Page - 153 - in Contributions to GRACE Gravity Field Recovery - Improvements in Dynamic Orbit Integration, Stochastic Modelling of the Antenna Offset Correction, and Co-Estimation of Satellite Orientations

Conclusion and Outlook 10 The presented research, building on the foundation of the state of the art GRACE level 1B to level 2 data pipeline at IfG, furthers diverse aspects of GRACE data processing. The improved stochastic model described in chapter 8 and the treatment of uncertainties in independent variables outlined in chapter 9 can bring immediate observable improvements to the processing of GRACE data. This is not the case for the first presented topic, the optimization of the dynamic orbit integration (chapter 7). The previously existing implementation of dynamic orbit integration is of sufficient fidelity for traditional GRACE processing. This was proven by variance propagation of anerrorestimate for thedynamicorbits to therange-rate domain,andcomparisonwith GRACE ll-SST noise models. The same investigation showed that the stability of the classical approach could prove insufficient for GRACE-FO processing, where the new laser ranging interferometer is expected to provide higher-precision observations of the inter-satellite separation. To ameliorate this deficiency, an improved dynamic orbit integration algorithm based on a modified Encke method using a rigorously optimized reference trajectory parametrised in equinoctial elements was introduced. Using this parametrisation, the stability of the integration is expected to be sufficient for GRACE- FO processing. This implementation was already used in the ITSG-Grace2016 series of gravity field solutions. It is also expected to be used for the upcoming ITSG-Grace2018 solutions. At the time of writing, GRACE-FO was successfully launched, and initial data from the early operations phase showed the satellite instrumentation to work as expected. Should GRACE-FO LRI data become available towards the end of 2018, the results presented in chapter 7 will be able to be verified using real data. The description and rigorous evaluation of the effect of the spacecraft orientation uncertainty on the ll-SST antenna offset correction, outlined in chapter 8, has improved understanding of the error spectrum of real GRACE observations. The derived stochas- ticmodel for theAOCexplainsspuriouspowerobservedintheestimatedmonthlyKBR PSDs, located in the frequency band associated with inter-satellite pointing variations. This power was shown to be the result of aliasing of the unmodelled non-stationary noise of the AOC into the estimated monthly stationary PSD. Disentanglement of these noise sources allows for the decorrelation of all observations in the month with a more appropriate PSD, unaffected by transient noise in only parts of the observations. Further, the variance factors used to weigh individual observation arcs were strongly correlated with the inter-satellite opening angle. Arcs with higher opening angles were consistently downweighted. This is of course desired behaviour, as the observations in these arcs are affected by higher noise. Describing the variations of the ll-SST uncer- tainty on an epoch-by-epoch basis is however highly preferable, as it allows for a more 153