Ruecker, Gernot
2.3 Backscatter study
Effects of fire on radar backscatter were investigated in a total of 56 test areas representing the classes of fire
damage introduced in section 2.2. Some of these areas were visited on ground surveys during and after the fire
event, while others were flown over in the aftermath of the fires by low flying air plane. Fire damage was
assessed visually from air photographs and videotapes taken during the flight survey.
Test areas were delineated in the geographic information system and then transferred to the image processing
software. Backscatter statistics for these test areas were calculated and evaluated.
2.4 Accuracy assessment
Validation of the ERS-SAR PCA burned scar map and damage assessment was done during four ground surveys
conducted between April 1998 and September 1999 covering more than 4500 km in the fire affected area (see
Figure 4 B). Locations in 10 timber concessions and 4 plantations were checked on ground and compared to the
ERS derived burned scar and damage map. Two methods were used to assess degree of accuracy of burn scar
and fire damage mapping:
- 143 Random samples of geocoded photographs and videotapes recorded during air surveys were assigned a
damage class by visual inspection and compared to the corresponding classification on the map.
- A block inventory of ground data by SFMP in some 178,000 ha of a forest concession was available for
interpretation in the GIS. Each block was assigned one of the damage classes explained in section 2.2 during
the inventory. This block inventory map was intersected with the damage map compiled from the radar
image product and each inventory block was paired with the corresponding area on the radar burn-scar map
according to the mapped class which covered most of it.
Mapping results and classes from the two reference datasets were than cross-tabulated and classification
accuracy was calculated for each class as well as for the overall discrimination of burned and unburned areas.
3 RESULTS
3.1 Backscatter Analysis
Changes in Radar backscatter induced by fire proved to be strong. In images acquired in April, during the
drought period, a decrease of 2-4 dB was observed in burned areas (figure 2 A). Under moist weather conditions
(in July, figure 2 B), radar reflectivity was slightly higher in areas severely affected by fire than in unburned
areas. In turn, the standard deviation of backscatter increases for all damage classes except for class 2 (> 80%
burned) from August 1997 to July 1998, while in the April 1998 images, standard deviation is slightly lower for
burned test areas than in August 1997, although this difference is not supposed to be significant.
-4 0.16
0.12
a
— =
a 3 0.08
9 G
©
£
= 004.
à 0 -
unburned 25-5096 50-8096 >80% >80% unburned 25-50% 50-80% >80% >80%
damage damage damage, damage, damage damage damage, damage,
standing burned standing burned
Figure 2 A. left: Mean backscatter for test areas in August 97 and April 98. Right: Mean of standard deviation of
backscatter in test areas in August 97 and April 98.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 1289