JTM
was
The
One
onn
d is
d in
cell
in a
hey
for
rom
the
Was
arch
)/4,
| of
lere
| 31
arch
rbit
o of
/ely
1 of
iS
0a
na
ain
the
1ge
ints
nts
the
offsets determination, one of the images is aligned to the
other by polynomial interpolation.
Next, the images are cross correlated to generate the
interferogram, which is a complex image by its self.
Selecting a flat terrain area in the image and performing
a 2D FFT on its interferogram, the azimuth and range
phase trends for a plane approximating the observed
surface are determined. These trends are then
subtracted from the interferograms. The interferogram is
then multi-looked by typically 10 pixels in azimuth and 2
pixels in range. The interferogram image is smoothed by
adaptive low pass filtering, i.e. moving box averaging and
directional gauss filter, before being unwrapped. For the
unwrapping the cuts placing algorithm was used. For
display purposes, the resulting wrapped and unwrapped
phases are also scaled to gray value images.
The unwrapped data are then converted to a ground
range DTM by selecting control tie points (GCP) from a
reference map of the area and the use of the orbit
information extracted from the CEOS format file header
information. For details of the DTM generation procedure
see (Kenyi and Raggam 1995).
4. RESULTS
The Weilerswist data, fringes could only be generated
from the first third of the scene. Therefor, the analysis
were only concentrated on this area. Luckily, for the area
around the city of Dortmund the coherency was
excellent, almost the whole quarter scene and all two
orbit pairs. Good quality fringes were generated which
implies generation of a DTM for the whole quarter scene
(partial quarter scenes INSAR DTMs are mostly
presented in the literature).
Normally, the wrapped or unwrapped phase images
follow the contour or shape of the topography of the
imaged terrain. However, as observed in the phase
images generated, it is not the case for Weilerswist area
test data. Deviations from the local topography were
observed in the fringe image of this area. These phase
deviations have been identified as atmospheric effects
(Kenyi and Raggam 1995). Figure 1 shows the flatten
fringe image of the Weilerswist area, where the phase
anomalies are clearly visible in the center of the image.
However, no such phase anomalies were detected in the
fringes of the area around the city of Dortmund; instead
small details such as artificial hills and man made
features could be clearly detected. Figure 2 shows a
flatten fringe of the dortmund area, namely orbit pair
13896 - 13939.
In figure 3, the INSAR generated DTM for the Weilerswist
area is shwon in a contour lines presentation. For
comparison a DTM was created from digitised contour
lines of a topographic map in a 1:50.000 scale. Figure 4
Shows this map-derived reference DTM in a contour lines
coding similar to that of the INSAR generated DTM. In
order to assess the errors in the INSAR DTM, a
difference of the two DTMs was computed. Figure 5
shows the difference DTM in contour lines presentation
443
too. Whereas, in tables 1 and 2 the statistical errors for
respectively, the generated DTM (GCPs) and the
difference DTM are presented.
5. DISCUSSION
For the entire area the standard deviation of the elevation
differences was about 11 meters. This is slightly worse
than the RMS height error of about 9 meters for the
GCPs (see Tables 1 and 2). Moreover, the
interferometric DTM is shifted on average by 2 meters in
comparison to the map derived DTM. The maximum
error was around 50 meters, whereas that of the GCPs
was about 14 meters. The 50 meters error turned out as
local deviations due to the atmospheric turbulence.
6. CONCLUSIONS
Although, work is still going on for the generation of the
DTMs from the unwrapped fringes of the Dortmund test
area by the time of preparation of this paper, it can be
concluded that even at relatively small baseline INSAR
derived DTMs gives height information with acceptable
errors. However, enough care must be exercise due to
the fact that atmospheric turbulence can introduce errors
of large magnitude in the INSAR height measurements.
Whereas, concerning fringe smoothing, the moving box
averaging filter performs best within largely spaced
fringes and conversely the Gaussian directional filter.
7. REFERENCES
Geudtner, D., M. Schwäbisch and R. Winter, 1994. SAR-
interferometry with ERS-1 data. In: Proceedings of
PIERS'94 Conference, Noordwijk, The Netherlands.
Kenyi, LW., H. Raggam and E. Kubista, 1996. Feasibility
of atmospheric effect in interferometric data and its
interpretation. ESA ESTEC-XEP Report, Prepared by the
Institutes of Digital Image Processing and Applied
Systems Technology, Joanneum Research.
Massonet, D., M. Rossi, C. Carmona, F. Adragana, G.
Peltzner, K. Feigl and T. Rabaute, 1993. The
displacement field of the Landers earthquake mapped by
radar interferometry. In: Nature, Vol. 364, pp. 138-142.
Prati, C., F. Rocca and A. Monti-Guarnieri, 1992. SAR
interferometry experiments with ERS-1. In: Proceedings
of 1st ERS-1 Symposium, Cannes, France, pp. 211-218.
Zebker, H., and J. Villasenor, 1992. Decorrelation in
interferometric radar echoes. In: IEEE Trans. Geoscience
and Remote Sensing, Vol. 30, No. 5, pp. 950-959.
Zebker, H., C. Werner, P. Rosen and S. Hensley, 1994.
Accuracy of topographic maps derived from ERS-1
interferometric radar. n: IEEE Trans. Geoscience and
Remote Sensing, Vol. 32, No. 4, pp. 823-836.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996