The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008 
1204 
is used to remove flat area as well as buildings. The use of color 
and texture features extracted from the aerial images make it is 
possible to merge segments in a semantic way and lead to 
delineate forest boundary automatically. 
2. MATREIAL AND METHOD 
2.1 Aerial images 
The true color ( RGB) aerial images in our database have been 
made available by Swiss Federal Office of Topography 
(Swisstopo) with an average scale of 1:30,000. The images are 
oriented block by block and orthorectified with a resolution of 
0.5m. The landscapes of Switzerland are very heterogeneous 
which can be divided into Jura, Swiss Plateau, Prealps, Alps, 
and South-western slopes of the Alps. Our test sites (green 
points in Fig.l) include mixed industry areas, agriculture areas, 
wines areas etc. which belong to Swiss plateau and Prealpine 
areas. 
0 50 100 Kilometers 
Fig. 1 Test sites 
2.2 LIDAR Data 
The LIDAR data was acquired between 2001 and 2004 by 
Swisstopo. Average flight height above ground was between 
1000m and 1500m and the footprint varies between 0.8m and 
1.2m. Fig. 2 shows the DTM of Switzerland where there is no 
LIDAR data above 2000m. Canopy Height Model (CHM) is a 
LIDAR-derived product for deriving relevant vegetation 
Fig. 3a Aerial Image Fig. 3b DOM 
Fig. 3c DTM Fig. 3c CHM 
Fig.3 Aerial image, DTM, DOM and CHM 
2.3 NFI Sample Plot Design and Forest Definition 
In the first NFI a 1km sampling grid was used. The 
intersections define the location of about 12'000 NFI1 sample 
plots in the forest. As in the NFI2 and NFI3, only half of the 
plots were investigated terrestrially. 
As shown in Fig. 4 the NFI-specific forest definition is applied 
to aerial imagery in a regular 500m grid. For the terrestrial 
sample plots a coarse grid with 1.4km grid was chosen. The 
discrimination of forest and non-forest areas in aerial 
photographs requires an unambiguous reproducible forest 
definition. The used NFI forest definition is the aspects of a 
stocking evaluated by the following stand criteria that can be 
measured in the aerial photograph: width, crown coverage and 
dominant stand height: 
• Width: The width of the stocked part of the interpretation area 
is at least 25m. 
• Crown coverage: The crown coverage of the stocked part of 
the interpretation area has to be larger than or equal to 20%. 
• Dominant stand height: The stocking has to have a dominant 
stand height of 3 m. 
For a positive forest decision the following conditions apply: 
The minimum width is 25m with crown coverage of 100% and 
the required dominant stand height. With increasing width the 
minimum crown coverage is allowed to decrease. The smallest 
acceptable threshold for the crown coverage is 20% at a 
minimum width of 50m. 
information. From the raw data points, the Digital Terrain 
Model (DTM) and the Digital Surface Model (SM) is D 
interpolated (2.5m x 2.5m). The CHM is obtained by 
calculating the difference between DSM and DTM. Fig. 3 
shows the DTM, DSM, and CHM of one analyzed aerial image.