'est 
Is 3 
Nas 
the 
-C 
the 
on 
ind 
es) 
dos Santos, Joäo Roberto 
  
according to the physiognomy of the savanna; all individuals of bush and/or arboreal features were identified 
botanically. As for the herbaceous strata, sampling to collect biomass and data on the percentage of soil coverage were 
also performed in at least 5 sections with 1m? size for each sample area. In those areas covered by forest formation (6 
samples), including the transition areas, the following measurements were made: DBH, height, percentage of crown 
cover as well as the botanical identification of individuals with DBH > 5 cm for areas of secondary succession and 
DBH > 10 cm for areas of primary forest, at sample areas of 1,000 m^ and 2,500 m? respectively. The estimation of 
biomass values was modeled by dendrometric parameters into allometric equations. At the biomass inventory a larger 
amount of samples was effectively obtained during the field survey, totaling 37 samples. The precise location of each 
field plot in the respective radar database was identified using the GPS system. 
During the SAR data processing, a Gamma Filter (window size 5x5) was applied to the JERS-1 image, and a MAP 
Filter (window size 3x3) was applied to the SIR-C image, to reduce the speckle noise of the image. The georeferencing 
of both SIR-C and JERS-1 images (including also the TM/Landsat image) was based on a bilinear method, with pixel 
resampling to a spatial resolution of 25 m. It was considered adequate to extract the digital number (DN) values of each 
radar image from previously selected sample areas inventoried in the field survey. The DN values were converted to dB 
values by appropriated equations (Rosenqvist, 1997; Kramer, 1999) considering the offset calibration factor of each 
sensor. The backscatter coefficient of the vegetation cover is a function of sensor parameters (wavelength, incidence 
angle and polarization) as well as of the geometric and dielectric characteristics of the vegetation type, according to the 
seasonal effects acting over this area. 
The empirical relation between backscatter (y) and biomass (x) values is based on the statistical analysis of their 
adequacy to a regression model, where these variables were adjusted. Diagrams showing the physiognomic-structural 
details of the vegetation types and the spatial distribution of the backscatter at primary forest , tall woodland, savanna 
woodland, tree and/or shrub savanna were made. After the definition of the appropriate equation for each sensor type, 
images were sliced in dB values found for each pixel, and the result was the definition of biomass intervals and its’ 
spatial distribution. 
4 RESULTS 
In a brief description, the forest typology section of Comodoro is included in the semi-deciduous category, where the 
climatic conditions are conceptually defined as seasonal (4 to 6 dry months or 3 months below 15°C), with 
xerophytic/hydrophytic physiology, including alic and distrophic soils (FIBGE, 1992). Within the common arboreal 
species identified, there are those of families Lauraceae, Burseraceae, Chrysobalanaceae, Euphorbiaceae, Anonaceae, 
Guttiferae and Sapotaceae, where one could find 570 individuals per hectare (DBH > 10 cm), with an average height of 
16 m. The estimated forest biomass of the sections inventoried is of 205.81 + 54.4 ton/ha (dry weight). As for the 
savanna formation, there is a variability on the content of estimated biomass, being generally 58.11 ton/ha (dry weight) 
for tall woodland (a forest-like physiognomy of savanna), 12.29 ton/ha for savanna woodland (cerrado strictu sensu) 
and 7.13 ton/ha for tree and/or shrub savanna (facies where grass stratum predominates, with sparse arboreal and/or 
bush strata). It is important to point out that the acquisition period of images, as well as of the time of field data 
collection should be concomitant, due to the strong effect of changes during the yearly seasons of this type of savanna, 
specially those where the grass stratum dominates. 
Based on the ground truth database related to the typical structure of forest and savanna formations, it is possible to 
show the empirical behavior of the backscatter derived from both sensors against biomass values (Figure 3). Using the 
logarithm regression model, the analysis shows that Lyy band of SIR-C data presents a little higher determination 
coefficient (r^ — 0.7406) when compared to JERS-1 data (r^ — 0.6791). Considering the Lay band, the SIR-C data (y 
3,3731 Ln (x) - 29,59 , with ’=0.8275) is already more significant, according to a separated analysis which was also 
performed. It is well known that cross polarization allows a better comprehension on the signal interaction with the 
complex structure of trunks, branches, twigs of distinct forest features. 
Taking into account an isolated analysis of JERS-1 data for the savanna, one can perceive that the largest backscatter 
values represent the class “tall woodland” (-6.9 dB), followed by -9.1 dB for “savanna woodland” and -10.5 dB for a 
“tree and/or shrub savanna”. All these savanna physiognomies occupy the space of attributes mentioned above, with 
values of + 0.5 dB around these average values. In the specific case of SIR-C (Lyp) the backscatter values are -9.1 dB, - 
12.0 dB, and -15.8 dB for the same topologic sequence found, presenting only a variability of - 1.0 dB around the 
average values. This can demonstrate a higher sensitivity of the SIR-C, compared to JERS-1 to discriminate small 
variations of inherent characteristics of each one of these classes. The same can be affirmed for the class “semi- 
deciduous tropical forest”, with average backscatter values of -7.2 dB and -8.1 dB for JERS-1 and SIR-C (Lun band) 
respectively. 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 379