International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 7-4-3 W6, Valladolid, Spain, 3-4 June, 1999 
186 
in this research work. These algorithms are generally known, so 
we refer to other literature for further reading (Pellemans et al„ 
1993, Chavez et al., 1991 for IHS; Welch and Ehlers, 1987, Pohl 
and van Genderen, 1998 for PC A; and Pohl, 1996 for Brovey). 
A relatively new fusion technique has been developed by 
Steinnocher (1997). He introduced the Adaptive Image Fusion 
(AIF) method that uses local object edges instead of image 
segmentation. This filter averages all pixels in a local 
neighbourhood, which could belong to the same distribution as 
the central pixel. The two unknown variables, window size and 
variance of the distributions, must be empirically estimated. 
Important prerequisites for a successful merging are high 
correlated sensor bands, high geometric accuracy and low 
temporal interval of the scenes (Darvishsefat, 1995). The aim of 
data merging of multispectral and panchromatic data in this 
project is to optimize the information content for visual 
interpretation. 
2. DATA AND METHODS 
2.1. Test site 
The test site Lörrach is located in Southwest Germany near the 
city of Basel, facing the river Rhine and France in the West and 
Switzerland in the South. The height variations in these southern 
foothills of the Black Forest is from 230 to 810m. 37% of the 
area is covered by forest, out of which 42% is coniferous and 
58% deciduous. The forest area consists of 21% state forest, 
40% community forest and 39% private forest. The community 
and state forest in this area is comprised of a large number of 
different stands mainly of smaller sizes (0.1 to 5 ha). 
The test site Tarvisio is located in Italy at the borders to Austria 
and Slovenia in the Camic Alps. It is characterised by very steep 
slopes. The heights vary from about 700m to 2800m. 
2.2. Satellite data 
The satellite data listed in Table 2 were available for this 
research project and have been used for the sensor fusion 
application. 
Satellite data 
Acquisition date 
Resolution (m) 
Test Site Lörrach: 
IRS-1C PAN 
14.10.1996 
5.8 
SPOT PAN 
20.07.1989 
10 
SPOT XS 
26.07.1995 
20 
Landsat 5 TM 
20.06.1995 
30 
Test Site Tarvisio: 
KVR 
14.05.1992 
2 
Landsat 5 TM 
18.08.1992 
30 
Table 2. Satellite data used. 
In addition to testing the potential of existing high resolution 
satellite data for forest stand delineation purposes, it was 
decided to simulate high resolution satellite data by fusing an 
aerial panchromatic orthophoto (scale: 1:10,000) with the 
multispectral information from Landsat 5 TM (band 5, 4, and 3). 
In order to simulate different ground resolutions, different scan 
resolutions have been used (Table 3). 
MOMS-02 P 
EarlyBird 
QuickBird 
Resolution 
5.3 m 
3 m 
1 m 
Scan Density 
48 dpi 
85 dpi 
254 dpi 
Table 3. Different scan resolutions of an aerial orthophoto to 
simulate high resolution satellite data. 
First, preprocessing, image restoration and geometric correction 
have been applied to the satellite data. Cubic convolution 
resampling has been applied for the panchromatic data with the 
intention of smoothing the image. 
2.3. Image selection and applied sensor fusion techniques 
After the description of the theoretical background of image 
fusion in section 1.3, the applied sensor fusion techniques on 
different input data and the final image selection for the visual 
interpretation will be given here. 
First, the panchromatic channel of IRS-1C from 14 th October 
1996 alone has been taken as input for the visual delineation of 
forest stand borders. A linear contrast stretch has been applied 
for the area under the forest mask. 
Another image was a BAV orthophoto scanned with 254 dpi. As 
the scale of the orthophoto is 1:10,000, the scanned image has a 
pixel footprint of lm. The orthophoto has been chosen, because 
the actual forest inventory is based on its analog version. The 
scan density of 254 dpi enables the interpreter to distinguish 
sufficient details and the data handling is well manageable. 
The first combination of sensors has been applied to the dataset 
of Landsat 5 TM of 20 th June 1995 and the panchromatic IRS- 
1C of 14 th October 1996. They have been combined with three 
different pixel-based sensor fusion algorithms: IHS colour 
transformation, principal component substitution and Brovey 
transformation. For the merging, the TM bands 5, 4, and 3 have 
been selected as input, because they comprise the highest 
spectral information. A visual comparison of the three different 
fusion methods showed best results for the IHS transformation. 
From SPOT the multispectral scene of 26 th July 1995 and the 
panchromatic scene of 20 th July 1989 were used. These images 
have on one hand a long acquisition time difference, but on the 
other, they are only separated by 6 days. In addition, the 
panchromatic scene is an early data-take from SPOT 1 with 
excellent radiometric characteristics. For this image set, 
different fusion techniques are combined: first the AIF has been 
applied and then an IHS transformation of the filtered result 
with the SPOT pan as high resolution input has been calculated.