Page - 141 - 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

management in 2011 does not seem to have a strong impact on most of the estimated misalignment angles, with the exception being the roll angle on GRACE-B. 9.3.5 Gravity Field Solutions This section presents the final full degree and order gravity field solutions first com- puted using the original ITSG-Grace2016 stochastic model, then incorporating AOC covariance matrices, and finally using the TLS approach. Degree Amplitudes Figure 9.8 shows degree amplitude graphs for two representative months. In a month of good data, shown in fig. 9.8a, the solution computed using the non-stationary AOC covariance matrices (blue) is almost identical to the standard old solution (brown). The TLS solution (pink) shows slightly lower amplitudes above degree and order≈50. The magnitude of the improvement (dotted pink) is on the order of the expected GRACE baseline accuracy. The improvement is much more pronounced for June 2010, shown in fig. 9.8b, where data quality is deteriorated due to larger than normal opening angles. Here, both of the new solutions differ from the old model by more than the GRACE baseline accuracy. Especially the TLS solution (pink) shows much smaller amplitudes at very high degrees, where noise is expected to dominate the recovered signal. In general, the TLS solution shows larger differences than the solution using the AOC covariances only. This also holds for the months not pictured here. Spatial Domain Figures 9.9 and 9.10 show equivalent water height maps for the same two example months of good and deteriorated data quality. The reference mean field and long-term signals — both from the GOCO05s model — were removed from the monthly solutions. A 300km Gaussian filter was applied to the datasets. The isotropic Gaussian filter was chosen instead of one of the more advanced non-isotropic filters (see e.g. Kusche, 2007; Swenson and Wahr, 2006) to retain information on the change of the potential in north-south direction. For similar reasoning, a conservatively small filter radius was selected, which allows for the retention of some high-frequency patterns in the signal after smoothing. For both shown months, the major spatial features of the monthly gravity variations are recovered in all three configurations (top and left columns). For the month of good data, the differences between the old stochastic model and the model including AOC covariance matrices, shown in fig. 9.9c, is small and mostly globally homogeneous. Figure 9.9e shows the differences of the TLS solution w.r.t. the old stochastic model, 9.3 Results 141