The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008 
spp, 
forest along the time, they are maps comparables to different 
scales. The more accurate classification method used in the 
generation of these maps was the algorithm Maximum 
Likelihood (with Kappa coefficient =0.87) (Figure 3). 
Figure2. Forest Map of Cuna Piru Reserve with nine units: 
green tones are different forest(six units),light green-ecotonal 
zone, beige-sacannah, yellow-grassland, black-shadow 
Figure 3. Forest Map of Cuna Piru Reserve (four forest units). 
Supervised classification of Landsat TM February 2007. 
topographic variables. 
The elaboration of a model of native forest of Cuna Piru zone 
from topographic variables was obtained. It explaines the 60% 
of the spatial distribution. The results of this model are 
preliminary. The Figure 4 shows this model, that if we 
compare it with the Figure 3 we can suggest that the units in 
the model have similar distributions than in classified satellite 
image. 
The best process in order to separate forest vs non forest was 
CnSIcp for MSS data (Kappa coefficient 0.88-0.85) and CnSI 
(Kappa coefficient 0.87-0.83). 
An effective comparison between the different forests and land 
use (agricultural and livestock) and the changes of the last 30 
years were detected, for example as shows the Figure 5, the 
native forest decreased 23% (435 ha) from 1976 to 2007. 
70.0 
Figure 5. Changes along time of different covers: beige=other 
covers, green= native forest, red=plantations.