417 
Figure 7. Quartimax rotation of three factors. The white and 
black areas represent areas of extreme change. 
seasonal change could be overcome; or possibly if a contextual classi 
fier were used. Such a classifier classifies a pixel with considera 
tion of the classifications of its surrounding neighbors, much like the 
process used in visual photointerpretation. (A contextual classifier 
is not yet available in the ORSER System.) The third method was also 
effective, but was hampered by sun angle and temporal change problems. 
It also required considerable data processing before analysis could 
begin. 
The FA method for detecting change worked well, with a few major excep 
tions. First, as with other methods, a change in cover type with no 
change in density cannot be detected. Perhaps visual interpretation of 
an enhanced final product could alleviate the severity of this problem. 
Another possibility would be to use the FA technique as a mask of un 
changed data and analyze the remaining data with another technique to 
identify the types of changes that have occurred. Second, the problems 
with sun angle and seasonal changes could be avoided by having the 
conditions on the day of the new photo mission match as closely as 
possible the conditions on the day the reference photo was taken. This, 
of course, is often impractical and unrealistic. Therefore, the analyst 
must be aware that shadow, reflectance, and climatic changes will be 
mapped, as well as areas of land use/land cover change. Problems of 
sun angle and seasonal changes are not unique to machine processing of 
the data. These problems are often encountered in manual aerial photo 
interpretation (Brew and Neyland, 1980). 
There are several questions that are raised by the FA method for the 
analysis of temporal information. First of all, is the technique appli 
cable to other data types? In particular, does it work well with MSS 
data? What affect will a change in pixel scale have on the technique? 
And, most importantly, how accurate is the analysis? Statistical sup 
port is needed to determine the degree of success of this method. These 
questions are being addressed at ORSER and the results should be forth 
coming within the next year. Based on this study, however, the FA 
technique looks promising as an enhancement method for analyzing tempo 
ral information in digital image data.