International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 7-4-3 W6, Valladolid, Spain, 3-4 June, 1999 
For the fusion with AIF, a variance of 0.01 and a window size of 
5x5 have been selected iteratively. The combination of the 
fusion techniques, compared to each one alone, showed the best 
results: edges were highlighted and spatial differences in forest 
stands were smoothed. 
A third fusion product for the comparison was the QuickBird 
simulation. An IHS transformation was used with input the TM 
band 5, 4, and 3 and the scanned aerial photograph with lm 
resolution. This product was used to evaluate the potential of 
high resolution satellite data, as well as for a comparison to the 
BAV aerial photograph. 
Table 4 shows a summary of the used image products (single 
images and fusion products) as input for forest stand border 
delineation in the test site Lorrach. 
Image 
Product 
Description 
Symbol 
1RS-1C pan 
Linear contrast stretch under forest 
mask, 5 m resolution 
Pan 
TM 5,4,3 + 
1RS-1C pan 
IHS colour transformation with con 
trast stretch of the panchromatic 
band, 5m resolution 
IHS_TM 
SPOT 3,2,1 
+ 
SPOT pan 
1. AIF with variance 0.01 and 
window size 5x5 
2. IHS colour transformation with 
contrast stretch of pan 
IHS_SP 
BAV 
orthophoto 
Scan with 254 dpi, contrast stretch 
under forest mask, lm resolution 
Ortho 
TM 5,4,3 + 
BAV 
orthophoto 
IHS colour transformation with 
contrast stretch of the BAV 
orthophoto, lm resolution 
Qsim 
Table 4. Used image products as input for forest stand border 
delineation in the test site Lorrach. 
For the Tarvisio test site, KVR and Landsat 5 TM data were 
available. KVR images have a resolution of 2 m (Figure 1). The 
problem of the highly different resolutions of both input datasets 
was circumvented by using the sigma filter and the AIF method 
(Steinnocher, 1997). 
The first step was to resample both datasets to an output 
resolution of 6m. The KVR data was processed with the sigma 
filter algorithm and the Landsat data with a nearest neighboor 
resampling. As input parameters for the sigma filter algorithm, a 
filter window size of 9 x 9 pixels, an output window size of 3 x 
3 pixels and a variance of 0.01 were used. Afterwards, a fusion 
of the filtered KVR image and the Landsat TM 5 data was 
carried out with the AIF algorithm. The input parameters for the 
fusion were set to a 9 x 9 filter window and a variance of 0.01 
for the first iteration. The variance was changed to 0.003 for the 
following five iterations. The output resolution was the same as 
the input images (6m). 
To get a ground resolution of 2m, in a last step a Brovey and an 
IHS transformation have been performed with input the AIF 
filtered images and the original KVR data. 
Fig. 1. KVR satellite image with 2m resolution. 
For the AIF method, it is necessary to perform a very accurate 
rectification of all the involved datasets. In the high 
mountainous test site of Tarvisio it was necessary to perform an 
orthorectification using the RSG software of Joanneum 
Research, Graz. 
2.4. Methodology of stand border delineation 
For the evaluation of the information content, the actual stand 
delineation was given to the test persons and they were asked to 
evaluate the image data in two steps: 
1. Visibility of the border in the image 
The interpreter should decide, whether the inventory border 
was visible in the satellite image. Three cases have been 
defined: 
Visible: stand border can be clearly detected in 
the image 
Not visible: stand border can not be detected 
Partially visible: stand border can not be detected, but can 
be implicitly inferred from external 
geometry (e.g. forest roads etc.) 
2. In the second step, in case of a visible border the 
interpreters should delineate the stand borders in the 
satellite data. If the border was visible, but had definitely 
another geometry compared to the inventory data, the 
interpreters had to digitize the geometry using the satellite 
images. 
2.5. Evaluation of the image products for visual stand 
border delineation 
The comparative evaluation is performed in two steps: (1) 
measuring the visibility percentage of forest stand borders in 
comparison to the official forest inventory maps and (2) 
calculation of quality measure criteria. 
Quality measure criteria proposed by Heipke et al. (1998) were 
used. Heipke et al. proposed a buffer method for the quality