b) 
  
C) 
  
Figure 1 — 1.3) Example of the Canopy Altitude Model, 1.b) 
the Digital Terrain Model, 1.c) and the Canopy Height Model (c 
— à - b). Brightness is proportional to altitude or height. 
Obtaining tree height from the Canopy Height Model 
The CHM gives interpolated height of all points in the canopy 
in the form of a regularly spaced grid having a 50 cm 
resolution. The height of a tree was defined as the pixel having 
the highest value in a high-valued pixel cluster corresponding to 
a crown. This "top pixel" is normally situated near the center of 
the crown put can sometime be found a few pixels from the 
center in the case of large hardwood trees. We believe that a 
  
   
    
    
  
   
   
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
    
    
   
    
   
   
   
  
    
    
      
    
    
   
    
International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 3W14, La Jolla, CA, 9-11 Nov. 1999 
more accurate method would be to find the difference between 
the maximum laser spot altitude and the underlying interpolated 
DTM altitude in order to preserve the preciseness of raw laser 
data. However, due to time limitation, we chose the simpler 
method described above. Improvements in the near future will 
comprise preserving raw laser data for vegetation and using a 
better interpolation method for ground laser points. 
a) 
b) 
  
Figure 2 — 2.a) Digital multispectral videography (initially in 
color) rectified according to the laser altimetry images 
presented in Figure 1, and, 2.b) overlay on the Canopy Height 
Model of figure 1.c). 
5 RESULTS 
Linear Regression 
Linear regression was performed between ground-measured 
height and laser predicted height on. The mean of the two 
height measures done in the field for these trees was regressed 
against the corresponding height read from the CHM for 36 
trees (12 hardwood and 24 softwood). The linear model yielded 
a R? of 0.90 (significant at &=0.01). The scatter of points in 
figure 3 does not bear any non-linear trends, an observation that 
was corroborated by the fact that the linear model fit gave the 
   
International 
best results. The pre 
height) is given by: 
Mean ground truth heig 
30 
  
25 
  
  
20 
  
Laser height (m) 
  
  
  
  
  
Figure 3 - Comparison « 
heights. 
Error assessment 
Tree heights predicted b 
3 were compared to the ı 
yielding an absolute di 
deviation of 1.15 m. The 
standard deviation of 9 ¢ 
two ground measuremen 
relative error (the absolu 
measurements divided by 
is 10% (SD = 8%). The 
absolute measurements c 
model, the laser predictic 
of ground measurements 
these results since the nu 
“truc” tree height, or 
centimetric accuracy is i 
laser predicted heights is 
truth data. The R? of equ: 
error associated with groi 
manual photogrammetic 
alleviate this problem. M 
laser-predicted height ei 
errors is only indicativ 
established with certainty 
attributed to ground mea 
use of laser predictions 
higher accuracy than grot 
The main factors de 
measurements, and thus 
believed to be laser spot 
laser penetration in veget 
Crown surface (obtained 1 
by using the average of