- 1782 - 
this problem assumed the general form 
V=-a+*b {{ch4+ch5) + (cn 6 + ch 7 ) + 
+ e ({ ch 4 + ch 5} - ( ch 6 * ch 7 ))? 
The proportion of the total variance of timber volume explained 
by this function (R2) was 0.88. An example of the pixel volumes 
estimated by this function is given in Pig. 2. 
4. PRELIMINARY RESULTS 
4.1 Soil-vegetation mapping 
One of the crucial problems tackled in classifying soil-vegetation 
units from Landsat data was the extraxtion of appropriate spectral 
classes to be interpreted as meaningful earth resource types. 
Among the factors disturbing the interpretation of timber and soil 
are, e.g. brush vegetation, mixture of tree species, variable 
moisture content of vegetation, condition of trees (stress) etc. 
The training fields chosen from Huittinen test site were analyzed 
statistically in order to find classes having maximum separabil- 
ity. The procedure involved B-distance calculation and resulted 
the theoretical upper bounds of the classification error as shown 
in Table 2. 
TABLE 2. The theoretical upper bounds of the classification 
error (%) (Huittinen test site). 
  
  
      
     
     
        
   
VEGETATION/SOIL 
CLASS 12 3 4 5 6*(7 8 (9) 319 1T 12 
l. Clear cut, recent/moraine 3 - e des. PEL ULR L -— tm 
2. Clear cut, old / " ==> ma = - 24 ">, mA 
3. Spruce stand Zi 33 
4. Pine stand / M" 
5. Mixed conifer yz" 
6. Birch stand Z5" 
7. Pine stand /till 
8. Birch stand /Ct-peat 
9. Pine stand LE 
10. Pine stand /St-peat (IR) 
ll. Pine stand / (TR) 
12. Pine stand /outcrop 
   
  
  
  
   
    
   
   
    
     
    
  
    
    
    
       
    
    
fui 
enc 
The 
sur 
pis 
TAI 
Lan« 
sat 
cla: 
fic: 
  
Spri 
Pine 
Mix 
Har« 
stai 
Out: 
pea: 
Pea: 
(pii 
Cle 
Mis: 
Tot.