Comparison of numerical solution strategies for gravity field recovery from GOCE SGG observations implemented on a parallel platform

摘要

The recovery of a full set of gravity field parametersfrom satellite gravity gradiometry (SGG) is a huge numericaland computational task. In practice, parallel computinghas to be applied to estimate the more than 90 000 harmoniccoefficients parameterizing the Earth’s gravity field upto a maximum spherical harmonic degree of 300. Three independentsolution strategies, i.e. two iterative methods (preconditionedconjugate gradient method, semi-analytic approach)and a strict solver (Distributed Non-approximativeAdjustment), which are operational on a parallel platform(‘Graz Beowulf Cluster’), are assessed and compared boththeoretically and on the basis of a realistic-as-possible numericalsimulation, regarding the accuracy of the results, aswell as the computational effort. Special concern is givento the correct treatment of the coloured noise characteristicsof the gradiometer. The numerical simulations show thatthere are no significant discrepancies among the solutions ofthe three methods. The newly proposed Distributed NonapproximativeAdjustment approach, which is the only oneof the three methods that solves the inverse problem in a strictsense, also turns out to be a feasible method for practical applications.Key words. Spherical harmonics – satellite gravity gradiometry– GOCE – parallel computing – Beowulf cluster