COST-G: Towards normal equation level combination

The very original motivation of the Combination Service for Time-variable Gravity fields (COST-G) has been a combination of monthly gravity fields computed by different Analysis Centres (ACs) on Normal EQuation (NEQ) level, thus, taking all formal correlations arising with the orbit and instrument
parameters directly into account. However, already early experiments within the EGSIEM project, where a prototype of the combination
service was developed, showed that the use of different nuisance parameters - set up to absorb observation and background model deficiencies - by each of the ACs leads to diverse formal uncertainty estimates of the spherical harmonic coefficients representing the Earth’s gravity field. Though formally possible, a combination on NEQ-level yielded degraded results. Even more, this rendered an automated combination process, applying Variance Component Estimation (VCE) on NEQ-level to derive relative weights for the individual AC’s contributions, impossible. Meanwhile, realistic uncertainty information is available for the majority of background models, and empirical noise modelling techniques leading to realistic uncertainty estimates are well established among the ACs processing the GRACE Follow-On data. In preparation for ESA’s Next Generation Gravity Mission (NGGM), where the combination of gravity field solutions from different ACs and a contribution from Satellite Laser Ranging (SLR) to stabilise the very low degree spherical harmonic coefficients is foreseen on NEQ-level, the combination strategy has been revisited with GRACE-FO NEQs of AIUB (rl03op), GFZ (preliminary RL07) and TUG (ITSG-Grace_operational) NEQs, as well as LAGEOS 1 and 2 and LARES 2 SLR-NEQs from AIUB and CNES, with very promising results, which we show in this contribution.
In this context, we also introduce the updated operational gravity field solution time series from the AIUB and investigate its impact on the COST-G combination. Background modelling has been revisited and the uncertainty characterisation has been improved by additionally co-estimating daily spherical harmonic coefficients constrained with AOe07 variance-covariance information.