Skógræktarritið - 15.05.2001, Page 155

Fig. 3. The forest on the north side of
Lake Tornetrask.
about 650 m above sea level, and
the mountain birch forest had
several times, since the 1950"s
suffered from insect outbreaks
that had killed many trees
(Bylund 1995).
Ground data
The test site (Figure 1A) was
established during the summer
of 1997, and consisted of 869
permanently marked circular
sample plots with a radius of 10
m, grouped into clusters of 9 or
15 in a regular grid with 30 to 500
m between plots (Figure 1B).
Distance between clusters was
500 m to 1 km. All plots were
located using GPS. On each plot
all trees were measured for diam-
eter above breast height (DBH),
coverage of shrubs estimated in
part of 10 per shrub species, and
soil type registered. Slope,
aspect and elevation were
extracted for each plot, from a
digital elevation model (DEM).
Biomass of trees and shrubs
were calculated for all plots
using allometric equations based
on the variables that were regis-
tered for each plot. These equa-
tions had previously been esti-
mated by regression analysis of
field-measured biomass of
mountain birch trees and shrubs
Fig. 4. Estimated
wood biomass
when the func-
tion was applied
to the satellite
image.
taken from the same study area
(Dahlberg et al. 2001). Since we
were only looking at mountain
birch forest in this preliminary
study, all plots without trees
were excluded.
Satellite data
The satellite data was one cloud
free IRS LISS 111 scene from
1 September 1998 (Table 1).
Table I. Characteristics of iRS LISS-III
Band Wavelength (pm)
1 Green 0.52-0.59
2 Red 0.62-0.68
3 Near IR 0.77-0.86
4 Mid IR 1.55-1.70
Although the image was geomet-
rically corrected using a 500 m
resolution DEM by the Swedish
remote sensing company
Observation, Mapping and
Monitoring (OMS-M), we found
residual distortions in the
image, which produced uncer-
tainties in the location of the
plots. This is a typical problem
with satellite images in moun-
tainous areas (Itten and Meyer
1993) that should be taken into
account when analysing the
results. To reduce the distortions
as far as possible within our test
site, the scene was further geo-
metrically corrected with ground
control points from aerial
orthophotos and resampled
using the cubic convolution
method (weighted average of 16
surrounding pixels to approxi-
mate the value of one pixel)
(Lillesand and Kiefer 2000). The
SKÓGRÆKTARRITIÐ 2001 l.tbl
153