Figure 2 - Results of the segmentation algorithm shown for TM 
bands 5 (left) and 4 (right) 
time of this writing, that these categories could be easily 
differentiated using TM data. 
A more detailed discussion of the segmentation and classification 
results considering the specific land use practices adopted in the 
region is presented in the following sections. Without 
consideration to region-specific information, a few more 
characteristics of the technique adopted could be mentioned: 
e. the segmented images typically present clearly delineated 
areas, with relatively few edge effects; however, the algorithm 
can "miss" areas characterized by texture without continuous, 
clear edges (e.g. some areas of abandonment in the middle of 
forest); at the same time, it can generate erroneous segments in 
the middle of areas of forest because of shade and other effects; 
e minimum area size and edge-threshold values are 
controlled by the user and the adequate definition of these 
parameters can minimize the effects described above; however, it 
is frequently difficult to establish the "ideal" values for one entire 
image, because of variations in terrain, vegetation characteristics, 
land use, etc.; the possibility of definition of different values for 
different parts of an image or an adaptive approach to calculate 
those values automatically could help to reduce the volume of 
corrections; 
* the region-oriented classifier produces the merging of 
"similar" neighboring segments; this procedure has the positive 
effect of reducing the number of segments but some classification 
errors are made more difficult to correct; conserving the richness 
of detail of the segmented images could be used by more 
16 
elaborate editing functionalities and reduce the time required for 
corrections. 
Finally, it can be noticed that although the work produced at the 
120-meter resolution lacks the detail of the original TM 30-meter 
resolution, experience has shown that they present less 
segmentation and classification errors due to relief, image texture 
and other effects. These results can be used as a reference or 
mask for more detailed work at the 30-meter resolution. 
Forest Clearing and Land Use Practices in the Region 
Most forest clearing in the area is made by two different 
techniques: clear-cut (all trees are cut) or the "quebradáo" 
practice (many large trees are left standing, to reduce clearing 
costs and to maintain part of the vegetation). Clear-cut areas can 
be easily identified on TM images for longer periods of time, 
even if they are abandoned. Areas of "quebradáo", however, can 
present a more vigorous regeneration; the authors found evidence 
that, in some cases, they can be confounded with forest just two 
or three years after the clearing. 
Another peculiarity of the region is that land can be cleared to 
assert ownership or to raise property value, without maintaining 
the property productive. Secondary vegetation in such areas is 
expected to regenerate faster than areas under heavier use (Uhl ef 
al. 1988), and thus reach a forest-like aspect within shorter 
periods of time. 
The use of multitemporal data can be valuable under these 
conditions. "Quebradáo" areas can sometimes be identified, 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996 
partici 
regene 
also b 
these 
defore 
carbon 
Chara 
Prelir 
show | 
of the 
checke 
1995. 
confiri 
contin 
older s 
These 
the fr 
betwe 
abande 
detaile 
TM c 
thorou 
produ 
aband 
One it 
of the 
forest 
classif 
discrir 
vegeta 
The a 
resear 
and in 
studie: 
evolve 
some 
approz 
studie 
causes 
ACKI 
by F/ 
numbe 
suppor 
Devel 
Silvan 
MORE 
Inform