Variance component estimation for co-estimated noise parameters in GRACE Follow-On gravity field recovery

Abstract

Temporal gravity field modelling from GRACE Follow-On has to cope with several noise sources contaminating not only the observations but also the observation equations via mis-modellings in the underlying background force models. One way to deal with such deficiencies is to extend the parameter space by additional quantities, such as pseudo-stochastic parameters, which are co-estimated in the Least-Squares Adjustment (LSA). These parameters are meant to absorb any kind of noise while retaining the signal in the gravity field and orbit parameters. In the Celestial Mechanics Approach (CMA) such pseudo-stochastic parameters are typically set-up as Piece-wise Constant Accelerations (PCA) in regular intervals of e.g., 15 min. The stochastic behaviour of these parameters is unknown because they reflect an accumulation of a variety of noise sources. In the CMA fictitious artificial zero-observations are appended to the vector of observations together with an empirically determined variance to introduce a stochastic model for the PCAs. In order to also co-estimate a stochastic model for the pseudo-stochastic parameters in the LSA we use Variance Component Estimation (VCE) as a well established tool to assign variance components to individual groups of observation. In the simplest case the magnitude of the constraints of the pseudo-stochastic parameters can be determined fully automatically. We present results for GRACE Follow-On gravity field recovery when extending the CMA by stochastic models for the piece-wise constant accelerations computed with VCE and provide noise and signal assessment applying the quality control tools routinely used in the frame of the Combination Service for Time-variable gravity fields (COST-G).