ciduous trees in autumn and winter, whereas the coniferous stand
exhibits a decrease of backscatter in these seasons. As stated
above, the theoretical study of Wegmüller et al. (1994) provides
an explanation of these phenomena. Similar experimental results
have also been found by other authors concerning leaffall influ-
ences to backscatter (Pulliainen et al., 1991, Ahern et al., 1993),
and decrease of the backscatter of conifers relative to deciduous
during the winter period (Ahern et al., 1993).
To summarise the observations for all the vegetated test areas,
Fig. 4 shows a plot of the ERS-1 backscatter range, defined as
the difference between maximum and minimum measured
backscatter of all acquisitions, versus the average ERS-1
backscatter value of all measurements. Generally, different target
groups ie. different combinations of high or low average
backscatter and high or low dynamic range can be distinguished.
Woody plants/forests form a group with high backscatter and a
low dynamic range of backscatter over the seasons. In this group
conifers show the highest dynamic range while young stands
show the lowest average backscatter. Herbaceous vegetation
(grassland) has a low average backscatter and medium dynamic
range. Agriculture has the highest dynamic range of backscatter
due to the influence of cultivation practices. A separation of
woody and herbaceous vegetation and also of different accumu-
lations of biomass seems to be possible using this representation.
-6 T T
+
eT old , mixed forests + 71
zm E m.
al + + + coniferous =
= +
cn
2 "
ER ici t —
5 —9 agriculture
© + +
2 iol young + gt]
x
o
d
S =11 = Hi 3 —
©
2 ep pes grassland m
+ +
—13— ei
-14 J L
2 3 4 5
Range of Backscatter [dB] +
Fig. 4: Average ERS-1 Backscatter Versus Range of ERS-1
Backscatter of All ERS-1 Acquisitions for Different Vegetated
Test Areas.
In order to confirm the conclusions drawn from Fig. 4, two sta-
tistical tests have been performed. The first test consists of the
computation of an index of separability (Dobson et al., 1992).
This test makes the assumption of Gaussian distributed classes.
The second test consists of the computation of the probabilities
of misclassification between classes, assuming that they are
Gamma distributed (Nezry et al., 1993). Results of both tests
showed up to be consistent.
As example table 3 summarises results of the probabilities of
misclassification between the major classes of vegetated areas
used until this stage of the study. It clearly appears that good dis-
crimination of all the classes considered can not be achieved us-
ing only one ERS-1 acquisition. In general, better conditions for
the separability of classes occur between October and May, due
to combined seasonal, phenological, and cultivation practices in-
fluences on ERS-1 backscatter.
50.0 11.8 27.9 à
50.0 10.7 50.0 50.0 133
50.0 14.0 36.7 50.0 112
50.0 21.4 40.0 50.0 17.6
50.0 224 50.0 50.0 214
30.1 354 50.0 50.0 314
30.7 20.3 1.3 36.5 21.9
31.5 21.3 5.9 50.0 254
26.8 19.4 17.7 40.1 28.7
27.8 18.2 13:3 26.7 230
24.6 14.9 7.5 34.2 17.6
232 13.3 11.5 27.4. 12.1
26.2 10.4 6.4 34.6 18.3
27.2 16.4 7.1 30.9 26.1
Table 3: Results (in Percentage) of the Test of Probability of
Misclassification for Pairs of Major Land Use Classes (50.0
Means No Separability).
6. RETRIEVAL OF FOREST STAND ATTRIBUTES
WITH ERS-1 AND SPOT/XS DATA
After multisensor/multitemporal data fusion, , the possibility of
retrieval of forest stands attributes was investigated using GIS
information and combined SPOT and ERS-1 data. Since, accord-
ing to the German forest taxation practice, biomass is not gen-
erally estimated for young forest stands, age information pro-
vided for all forest stands by the GIS was considered for this
study. Stand age is known to be, in general, correlated to impor-
tant physical forest stand attributes.
180
170
160
150
140
130
120
110
100
90
80
70
60
50
40
30
20
xs1 xs2 xs3 211 229 232 235 262 271 292 304 310 319 334 340 123 38
SPOT Bands Day of Year for ERS=1 Acquisitions
DN
Fig. 5: SPOT Reflectivity and Temporal Sequence of ERS-1
Backscatter Averaged for Forest Age Classes 1-10, 11-20, 21-
30, 81-90, 91-100, 101-110 Years (Bottom to Top). For Better
Comparison SPOT and ERS-1 Signatures are Represented in
Digital Numbers.
The response of forest age classes to SPOT spectral bands and to
the ERS-1 time series are depicted in Fig. 5. SPOT XS1 and X
bands show no apparent sensitivity to age of forest stands. The
SPOT/XS3 band shows decreasing reflectance for increasing
forest age. On the contrary, ERS-1 radar reflectivity increases
with forest age, until saturation is reached after about 30-40
years. A very interesting fact, and also potentially useful for
applications, is that this behaviour remains stable. Seasonal and
334
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996
