Ruecker, Gernot 
  
  
  
  
  
  
    
  
  
  
  
  
-4 0.16 
-6 J 0.12 
e 
u— [- 
m -8 
©, 8 = 0.08 
© 
£ 
=10 - 0.04 - 
-12 - 0 - i: 
unburned 25-50% 50-80% >80% >80% unburned 25-50% 50-80% >80% >80% 
damage damage damage, damage, damage damage damage, damage, 
standing — burned standing burned 
Figure 2 B. left: Mean radar backscatter in August 97 and July 98. Right: Mean of standard deviation of 
backscatter in test areas in August 97 and July 98 
Seperation between classes by mean backscatter is clear in the comparison between August 1997 and April 1998 
images only between the classes 1 (25-50%), 2 (50 —80% damage) and 4 (>80% damage with vegetation 
burned), while the class 3 ( >80% damage with unburned biomass) cannot be separated from its neighbouring 
class 2. Standard deviation of backscatter in this class is 
lower than in all other classes before and after the fire, 
making a discrimination of this class possible (figure 2 
A). 
Mean backscatter for images from August 1997 and July 
1998 is almost equal for all test areas. In all fire damage 
classes it is slightly higher in July than in August, except 
for areas with more than 80% damage and most trees still 
standing, where backscatter is slightly lower. A better 
discrimination of burned from unburned classes is 
achieved with standard deviation: this texture-related 
parameter increased threefold between August 1997 and 
July 1998 in all classes except for the class 3 (figure 2 B). 
This indicates that for the other classes, differentiation of 
burned from unburned surfaces is good, but separation 
between classes may be worse. 
The increase in standard deviation, which can be 
interpreted as an increase in image texture is illustrated in 
Figure 4: panel A shows the texture-filtered image mosaic 
of Kalimantan Timur as described in 2.1, generated from 
the ERS-2 SAR image from August 1997 (before the 
fires), while panel B is the same image product from July 
1998 (after the fires). While in the central basin (the 
Mahakam area) green image tones predominate in both 
images, the hilly areas at the central coast as well as north 
of the Mahakam basin appear in green tones in the August 
1997 image and in pink tones in the July 1998 images. 
These pink tones are due to high values in image variance 
enhanced by textural filtering. The change in image 
variance indicates that these areas have been burned. 
  
Figure 4. Artificial image composed from 
texture-filtered and gamma-filtered images from 
August 1997 (A) and July 1998 (B). For details 
In lowland areas more difficulties are therefore to be 
see text. 
expected for assessing burned and unburned areas as 
well as degrees of damage with images from the moist 
season than in hilly regions. However, most of the lowland areas are covered by the image from April 1998, 
which according to the presented backscatter analysis is expected to provide for better discrimination between 
damage classes as well as between fire-affected and unburned areas. 
  
  
1290 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000.