Andrade, Nilo Sergio de Olive 
  
  
  
  
  
  
  
  
  
1995 SITUATION 1997 SITUATION | COLOR | CHANGED AREA | PERCENT OF CHANGE 
Mature and secondary forest Bare soil Red 12.54 Km” 2.32% 
Mature and secondary forest Pasture White 25.84 Km” 4.77% 
Bare soil Pasture Blue 5.98 Km” 1.11% 
Pasture Bare soil Yellow 9.27 Km? 1.71% 
Regrowth Pasture Cyan 0.00 Km“ 0.00% 
Burned areas Bare soil Black 0.42 Km? 0.08% 
Burned areas Pasture Brown 0.41 Km? 0.08% 
  
  
  
  
  
  
  
Table 4 —Evolution of changes obtained with the unsupervised classification (K-means). 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
1995 SITUATION 1997 SITUATION | COLOR | CHANGED AREA | PERCENT OF CHANGE 
Mature and secondary forest Bare soil Red 12.53 Km” 2.32% 
Mature and secondary forest Pasture White 28.00 Km” 5.17% 
Bare soil Pasture Blue 6.36 Km‘ 1.18% 
Pasture Bare soil Yellow 0.84 Km” 0.16% 
Regrowth Pasture Cyan 1.74 Km” 0.32% 
Burned areas Bare soil Black 0.28 Km" 0.05% 
Burned areas Pasture Brown 0.78 Km“ 0.15% 
  
Table 5 —Evolution of changes obtained with the supervised classification (Maximum Likelihood). 
  
Figure 5. Regions of Interest (Land use and cover … Figure 6. Regions of Interest (Land use and cover 
changes) obtained with unsupervised classification. changes) obtained with supervised classification. 
In this part of the study the values presented in Table 3 and the Regions of Interest shown in Figure 3 are used as 
ground truth. It can be noticed that the results presented in Table 3 (ground truth), Table 4 (K-means) and Table 5 
(maximum likelihood) are quite similar. But, when we analyze the regions of interest (showed in Figures 4, 5 and 6), 
detected from the visual analysis of TM images, unsupervised and supervised classification, respectively, that the 
unsupervised classification presents a lot of misclassification areas, mainly for the region of pasture in 1995 that 
became bare soil in 1997 (yellow color). 
The results obtained with the supervised classification is quite reasonable when compared with the ground truth results, 
and represent an easy and fast way to classify images with the purpose of monitoring changes in natural areas. The 
  
68 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000.