Page - 154 - 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

granular weighting of observations, not necessarily downweighing parts of one arc because anomalous data was detected at other times during the same arc. Even though the method of deriving AOC covariance matrices from the results of an in-house SCA/ACC sensor fusion was proven to be helpful in the analysis of GRACE data, its applicability to e.g. the ITSG-Grace2018 release is uncertain. The ITSG- Grace2018 series of gravity field solutions will be based on the release 3.0 dataset of GRACE level 1B data provided by JPL. The SCA1B product of the release 3.0 data will contain satellite orientations derived from a SCA/ACC sensor fusion computed at JPL (GRACE, 2018). The fusion computed at JPL will be based on a Kalman filter approach, and full covariance matrices will not be available to the processing community. This prohibits the computation of a meaningful SCA/ACC sensor fusion at IfG, as the necessary raw data is no longer available in unprocessed form. It could be possible to compute an approximate orientation covariance matrix using available data and use this information to derive the AOC covariance matrices. This approach has not yet been studied in detail. For GRACE-FO, full level 1A data is expected to be released (Wen et al., 2018). This will again allow for the determination of the spacecraft orientation through sensor fusion, but now based on directly employing level 1A data. The resulting covariance matrices can then be used in the determination of AOC covariance matrices for GRACE-FO. Building on the full arc-wise information of the orientation uncertainty, the satellite orientation was co-estimated in the least squares adjustment for the Stokes coeffi- cients. In previous solutions up to ITSG-Grace2016, the orientations of the spacecraft were assumed fixed and error-free, which is clearly not true. The equivalence of two formalisms for the treatment of uncertainties in independent variables was proven, namely the TLS algorithm as formulated by Reinking, 2008 and a formulation based on classical parameter elimination. Employing this apparatus, co-estimation of spacecraft orientations led to a more stable estimate of the KBR antenna phase centres, especially improving on the previously strongly biased length estimate. The main impact of the changed parametrisation on the recovered gravity field solutions was a reduction in temporal variability over the ocean. This can reasonably be interpreted to correspond to a reduction in noise. In the spectral domain, the reduction in variability was especially prominent for some Stokes coefficients whose spatial patterns correspond to the domi- nant pitch-motion of the GRACE satellites. Spectral analysis shows an abrupt drop in temporal variability at spherical harmonics of order 61. This is cause for caution, as it implies that the chosen processing strategy — co-determination of the stochastic model and initial approximation of the satellite orientations in an adjustment up to D/O 60, followed by a full adjustment up to D/O 120 — is the cause of these processing artefacts. Further studies are needed to investigate this effect, preferably recomputing the stochastic model and orientations together with a full D/O 120 solution. The conclusions drawn in the spatial domain should however remain largely unaffected by this artefact, as the 300km Gauss filter applied to the solutions prior to analysis has a cut-off far below the D/O 60 threshold. As the TLS algorithm depends on somewhat accurate covariance information, this approach will not be used in the ITSG-Grace2018 gravity field solutions, due to the Chapter10 Conclusion and Outlook154