spectral identification has been the varying degrees of vegetative cover. 
A related phenomenon, degrees of weathering of impervious surfaces, is a pro 
duct of wet versus dry and is analogous to the chemical/physical weathering 
differences recognized in geomorphology. 
From the view point of the spectral interpreter, the differences in 
environment are both help and hindrance. Knowing how to capitalize on advan 
tages and how to overcome difficulties requires a full knowledge and apprecia 
tion of the spectral nature of surfaces and land covers and how they behave in 
differing environments.