modeling methods 
table 3. As base 
ılts indicated that 
best among them, 
9), and ANN (0.58) 
loped only from 
other independent 
e 4, predictions of 
ed high potential 
the area of field 
entire study area 
| for predicting the 
‘those models (see 
lel, we extended 
' and assessed the 
; SD-2 and SD-3, 
ly from 0.58-0.70 
' near zero, with 
>.0 km as the four 
5 from Tong-feng, 
uilt based on four 
han and Kuan-Dau 
y Tong-Feng test 
that the models 
are most easily 
; used because of 
5 patterns in small 
e a combination of 
. climate, rainfall, 
). However, the 
ctrapolation from 
aphic attributes of 
ach other. Then, 
iables can only be 
without significant 
d 
watershed 
ax Min 
06 1157 
6 1 
9 2 
| 2 
7 20 
3 64 
  
  
Kappa coefficient 
  
  
Sampling Design (SD) Test Data 
MAXENT DOMAIN GLM ANN 
E sp Tong-Feng 0.70 0.62 0.59 0.58 
SD-2 Yo-Shan 0.37 0.30 0.39 0.23 
SD-3 Kuan-Dau 0.00 0.03 0.00 0.00 
  
Table 3. Comparison of the accuracies of four models for predicting CGTs potential habitats with three sets of test data 
  
  
  
  
MAXENT DOMAN GLM ANN 
Cos Area (ha) % Area (ha) % Area (ha) Yo Area (ha) % 
Habitat 1,051.27 6 694.47 4 719.44 4 569.54 3 
Non-habitat 16,084.73 94 16,441.53 96 16,416.56 96 16,566.46 97 
Sum 17,136.00 100 17,136.00 100 17,136.00 100 17,136.00 100 
  
  
Table 4. The distribution statistics of three models predicting the potential habitat of CGTs 
    
  
Figure 3. Perspective-viewing map showing the Huisun Forest Station 
  
Q CGT training data 
» CGT test data 
Habitat 
8D Non-Habitat 
Figure 4. Sampling design 1: four models for mapping the potential habitat of CGTs in the study area