International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
  
essentially, VI and BI have high correlation of negative. 
After that, set the scaling of zero percent point and a hundred 
percent point. Details in (A. Rikimaru, 1996) 
>  Scaled Shadow Index; SSI 
The shadow index (SI) is a relative value. Its normalized 
value can be utilized for calculation with other parameters; 
The SSI was developed in order to integrate VI values and SI 
values. In areas where the SSI value is zero, this corresponds 
with forests that have the lowest shadow value (i.e.0%). In 
areas were the SSI value is 100, this corresponds with forests 
that have the highest possible shadow value (i.e.100%).SSI is 
obtained by linear transformation of SI. With development of 
the SSI one can now clearly differentiate between vegetation 
in the canopy and vegetation on the ground. This constitutes 
one of the major advantages of the new methods. It 
significantly improves the capability to provide more 
accurate result from data analysis than was possible in the 
past. 
> Integration process to achieve FCD model 
Integration of VD and SSI means transformation for forest 
canopy density value. Both parameter has dimension and has 
percentage scale unit of density. It is possible to synthesize 
both indices safely by means of corresponding scale and unit 
of each 
FCD=~VD*SSI+1-1 ©) 
5. FOREST CANOPY DESITY MAP FOR STUDY 
AREA 
The degree of forest density is expressed in percentages: 10% 
FCD, 20%, 30%, 40% and so on. The Fig.S indicates forest 
canopy density map of the study area for ETM+ image. 
  
0% 
Figure 5. Forest canopy density map for ETM+ (2002). 
For accuracy assessment and collected ground truth, the 
distance between classes were changes to the form below: 
Class!) Water & Cloud -W&C 
Class2) No Forest =NF (0-5%density) 
Class3) Low Forest = LF (5-40%density) 
Class4) Middle Forest =MF (41-70%density) 
Class5) Dense Forest =DF (71-100%density) 
Figure 6 indicates this map with these classes. 
  
  
37 Legend 
Dens Forest 
[. ]Low-Forest 
[...]Middle-Forest 
[.]NO-Forest 
| MESI 
    
  
Figure6. 4 classes density map. 
Table 2 shows the confusion matrix, overall accuracy and 
kappa coefficient, which is calculate for 2002 images. 
  
  
  
  
  
  
  
  
  
NF LE MF DF Total U.A 
(96) 
CI 10328 1622 18 0 11968 86 
C2 273 8163 3311 3 11750 69 
C3 9 246 6852 1227 834 | 8222 
C4 0 0 127 8984 9111 | 98.61 
Total | 10610 10031 10308 10214 | 41163 | 41163 
P.A 97.34 81.38 66.47 87.96 
% 
  
  
  
  
  
  
  
250 
Table 2. Confusion matrix 
Kappa Coefficient =0.78 
Overall accuracy =83% 
UA : User’s accuracy P A: Producer’s accuracy 
6. Forest density & Area changes in the studied site 
during 1991-1998 
For change detection, the images should be taken at à 
simultaneous date and season also, they should be 
dereferencing exactly. Since the season for the 2002 image, 
differs from 1991, so we use 1998 and 1991 images. They 
have been dereferencing to 2002 image. Then we prepared 
density map, using these two images for the years 91 and 98. 
Fig 7 & 8 shows the forest density map for these years. 
  
  
Internatio 
BCE 
* 
sa 
e 
Figure7. | 
  
  
  
  
Figure 
Pixel size a 
Since every 
in the chan 
changes. T 
forest chan 
density cha 
  
  
Figure 9 
The table 3