d NUS Sn 
Figure 3: ERS-2 Intensity image 
(information provided by ESA). The lengths of these 
components satisfy the wish to have a limited baseline 
decorrelation. 
The window that is selected from the images reflects an 
area in the central part of the Netherlands. The area 
consists of a relative new polder with large agriculture fields, 
a developed moor land including reed fields, an area with 
small agriculture fields on the "old" land and parcels with 
forest. The old and the new land are separated by a lake. In 
the area there are no relief differences, it is a very flat 
terrain. 
2.2 Field data collection 
In order to classify the images, samples were taken that are 
supposed to be representative of the various types of land 
cover, and, at the same time, in order to check the accuracy 
and reliability of the classification another independent set 
of samples was acquired. Field data was acquired a few 
weeks later than the image acquisition dates. The eight 
classes that are selected did not change between the 
acquisition date and the field check. Only some grass field 
may have been cut in that period and some stubble fields 
plowed. 
The following classes are recognized : 
. Bare soil. 
. Sugar beets. 
. Stubbles. 
. Forest. 
. Maize. 
. Grass. 
. Water. 
. Reed. 
o 100120 mw -— 
2.3 SPOT georeferencing 
The Spot image is georeferenced using the ILWIS image 
processing software applying an affine transformation 
including a nearest neighbor "interpolation". The overall 
accuracy, expressed in RMSE, of the reference points used 
for the creation of the transformation polygons was less 
than 1 pixel. 
2.4 ERS1/2 intensity images 
From the SLC images, intensity images are created. The 
one with the highest number of clearly visible objects (in this 
case the ERS-1) is selected to support the registration of 
the coherence map to the SPOT image. 
2.5 Coherence map creation 
The coherence map (figure 4) is created using software 
developed at DLR in Oberpfaffenhofen, Germany. The 
interferometric software requires some information which is 
available in the leader file. 
    
Figure 4: Coherence map 
For an accurate registration of two SLC data sets it is 
necessary to measure the shifts between the images in 
azimuth and range direction which is done in the intensity 
images. These shifts are introduced in the data processing 
for a coarse registration. The WGS84 is chosen as 
reference system for the resampling of the second scene. 
To inspect the coarse registration, the fringes and the 
coherence map are calculated from the coarse registered 
images. If the registered images cannot produce a sufficient 
number of fringes, the coarse registration has to be 
repeated with more accurate shifts. The next step is the fine 
registration of the images. This is one of the most important 
steps during the whole data processing because it is the 
basis for the quality of the later derived interferometric 
products. The master image remains unchanged while the 
slave image is fitted to the master. To reduce unwanted 
variation in the coherence map, a window of 20 azimuth * 2 
range elements is used for the computation of the 
correlation. Based on this filtered map the interferogram is 
calculated, i.e., the resulting fringes are also corrected 
under assumption that the Earth is flat. 
2.6 Coherence map registration 
The coherence map and the ERS-1 image have apart from 
minor shifts, which are caused by the slightly different orbits 
of the ERS-1 and the ERS-2, the same geometry. So 
registration of the coherence map and ERS-1 image will 
follow the same transformation steps. The following steps 
are performed, partly with ILWIS and partly with PCI's 
EASI/PACE. 
- resample the intensity image to an azimuth resolution of 
20 meter by averaging five columns elements into one. 
- select the sub image of the ERS-1 that covers 
approximately the same area as the SPOT sub image. 
- transform the image from slant range into ground range 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996 
1 
as Ar 
1 
p — rm 
3.1 | 
Opti 
For 
banc 
the | 
were 
the | 
the 
map 
the r 
take 
stati: 
Yi Pe 
were 
Mah 
were 
to be 
and 
shov 
Opti 
The 
clas: 
Only 
In ta 
3.21 
Mas 
Neu 
auto 
appc 
So a 
class 
crea 
clas: 
Opti 
For 
(vers 
inter 
layei 
optic 
com 
hidd 
(199 
using 
accu 
and