-B8, 2012 
ed on the DSM, 
infra-red image 
2 201 1). At first, 
nt land in the 
tically extracted 
1g Normalized 
(1)) from ortho- 
tree height is 
hts of extracted 
egions. At last, 
om the formerly 
region as a clue 
(D 
ments 
ements 
zht of ground 
(known) 
  
Estimation 
ght of ground 
rest (unknown) 
  
imation 
| obtained from 
DTM generated 
[ generated by 
periment forest 
shows that the 
DAR is almost 
neter. And the 
AR is less than 
as stereoscopic 
vides a much 
ound resolution 
n also provide 
necessary for 
DTM generated 
yy for practical 
  
TM 
  
  
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B8, 2012 
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia 
  
  
  
TEE HRY BS 526 45: BA EYRE 
      
X mm 3? Xx 
Figure 5. Comparison of DSM and DTM 
  
iod x 78 
  
  
3.12 Tree Species and Density Estimation using Image 
It is important to estimate tree species and density for forest 
resource estimation. For example, in the beginning of tree 
planting in the artificial forest, single species such as Japanese 
cedar or Japanese cypress is planted. But after a long time, 
surrounding broad-leaved forest intrudes into the sub- 
compartment. In this case, the single species described in the 
forest registration is not correct any more. Similar problem 
exists for the tree density. For some sub-compartments under 
frequent investigation, the forest density information is kept 
updated. But due to the increase of tree un-thinning area in 
these days, tree density is not accurate at all and also becomes 
not uniform inside the sub-compartments. 
Therefore, we make experiments to examine the possibility of 
estimating tree species and density from images. At first, we 
conduct a field survey in the natural forest in Tatera Mountain 
in Tsushima, and then we investigate the possibility of tree 
species classification based on 3D shape and colour distribution 
of tree crown using aerial photograph in this area. Figure 6 
shows that tree species is able to be classified based on R and B 
value of the images. 
Next, we compare the number of trees acquired by 
photographic interpretation with that from object-based image 
analysis. The result shows that there is strong correlation 
between these two numbers obtained from different methods. 
  
  
155 [ «S 
9 135 [ 
= 
ë o8 
o 115 A 
= % 
S 
> 
B ws U 
© á X Castanopsis sieboldii 
« 75 OQ Quercus salicina Blume 
vw @ Distylium racemosum 
55 L_@— 2] 
  
55 75 95 115 135 155 
Average value of Red 
Figure 6. Result of Species Classification by Colour 
32 GIS system 
As described in chapter 3.1, it is proved that estimation of tree 
height, density and species has reached certain accuracy level 
for practical use. But next problem is what kind of system 
should be provided for the practical work like the work in the 
tree farm. In many cases, the GIS systems owned by forest 
owner's cooperative and so on are equipped with various 
general functions but users usually only use the simple function 
like map viewing. 
Then, we divide the whole system into two stages, firstly to 
generate basic data, and then to estimate resources by using the 
basic data. In the first stage, that is, the stage to generate DSM 
and DTM from aerial photograph, it is better to conduct aerial 
photography and data processing in a large scale considering 
the cost. So, not individual forest owner’s cooperative, but 
prefecture or federation of forest owner’s cooperative 
association or service vender should carry out this stage for all 
the concerned forest area. On the other hand, resource 
management including the lumber volume estimation in certain 
sub-compartment is useful for various planning trials of tree 
thinning in helping the individual forest owner’s cooperative to 
carry out a massive work plan. 
Therefore, in the proposed forest management GIS, we provide 
a simple interface targeting estimation of forest resource 
harvested from a sub-compartment for the end users (Figure 7). 
The input of the system is the compartment data (shape format), 
that include tree species, forest age, site class, area, and so on, 
forest base map as a background image, ortho-photo generated 
from aerial photographs, DSM and DTM generated by stereo 
processing. These input data is managed in layer level and also 
displayed as layers on the display similar to common GIS 
system. 
‘Database 
# «Compartment data 
# “Forest road, Loading area; 
& Forest base map f 
% -DSM 
*DTM 
i. "Orthophoto 
m NS 
Merc * 
» H 
P 
ia M 
x 
Tree Density Estimation Forest Road Setting 
Figure 7. Forest Resource Management GIS 
  
  
      
  
   
  
Output 
s *tree volume 
Harvest *tree density 
E: Estimation -number of timber 
*volume of timber remnants 
*cost 
*income 
   
  
  
  
  
  
   
  
  
  
  
  
  
  
  
  
With the proposed forest resource management GIS system, it 
is possible to carry out estimation of tree density and then forest 
resource in each sub-compartment. Operation procedures are 
described as follows. 
3.2.1 Tree Density Estimation 
The processing unit of this system is sub-compartment. At first, 
user selects the target sub-compartment and click “density” 
button. Then, the window of tree density estimation appears, 
shown as the left figure of Figure 8. In this window, the 
selected sub-compartment is wholly displayed. Then user sets 
the plot region inside the sub-compartment. The area of plot 
region can be selected from 5 candidates, 0.04ha, 0.05ha, 0.1ha, 
0.2ha and 0.25ha, similar to the conventional method of tree 
density estimation on photographic surveying. For example, 
when 0.1ha is selected, rectangle that corresponds to 0.1ha is 
automatically displayed on the screen, obtained on the basis of 
ground resolution of orthophoto. User sets the location of the 
plot region in the target sub-compartment by dragging the 
rectangle by mouse on the screen. Then magnified view is