Schardt, Mathias 
  
  
  
  
40 
o 
304 o R oc gn 
2 $ 
n 
= n "a Wen o of 
= n [d 9 
i p 
= ge Xy de a 
£ 20] p Mr 2 | 
2 a Po "a Bog 
3 Jr" a efus aD 
ë ng tly "500 
B $a 9 
10 % $9 8 
on 
a ao 
0 
0 10 20 30 40 
  
weighted Top Height 
Figure 5: Predicted top heights vs. weighted top heights for both test sites using equation 3 
  
h hf 
timber volume | 0.779 | 0.85 
Table 5: Correlation coefficient between laser-derived features and timber volume 
  
  
  
  
  
  
Assessment of timber volume 
Assessments of timber volume were carried out within the hilly test site. The following features have been taken into 
account: 
— predicted top heights by equation 2 (hf) 
— laser-derived mean tree-heights (h - from max laser) 
Table 5 depicts the correlation coefficient between laser-derived features and timber volume from forest inventories. 
Clearly, it can be seen that the fitted top height (hf) was the best predictor for timber volume. The regression model based 
on this predictor was formed and the obtained model, coefficient of determination (R square) and standard error (SE) of 
the model are shown in figure 6. 
Laser-derived predicted top heights explain 7296 of the variability of timber volume in forest inventory data (R square 
= 0.72). The corresponding obtained standard error is 76%, Typically, stand-wise forest inventory is carried out with 
15 - 30 % error concerning main forest attributes. In this case an error of 55a for timber volume is assumed. That 
error affects the outcome of the previous analysis, deteriorating the obtained R square and standard error. Assuming that 
these two errors are independent of each other, the corrected laser-based error can be estimated by taking the root of the 
difference of the squared errors, denoted by s, 
762 — 552 (4) 
which is slightly better than the error of the conventional forest inventory (s — 52.570). (Hyyppä et al., 1999) stated a 
: 3 ; 
corrected laser-based error for timber volume assessments by means of laser data of 21.5 7—. This result is due to nearly 
homogenous spruce stands. 
In the Austrian case the hilly test site is characterized by a mixture of coniferous and broad-leaved trees with differences in 
top height between 1 and 3 meters within one stand, summarized within the weighted top heights. Tree species information 
from satellite data (e.g. with an resolution of 10 m) could provide the necessary information to decrease the standard error. 
Another problem in the Austrian test sites occurs, as within stands with high trees and low crown closure laser-derived 
mean tree heights, even predicted top heights (laser-derived), underestimate top heights from forest inventory significantly, 
see figures 4 and 5. In order to solve this problem and to increase the estimation accuracy for timber volume, algorithms 
for segmentation of single trees were developed. 
However, this study demonstrates that the stand-wise assessment of timber volume for the forest inventory is feasible with 
good accuracy by means of laser scanner data. 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 1321