3D 
nce 
lave 
nce 
nts, 
the 
that 
with 
adi- 
are 
ach 
the 
'een 
inal 
TOrs 
ints 
the 
cing 
  
  
  
E + 80 m 
co -50m Im +50m 
[31-90m [B Om 
  
  
Figure 2: DEM (InSAR and SPOT data fusion) versus reference DEM: map of the height differences. 
The error map is shown in figure 2. The accuracy of 
this DEM is much better than the one of the InSAR 
DEM (especially in mountainous areas). On the con- 
trary, it is very similar to the one of the SPOT DEM. 
It is important to notice that only a portion of the 
SPOT stereo pair has been used for the data fusion. 
Clouds cover other portions of the same pair: for 
these areas the data fusion would be actually effective 
to improve the accuracy and the completeness of the 
generated DEM. 
3. Data Fusion for Land Use Classification 
The potential of multi-temporal SAR amplitude im- 
ages for Land Use Classification and agricultural crop 
mapping has been investigated in the last years at the 
Olsztyn University (Kurczynski and Mr6z 1998). 
The classification procedure has been recently ex- 
tended to optical and interferometric SAR data. The 
very first results, presented in this paper, have been 
obtained fusing an amplitude SAR image, a coher- 
ence image and a SPOT panchromatic image in order 
to perform the Land Use Classification over the same 
test area used for the DEM generation. All the used 
data belong to the ORFEAS data set. 
The data set (3 layers) used for the classification is 
rather poor. Up to now, the work has been mainly 
focused on the procedure to fuse the InSAR and 
SPOT data that involves the geocoding of the images. 
The classification results should improve sensibly 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998 
using more layers (e.g. time-series of coherence im- 
ages, multi-spectral SPOT images, etc.). 
In next paragraphs, the data used for the classification 
are briefly described; the classification procedure is 
explained and, finally, the results are analysed. 
3.1 Description of the Input Data 
The Land Use Classification procedure implemented 
by the authors is based on two complex SAR images 
and a panchromatic SPOT images. The complex SAR 
images have been used to estimate the interferometric 
correlation (coherence) and the backscatter amplitude 
of one of the two images. 
Assessing the applicability of multi-source data syn- 
ergy, the creation of data sets in a common ground 
reference system has to be performed. The geocoding 
can be defined as the procedure to convert data from 
the imaging geometry into a map projection. 
A large number of people in Remote Sensing still 
perform geocoding using very simple transformation 
functions (e.g. polynomial, kriging, etc.), not consid- 
ering the effects related to the terrain relief. Very 
often this simplified procedure has a big impact on 
the data fusion and feature extraction, because the 
resulting errors create many ‘mixed pixels’ in the 
radiometry of the composite image, thus generating 
wrong information or artefacts which do not corre- 
spond to the physical realities. An accurate geocoding 
requires a precise and robust geometric processing 
701