4.2 Coping with Layover 
Since the polar orbit of Magellan was highly elliptical, the 
probe was much closer to the planet when passing the 
equator than over the poles. To obtain image-strips along 
the meridians, the inclination of the radar-beam was 
adjusted during image acquisition. So, the initial Cycle 1 
had images taken at 45° off-nadir near the equator and 
about 11° near the poles. The same applies to Cycle 2 
images, taken in similar way from the opposite direction. 
The stereo partners to the initial coverage were acquired 
with Cycle 3 and had look-angles ranging between 25° 
off-nadir at the equator to 7° off-nadir near the pole. This 
resulted in stereo-intersection angles between 20° and 4°. 
While these angles are fairly small and represent a 
limitation to the accuracy at which a DEM can be 
extracted from the SAR images, there is another issue 
caused by the steep  look-angles: layover. In 
topographically accentuated terrain a very large portion of 
the slopes may be laid over. As shown in Figure 3, such 
layover areas do not lend themselves easily to the 
reconstruction of surface slopes and elevations. Connors 
(1994) has shown that it is in principle possible to employ 
overlapping stereo coverage to determinate whether a 
layover situation exists or not. There is an ambiguity 
which of two possible slopes may cause a particular image 
situation. The ambiguity can be resolved with the third 
opposite coverage. Gelautz et al. (1996, in print) are 
working on automating the process used by Connors 
(1994) in a manual manner. This would lead first to an 
identification of layover areas in all of Magellan's 
coverage; and it would secondly use the layover to 
improve the slope measurement of the terrain that is giving 
raise to the laid-over images. 
  
Figure 3: Example of laid-over terrain features in Magellan 
images taken at 11? look-angle off-nadir. 
Area covered is 15 km x 2 km. 
494 
4.3 A System for Reprocessing Stereo Data 
The use of the Magellan Stereo Toolkit, the elevation and 
slope errors caused by erroneous ephemerides and the 
problems arising from laid-over features cause us to 
suggest that extraction of a detailed topographic relief 
from Magellan images should be based on a complete 
processing chain. It should begin at the raw signal 
histories received at the ground receiving stations on Earth 
from the satellite. We describe a sequence of procedures to 
accomplish an optimum extraction of topographic relief. 
(a) The ephemerides need to be reprocessed in the manner 
described by Chodas et al. (1992). A total of about 
5,000 orbits are at issue. For each orbit a number of 
tie-points needs to be identified between an image of a 
particular orbit and images from other cycles. Perhaps 
10 to 20 tie-points would be needed per orbit. The 
improvement of the ephemerides can also take 
advantage of the most recent gravity model of the 
planet. Its quality was vastly improved when the orbit 
was circularized just prior to the satellite dipped into 
the atmosphere and perished. That accuracy may be 
sufficient to obtain an ephemeris as accurate as that 
which could be obtained with tie-points in a type of 
photogrammetric block adjustment. 
(b) Reprocessing raw signal histories can be based on the 
improved ephemeris and produces full-resolution 
image strips. These will not be corrected for 
topographic relief obtained from altimetry. 
(c) Stereo matching uses new images as each orbit of 
Cycle 1 crosses over orbits of Cycle 3. Such stereo 
matches should have errors in the range of open 0.6 to 
2 pixels depending on the type of features and the 
dissimilarity between the images (Leberl et al., 1993). 
Various authors have argued that image matching 
should be performed on the mosaics that are currently 
being processed in the form of F-MAPs. Match points 
in each image could then be converted to time, 
Doppler frequency and range which could then be 
attached to the improved ephemeris. This proposal 
would skip step (b) until such time that the stereo- 
derived topographic relief has become available. But 
in that event one would not use the best and highest 
resolution images for matching. Given current parallel 
processing technologies one could argue that going 
through all signal history records and creating a new 
intermediate set of 5,000 full-resolution images is no 
longer the monumental task it was during the Mission. 
(d) Intersecting surface XYZ-values is based on the 
Doppler frequency and range measurements of 
homologue features in Cycles 1 and 3. This produces 
surface locations in XYZ at an accuracy of the 
ephemeris. In areas where no stereo observations can 
be made one can employ the altimetry data that were 
independently obtained from a vertical looking 
antenna (Ford et al., 1994). 
(e) Gridding converts the stereo-derived surface points at 
irregularly spaced position into a regular grid. This is 
based on interpolations; such points are regularly 
spaced in latitude/longitude or in a map projection. 
Elevations obtained in this way can also be resampled 
to be attached to the individual image pixels. 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996