J. Geophys Res., 106(E10), 23753-23768, October 25, 2001.
ABSTRACT
In its first year of continuous operation, the MOLA instrument aboard Mars Global Surveyor ranged to Mars over 300 million times, generating more than 5000 orbital profiles, with a ranging precision of 0.4 m over smooth terrain. The accuracy of the profiles depends on knowledge of the spacecraft position, orientation, and observation time, which are subject to errors. We model these errors via the analysis of over 24 million altimetric crossovers. A quasi-periodic adjustment of the ground tracks as a function of time in three locally orthogonal directions minimizes the altimetric residuals via least-squares. A sparse matrix technique reduces the bandwidth of the least-squares problem, so that computational effort scales linearly with the number of crossovers. Altimetry along the adjusted tracks generates a topographic model whose accuracy is typically better than one meter vertically and 100 meters horizontally with respect to the center-of-mass of Mars. Spatial resolution is limited only by the spot size of the instrument and by track-to-track spacing. Terrain models from accurately located lidar data can be gradient-shaded to illuminate geological structures with slopes much less than 1 in 1000 that are invisible to cameras. Temporal changes in elevation (CO2 frost deposition/ablation) at decimeter levels may also be assessed using crossovers but results must be interpreted with caution due to uncertainties in the range walk correction.
Highlights
Figure 8 shows histograms of the crossover error before and after adjustment, from analysis of 24 million crossovers using eight parameters per revolution for each dimension. Of the inital set of crossovers, only 1.6% could not be fit within 10 m, and were rejected. Of the remaining crossovers, the initial crossovers are long-tailed, only the central portion being shown. More than 15% exceed 10 m, and more than 4% exceed 20 m. The final residuals are highly compact. The root mean square (RMS) erros is reduced nearly fivefold, while the median absolute error, scaled to one standard deviation for a normal distribution, is reduced nearly threefold, from 2.7 to 0.96 m. These numbers mainly reflect reduction in error in the central portion of the histogram, but we note that the intital errors were as large as 300 m before adjustment.
Contact Gregory Neumann (neumann@tharsis.gsfc.nasa.gov) for further information or for a preprint of this paper.