The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008
977
thousands for Magellan) and random rather than north-south
orientation made it more convenient to store the beam-and-burst
information as a raster map in the same projection as the image,
rather than as a tabular database. The needed ancillary data for
each burst, including both its footprint boundary and the space
craft position and velocity, are obtained from a binary table
known as the SBDR (Stiles, 2008a). A final difference between
the Magellan and Cassini processing comes about because there
is no pre-Cassini topographic information for Titan. Cassini
BIDRs are therefore projected onto a spherical reference
surface rather than onto a low-resolution DTM. Because the
equations of projection onto a sphere can be expressed
analytically, no resampling coefficients are required. Care is
needed, however, to use the adjusted spacecraft position and
velocity to calculate range-Doppler coordinates from the
ground point location, but to use the original position and
velocity estimates to calculate map coordinates from range and
Doppler, in order to be consistent with the way the BIDRs were
generated.
Our approach to processing the Cassini RADAR stereopairs has
also been informed by our experience with Magellan. In
particular, the practice of “seeding” the automatic matching
process with a loose set of surface points selected interactively
has once again been shown to reduce the need for final editing.
The initial mapping results reported below were obtained
without any bundle adjustment, but we expect that as we
analyze a larger number of overlap areas, some adjustment will
be necessary to achieve consistent results at the sub-kilometer
level of precision.
3.3 Results
Evidence about the topography of Titan prior to our beginning
radargrammetric mapping with SOCET SET came from a
variety of sources, all of which suggested that both local and
global relief is low, with elevation variations greater than about
1000 m rare. Radarclinometry (shape-from-shading) initially
revealed only a few hundred meters relief in areas where it
could be applied (Kirk et al., 2005) though more recent results
reach 1500-2000 m for some mountains (Radebaugh et al.,
2007). Topographic profiles have been obtained over a limited
number of short arcs by operating the RADAR as an altimeter
(Johnson et al., 2007) and over longer arcs by an ingenious
method that compares the signal strength from adjacent
overlapping beams to determine heights along each SAR image
(Stiles et al., 2007). The profiling methods agree well where
they have been compared (Gim et al., 2007), and both show
relief of a few hundred meters or less. Finally, preliminary
estimates of topography from stereo, again indicating relief of
hundreds of meters (Kirk et al., 2007) were based on automated
image matching by Scott Hensley at JPL and on manual
parallax measurements by us, but in either case a simple
parallax-height scaling based on the two radar incidence angles
was used in lieu of a rigorous sensor model to estimate relative
elevation differences. Hensley has since implemented a
rigorous Cassini RADAR sensor model for his Magellan-
derived matching software (written communication, 2007).
Our plans to map the complete set of RADAR overlap areas
now that a rigorous sensor model is available for SOCET SET
have been delayed somewhat by the discovery of substantial
(up to 30 km) positional mismatches between many of the
image pairs that would introduce spurious parallax and/or
prevent stereo matching altogether. These offsets have been
traced to the need for an improved model of Titan’s rotation,
and have been reduced to sub-km levels by adjusting the
orientation of the spin axis, the rotation rate, and the first
derivatives of these parameters (Stiles et al., 2008). The
nonsynchronous spin rate, in particular, implies that the ice
crust of Titan is decoupled from the deep interior by a
subsurface liquid water “ocean” (Lorenz et al., 2008b)—an
interesting example of a significant geophysical discovery
arising from a routine use of radargrammetry to improve
cartographic products. Reprocessing of the complete set of
Cassini BIDRs based on the new rotational model will be
completed in the late spring or early summer of 2008.
Meanwhile, several overlapping images obtained in 2007
February to April were made at the same rotational phase, so
that the misregistration caused by using the older rotation
model is negligible. Fortunately, the overlap between these
images covers one of the most interesting regions of Titan, an
area of extensive dark areas interpreted to be lakes and seas
(Stofan et al., 2007) near the north pole.
Figure 3. Color-coded topographic map of part of Titan’s north polar “lake country” based on stereoanalysis of RADAR images
from flybys T25 and T28. Polar stereographic projection, north approximately at top. Island at right center is Mayda Insula,
discussed in text.
Figure 3 shows our topographic model of the overlap between
the images from the T25 (2007 February 22) and T28 (2007
April 10) flybys. Both flybys illuminated the area from the
south, yielding same-side stereo with vertical precision
typically on the order of 100 m for single pixel (175 m)
matching error. The procedures for making this DTM were
tested by initial mapping of Mayda Insula, a 90x150 km island
centered near 78° N 312° W (Kirk et al., 2008c). Bundle
adjustment was not needed for this data set, because cross
stereobase misregistration was less than one pixel, and the
unadjusted stereo elevations along the SAR topography profile
agreed at the 50-100 m level with the absolute elevations of
the latter data set. We note that that this agreement is obtained
even where the two incidence angles are similar, at the east end