SAR INTERFEROMETRY: A COMPARATIVE ANALYSIS OF DTMs 
Lado W. KENYI and Hannes RAGGAM 
Institute for Digital Image Processing, JOANNEUM RESEARCH 
Wastiangasse 6, A-8010 Graz, Austria. 
Commission IV, Working Group IV/2 
KEY WORDS: SAR Interferometry, DTMs, Fringe Smoothing, Comparative Analysis. 
ABSTRACT: 
In this paper results of a comparative analysis of interferometrically derived DTM and topographic map digitised DTM 
are presented. The RMS error of the INSAR generated DTM was about 11 meters, while that of the control points was 
9 meters. The maximum value was found to be 50 meters which has been attributed to atmospheric effects. The 
INSAR DTM was found to be shifted by 2.3 meters in general. 
1. INTRODUCTION 
SAR interferometry (INSAR) is a recent promising 
technique for the application of remote sensing data. It 
allows the production of detailed and accurate three 
dimensional relief maps of the Earth's surface directly 
from two SAR complex image data that can be acquired 
simultaneously by two SAR receivers in a single pass or 
by one SAR receiver at different times in multiple passes 
(Prati et al. 1992) and (Zebker et al. 1994). The 
technique can also be used to detect very small 
movements of land surface features in the cm-range, 
which is known as differential interferometry (Massonet 
et al. 1993). However, there are limitations in the 
practical exploitation of the data in the multiple pass 
(repeat orbit) case, which are: geometrical decorrelation 
due to the imaging geometry and imposes limits on the 
baseline; and temporal decorrelation caused by the 
physical changes in the imaged earth surface which show 
up as incoherency between the SAR images (Zebker and 
Villasenor | 1992). Although, satisfactory results 
concerning the validation of interferometrically derived 
digital terrain models (DTM) have been reported in the 
literature (Zebker et al. 1994), a comprehensive analysis 
of the results especially the reproduction of the results 
from the same test area with different data sets and the 
comparison of the fringe smoothing filters on the 
accuracy of the results have not been performed. 
Therefore, the intention of this paper was to present 
results of a comparative analysis of interferometrically 
derived DTMs and topographic maps digitised DTMs of 
two different test sites. The work was also to include the 
assessment of two fringe smoothing filters, namely the 
moving box averaging and the directional adaptive 
Gaussian filter (Geudtner et al. 1994), on the accuracy of 
the DTMs. But, due to lack of results by the time of the 
deadline for the submission of the papers for inclusion in 
the proceedings, only the results of an INSAR DTM 
generated for one of the test sites is presented. The full 
results, however, will be orally presented at the congress. 
2. TEST AREA AND DATA 
2.1 Test Areas 
Basicaly, two test areas have been treated. One 
covering a more flat area to the west of the city of Bonn 
in Germany, the city of Weilerswist, and the second is 
also a flat terrain area around the city of Dortmund in 
Germany too. Reference digital terrain models with a cell 
size of 50 meters digitised from topographic maps in a 
scale of 1: 50000 of these areas were available. They 
were resampled to cell sizes of 40 x 40 meters for 
matters of comparison with the interferometric products. 
2.2 INSAR Data 
For the interferometric test data, suitable baselines from 
ERS-1 phase B and D data were selected. For the 
Weilerswist area one phase B interferometric pair was 
available. The pair was acquired on the 14 and 29 March 
1992 corresponding to ERS-1 orbits 3459 and 3674, 
respectively. Whereas, for the area around the city of 
Dortmund 4 phase D interferometric quarter scenes were 
selected. The acquisition dates for these images were 31 
December 1993, 03 January 1994, and 13 and 16 March 
1994, which correspond to the respective ESR-1 orbit 
numbers 12864, 12907, 13896 and 13939 in the frame of 
the area. The baselines for these scenes were relatively 
small ranging from 60 - 160 meters. 
3. INSAR PROCESSING 
The INSAR processing starts with the co-registration of 
the images to a subpixel accuracy of 1/30. This is 
achieved by first correlating patches of 25x25 pixels to a 
pixel accuracy and by a subsequent surface fitting in à 
3x3 window around the maximum point in order to obtain 
subpixel accuracy. The correlation measure used is the 
complex correlation function, which sensetive to fringe 
visibility. This process is repeated for a number of points 
covering the whole image, where only those points 
showing high correlation values are considered. After the 
442 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996 
  
offs 
oth 
Ne» 
inte 
Sel 
pha 
surf 
sub 
ther 
pixe 
ada| 
dire 
unw 
disp 
pha: 
The 
rang 
refer 
infor 
infor 
see | 
The 
from 
were 
aroul 
exce 
orbit 
implie 
(parti 
prese 
Norm 
follow 
image 
image 
test ( 
obser 
deviai 
(Keny 
fringe 
anom 
Howe 
fringe: 
small 
featur 
flatten 
13896 
In figu 
area 
compe 
lines o 
shows 
coding 
order 
differe 
shows