Data-driven gap filling and spatio-temporal filtering of the GRACE-GRACE-FO records

The Gravity Recovery And Climate Experiment (GRACE; April-2002-June 2017) and current GRACE-Follow On (GRACE-FO; June 2018-present) missions have provided monthly global measurements of the space and time varying Earth’s gravity field, monitoring changes in the ice-sheets and glaciers, hydrological water storage, sea level and within solid Earth. Yet, temporal gaps, including the long 11 months gap between missions, prevent the interpretation of long-term mass variations. Moreover, despite the data processing strategy adopted, GRACE and GRACE-FO solutions show high level of distinctive unphysical noise.  
Consequently, we use the Multi-Channel Singular Spectrum Analysis (MSSA) and exploit both spatial and temporal information contained in multiple solutions of GRACE and GRACE-FO to fill the observational gaps and develop a data-driven spatio-temporal filter to enhance the data signal-to-noise ratio. 
First, we use the well-established decorrelation DDK7 filter to remove a large part of the distinctive noise in a North-South striping patterns. Because we detect persisting noise at high orders, we develop a complementary filter based on the residual noise between fully-processed data and parametric fit to the observations. We then fill observational gaps using an iterative M-SSA approach and series of equivalent water height from four Level-2 solutions (CSR, GFZ, JPL, TU-GRAZ). The method is validated on a synthetic test, where we remove and reconstruct one year of the time series. By using multiple solutions in the process, we form a combined solution based on their common modes of variability. Finally, we take full advantage of the M-SSA to reduce residual spatially uncorrelated noise, namely stripes, by conserving common signals between times series of each point on the globe and its neighbors.  
Comparison of the GRACE-GRACE-FO M-SSA solution with independent observations of the low-degree Earth’s gravity field via Satellite Laser Ranging validates the method’s potential to recover missing observations. Furthermore, comparisons with other solutions show a significant noise reduction compared to spherical harmonic solutions, and the ability to retrieve short-wavelengths geophysical signals masked by Mascons-type processing strategy.