416 
Figure 6. Density slices of axis 1 of the varimax rotation. The 
white and black areas represent areas of extreme change. 
The FA method resulted in the same problem with identification errors 
in the less extreme density changes, as was experienced with the third 
method. However, this method had fewer major detection errors. For 
example, the addition to the incinerator building, which was identified 
as unchanged in the third method, was correctly detected as having 
changed in the fourth method. The errors with the less extreme density 
changes could be caused, in part, by the sun angle and/or the temporal 
change problems discussed earlier. This method does, however, identify 
the major changed and unchanged reference features present in site 3. 
The results from the FA method applied to the remaining test sites were 
similar to those found for test site 3. The four sites were also af 
fected by the problems of sun angle, and seasonal and climatic change. 
Again, there were errors in identification of those areas of change that 
had less extreme density differences between photos. In general, though, 
the FA technique did enhance the areas of change where a shift in photo 
graphic density had occurred. (A more complete description of these 
test sites is given in a recent research report by the authors (Luce 
and Turner, 1981).) The quartimax rotation proved to be the best for 
less than the maximum number of initial factors, for test site 3. How 
ever, it yielded results less satisfactory than rotating all factors. 
In fact, the rotation of 2 out of 4 initial factors gave poor, confus 
ing results. The rotation of 3 out of 4 initial factors can be seen in 
Fig. 7. It shows results similar to those from rotation of all factors. 
It includes some omission errors that the four-factor rotation does not 
commit but "corrects" some errors that the four-factor rotation commit 
ted. Overall, there seems to be little or no advantage to rotating 3 
factors, instead of 4, for this test site. 
6. CONCLUSIONS AND RECOMMENDATIONS 
Using the currently available software at ORSER, the best technique for 
detecting change in microdensitometer data was the FA approach. It was 
the simplest method to use, requiring no specific knowledge of land use 
change, and it gave the best results in the initial test site. The 
first and second methods would have given better results if the classi 
fications in each method could have been improved. This could be 
achieved if some of the problems discussed, such as sun angle and