Lon 
Sd 
nage 
Ba 
lage 
Studentized 
  
  
  
  
  
25 difference 
20 
15H 
10f | 
E. [> 3 
sL IE Calib 
error 
| 
90-91 91-92 92-93 
Implementation period 
Figure 3. Spectral change of preparatory cut on the difference 
image H93-90, TM channel 3. 
The possibilities for applying generic training data for 
supervised change detection was studied by comparing the 
spectral change of different treatment classes between 
different regression calibrated/studentized image pairs. All 
the spectral changes between the three image pairs form the 
same geographic location were at comparable level except on 
channel four (Figures 5, 6, 7 and 8). However, the image 
  
  
  
  
Studentized 
ao (difference 
sol 
20} 
10} | 
L HE lib 
error 
| Standard 
9 levint 
90-91 91-92 92-93 
Implementation period 
Figure 4. Spectral change of preparatory cut on the difference 
image H93-90, TM channel 5. 
pair from different geographic locations showed generally 
lower spectral change on all the treatment classes. The 
treatment class *uncommercial thinning' was not clearly 
separable on any channel. Channel four showed varying DN 
change between the image pairs (Figure 7.). The same effect 
was noticed on channel 7 with treatment class ‘uncommercial 
thinning’ (Figure 8). 
  
  
  
  
  
  
  
  
  
  
  
Studentized 
25 SS 
20H 
15 + 
10H B Mean 
P N1988—1990 
E. Mean 
Sf " H1990-1993 
[ Mean 
0 H1990- 1992 
E [ ] Mean 
-5L H1992-1993 
Unt C. thinn. Reg. cut Soil prep. 
Unc. thinn. Prep. cut Clear cut 
Figure 5. Studentized differences of treatment class means on different image pairs on TM cannel 3. 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996