40°-50° incidence angles (Ulaby and Dobson, 1989). The value 
of S was 6.6 dB. This agrees with Freeman and Durben (1996) 
data, who presented a difference between VV and HV equal to 
6.6 dB. The authors worked with the following classes: Forest, 
Swamp forest, Flooded forest, Regrowth, Open water, etc., also 
in C band with incidence angles ranging between 40°-50°. 
3.3 Biophysical Indices 
Three biophysical indices proposed by Pope et al. (1994) were 
applied to data in order to improve the interpretation of the 
vegetation physiognomy. This indices are meaningful only 
when taken in the context of the type of interaction between 
microwave radiation and canopy. The indices used are: 
VV 
- CSI canopy structure index = ——————— 
VV + HH 
- tt = DEAN 
volume scattering index Silk 
- BMI biomass index = LK 
where: 
HV + VH VV+FHH 
-CS = yim “EK = nl 
CSI is a measure of the relative importance of vertical versus 
horizontal structure of the vegetation. Ecosystems dominated 
by nearly vertical trunks or stems will have higher CSI values 
than will ecosystems dominated by horizontal or near- 
horizontal branches. It also should be noted that ecosystems 
with a high percentage of double bounce will have lower CSI 
values than ecosystems with similar structure but little double 
bounce interaction. 
VSI is a measure of the depolarisation of the linearly polarized 
incident radar signal. High values of VSI result when the cross- 
polarized backscattering (CS) is larger than the like-polarized 
average (LK). Therefore, VSI is an indicator of canopy 
thickness or density. 
BMI is an indicator of above-ground biomass. A critical aspect 
of BMI is the relationship between the radar wavelength and the 
size of vegetation components, which determine whether or not 
a given component acts as a scatterer or absorber. An increase 
in total, above-ground biomass can cause either an increase or a 
decrease in BMI, depending upon how this biomass is 
distributed. 
The indices were calculated for all classes, excepting for Open 
water, using the digital number values. 
4. RESULTS AND DISCUSSION 
Table 3 presents the number of sample (7), the number of 
pixels, the mean o? and the standard deviation, for all studied 
classes. The Table 4 shows the classes dynamic range. 
Low dynamic range values denote a smaller c? variation inside 
of the class, and a higher homogeneity between the samples. 
This allows to a better class characterisation (Ahern et al., 
1993). Lower dynamic range values were determined for forest 
in all polarization. In the average the HH polarization presented 
lower dynamic range values, denoting better classes 
discrimination possibility. 
The variation of the mean c? values are presented in the Figure 
3. The large o? variation inside the classes are due to radar 
parameters, like speckle, and variations in the canopy 
morphologic and structural characteristics. 
The aquatic vegetation stands may present variation in the 
density and homogeneity. This fact may affect the Spacing 
between the leaves and stalks, and consequently the 
backscattering. Differences in age and height may contribute 
for the variation in the dynamic range too. 
In Figure 3 it is possible to evaluate the polarization behaviour 
inside the classes. Forest and Eichhornia sp. presents VV 
values bigger than HH. This may be due to the vertical 
orientation of the branches and stalks, respectively. Ulaby and 
Dobson (1989) also founded bigger o? VV values for trees and 
grasses. 
Figure 4 shows the dynamic range in the four polarizations, for 
each class. Figure 4 indicates small possibility of discrimination 
among the classes in the four polarizations. 
Figure 5a indicates the possibility of discrimination among the 
classes using de mean o? values. In VV polarization Forest, 
Scirpus sp, and Eichhornia sp. can not be discriminated. The 
HH polarization seem to present the best discrimination 
capability for all classes. It is important to point out that only 
the mean value is been considered, and that the mean c? values 
and the coefficient of variation were derived from large 
samples. This fact bring the mean near to the population mean. 
The coefficients of variation of the classes for all polarizations 
are showed in the Figure 5b. Its analysis show a different 
behaviour when compared with mean o? values. There are a 
smaller variation between the polarization inside the classes, 
and smaller discrimination possibility between the classes. 
The plot of the four polarizations with the respective coefficient 
of variation are presented in Figure 6. It's possible to observe 
well defined groups, mainly in HH polarization. In the other 
polarizations the classes Scirpus sp., Eichhornia sp. and Forest 
make the groups less evident. 
Table 5 presents the biophysical indice values for all classes. 
Although of the low variation between the indices values it is 
possible to observe large CSI values for Forest and Eichhornia 
sp.. This confirms the larger influence of the vertical structures 
in the larger VV backscattering. The Forest presents larger VSI 
values indicating a larger canopy wave penetration. Probably 
this fact is due to the small wavelength (5.6 cm) in relation to 
the forest scatterer size distribution. Scirpus sp. have the bigger 
BMI values, indicating larger biomass. Although of the high 
forest biomass it is not showed by the BMI index, probably, 
because the C band wavelength interact only with the upper 
forest canopy, not been influenced by the trunk biomass. 
530 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996