Seite - 146 - 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

Table 9.3 shows that for all areas, the RMS of the solution using AOC covariance matrices isnotmuchsmaller thanthatof thesolutioncomputedusingtheoldstochastic model. For June 2010, it is even larger in all areas, though within a margin that is difficult to attach a qualitative meaning to. The TLS solution shows reduced RMSs for all study areas in both months. Here, the RMS reduction over the continents is smaller than that over the ocean, and it is largest for the pacific patch area. This supports the conclusion that the TLS solution reduces orientation-induced noise in the solution without overly damping the magnitude of the recovered signal. Temporal Domain Thetimeseriesofcomputedmonthlysolutions isalsoanalysedin the temporaldomain. Some months of reduced data quality were excluded from this analysis. The inclusion of these outliers strongly skewed the results and prohibited meaningful interpretation of the created time series and statistics. Specifically, the excluded months were 2004-01, 2004-09, 2012-04, 2012-06, 2015-01, and 2015-02 due to repeat or near-repeat orbit geometry; as well as 2017-03 and 2017-06 due to unavailability of ACC observations for GRACE-B and the subsequent use of “transplanted” ACC observations from GRACE-A. Temporal variability - spatial domain Figure 9.11 shows the temporal variability of the three computed time series. The variability was computed as the RMS for each point on a 1°×1° geographical grid, once for each complete time series. Only the inter-monthly variability was considered. Annual, semi-annual, and secular signals were removed prior to computation of the variability. A 300km Gaussian filter was applied. Figure 9.11e shows that the variability in the TLS solution is clearly reduced in distinct bands in the northern and southern mid-latitudes, and around the equator. The reduction in variability is larger over the oceans than over land. For the solution using the AOC covariance matrices, the reduction has a mean of 0.015cm over land and 0.020cm over the ocean. For the TLS solution, the reduction is 0.111cm and 0.165cm over land and the ocean, respectively. In opposition to the overall trend of a reduction in variability, some areas of large signals, especially the Amazon basin and southern Africa, show a slight increase in variability. Overall, the magnitude of the reduction in variability is not very large. It is encouraging that variability over land is reduced by a smaller factor than over the ocean. This supports the theory that correct handling of the non-stationary errors due to the satellite pointing leads to a better recovery of the sought signal, while noise in the recoveredsolution isdamped.Thereduction is larger for theTLScase thanfor theAOC covariance only case. As a larger number of parameters is estimated, it is reasonable to expect less noise remaining in the solution. What cannot be seen in fig. 9.11e is a systematic damping of temporal variability where large hydrological signals are expected, which is encouraging. Chapter9 Co-Estimation of Orientation Parameters146