The 
the 
"fic 
the 
rial 
ned 
ger 
idy. 
rial 
and 
r, it 
tion 
lues 
| al, 
nits 
ts a 
nré, 
the 
for 
the 
rain 
her 
e is 
the 
use 
ber 
Barbosa, Paulo 
  
Only the area of the ARVI difference image corresponding to the burned areas was used for the computation of the 
mean and standard deviation in order to exclude other variability factors outside of the burned areas. The ARVI 
difference images (e.g.: 91-92) were density sliced to put in evidence all the negative/positive values that were located 
on the extremes of the distribution. 
The validation of the mobilisation maps was done using the 1995 ortho-rectified aerial photographs on a number of test 
areas. Two types of test areas were chosen: areas that suffered terrain mobilisation and areas that didn’t suffer terrain 
mobilisation. Since the ortho-photos were from 1995, and the period under analysis spanned from 1991 to 1995, it was 
necessary to use them together with the support of the RGB Landsat images, in order to define the test areas. 
Apart from the production of terrain mobilization maps, a number of test areas was retrieved using the 1995 ortho- 
rectified aerial photographs to follow up the land cover change over forest burned areas and its spectral evolution with 
time. 
5 RESULTS AND DISCUSSION 
5.1 Production of terrain mobilization maps 
The initial output data was a series of ARVI difference images: 1991-1992, 1991-1993, 1992-1993, 1992-1994, and 
1993-1994, 1993-1995. Only the area of the ARVI difference images corresponding to the burned areas was used for 
the computation of the mean and standard deviation, since the areas that had been previously masked had zero value 
and were excluded from the computation. The mean and standard deviation of the masked ARVI difference images is 
presented in Table 1. 
Table 1. Mean (ju) and standard deviation (c) of the ARVI difference images for post-fire change detection. 
  
  
  
  
  
  
  
ARVI image difference u o 
1991-1992 - 0.042 0.165 
1991-1993 -0.210 0.203 
1992-1993 -0.123 0.149 
1992-1994 -0.144 0.163 
1993-1994 -0.094 0.143 
1993-1995 -0.145 0.168 
  
  
  
  
  
The ARVI difference images) were density sliced to put in evidence all the negative/positive values that were located 
on the extremes of the distribution, in the following way: 
auto 
b) u + 1.56 
utile 
The part regarding the terrain mobilisation will be on the right side of the mean value since the ARVI value for terrain 
mobilisation will be higher than for burned areas. These different output images were checked against visual analysis of 
the original Landsat images to assess which one was reproducing better the areas that had been subjected to 
mobilisation. After careful comparison with the Landsat TM images (RGB — 5, 4, 7) u + 1.56 was chosen to derive the 
final map. Because the objective of the study is to detect reforestation actions that might be infringing the law, it was 
found desirable to use a relatively small number of standard deviations, even if it will increase the commission errors. 
The ARVI difference images (e.g.: 91-92) were density sliced using the thresholds derived from p + 1.56. The part 
regarding the terrain mobilisation, on the right side of the distribution was set to 10 and the rest of the image was set to 
An example of a terrain mobilisation area detected in the output map is shown in comparison with the sequence of 
Landsat TM images from 1990 until 1994 (Figure 2). 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 129