Schardt, Mathias 
  
  
  
  
  
  
  
number of stands | mean difference weighted top heights - h | Standard deviation of differences 
Alpine test site | 139 - 0.3256 3.1259 
hilly test site 49 - 0.9099 5.2412 
both test sites 188 - 0.4949 5.0520 
  
  
Table 3: Statistics for the assessment of tree heights 
  
  
  
  
R Square | Std. Error of Estimation 
Alpine test site | 0.633 3.9833 
hilly test site 0.823 3.0045 
both test sites | 0.715 3.8018 
  
  
  
  
Table 4: Statistics for the fitted top height models 
The following features were extracted for each stand from the maximum filtered laser data (max laser): 
— arithmetic mean of the laser-derived tree heights (h) 
— crown closure, assuming heights of more than 6 m to be crown hits (c) 
The stand-wise average tree height values were compared with weighted top heights (one top height for different top 
heights from different tree species within one stand) from forest inventory data. Table 3 depicts the statistics for each test 
site alone and for both test sites together. 
The assumed dependence on crown closure is shown in figure 4 for both test sites. In cases where the crown closure is 
below 65 %, the laser-derived tree heights (h - from max laser) are significantly lower than the weighted top heights from 
forest inventory. Hence, weighted top heights were predicted by laser-derived tree heights (h - from max laser) and crown 
closure (c) for the Alpine test site 1, the hilly test site 2 and both test sites together 3 
predicted top height = 16.16 4- 1.35 « h — 29.3«c (1) 
predicted top height = 12.366 + 1.619 «xh — 31.889 x ¢ (2) 
predicted top height = 15 -- 1.43* h — 29.5 «c (3) 
Table 4 depicts the statistics for the fitted top height models and figure 5 shows the predicted top heights for both test sites 
using equation 3. 
Statistics show, that 72 % (R square 0.715) of forest inventory top heights can be predicted by laser-derived mean tree 
heights and crown closure with the same model for different test site conditions concerning tree species mixture and 
terrain. However, mean errors from forest floor model between 18 and 100 cm, deteriorating the tree-height map, have to 
be considered. Assuming that forest floor DTMs can be improved (Ruppert et al., 2000), accuracy for laser-derived mean 
tree heights cannot be stated as the final results in this study. 
  
40 
Q.0 
$9 o nl 
] + 
30 boat. ü 
30 + 
ooo 
201 + prr 
o 9 P t 
weighted Top Height 
o,, $92 1 
o 9 ee t 
Hn 
oca * 
10 Te 
H+ 
oË ++ Crown closure 
[*] o * + 
o® + 
sd + >= 65% 
0 O « 6596 
  
  
  
0 10 20 30 40 
h max Laser 
Figure 4: Weighted top heights vs. laser-derived tree heights in two crown closure classes. 
  
1320 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 
AS 
acc 
Tal 
Cle 
the 
whi 
cor 
hon 
In tl 
top 
fror 
Anc 
mea 
see 
for : 
Hov 
goo