Seite - 107 - 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

from the orientation uncertainty to the AOC is purely linear, this variance factor can also be taken to represent a scaling of the uncertainty in the satellite orientation, as obtained from the SCA/ACC sensor fusion. The estimation of the ll-SST covariance function as described in section 6.5 remains unchanged, with one variance factor per time lag. The total number of estimated variance factors for all ll-SST observations in one month is thus: Nmax=2160 variance factors, one for for each time lag (3h arcs at 5s sampling); M=248 arc-wise variance factors (31 days with eight 3h arcs each); and 2 additional variance factors for the AOC covariance matrices. 8.3 Results The GRACE time series was reprocessed with the updated stochastic model described in the previous sections. For some months, especially in 2002, this was not possible due to issues with the release 2.0 level 1B data. In December 2002, for example, the sca idfield in the SCA1B data file is set to the value 5, which is not defined in the level 1B user documentation (Case, Kruizinga, and Wu, 2010). As this is the flag that describes which SCA heads were used in the determination of the satellite attitude no meaningful AOC covariance can be computed for this period. These months were thus excluded from further analysis. In total, 152 monthly solutions were processed spanning the period from February 2003 to June 2017. Thefollowingsectionswillgiveanoverviewof theAOCcovariancematrices’ impacton several aspects of the stochastic model. Further, the post-fit residuals of one particularly interesting month will be analysed. As the improved stochastic model does not have as strong of an impact on the overall monthly gravity field solution as the co-estimation of the satellite orientation later introduced in chapter 9 this aspect will not be discussed here explicitly. For compactness, it will be illuminated together with the results in section 9.3. 8.3.1 AOC Variance Factors The distribution of the monthly AOC variance factors for the processed GRACE time series is illustrated in fig. 8.5. The displayed probability density functions (PDFs) were determined using a non-parametric kernel density estimator (e.g. Rosenblatt, 1956). If the a priori orientation uncertainty estimate from the SCA/ACC sensor fusion were accurate, and no other unmodelled effects were present, the expected values for the monthly AOC variance factors would be 1. The estimates, however, show a mean of slightly above 2 (2.30 for GRACE-A and 2.26 for GRACE-B). These increased means indicate that the estimates of the orientation uncertainty from the SCA/ACC sensor fusion are possibly too optimistic. 8.3 Results 107