Skógræktarritið - 15.05.2001, Page 156

pixel values for each field plot
were extracted using cubic con-
volution method.
Regression
Multiplicative linear models were
estimated for leaf - and wood
biomass, where the pixel values
from all bands (Table 1) of the
image, as well as sin(slope),
sin(aspect), their interaction
term and elevation were tested
as explanatory variables. This
model can be represented as:
a b. b?
y - e -xx • x2
This was transformed using a
logarithm:
n
ln(y) = a + y^bt ln(jc,.) + ln(£)
i=l
where:
ln(y) becomes the response vari-
able,
a and b{ are parameters to be
estimated,
%j are the explanatory variables,
and
ln(e) is normally independently
distributed error term with a zero
mean and an unknown constant
variance (ln(e) ~NID(0, a2)). lt is
assumed that the explanatory
variables are not subject to ran-
dom variation.
Results
Only bands 3 and 4 of the LISS III
image were found to be signifi-
cant as explanatory variables for
biomass (Table 2). We did not
find the variables sin(slope),
sin(aspect), the interaction term
or elevation significant.
When the resulting functions
were applied to the IRS LISS III
image, we got biomass values
on the lake, and quite high bio-
mass values on snow beds as
the regression function simply
extrapolates the function to all
the pixel values. Figure 4 shows
the result when the function for
wood biomass is applied to
band 3 and 4 of the IRS LISS III
image, after snow and water
have been masked out. The tree
line appears clearly on the
image, which also shows higher
biomass at south facing slopes.
This is consistent with the
observations made in those
areas. The estimated mean
value of wood biomass within
the whole area of the test site
was 9100 kg/ha, and the mean
leaf biomass was 800 kg/ha.
Discussion
The standard statistical tests
(ANOVAand adjusted R2) indi-
cate that there could be a rela-
tionship between biomass and
IRS LISS data in the study area.
The adjusted R2 is rather low
(0.30 and 0.21) which could be a
result of many variations related
to geometric and radiometric
distortions in the image and also
because of noise. The only
explanatory variables that were
significant for predicting biomass
were bands 3 and 4 of the IRS
LISS III data. These bands were
expected to predict biomass
since vegetation highly reflects
the energy in the near infrared
(band 3) and mid infrared (band
4) bands while energy in the visi-
ble region (bands 1 and 2) are
much less reflected (Lillesand
and Kiefer 2000). In mountainous
areas there are often differences
in solar irradiance on adjacent
slopes up to 900 W/m2 (Parlow
1996). This means that the same
vegetation at different slope
angles can have significantly dif-
ferent reflectance. The interac-
tion term should have been sig-
nificant but they were not. There
are two possible explanations for
this. The range of slopes avail-
able for the test area for trees
was not sufficiently large to pro-
duce a significant result. Further-
more, topography can also have
a direct influence on the amount
of biomass, which could be large
enough to make the interaction
term not significant. These will
be investigated further in the
future.
The biomass map produced
when the regression functions
were applied to the IRS LISS III
image is very similar to the truth.
The biomass on the north side of
the lake is high as the forest
there is rich and trees are much
larger. There are also more
undergrowth and shrub (Figure
3), which made the total biomass
higher. Furthermore, the mean
values of wood and leaf biomass
(9100 kg/ha, 800 kg/ha) for the
test site are comparable with the
mean value that was calculated
from the 869 test plots (8000
kg/ha, 700 kg/ha) (Dahlberg et al.,
in prep), that are representative
for the area.
This study has demonstrated
the information content of IRS
LISS III data for estimating
biomass at landscape level.
Although most of the known
distortions of the images were
not fully corrected, a good fit
for regression models was
obtained.
Table 2. Estimated parameters for regression functions for wood (BiOMWood) and
leaf biomass (BIOMLeaf).
Response R2 MSE Constant lnIRS3 lnIRS4
ln(BIOMWood) 0.300 0.522 11.192 3.3857 -3.7947
ln(BIOMLeaf) 0.214 0.636 2.760 2.7849 -1.7587
154
SKÓGRÆKTARRITIÐ 2001 l.tbl