Barbosa, Paulo 
  
The overall accuracy obtained was high, and the Kappa coefficient that accounts for the randomness of the accuracy test 
is also quite high. From the point of view of the user (DGF) 93% of the pixels found in the output map as mobilisation 
are correct. For the control of reforestation actions, there should be a good balance between commission and omission 
errors, allowing DGF to perform the monitoring of the reforestation areas without a big increase in the workload of 
forest officers. 
5.2 Land cover change spectral analysis 
The temporal evolution of the land cover was analyzed for test areas that appeared burned in the 1992 Landsat TM 
image although the fires occurred in the 1991 fire season. The analysis was done from 1990 until 1998 using ARVI 
(Figure 3). 
0.9 
  
—e- Pi-Eu-93 -«- Pi-Eu-92 — Eu-Eu —Pi-Sh —- Sh-Sh 
0.8 - 
107 - 
0.6 - 
0.5 4 
ARVI 
0.4 - 
0.3 - 
0.2 4 
04 - 
  
  
  
1990 1991 1992 1993 1994 1995 1996 1997 1998 
Figure 3. Comparison of the spectral evolution of different land cover transitions using ARVI. 
Five different types of test areas were collected in order to follow up the spectral evolution of the land cover over forest 
burned areas: Shrubland — Shrubland (Sh-Sh), Eucalyptus —Eucalyptus (Eu-Eu), Pine- Shrubland (Pi-Sh), and Pine — 
Eucalyptus (Pi-Eu). The last test area type was divided in two different sub-types depending if the terrain preparation to 
plant Eucalyptus was done in the first (Pi-Eu-92) or in the second year (Pi-Eu-93) after the fire. 
It can be observed that in the 1992 image, all the land cover types have an abrupt fall in ARVI due to the burned area. 
However, in the case of Pi-Eu-92 the fall in ARVI is much higher due to the terrain preparation. In the case of Pi-Eu-93 
there is a further fall in ARVI from 1992 to 1993 due to the terrain preparation. These observations explain why the 
ARVI difference method can give good results for detecting terrain preparation on previously burned areas. 
Regarding the evolution of the burned area with time it can be seen that while the regrowth of Shrubland is rather slow 
both in Sh-Sh and in Pi-Sh, the regrowth of Eucalyptus in Eu-Eu is relatively faster. This can be explained by the fact 
that Eucalyptus is a species that has a vegetative reproduction from stumps that allows for a rapid regeneration after a 
clear cut or after fire. However, the fastest regrowth of all is the one of Eucalyptus plantations (Pi-Eu-92 and Pi-Eu-93) 
which reaches stabilization in terms of ARVI within two to three years after the plantation. 
6 CONCLUSIONS 
The change detection methodology to identify terrain mobilization after forest fires was found good enough for the 
purposes of identifying potential illegal forest plantations after fires. The overall accuracy of the change detection 
method was 95%, while the Kappa index was 79%. Although from the point of view of the map producer only 74 96 of 
the terrain mobilization areas were detected, from the point of view of the user (DGF) 93% of the areas detected as 
terrain mobilization were right. 
Furthermore, the analysis of the multitemporal time series of the different land cover transitions showed that different 
types of land cover have different spectral behavior after the fire. These different temporal spectral characteristics can 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 131