■■■i
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008
976
model was adopted. Mapping to the east of the Joliot-Curie
quadrangle revealed that the quality of the Cycle 3 images
declined dramatically toward the end of the cycle in terms of
both signal-to-noise ratio and data dropouts, increasing the need
for interactive editing by a factor of several.
A particular concern of the geologists interested in using
Magellan stereo DTMs was whether artifacts could occur at
high-contrast boundaries, either directly by effects on the SAR
images or as a result of the errors that such boundaries were
known to induce in the altimetry used for vertical control. We
addressed these issues by test mapping of a 10°x3° region in the
rugged highlands of Ovda Regio containing sharp boundaries
between high and low radar backscatter. These boundaries are
caused by a temperature-dependent change in the equilibrium
mineral phases on the surface, so they are expected to form at
nearly constant elevation. The images clearly show the surface
to be smoothly sloping near the boundaries, but the altimetry
contains artifactual “pits” as deep as 3 km. We found, as
expected, that our smooth (linear with time) adjustment to the
spacecraft ephemerides did not allow the stereo DTM surface to
deform to follow altimetry artifacts. Vertical control points
placed in the “pits” were readily identified as outliers, and the
stereo DTM indicated highly consistent elevations along the
contrast boundary, with variations of -200 m locally, compared
to -500 m variation previously estimated by mapping with
uncontrolled images in MST (Arvidson et al., 1994). The
constancy of the transition elevation provides as good a test of
the precision of our stereo mapping as is likely to be obtained
until a future mission supplies higher resolution data.
2.4 Lessons Learned
Overall, our Magellan experience showed that digital stereo-
grammetric processing techniques, including automatic image
matching, could be applied successfully to planetary SAR data,
and that a custom sensor model could be used to bundle adjust
and work with images that had already been map projected and
even mosaicked. Our tests showed the utility of a “seed” DTM
for automatic matching and the value of opposite-side radar
stereopairs for mapping areas of subtle relief, and suggested
what may be a general rule, that accurate mapping requires both
the best available reconstructed ephemerides and further bundle
adjustment based on image tiepoints. Finally, we learned not to
underestimate the additional complications and difficulties that
arise in the mapping of each new area of Venus.
3. TITAN
3.1 Cassini Mission and Data
The Cassini-Huygens mission consists of the NASA Cassini
spacecraft, which went into orbit around Saturn in 2004, and the
ESA Huygens probe carried by Cassini, which landed on the
giant satellite Titan in 2005. Investigation of Titan, which is
larger than the planet Mercury and wrapped in a smoggy
nitrogen atmosphere four times denser than Earth’s, is a major
objective of the Cassini instruments as well as the sole goal of
Huygens. Prior investigations of the organic chemistry of
Titan’s atmosphere raised the strong possibility of reservoirs of
liquid methane and ethane on the body’s surface, where they
might be expected to form lakes and even participate in an
exotic equivalent of Earth’s hydrologic cycle. The RADAR
instrument (Elachi et al., 2004) uses Ku band (2.17 cm X)
microwaves to penetrate the atmospheric clouds and haze, and
has provided the highest resolution images of Titan's surface
apart from very localized coverage from the cameras on
Huygens. On selected flybys of Cassini past Titan, the RADAR
obtains a SAR image strip 200-500 km wide and as much as
6000 km (130° of arc) in length. So far, approximately 25% of
Titan's surface has been imaged, at resolutions from about 0.3
to 1.5 km; a grid spacing of 175 m (1/256°) is used to ensure
oversampling of the data. Beginning in late 2006, most new
SAR image strips partly overlapped one or more earlier images,
and by the end of 2007, more than 20 image overlaps covering
more than 1% of Titan in stereo were available (Figure 2).
Operating in other modes, the RADAR also provides lower
resolution altimetry, scatterometry, and radiometry data.
As in the Magellan mission, the image strips are known as
BIDRs (Stiles, 2008b) and are made available in map projected
form. The pattern of Cassini flybys (Fig. 2) is much less regular
than the north-south arrangement of orbit strips on Venus,
however, so each BIDR uses an Oblique Cylindrical projection
oriented along its individual flyby ground track. Mosaics of
multiple BIDRs transformed to a common global map
projection are being made, but these are not useful for
stereoanalysis because there is no equivalent to the Cycle-1 and
Cycle-3 data sets of Magellan. Instead, we use the individual
BIDRs and map the overlap areas of pairs of them.
Figure 2. Mosaic of Cassini RADAR image coverage of Titan.
Polar Stereographic projections of the northern (left) and
southern (right) hemispheres with 10° parallels. Longitude 0° is
at the bottom on left, at top on right. Stereo overlaps with same-
side illumination and viewing are shown in green, opposite-side
in red, and high-angle overlaps in yellow. Total coverage to
date is -25%, total stereo -1%.
3.2 Methodology
Our approach to radargrammetry with Cassini data closely
follows that outlined above for Magellan, in broad outline using
ISIS to ingest and prepare the data and SOCET SET with a
custom sensor model written by us to perform bundle
adjustment, automated matching, and DTM editing. The inputs
for mapping are BIDRs in Oblique Cylindrical projection rather
than BIDRs, MIDRs, and FMAPs in Sinusoidal projection, but
the sensor model follows the same steps of identifying the
relevant radar burst for a given ground point, calculating the
range-Doppler coordinates of the point for that burst, and then
duplicating the transformation of range-Doppler into BIDR
pixel coordinates. The Cassini BIDRs must be treated as
mosaics, however, because they contain five parallel swaths of
data obtained by the five separate beams of the instrument. It is
thus necessary to identify first the beam and then the burst
whose parameters should be used in sensor model calculations
for any given pixel. The small number of BIDRs (tens versus