- 1768 - 
the initial 100 stand types are compared with the digital analysis 
results, it is evident that many unclassified areas could have been 
identified if the species grouping used to define the three major 
types had been slightly different. Hence, the more detailed the 
level of stand classification one chooses, the closer the fit of 
the types delineated by the two methods. 
Table 3 indicates the comparison of the total areas of the 
three types obtained using the digital analysis method and aerial 
photographic interpretation. 
TABLE 3 
Comparison of Area Measurements of Forest Types 
Obtained from Landsat and Aerial Photography 
  
Type Landsat Aerial Photography Difference 
1 142 ha 158 ha 16 ha 10% 
2 88 ha 153 ha 65 ha 42% 
3 182 ha 189.6 ha 7.6 ha 4% 
It will be noted that the major difference in estimates for Types 1 
and 2 are defined on the basis of differing proportions of their 
combined poplar and white birch representation. These two types can 
be identified consistently by airphoto interpretation but with diff- 
iculty by digital analysis of Landsat imagery because of a close 
similarity in their reflectance values. Since Types 1 and 2 would 
probably be combined for most inventory projects of the type contem- 
plated, the maximum expected variation in the total area estimates of 
the two methods would be about 25%. If the range of stand conditions 
within each type were broadened slightly, many of the small unclass- 
ified areas of this comparison would be included in the combined 
Types 1 and 2 category, and the total area of this category estimated 
by the two methods would be more similar. 
To eliminate the effects of variation in sun angle and atmos- 
pheric conditions, this digital analysis would not be extrapolated 
beyond one Landsat frame. However, the reliability of the extrapol- 
ation decreases as the area considered increases because of the 
increasing probability of including stands which differ from those 
of the training area. Although tests of accuracy of extrapolation 
have not yet been completed, it is believed that the accuracy of 
extrapolating the result to the area of one Landsat frame will be 
within acceptable limits for an inventory project of the type which 
is contemplated. A broader generalization than desirable of the 
boundary of forest stands may result from the condition where the 
boundary between two stands occurs within pixels and the reflectance 
    
  
   
  
  
  
   
   
   
    
     
    
    
    
   
  
  
  
  
  
     
   
   
   
  
  
     
   
  
  
  
val 
dom 
met 
The 
tes 
for 
As 
ana 
met 
lev 
tha 
256 
ana 
of 
the 
ing 
nun 
exc 
Dec