grown-up coniferous forest and water, but not 
agricultural land and young forest. Areas covered 
by clouds, cloud shadows or technical errors in 
any of the 36 channels used were excluded from the 
mask. Under the mask median values were extracted 
from all the 36 channels in 1730 systematically 
located windows of 2 by 2 pixels. All the field 
surveyed thinnings, except the two stands with 
more than 50 % deciduous forest and the seed tree 
stands after cutting, were covered by the mask. 
The calibration functions, and with them the 
measured DN changes, will thus be less 
representative for deciduous forest and seed tree 
stands. 
For each channel at the Landsat acquisitions 1985- 
1989, the same channel the previous year were used 
as regressor in a linear regression. The 
calibration functions where thus of the type: 
X(t2,ch i ) = a + b * X(tl,ch i ) (1) 
Where: 
X = a DN value 0..255 
t2 = an aquisition year 1985-1989 
tl = t2-l 
ch^ = one of the used TM channels 
The influence of different definitions of forest 
masks and calibration algorithms was tested on 
the field surveyed stands before thinning. The 
functions finally used were estimated with robust 
regression, procedure RREG;NORMAL in Minitab 
version 7.2. This procedure is less sensible for 
outliers than least squares regression. 
2.5 Calculation of standwise spectral changes 
The analysis of change in DN values was in this 
first version only made when a field surveyed 
stand had a valid acquisition both the first 
summer before (tl) and the first summer after (t2) 
the reported month of cutting. By use of the 
calibration fuctions (1), was a mean value at (t2) 
predicted from the stand mean value at (tl), for 
each stand and channel. This was done for all 
valid pairs of aquisitions. The number of 
cuttings observed was 41. The number of valid 
change measures on the field surveyd stands before 
cutting was 105. 
The changes in DN is the difference between the 
actually measured mean values at (t2) and the 
predicted values. Since the stands were cut in 
different years, the change values obtained are 
mean values for the 5 pairs of summer acquisitions 
used. 
3. RESULTS 
3.1 DN change as a function of thinning grade 
Table 3 shows a summary of the field inventory and 
the changes in DN. The table is grouped by degree 
of thinning. The differences in DN mean values 
between the three groups of thinning is small. The 
standard deviations fore the thinnings are of the 
same order as standard deviations before cut. The 
measured DN changes as a function of thinning 
grade is illustrated in figure 1. 
Table 3. Summary of field inventory and DN changes, grouped by thinning 
grade. The measured DN differences for all stands before the year of thinning 
are shown in the left column. 
All stands 
unthinned 
Thinnings, grade 
19.0 - 30.0 - 
29.9 % 39.9 % 
40.0 - 
51.2 % 
Seed tree 
stands 
74.1-90.4% 
Number 
of observations: 
121 
15 
14 
7 
5 
Before cutting: 
Basal 
area (m /ha) 
24.0 
22.1 
26.3 
26.5 
18.3 
Spec. 
dist. pine 
63 % 
68 % 
56 % 
47 % 
76 % 
spruce 
25 % 
23 % 
30 % 
25 % 
20 % 
deciduous 
12 % 
9 % 
14 % 
28 % 
4 % 
After cutting* 
Basal 
area (m /ha) 
- 
16.7 
17.0 
14.4 
3.3 
Spec. 
dist. pine 
- 
72 % 
64 % 
59 % 
100 % 
spruce 
- 
23 % 
28 % 
26 % 
0 % 
deciduous 
- 
5 % 
8 % 
15 % 
0 % 
Mean height (m) 
- 
14.9 
15.6 
15.4 
16.1 
Measured-Predicted DN: 
TM 1 
mean 
-0.08 
0.43 
0.50 
0.58 
7.84 
std 
1.30 
0.80 
1.14 
0.98 
3.66 
TM 2 
mean 
0.25 
0.51 
0.35 
0.49 
4.37 
std 
0.82 
0.63 
0.59 
0.60 
1.12 
TM 3 
mean 
-0.03 
1.57 
1.12 
1.45 
14.48 
std 
0.73 
1.26 
0.72 
1.39 
8.86 
TM 4 
mean 
0.46 
-0.92 
-3.09 
-1.81 
5.53 
std 
2.64 
2.38 
3.74 
2.88 
2.28 
TM 5 
mean 
0.34 
4.82 
3.55 
4.38 
29.80 
std 
1.36 
2.45 
2.20 
3.41 
8.18 
TM 7 
mean 
0.01 
2.15 
2.00 
2.55 
13.79 
std 
0.60 
1.17 
2.70 
1.80 
3.72 
852