The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008 
Fig.4 Swiss NFI sample design 
The forest boundary line (FBL) is interpreted at each sample 
plot. It determines the border line between normal forest and 
non forest. It allows the evaluation of the forest width and forest 
interspaces which are needed to reach a Forest/Non-forest 
decision. Fig. 5 shows the forest boundary line on a convex and 
concave forest edge. 
Interpretation area 
• Sample plot center 
Forest boundary line 
# Stocking element 
Fig.5 Forest boundary line on a convex and 
concave forest edge 
2.4 Forest Detection with LIDAR Forest Mask 
LIDAR is a well-established technique in terms of its capability 
of direct measurements on canopy structures (Maltamo, et al., 
2004, Naesset, 2002). From the obtained CHM data, we 
develop a forest mask for the forest area detection. This is 
performed by using a moving window approach. (Fig. 6) 
high than 3m, let p = T /(K x K) , and if p > 20% , then the 
current pixel at position (/, j) will belong to forest, otherwise it 
will belong to non-forest. At last, we shrink the edge of the 
results obtained from the forest mask with a small size window. 
Fig. 7 shows the detection result of the forest mask, (green color: 
detected forest area; red line: manual forest boundary) 
Detected 
forest 
Manual 
delineation 
boundary 
Not- 
detected 
forest area 
Fig. 7 Forest detection with 
forest mask (moving window) 
However, the forest area will not be detected for low quality 
CHM area (area with yellow circle in Fig.7). This forest mask is 
not practical when the CHM is not well distributed or under 
certain canopy conditions. 
2.5 Forest Delineation with Integration of Aerial Image 
and LIDAR data 
Fig. 8 Schematic workflow of overall process 
Fig.6 Forest mask with 
a moving window 
Let j\. K is the pixel value in CHM within the 
KxK window centred at position (i,j) , then according to 
NFI forest/non-forest definition that the tress in forest should 
higher than 3m as well as with 20% crown coverage, we 
calculate the percentage of the pixels which are high than 3m 
within the current window. Let T is the sum of pixels which are 
High resolution aerial images can improve efficient forest 
management at fine scale(J. Hyyppa, 2000, M. A. Lefsky, 
2001). Benefit from the high spatial resolution of NFI aerial 
images, we apply JSEG (Deng and Manjunath, 2001)image 
segmentation method to obtain homogenous sub-areas. This 
method is one of the color image segmentation methods which 
provide robust segmentation results on a large variety of color 
images. From the CHM, we calculate curvature feature for 
building’s remove. A vegetation index named GVI (Green 
Vegetation Index) is calculated for non-green fields removing.