Treatment of ocean tide background model errors in the context of GRACE/GRACE-FO data processing

Abstract

SUMMARY Ocean tide (OT) background models (BMs) used for a priori de-aliasing of GRACE/GRACE-FO observations feature distinct spatial uncertainties (primarily in coastal proximity and in latitudes above ±60°), and therefore pose one of the largest contributors to the overall retrieval error. The retrieval performance can be expected to increase if this underlying spatial error distribution is stochastically modelled and incorporated into the data processing chain. In this contribution, we derive realistic error variance-covariance matrices (VCM) based on a set of five state-of-the-art OT models. The additional value of using such VCMs is assessed through numerical closed-loop simulations, where they are rigorously propagated from model to observation level. Further, different approximations of the resulting VCM of observations are assumed, that is full, block-diagonal and diagonal, in order to evaluate the trade-off between computational efficiency and accuracy. It is asserted that correctly weighting the OT BM error can improve the gravity retrieval performance by up to three orders of magnitude, provided no further error contributors are considered. In comparison, the overall gain in retrieval performance is reduced to 75 per cent once instrument noise is taken into account. Here, it is shown that simultaneously modelling the OT BM and the instrument errors is critical, as each effect induces different types of correlations between observations, and exclusively considering covariance information based on the sensor noise may degrade the solution. We further demonstrate that the additional benefit of incorporating OT error VCMs is primarily limited by the de-aliasing performance for non-tidal mass variations of atmosphere (A) and oceans (O). This emphasizes the necessity of best-possible AO-de-aliasing (e.g. through optimized processing techniques and/or improved BMs) in order to optimally exploit the OT BM weighting.