2. MATERIALS AND METHODS 
2.1 Ground data collection 
The studied area is 40 km by 30 km. It is 
situated close to Umea in northern Sweden, with 
the center at 64°05' N and 20°30'E. 
Representatives for all owners of forest land in 
the area were contacted and asked for thinned 
forest stands and the dates for thinning. All 
known stands that were thinned once between 
august 1984 and july 1989 and were larger than 2.5 
ha and more than 100 m wide were selected. After 
excluding 2 extremely heterogeneous stands did 44 
thinnings remain. The field survey of the stands 
were carried out in October 1989. 
The tree species were Scots Pine (Pinus 
Silvestris), Norway Spruce (Picea Abies) and 
Birch (Betula spp.). Only 1 % of the stems were 
other deciduous species than Birch. 
In each stand nominally 10 sample plots were 
placed in a regular grid. The distance between the 
field plots were 40-80 m depending on stand size. 
A 50 m grid was chosen in the majority of the 
stands. No plot was placed less than 25 m from 
the stand border. 
Stem and stump diameters were measured in 
circular sample plots of 154 m 2 . About 20 more 
variables were also recorded. In 20 % of the 
stands, surveyed in the beginning of the study, 
the stump diameters were measured for 128 m 2 
plots. The basal area in these stands were 
measured by relascope and a smaller number of 
field variables were recorded. 
The stump diameters were transformed to diameter 
at breast height by regression functions from the 
NFI. The basal area is defined as total stem 
cross-section area at breast hight in m 2 per 
hectare (ha). The basal area for the remaining 
trees at the year of thinning were calculated 
from the measured basal area 1989, using functions 
from the NFI. Thus, all basal areas refer to the 
year when the individual stand were thinned. The 
thinning grade is then defined as basal area cut 
divided with total basal area before thinning. 
To illustrate the spectral changes for stronger 
cuttings, five seed tree stands were also 
included in the analysis. These stands were 
subjectively chosen and field surveyed with the 
same methods as the thinnings. 
2.2 Satellite data used 
The Swedish Space Corporation (SSC) provided the 
project with digital Landsat TM data from the 
summers 1984 - 1989, (table 1). SSC-Satellitbild 
did also make the geometrical precision correction 
to the Swedish national grid, with 20 m pixel 
size. 
Table 1. Satellite acquisitions used. 
Satellite Sensor Reg. date Scene 
Landsat 
5 
TM 
1 
aug. 
1984 
194/15 
Landsat 
5 
TM 
26 
june 
1985 
193/15 
Landsat 
5 
TM 
6 
june 
1986 
192/15 
Landsat 
5 
TM 
12 
aug. 
1987 
192/15 
Landsat 
5 
TM 
27 
june 
1988 
192/15 
Landsat 
5 
TM 
21 
june 
1989 
193/15 
The satellite scenes were radiometrically 
calibrated, using the calibration lamp onboard the 
satellite. The digital numbers (DN) measured in 
the satellite scene might be converted to physical 
units for spectral radiance using the following 
formula: 
Lmax - Lmin 
Spectral radiance = Lmin + DN * 
255 
The values of Lmin and Lmax used in the pre 
processing of this data set are found in table 2. 
Table 2. Calibration values for converting DN 
values to physical units for spectral radiance, 
expressed in W/ (n? *sr*pm) . 
channel 
colour spectral range(um) 
Lmin 
Lmax 
TM 1 
blue 
0.45 - 0.52 
-1.50 
152.10 
TM 2 
green 
0.52 - 0.60 
-2.80 
296.80 
TM 3 
red 
0.63 - 0.69 
-1.20 
204.30 
TM 4 
near-IR 
0.76 - 0.90 
-1.50 
206.20 
TM 5 
middle-IR 
1.55 - 1.75 
-0.37 
27.19 
TM 7 
middle-IR 
2.08 - 2.35 
-0.15 
14.38 
Thus, a data set of 36 channels were available. 
In the analysis phase, the TM data were handled as 
12 blocks of 500 by 500 pixels, corresponding to 
10 km by 10 km. 
For each block the data from one year was adjusted 
to 1:10 000 maps. The remaining scenes were 
manually adjusted to the reference scene with best 
integer pixel precision. For best fit compared to 
1:10 000 maps, the data on an average had to be 
moved 22 m pooled distance, with a standard 
deviation of 22 m. One of the reasons for the 
remaining geometrical errors is obviously height 
parallaxes due to scenes from different paths, but 
also the 16 lines sensor swaths were sometimes 
displaced. 
2.3 Extraction of standwise DN mean values 
Map coordinates were assigned to each field 
surveyed plot. For each plot the DN values for the 
pixel containing plot center, and the eight 
neighbour pixels, were extracted in all the 36 TM 
channels. Pixels already assigned to a plot were 
not used again. 
The location of the plots were now for the first 
time checked together with the satellite images. 
If a stand for a given acquisition was covered by 
clouds, cloud shadows or technical errors, no 
data were extracted for that year. To reduce the 
problem with mixel values at stand borders, all 
boarder pixels that in any channel were above the 
99.9 % level in a t-test were excluded for all 
years and channels. This procedure eliminated 
5.6 % of the stand pixels. On an average 60 
pixels per stand remained, which corresponds to 
2.4 ha. 
2.4 Satellite data calibration functions 
For the between-scene calibration functions, an 
independent set of pixel values was systematically 
sampled from the studied area. The values were 
sampled under a forest mask obtained from 
thresholding in TM 5, 1989. The mask covered