Spatio-temporal Wiener filtering of GRACE/-FO solutions

Monthly estimates of the Earth’s gravity field based on the GRACE/-FO missions have enabled decisive progress in the quantification and understanding of mass transfers in the Earth system. However, raw level 2 GRACE/-FO solutions are affected by large errors in the form of North-South stripes. Several filtering methods have been developed to isolate the gravity signal from these errors, based on either spatial covariance models of the signal and errors (gaussian filters, DDK filters), or on statistical decomposition techniques (PCA, ICA, SSA, M-SSA). This presentation introduces a new filtering method that relies on both the temporal and spatial structures of the signal and errors.

In the temporal domain, a spectral analysis of the GRACE/-FO Stokes coefficient time series suggests that they may be modeled as the sum of:

• linear trends,
• seasonal (annual and semi-annual) variations,
• ~160-day periodic variations,
• temporally uncorrelated aperiodic variations (white noise),
• temporally correlated aperiodic variations that may be described by a power-law stochastic process.

Assuming that the linear trends, seasonal variations, and time-correlated aperiodic variations reflect gravity signal, that this signal is spatially isotropic, while the ~160-day variations and white noise are errors, we construct a spatio-temporal Wiener filter of the GRACE/-FO level 2 solutions. We then compare the results of this Wiener filter with those from the DDK5 filter and Gauer et al. (2023)’s M-SSA filter. The advantages, drawbacks and prospects for improvement of the Wiener filter are finally discussed.