The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008
974
(PDS) Geosciences node and at the USGS in Flagstaff, Arizona.
For ease of use, mosaics of BIDRs with various scales and
formats covering areas ranging from 5°x5° to 120°xl20° were
made both by the Magellan mission and, later, by the USGS
(Batson et al., 1994). The mosaic series (known as MIDRs and
FMAPs, respectively) have received wide distribution though
the PDS and are available online (http://pds-geosciences.wustl.
edu/missions/magellan/index.htm, ftp://pdsimage2.wr.usgs.gov/
cdroms/magellan/) Both the BIDRs and the various mosaics
were prepared in Sinusoidal projection for most of Venus, with
additional projections used to represent the poles.
2.2 Radargrammetry Implementation
Numerous approaches to radargrammetric processing of the
Magellan images have been proposed (e.g., Hensley and
Schafer, 1994; Herrick and Sharpton, 2000). Although the
USGS briefly considered using an analytic stereoplotter to work
with hard copies of Magellan BIDRs (Wu et al., 1987), the
large volume (30 GBytes) of same-side stereo data makes a
digital or “softcopy” approach desirable if not essential. After
working with two such systems, the VEXCEL Magellan Stereo
Toolkit (MST; Leberl et al., 1992; Curlander and Maurice,
1993), and the SAIC Digital SAR Workstation-Venus (DSW-V;
Wu and Howington-Kraus, 1994) we set out to develop a
processing capability that would combine the best features of
each, the automated image-matching capability of the MST and
the geometrically rigorous sensor model of the DSW-V. To do
so, we made use of both the USGS digital cartography system
ISIS (Eliason, 1997; Gaddis et al., 1997; Torson and Becker,
1997; see also http://isis.astrogeology .usgs.gov), and the
commercial digital photogrammetric software SOCET SET (®
BAE Systems) (Miller and Walker, 1993; 1995). We use ISIS
to ingest the raw images, prepare them for use (e.g., by
decompression, radiometric calibration, geometric distortion
correction, as needed for a particular sensor), and export them
and their a priori orientation metadata in formats that can be
ingested by SOCET SET. SOCET SET then provides tools for
bundle adjustment to improve the geodetic control of the
images, production of digital terrain models (DTMs) by means
of flexible and continuously evolving algorithms for automatic
image matching (Zhang and Miller, 1997; Zhang, 2006),
display of the images and overlaid DTM data on a stereoscopic
monitor for interactive quality control and editing with point,
line, and area tools, and production of orthoimages and
orthomosaics. We normally export the DTMs and orthoimages
back into ISIS for final processing and analysis. This workflow
draws on the strengths of both systems (rapid in-house
adaptation to new planetary missions for ISIS; rigorous
stereogrammetric calculations and 3D display and user input
with special hardware in SOCET SET) and forms the basis for
our processing of numerous types of optical images from lander
cameras (Kirk et al., 1999) to orbit (Kirk et al., 2008a). It is
also the basis of our approach to processing SAR data from
multiple missions described here. The main difference is that
the three “generic” sensor models for different camera types
(frame, pushbroom, or panoramic) that are provided with
SOCET SET suffice to process the full variety of optical images
we have encountered so far. In contrast, each of the radar
systems described here has unique characteristics that require
the development of a separate sensor model. Although the
geometry of SAR image formation is the same in each case,
differences in how the data have been projected, combined, and
catalogued make it necessary to handle each case individually.
Sensor Model. Mathematically, a sensor model is a function
that specifies the transformation between image space (lines,
samples) and object or ground coordinates (latitude, longitude,
elevation). As implemented in software, a sensor model must
also include “bookkeeping” functions to obtain all the
information needed to carry out the mathematical transfor
mation and to communicate with the rest of its software
environment. The Developers’ Toolkit (DevKit) makes it
relatively straightforward to implement new sensor models as
“plug ins” to extend the native capabilities of SOCET SET.
Our goal in creating a SOCET sensor model for the Magellan
SAR (Howington-Kraus et al., 2000) was to make it both
physically rigorous and flexible enough to work with all types
of Magellan data.
The variety of data formats, including multiple types of mosaics
as well as single orbit strips, is only one obstacle to working
with the Magellan images. This can be handled by defining a
Magellan data set for use in SOCET SET as a collection of one
or more BIDR strips in Sinusoidal projection, with no
restrictions on scale, extent, or center longitude. An additional
complication arises because all of the images have been map-
projected based on whatever spacecraft trajectory data were
available at the time of processing and partially orthorectified
based on a low-resolution, pre-Magellan model of Venus’s
topography. Our sensor model, based on the one we helped
develop for the DSW-V, deals with this processing by using a
database containing metadata obtained partly from the mosaic
being used and partly from the BIDRs in that mosaic.
Specifically, for a given ground point, the sensor model first
determines which orbit strip (BIDR) the ground point is
contained in, and then which radar burst from that BIDR, by
comparing the lat-lon coordinates to strip and burst outlines in
the database. Once the radar burst is identified, the burst
resampling coefficients and spacecraft position and velocity at
the time of observation are obtained from the database. Next,
the spacecraft position and velocity are used to calculate the
range and Doppler coordinates at which the ground point would
be observed. This is the physical process of image formation
that we must model, and, unlike the approximate rectification
that was done in the original processing, it can incorporate
adjustments to the spacecraft trajectory. In this way, we allow
for bundle-adjustment of the BIDR strips to improve the
positional accuracy of the resulting DTM, even when using
images that have been combined in an uncontrolled mosaic.
The geometric range just calculated is next corrected for
atmospheric refraction. Finally, the resampling coefficients
associated with the burst are applied to the range and Doppler
coordinates to determine the image coordinates at which this
range and Doppler point would have been put into the image.
Procedures. Topographic mapping with Magellan data begins
with ingestion of the BIDR, MIDR, or FMAP images into ISIS.
The full-resolution FMAP mosaics can be used for most DTM
production, but in potential problem areas within a mosaic,
where pixels are lost at F-BIDR seams, it is necessary to collect
DTMs from the unmosaicked F-BIDRs. The BIDRs are also
essential if strip-to-strip ties are to be collected for bundle
adjustment, and they must be read in the first time a new area is
mapped, because they contain the auxiliary data needed to
populate the database described above. Only the image data for
the latitudes being mapped needs to be retained from the pole-
to-pole BIDR strips. Information about the spacecraft position
and velocity can be taken either from the BIDR headers or from
separate NAIF SPICE kernels (Acton, 1999; data are available
from ftp://naif.jpl.nasa.gov/pub/naif/MGN/kemels/), letting us