International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
  
  
   
  
     
    
   
<<Display>> | External 
GUI | | software 
<<File>> | 
Image data | 
| / 
7 '««API»» 
L / 
; = SE TT 
<<File>> | "«. |Forestime <<File>> 
Ground | ———— server — ——— Estimate 
data | 
= 
1 Forestime 
<<File>> Analysis | 
Feature Analysis | 
bank 
| <<Command Line Call>> 
   
  
| Analysis 
| tool 
Analysis 
Figure 5 Deployment diagram of Forestime. 
4. RESULTS AND DISCUSSION 
The Forestime system performance in forest variable estimation 
was tested with Ikonos image acquired on 5. Sep 2003 from 
Suonenjoki, and with FFRI sample plot data from the same 
area, measured 2001. The tests included producing estimates for 
total stem volume (V), average stem diameter weighted with 
basal area (DW), stem number (Nstems), and for the individual 
stem volumes for pine (Vma), spruce (Vku) and broad-leaved 
trees (Vlepu). The last three estimates were then used to 
calculate the tree species proportion estimates with an external 
program (ErMapper). 
The root mean squared error (RMSE), the relative root mean 
squared error (RMSE%) wrt. mean of the variable, and the 
correlation of the variable estimate with ground data were then 
computed to obtain the estimation accuracy. Due to the small 
size of the ground data sets, a 5-fold cross-validation was used 
to the model validation. The shown figures of merit are the 
mean values of the values calculated for K=5 partial data sets. 
Table 1 shows results for the estimation obtained with spectral 
features only (baseline set). 
Table 2 shows the corresponding figures for estimation with 
feature set vol 1 = spectral channels + Haralick entropy. 
  
  
  
  
  
  
  
  
  
VARIABLE RMSE RMSE % Corr, 
|% of mean] coeff 
Total stem 54.4 m3/ha 37.4 0.82 
volume 
Mean diameter 4.1 cm 23.4 0.87 
Stem number 917 87.1 0.50 
stems/ha 
Pine 23.4 96 111 0.74 
percentage 
Spruce 25.195 47 0.79 
percentage 
Broad-leaved 21.8% 137 0.53 
percentage 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
VARIABLE RMSE RMSE % Corr. 
[% of mean] coeff 
Total stem 56.6 m3/ha 38.9 0.81 
volume 
Mean diameter 4.2 cm 24.0 0.86 
Stem number 953 90.6 0.45 
stems/ha 
Pine 26.9 % 732 0.64 
percentage 
Spruce 30.5 % 60 0.69 
percentage 
Broad-leaved 25.0 95 166 0.49 
percentage 
  
  
Table 1 Figures of merit for the baseline estimation 
Table 2 Figures of merit for feature set vol 1 
Examples of scatterplots for parameter estimates vs. ground 
reference data are presented in Figure 6, Figure 7 and Figure 8 
(stem volume (V), average stem diameter (DW), and spruce 
percentage (Vku) respectively). The estimate images for the 
same variables are presented in Figure 9, Figure 10 and Figure 
11 respectively. 
Estimated stem vol vs. gnd data, FS1-BL-111203 
500 
ag R = 0.85 
400 
'"w 350: 
X 
© + 
£ 300 ne 
2 + 
© 
E 250 . $ a. e e d e 
3 . EE "e 2" . 5 . 
9 200 es S 1*5 b as s e 
«**24 ob 
= . e sa te ® +b e . 
eed eap " * % € ec ® 
= 150 Srey fT 
e* Res ta % 8 e 
s* 9. t : 
100 ‘ 
50* 
3^5, 
oe 
0 .* 
0 100 200 300 400 500 
stem volume [m3/ha] 
Figure 6 Stem volume(V) estimate plotted against ground truth 
data (N — 296), baseline feature set 
338 
  
Internati 
stem diameter estimate [cm] 
35 
30 
— N N 
a1 © c 
e 
ces ac 
Figure 7 
spruce percentage estimate [%] 
Estim 
120 
100 
80% 
Figure