Ms from
A. In the
d on the
age pair,
includes
/ and the
se of the
of view.
e at JPL
imber of
iceborne
e highly
ia, ETH
oncerted
ject was
id DEM
as made
ration of
used on
ating the
ehensive
sis of the
sts of the
Michele Crosetto
- generation of the irregular grid of 3D points
- interpolation of the regular grid
The first three processing stages are based on the ISAR-Interferogram Generator software (distributed, free of charges,
by ESA-ESRIN), an effective tool to obtain good filtered interferograms and the related coherence images (Koskinen,
1995). The phase unwrapping is based on the so-called “branch cuts” approach (Goldstein et al., 1988). The most
original parts of the procedure are the rigorous model for the conversion from interferometric phases to terrain heights
(used in the generation of the irregular grid of 3D points) and the calibration of the InSAR geometry (Crosetto and
Crippa, 1999). The calibration is based on ground control points (GCPs), where either full GCPs or height GCPs may
be used. It allows achieving an accurate geolocation of the INSAR generated DEMs. The implemented procedure allows
fusing data coming from multiple InSAR pairs (e.g. ascending and descending pairs). In fact, it includes the
simultaneous calibration of the geometry of different InSAR pairs based on the use of tie points (in full analogy with
the procedures adopted in photogrammetry).
An important aspect of InSAR is the influence of atmospheric effects on the generated DEMs. These effects are mainly
due to variations of atmospheric relative humidity between two SAR image acquisitions (Hanssen, 1998). They result in
artefacts (e.g. depressions) in the generated InSAR DEMs interpreted as terrain relief. A single SAR pair can not check
the presence of such artefacts, and this represents a very important limit of the InSAR technique. In order to reduce the
influence of atmospheric artefacts, we adopt a strategy based on the use of auxiliary low-resolution height data (e.g.
with a resolution 10 times lower the one of the InSAR DEMs), see (Crosetto, 2000). Firstly, the InSAR and auxiliary
data are accurately geolocated with respect to the same reference system. The fusion procedure employs a
multiresolution data analysis in the space domain adopting two resolution levels: the first one corresponds to the high
frequency components of the terrain topography contained in the InSAR data and the second one corresponds to the low
frequency components contained in the auxiliary data. The output DEM contains the high frequency components of the
original InSAR DEM and the low frequency components (not affected by atmospheric effects) of the auxiliary data. The
effectiveness of the atmospheric distortion compensation is shown in the analysis of the InSAR DEM quality.
2 ANALYSIS OF THE RESULTS
Over the ORFEAS test site different DEMs were generated using stereoscopic techniques (with SPOT and Radarsat
stereo pairs) and InSAR procedures (ERS-1 interferometric pairs). The DEMs generated with our procedure were
validated using a suited reference DEM (coming from aerial photogrammetry) whose precision is one order of
magnitude better than that obtainable by InSAR DEMs. All DEMs analysed in the following cover the same area
(approx. 25 by 35 km), which includes the flat plain crossed by the Ebro River and a set of mountain chains (the
maximum height difference is about 1150 m). From the viewpoint of SAR images, this area includes many portions
affected by foreshortening, layover and even shadow effects.
2.1 Ascending Image Pair
The characteristics of the ascending SAR image pair chosen for the processing are summarised in Table 1. The baseline
length is about optimal for InNSAR DEM generation and due to the quite high coherence a good interferometric phase
quality can be expected. From the ascending pair we generated a 30 m spacing DEM that was compared with the
reference one (see statistics of the processing type “without atmospheric corrections” in Table 2). The global bias (mean
error) of the grid can be considered satisfactory, i.e. the calibration with 14 GCPs resolves quite well the geo-location of
the generated 3D grid. One may notice an important decrease of the DEM precision between flat and mountainous areas
(where unwrapping errors occur). In the following three important aspects of INSAR DEMS are discussed: the influence
of the image coherence, the degradation of the DEM quality in mountainous areas and the atmospheric effects.
Acquisition Date 12 and 15 September 1991
Baseline Length 161.5 m
Sub-image range dimension 1500 pixels
Sub-image azimuth dimension 5000 pixels
Mean coherence of the SAR filtered images | 0.57
InSAR geometry calibration 14 Ground Control Points
Table 1: Characteristics of the ERS-1 ascending InSAR pair.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part Bl. Amsterdam 2000. 47
