v 
es 
on 
he 
in 
Data 
mit 
of the 
erman 
nland- 
of this 
forest 
est is 
south, 
forest 
small 
of the 
  
Figure.1: Landscape unit in the south Palatine Forest 
Landinformation System 
For the spatial and spectral analyse of the remotely 
sensed data and for verification of classified images it 
was necessary to build up a landinformation system with 
reference data for the forest and non-forest areas. 
Based on the nomenclature table , which was developed 
together with the nature park administration and the 
technical project partners for the classification process a 
landinformation system was build up using forest 
management data from 1996 for forest areas and 
topographic and vegetation data. 
In addition main thematic information layers listed below 
were integrated in the landinformation system .: 
- geological map 1:200.000 
- map of climate 1:200.000 
- digital elevation model (DGM) 
- topographic maps 1:25.000 
- maps of biotops 1:25.000 
- forest map 1:25.000 
- forest soil maps 1:10.000 
- grassland maps 1:25.000 
- landuse maps 1:25.000 
- CORINE-landcover classification 
- administration units 
Based on this geo data the description of the 
investigation area and the selection of training data sets 
were carried out. 
Satellite Data 
For the research work two SPOT XS data sets from June 
1992 and April 1995were ordered. The data sets are 
characterised by a side looking data take. According to 
the view angle of -16° it was necessary to run an 
orthorectification on the data sets. 
In addition one LANDSAT TM data set from July 1994 
and one panchromatic IRS-1C data set from April 1997 
was provided. 
Pre-Processing 
The pre-processing the above listed data set were 
analysed regarding to radiometric errors. The TM data set 
( processed in Fucino, Italy) was characterised by un- 
systematically distributed double lines with the same 
radiometric values and a forward/reverse scan banding 
( Crippen, 1989). The correction of the unsystematically 
double lines was impossible with available algorithms and 
in an appropriate time. The use of the TM-Band- 
Correction algorithm based of the work of (Crippen, 1989) 
was not successful. Some parts of the dark banding lines 
were still overcorrected while other parts were sufficiently 
corrected. 
The banding of the IRS-1C PAN image was in 
comparison to the 1996 data strongly reduced but still 
visible in the data set. Based on the processing steps 
after data reception the IRS-1C data set is coarsely 
georectified to the German Transverse Mercator 
coordinate system, which rotates the banding lines in the 
IRS-1C image. 
Georectification 
The georectification of the data sets was carried out using 
a map-to-image rectification for the SPOT-XS data set 
from 1995. The other images were rectified by a image- 
to-image rectification. 
The map-to-image rectification with the spatially high 
resolution image was not possible according to the late 
delivery of the data set in December 1997. 
Based on 32 control points the satellite data sets were 
rectified. The errors of rectification were calculated with 
0.3 to 0.5 pixels for the SPOT- and LANDSAT data sets 
and with 0.7 pixels for the IRS-1C PAN data set. 
Topographic Normalisation 
Topographic Normalisation on the cosine basis was 
applied. Due to the overcorrection of shadow areas this 
was rejected and substituted by illumination masks. 
For the classification 4 data sets from LANDSAT TM (July 
1994), SPOT XS (June 1992, April 1995) and IRS-1C 
PAN (April 1997) were available. 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998 371