International Archives of the Photogrammetry, Remote
Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
and 52653629. where it was necessary to use older values to get
a reasonable solution). This better a priori camera station
information should result in a better solution, particularly since
we do not adjust the exposure epoch or spacecraft position.
Solutions using this new information do indeed show at least a
5% lower overall RMS, changing (in image space) from 17.8
um to 16.9 um.
New camera reseau-finding procedure. An improved
algorithm has been created in the USGS ISIS (Eliason, 1997:
Gaddis, et al, 1997; Torson and Becker, 1997; also sce
http://isis.astrogeology.usgs.gov/) software for determining the
locations of the reseau marks on VO images. In the cases
where we have the original RAND and USGS pixel VO image
measurements of control points (which is the case for 77,225
measurements), these new locations have been used to
recalculate (mm) control point locations in the image plane
prior to adjustment. In addition, a number (329) of
measurements of control points near the edges of the images
and outside the available reseau information (and therefore of
questionable value) have been removed. Solutions with these
changes show a 4% lower overall RMS, changing from 16.9 um
to 16.2 um, although some of this decrease is simply due to a
reduced number of observations.
The radii of all 37,652 control points (Figure 1) have been
derived by interpolation of a MOLA global radii grid (see
http://Awufs.wustl.edu/missions/mgs/mola/egdr.html). The
MOLA radii should be accurate to ^10 m vertically and =100 m
horizontally (Neumann, et al., 2001). This procedure has been
iterated a number of times so that as changes are made in the
solution, or new data are introduced and new horizontal
coordinates are derived for control points, new a priori radii
information is obtained from the MOLA dataset. Again, that
there is an improvement in using the MOLA data in these
successive steps is shown by an 11% decrease in the overall
control network solution RMS.
Measures from additional images are included. Measures
of 52 images that were used in MDIM 2.0 but not rigorously
included in the previous RAND adjustment for MDIM 2.0 have
now been included in this solution. There are 406 such
measurements of 203 control points on 102 images (including
the new images and images that overlap them).
Horizontal positions of a number of control points have
been fixed to MOLA-derived values. This in effect provides
equally spaced “ground control” for Mars globally. Our
procedure was to match high resolution MOLA DIMs (as
derived by Duxbury) with VO images, and measure the
positions of existing and new control points on both. Such
measurements were made using an annulus cursor centered on a
crater rim in order to avoid parallax problems in measuring the
position of the center of a crater. In the network solution, the
latitudes and longitudes of these points, as derived from the
MOLA DIMs, were held fixed. A grid of such points has been
measured globally on Mars, with 15° latitude and 30° longitude
spacing. Some additional points were also measured on the
area to the west of Olympus Mons, due to the difficulty of
finding suitable points on both the MOLA DIMs and on Viking
images in this area of mantled terrain (Figure 2). We have
assumed that at the locations of these points the horizontal
positions are therefore similar in accuracy to the inherent
accuracy of the MOLA DIMs, or about 100 to 200 m, with most
of the uncertainty resulting in the correct measurement of the
VO images and the MOLA DIMs. The accuracy will obviously
be less as one moves to areas away from these MOLA tie
points, but we are planning to verify (below) that the horizontal
positional accuracy does not degrade substantially from these
estimates.
864
Existing and new image measurements have been verified.
Measurements with solution residuals having pixel values over
4-5 Viking-sized pixels (85 pixels/mm) were carefully checked
in order to reduce such residuals. In the final solution, the
largest measurement residual was less than 4.7 pixels. Out of
90.130 measures, cumulatively only 31 measures had residuals
over 4 pixels, 553 over 3 pixels, 3,423 over 2 pixels, and 25,590
over | pixel. This is in comparison to previous (RAND)
solutions where the largest residuals were about 7.5 pixels. The
last RAND solution, with 88,325 measures, had 2 measures
with residuals over 7 pixels, 4 over 6 pixels, 21 over 5 pixels,
140 over 4 pixels, 883 over 3 pixels, 4.326 over 2 pixels, and
26.531 over 1 pixel. Many measurements have been redone,
while others have been removed from the solution in cases
where it was felt the control point in question could not be
adequately remeasured (e.g. because of a poorly defined
feature, a low contrast image, or a point near the edge of an
image) We additionally prepared large-area test MDIM 2.1
mosaics based on our solutions, which were carefully examined
for any problems. We added MOLA-derived contours to these
mosaics (Figure 3 shows an example using the final MDIM 2.1
mosaic) to check the registration of the mosaic to the MOLA
data. In cases where the registration showed differences (at the
more than a few hundred meter level) or in cases where there
appeared to be any misregistration of VO images with each
other. we made additional image and MOLA control point
measurements, and improved the solution with these
measurements in order to eliminate the problems. This process
was repeated using the final solution and MDIM 2.1 mosaic,
and no significant differences were seen in the registration of
MOLA contours with features on the mosaic.
3. RESULTS
We still plan to do additional checks on the overall
horizontal accuracy of the control network by checking the
locations of additional MOLA tie points and also of the Viking,
Pathfinder, and MER landers (whose horizontal positions are
also known to high accuracy via spacecraft tracking (Folkner, et
al. 1997: Golombek and Parker, 2004a, 2004b)). This will be
done not by fixing their coordinates in the control network
adjustment, but rather by comparing their solved-for
coordinates with the known locations.
The final MDIM 2.1 Mars control network solution
contains 90.130 measurements of 37,652 control points on
6.371 images. Of these measurements, 77,621 are on 5.317 VO
images, whereas 12,509 of the measurements are on 1,054
Mariner 9 images, as a carry-over from the original RAND
networks. The Mariner 9 image measurements had generally
lower residual values than the highest residual VO image
measurements, so were maintained in the solution both to add
geometrical strength and also to allow for the production of
updated Mariner 9 camera pointing information. A total of
1,232 control points were tied to MOLA DIM tiles, and it is the
coordinates of these control points that were held fixed (to the
appropriate MOLA-derived latitude and longitude). The
solution RMS is 15.8 um or about 1.3 Viking pixels.
4. CONCLUSIONS
We have completed a new global Mars control network,
extending earlier work done at RAND and USGS. This new
network is consistent with the IAU/IAG 2000 Mars body-fixed
reference system, and in particular, topography derived from
MOLA data in that system. The overall accuracy of positions
derived is expected to be similar to that of MOLA in both the
Int
