When comparing the results of a multitemporal (three
dates) per field classification based on ERS-1 data with
the results for JERS-1 data, varying results were obtai-
ned (cf. table 2). Some crops were classified more
accurately using JERS-1 data, other crops were classi-
fied worse in comparison to the use of ERS-1 data. For
instance, grassland and potatoes were classified most
accurately using ERS-1 data. On the contrary, sugar
beet, maize and barley, for instance, were classified
more accurately using JERS-1 data. The difference in
overall accuracy between JERS-1 and ERS-1 was only
minor. Due to the restricted data set the overall accu-
racies are low in relation to the results with an optimal
data set (about 80% overall classification accuracy).
5. FOREST CLASSIFICATION
5.1 Introduction
For the classification experiment a selection of species
from the two forest test sites was made. All species
and species varieties listed in table 3 occur at least 5
times, excluding stands with extreme values, i.e.
stands for which the stand averaged o° deviated from
the species averaged 0° more then 2 times the stand-
ard deviation of the stand averaged values.
Table 3. Selected classes for classification experiment.
‘Dorschkamp”
'Flevo'
'Robusta'
Populus 'Zeeland'
Populus 'Oxford'
Acer pseudoplatanus
Fraxinus excelsior
Quercus robur
Pinus sylvestris
Populus
Populus
Populus
OO o 100100 Nb =
5.2 Classification Method
To assess classification possibilities, temporal signatu-
res were visualized to get an impression of class sepa-
rability. Subsequently, a maximum likelihood classifica-
tion was made using Bayes criterion. Both approaches
were applied on the ERS-1 and the JERS-1 data sets
separately.
5.3 Classification Results
The temporal signatures of the selected classes are
shown in figure 3 for the ERS-1 images. In February all
poplar classes are clearly differentiated from the other
tree species. This may be a result of the relatively
large trunk size, the relatively high soil moisture con-
tent at that date and the absence of leaves and under-
growth. In March the soil is drier and the contrasts are
lower. In April leaves start to develop and contrasts
start to change. Especially in May the contrast be-
tween classes is large and three groups of classes can
be differentiated. In August the contrasts decrease
again. It appears to be impossible to properly differen-
tiate coniferous (Pinus sylvestris) from the deciduous
species.
A maximum likelihood classification using four ERS-1
images resulted into an overall classification accuracy
of 6596.
: Sigma 0 (dB) | ———— pop dorschkamp
| ——Q-—— pop flevo
— + pop robusta
——Óo——— pop zeeiand
| ———— pop oxford
| ——*—— fraxinus excelsior
|| ——O—— quercus robur
|
930810 | ——X—— pinus sylvestris
Acquisition dates |
930523
930427
930216 930323
Figure 3. Class averaged backscatter as a function of
time for ERS-1.
4;
Sigma 0 (dB)
——À——- pop dorschkamp
—{— pop fievo
"op robusta
——o0— — pop zeeland
De
"s NORMEN US
S.
—— —4—— pop oxford
— —t——- Qcer pseudoplatanus
4 - E c rM Lee fraxinus excelsior
L10 = E ere Quercus robur
930212 930328 930512 930807 —— X-— pinus sylvestris
Acquisition dates
150
Figure 4. Class averaged backscatter as a function of
time for JERS-1.
The temporal signatures for JERS-1 are shown in
figure 4. Though the wavelength is longer the back-
scatter level is comparable or even slightly higher. This
behaviour may be caused by the fact that the longer
wavelength radiation penetrates deeper into the cano-
py and is reflected by larger canopy components, i.e.
by large branches and through the trunk-ground inter-
action, rather than by leaves and twigs as may be the
case for the C-band. In the winter observations Popu-
lus 'Flevo' and Populus 'Dorschkamp' show a relatively
low backscatter level. These may be caused by the
low trunk density of these stands and/or the non-verti-
cal orientation of the trunks. The latter is caused by
dominant wind and theoretically results in a lower
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996
| ——— acer pseudoplatanus
