forest cover types that were in a optimal phenological 
stage (summer) got such a good classification accuracy. 
Lannelongue & Saint (1981) investigated simulated 
SPOT-data, and stated that "the geometry is not repre 
sentative of the SPOT-system". I do hope that the real 
SPOT-data will give us a possibility of improved detec 
tion and mapping of forest cover types, but I don't 
expect the optimal use of the SPOT-satellite system for 
vegetation mapping purposes before the HRV-sensor get 
an additional channel in the MIR-area of 1.5 - 1.7 um. 
This is also the area where the vegetation cover types 
with lichen cover is best detectable and mapped. 
3.5 Visual interpretation and classification 
Visual classification and interpretation was done on 
the imageries. Ratio methods were elaborated using 
channel-ratios of 5:3, 4:3, 5:4:3:2 and 4:3:2:1. In 
addition to this, several channel-combinations were 
interpretated in order to detect very hardly detectab 
le vegetation cover types by digital classification. 
The most successfully results were obtained by using 
the ratios and channel-combinations in the following 
subheadings: 
3.5.1 Ratio 
Landsat 5 TM 1984-06-03 
Ratio 5:3 gave a good delineation of mixed spruce 
seedling stands (01) from the surroundingdeciduous 
forests (figure 3). In addition this ratio gave a good 
delineation between lowland vegetation and alpine vege 
tation (T0mmervik 1985b). This was also the result by 
using the ratio of 5:4:3:2. 
Ratio 4:3 gave a good delineation of pine forests 
(A4a) from other forests. 
SPOT-simulation 1982-06-30 
Ratio 3:2 (NIR-quotient) gave a good delineation of 
clearcuts, roads, in addition of detection of the var 
iation in the farmland. Spruce seedling stands (homo 
geneous) were good delineated from the surrounding 
forests. 
3.5.2 Channel combinations 
Landsat 5 TM 1984-06-03 
The channel combinations of CH 456, CH 532 and CH 
754 gave a good delineation of the mixed spruce seed 
ling stands from the surrounding deciduous forests, 
but not so good as the ratio 5:3. This combination gave 
a good delineation of G7 Birch forests of meadow type 
and E5a/E5c Grey alder forests (rich type) from poorer 
forest cover types. Pine forests were also well detec 
ted. 
Landsat 5 TM 1984-10-02 
The channel combinations of CH 432, CH456 and CH 543 
gave a good result in detection and delineation of C5 
Swamp forests at the river banks and E5a/E5c Grey alder 
forests at the hillsides,due to the litter. The chann 
el combination of CH 432 was a good basis for inter 
pretation of H/ Rich shrub and snowbeds (K2, K6 and 
K6'). The channel combination of CH 456 gave a good 
délinéation of pine forests (A4a), Spruce seedling 
stands (01), Snowbeds (K2 and K6) and Farmland (AA1/ 
AA2) . 
SPOT-simulation 1982-06-30 
Channelcombination of CH 321 gave almost same result 
of interpretation and detection of vegetation cover 
types as Landsat 5 TM, but as a result of the improved 
spatial resolution, texture and variation within the 
vegetation cover types can be seen. Channel combinat 
ion of CH 432 gave a good ability to study texture and 
pattern within the vegetation cover types. This is due 
to the association between the panchromatic channel 
(4) and the multispectral channels 3 and 2. As a res- 
Figure 3. Ratio of channel 5 and 3 (5:3) shows a good 
delineation of mixed spruce seedling stands from the 
surrounding deciduous forests. The area is marked with 
an arrow. 
ult of improved spatial resolution (10 m), clearcuts 
with shrubs were detectable. 
3.5.3 General discussion 
Hame (1984) has stated that the best results in deline 
ation of mixed spruce seedling stands and other very 
hardly detectable vegetation cover types, were obtained 
using parallelepiped classifications with the first two 
principal components calculated of Landsat MSS-imagery. 
Jaakkola (1985) has done a similar study on SPOT-sim 
ulated imagery and stated that the best results of the 
classifications of the imagery, were obtained by using 
multi-point (contextual) classification techniques. I 
have shown by simple ratio-methods and channel combin 
ations elaborated on TM-imagery, that we can delineate 
very heterogeneous forest cover- and vegetation cover 
types easily. 
4. CONCLUSIONS 
Second generation satellites such as Landsat 5 TM and 
SPOT HRV, providing high spatial resolution between 
10 m and 30 m and in the case of Landsat 5 TM, new 
spectral channels, will increase the level and accuracy 
of digital classification. This study has shown that it 
was possible to classify the vegetation cover types 
which were furthest phenological develloped, with an 
overall accuracy of 90 percent or more, in spite of the 
unfortunately fact that the scenes were taken too early 
in the springtime or taken to late in the autumn. In 
addtition to this the digital classification even went 
good in an area with a fairly wide variation in eco 
logical niches and a heterogeneous vegetation. 
Visual interpretation and classification based on 
ratio-techniques and channel combinations was a very 
good tool to improve the interpretation of imagery. The 
best results in this case were obtained by the TM-sen- 
sor of the Landsat 5 satellite. 
The middle infrared (MIR) channels of TM were useful 
for vegetation classification, especially for deline 
ation of heterogeneous forest cover types. 
Comparison between Landsat 5 TM-sensor and the simu 
lated SPOT HRV-sensor has shown that the two sensor- 
systems have almost the same ability to detect and map 
vegetation cover types within the area, due to the 
higher radiometric resolution for the TM-sensor compa 
red to the simulated HRV-sensor. Classification of the 
SPOT-simulated imagery showed that vegetation units