Ruecker, Gernot 
  
units using an algorithm provided by ESA (ESA, 1998). For some images calibration had to be corrected for 
replica power. 
For the backscatter study, the images from the three different dates were combined to form a three channel 
image with floating point datatype. 
For visual change detection, each of the two images from April and July (post-fire) were subject to principal 
component analysis with the pre-fire image from August 1997. In a two-band image, principal component two 
holds the change information, while principal component one holds information on common properties of the 
two images. Principal component one was then discarded, while principal component two was combined with 
the speckle-filtered datasets from August and April and August and July, respectively (Siegert and Ruecker 
1999). The images were assigned to the colour guns as follows: Red: PC band 2, Green: image one (August), 
Blue: image two (April or July, respectively). The images were contrast-stretched and converted into a 8-bit 
RGB-image. These two images were then georeferenced to UTM co-ordinate system, Zone 50 N, WGS-84 
ellipsoid. Due to lack of ground control points and reliable maps this had to be done using the orbital information 
from the ERS-satellite. In the central area of the mosaic (Mahakam basin and surroundings) GPS-measurements 
were available for assessing registration accuracy. From this data it is estimated that accuracy of registration is in 
the order of four pixel sizes, i.e. 100 m. 
To enhance changes in image texture induced by burning, another series of image products was generated 
following a method developed by Siegert and Kuntz (1996) for radar-aided classification of land-cover. This 
method is based on occurrence-based textural filtering of the raw image with very large kernels (15 x 15 and 25 
x 25) in order to enhance image texture. The results are assigned to the red and blue colours while the green 
channel is the adaptively filtered radar image is assigned to the green channel. This allows for identification of 
areas deprived of their vegetation cover when comparing two images acquired on different dates due to unveiling 
of the underlying relief (Siegert et al. 1999). 
This image product served as an additional reference for burn scar assessment, while the PCA image was the 
base image for mapping. Images were loaded into a GIS and burn scars were mapped by visual interpretation to 
comply with a final map scale of 1:200,000. 
2.2 Damage classification 
A classification scheme to map fire damage in timber concessions was established by the Sustainable Forest 
Management Project (SFMP), a partner organisation. This scheme was adapted for the SAR analysis. Depending 
on the type of vegetation and fire impact it was possible to discriminate 4 different damage classes (Figure 1). In 
class 1 25-50% and in class 2 50-80 % of the vegetation have been killed by the fire. This type of damage was 
found predominantly in logged over Dipterocarp forests in which ground fires of variable intensity killed tall 
trees but left most of the biomass unburned (Figure 1 A and B). Class 4 indicates that more than 80 % of the 
vegetation cover has been consumed by the fire. This class was typical for strongly degraded forests and 
Imperata cyclindrica (alang-alang) grasslands. For ecological and fire prevention reasons we introduced a fourth 
class (class 4) which could not be discriminated using backscatter values alone. In peat swamp forests almost 
100% of the trees have been killed by the fire but most of the above ground biomass was left unburned thus 
causing a high fire risk in future (Figure 1 C). 
  
Figure 1. Fire damage classes. Fig. 1 A corresponds to damage class 1 less than 50% of the trees have been 
killed. Fig 1 B to damage class 4 with more than 80% of the tress dead and the lower left part of Fig 1 C to 
damage class 3, with more than 80% dead, but the canopy structure almost remaining intact, while the upper 
fight pewsaf the image shows undamaged swamp forest. 
  
1288 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000.