ation
f the
1etric
te of
f the
been
ram-
SAR
ance
ising
wide
gen-
shed
nter-
hape
ques
radar
one
POT
tech-
f the
ysed.
used
ning
data
d its
sing
The
and
SAR
rent
gs a
n of
nore
and
The InSAR procedure implemented by the authors
includes the following steps [Crippa et al. 1998]:
1) acquisition of an interferometric image pair;
2) precise image registration;
3) calculation and filtering of the interferogram;
4) unwrapping of the interferometric phase;
5) sensor parameter calibration and DEM generation
(i.e. transformation from phases to heights and
geocoding of the DEM).
Phase unwrapping remains the most complex problem
of the entire procedure; it greatly influences the qual-
ity of the generated DEM. The last step, and in par-
ticular the sensor parameter calibration (based on
GCPs, Ground Control Points), is very important to
get the final INSAR product (i.e. the geocoded regular
grid of 3D points) and hence to assess its quality.
2.1.1 Characteristics of InSAR DEMs
The InSAR technique can generate DEMs of good
quality (i.e. high spatial resolution, e.g. 30 m mesh
size, and good accuracy), assumed at least a medium-
high coherence (e.g. bigger than 0.5) over the entire
interferogram and gentle terrain variations within the
covered areas.
Dealing with more complex topography or with low
coherence, many problems arise. In fact, the slant
range nature of the SAR data implies big distortion
effects (foreshortening, layover and shadowing) when
mountainous and hilly terrain is imaged. Where fore-
shortening and layover occur, the interferometric
phase signal is under-sampled, producing aliased
phase differences between adjacent pixels. If in the
phase unwrapping the lines of aliasing (called ghost-
lines) are not properly detected, the unwrapping
(based on phase difference integration) generates
aliasing errors (multiple of 2x). These errors degrade
the DEM quality (e.g. with a baseline of 150 m, an
aliasing of 2x in the phase results in about 50 m
height error in the generated DEM ).
Changes in the terrain surface during the two image
acquisitions can cause low coherence in the inter-
ferometric pair; low coherence (e.g. less than 0.1)
means bad phase quality and can engender many
problems for the phase unwrapping. In such areas the
DEM quality is degraded.
In the following, an example of InSAR generated
DEM is shown. The Polytechnic of Milan is involved
in an European Union Concerted-Action called OR-
FEAS (Optical-Radar sensor Fusion for Environ-
mental ApplicationS), including several European
research groups (University of Thessaloniki, Carto-
graphic Institute of Catalunya, ETH Zurich, Technical
University of Graz and Polytechnic of Milan). An
interesting data set, covering south Catalunya - Spain,
(ascending and descending ERS-1 SAR images,
SPOT images, orthophotos, reference DTM, land-use
map, etc.), is available for ORFEAS participants. All
International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998
the results presented in this paper have been obtained
through the ORFEAS data set.
Two ascending ERS-1 sub-images (1500 pixels in
range by 5000 pixels in azimuth) with a baseline
length of about 160 m have been processed. The
mean coherence over the entire (filtered) interfero-
gram equals 0.57 (i.e. the interferometric phase qual-
ity is globally quite good). The generated DEM (30 m
mesh size) covers an area of about 35x25 km?. The
maximal height difference is approximately 1120 m.
Comparing the InSAR DEM with the reference one
(coming from aerial photogrammetry with a RMS
error of about 1 m) gives:
Mean error =0.9m
Standard deviation = 18.8 m
Maximal abs. error =230.1m
In the error map (i.e. the map of the height differences
between the generated and the reference DEMs, see
figure 1) are clearly visible the areas where big height
errors occur. These are mainly mountainous areas
where the phase unwrapping fails, originating aliasing
errors.
2.2 SPOT stereoscopy
The stereoscopy based on optical images is a well-
established technique to generate DEMs. It is em-
ployed operationally both with aerial images (aerial
photogrammetry) and remote sensing images (e.g.
SPOT images) This kind of DEM generation is
highly automated; see, for instance, [Chen and Rau
1993]. The homologous points are matched using
image correlation techniques. The human operator
performs only the measure of GCPs.
The quality of the SPOT generated DEMs is not
strongly dependent on the terrain topography (the
InSAR DEM quality, on the contrary, depends very
much on the topography).
A SPOT derived DEM generated at the Institute of
Geodesy and Photogrammetry - Zurich Institute of
Technology (Switzerland) has been processed.
The original data coming from Zurich consist of a
regular grid of 3D points generated with the Helava
Digital Photogrammetric Workstation (DPW) 770; to
each point the DWP 770 assigns a quality factor.
According to this factor, the unreliable points have
been eliminated. Using the same interpolator and the
same grid of InSAR (see previous paragraph), the
generated SPOT grid has been compared with the
reference one:
Mean error = 10m
Standard deviation = 12.2m
Maximal abs. error =367.5m
The DEM is not biased. The error distribution does
not present systematic errors.
699
