EH +100 m 
[1-50m pgm +50m 
[7] -100m Om 
  
  
Figure 1: InSAR DEM versus reference DEM: map of the height differences. 
With the exception of very steep terrain (where prob- 
lems for stereo matching occur), the obtained accu- 
racy is very high. This is probably due to the well- 
textured images (that means good image quality for 
the point matching) and to the absence of cloudy 
zones in the imaged scene. In fact, the presence of not 
well textured areas or cloudy zones in the imaged 
scene represent the major degradation factor for the 
SPOT DEMs. 
Dealing, for instance, with SPOT images corrupted 
by clouds, in the corresponding DEM appear “holes” 
that deteriorate the DEM quality. In this case, the 
fusion with other kind of height data can be very 
effective. 
2.3 Data Fusion Procedure 
The integration of SPOT and InSAR derived height 
data for DEM generation can offer a solution to over- 
come the above-mentioned limitations of the two 
techniques. In fact, SAR and SPOT data seem to be 
quite complementary: in rugged areas stereo SPOT 
can deliver quite good height data (on the contrary, in 
these areas InSAR data are usually much less accu- 
rate); InSAR can work good in areas where SPOT 
images can be corrupted by clouds; etc.. 
The InSAR procedure implemented by the authors is 
quite flexible to allow the integration (fusion) with 
SPOT derived height data. 
Each source of data (e.g. ascending SAR pair, de- 
scending SAR pair and SPOT stereo pair) is sepa- 
700 
rately processed to generate different sets of 3D 
points. The points are given in the same reference 
system (the GCPs used for SPOT and for InSAR have 
to be in the same reference system). A weight is as- 
signed to each point. 
The weight can be a function of the local coherence 
for the INSAR points and a function of the local im- 
age correlation for the SPOT points. All the points, 
with the relative weights, are used to estimate the 
final DEM grid (with an interpolation procedure that 
takes into account the point weights). 
For the grid interpolation, the terrain is modelled with 
linear splines (with a constraint on the surface gradi- 
ent in order to avoid oscillations). The splines are 
estimated by least squares adjustment so that to each 
estimated height its theoretical standard deviation can 
be associated. 
The joint estimation of the DEM grid based on the 
above-described InSAR and SPOT data has been 
performed. Comparing the InSAR and SPOT original 
grids, the InSAR zones affected by large errors 
(mountainous areas) have been masked. The points 
that belong to these areas have not been used for the 
data fusion. 
Performing the joint estimation of the 30 m spacing 
grid and comparing it with the reference one, gives: 
Mean error =12m 
Standard deviation = 12.3m 
Maximal abs. error =341.5m 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998 
this 
DEN 
trary 
It is 
SPO 
Clou 
thes: 
to in 
gene 
3.I 
The 
ages 
map 
Olsz 
The 
tend 
very 
obta 
ence 
top 
test 
data 
The 
rath 
foci 
SPC 
The