0 
le 
It would make the data range smaller, and 
the estimated spectral age might be not 
accurate. Thus the correlation 
coefficients between winter images became 
smaller than that between summer images. 
However, the good correlation 
coefficients among :3. images without D933 
would suggest that the exponential curves 
were probably appropriate Eo define 
training data, and the three TM channels 
(TM3, TM4, and TM5) were also appropriate 
to classify spruce sistands into: age 
classes spectrally. However, the spectral 
age doesn't show exact stand age, but the 
forest ticonditieon, which ds... typical. (op 
ideal in each age for forest management. 
identify relatively 
stands with spruce, 
Tt" was possible to 
dense or sparse (poor) 
when the spectral age was compared with 
the stand age. If spruce is densely 
populated, the. spectral. age will appear 
older than its stand age. Meanwhile, if 
spruce 1s sparsely populated with many 
deciduous broad-leaved trees, the 
spectral age will appear younger than its 
Stand . age... Lf . two «spectral |.ages . are 
compared, changes in forests are 
evaluated as changes of spectral age. For 
example, dots, which appear far above the 
regression lines, show stands with 
increasing of spruce dominance during 8 
years Priqurs 7' cer. Tt meant tbat 
Spruce trees surpassed broad-leaved trees, 
or broad-leaved trees were selectively 
Gut. On the dots, which 
other hand, 
a) 1004 
804 
60 
40- 
20-4 
Spectral Age (1985/3/12) 
  
  
  
e 
I | | 
-20 0 20 40 60 80 
Stand Age in 1985 
c) 1004 
804 
60 
404 
20-1 
Spectral Age (1993/7/8) 
  
  
  
T T 
-20 0 20 40 60 80 
Stand Age in 1993 
Spectral Age (1985/8/10) 
29 
appear far below the lines, show stands 
with decreasing of spruce dominance. It 
meant that spruce trees were suppressed 
by. broad-leaved trees, or spruce trees 
were clearly or selectively ‘cut during 8 
years. 
Since. the trajectory :reaches: asymptotic 
Values in every-cases/ at around 30: years 
old, accurate successional, monitoring 
wouldcsbe difficult 4n:old growth forests 
based on spectral signatures. The smaller 
slope of the regression lines in Figure 3 
may be caused by the asymptotic nature of 
the successional spectral trajectory. 
However, if the spectral age is compared 
with the stand age (Figure 33, 
differences between two age classes 
suggest ‘forest condition, ex. dense or 
Sparse “spruce jiiclearly. TF the spectral 
age is much bigger than stand age (ex. 
stands {with stand age: 25, and spectral 
age: 405 to 70 ina and. b), spruce 
canopies closed completely and the stands 
may need selective logging. It would hold 
true even for older stands. 
On the over hand, if the spectral age is 
much smaller than stand age (ex. stands 
with stand age: 50 to 70, and spectral 
age: less than.42043, 4the stand ‘may ide 
suppressed by broad-leaved trees and need 
any nursing operation. 
When 
images are compared, 
spectral ages of winter and summer 
changes in deciduous 
  
  
  
1004 b) 
80 
604 
404 
204 
0 T T T T m3 
-20 0 20 40 60 80 
Stand Age in 1985 
Figure 3. Comparison of stand age and 
spectral age 
The: solid line -has 1. and..0 sof the: slope 
and offset respectively. Dashed lines are 
regression lines with the equations 
bellow. 
a)>vesi 4 .961U0b0.73844. 75x r=0.725 
bi y. > 8.452 + 0.7325: '% r=0.662 
C)ityoe 13.079 4.0.5559. x r-0.643 
r: correlation coefficient 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996