Assessing the effects of filtering on GRACE-FO signal: a case study of European river basins

The provided gravity fields of the Gravity Recovery and Climate Experiment (GRACE) mission are noisy. In order to restore the true geophysical signal, it is essential to apply an appropriate filter. A filtering minimizes the noise, as well as affects the real signal inducing uncertainty in the final results. To date, to restore the suppressed geophysical signal the majority of studies have relied on determining the scaling factor using a hydrological model. To estimate the magnitude of the leakage effect we use only the filtered GRACE Follow-On (GRACE-FO) gravity fields. We study two different data-driven approaches, i.e. method of (i) scale and (ii) deviation. For the first method, to derive the scale factor a uniform layer approximation is used. The factor depends on the filter kernel used and the basin mask i.e., its shape and size, and is used to counter the attenuation of the basin-confined signal. For the second approach, the restored signal is independent of the river basin size. It attempts to obtain the deviation integral, which is determined from the filtered deviation field and the regional average estimated for the selected area. We analyze the signal leakage for GRACE-FO monthly gravity fields based on recent the Science Data System (SDS) solutions, i.e., the Center for Space Research (CSR), the German Research Center for Geosciences (GFZ) and the NASAs Jet Propulsion Laboratory (JPL). We use the available GRACE-FO data for 67 months (June 2018 to February 2024) in the form of spherical harmonics up to degree and order 96. For detailed analysis we choose 24 river basins over Europe that are larger than 50,000 km2 and experienced extreme hydrometeorological changes in recent years. To assess the reliability of obtained results we use total water storage (TWS) from GRACE-FO JPL mascon solution and the Global Land Data Assimilation System (GLDAS) hydrological model. For spherical harmonics we find the largest differences between original (Gaussian-filtered) and data-driven fields in the parts of Europe indicating disparate TWS signals within small basins. We observe changes in trend values by 1-2 mm/yr and amplitudes by up to 5 cm in the western and the northeastern river basins. The regions with extreme values are spatially coherent with the mascon solution results and are underestimated by the hydrological model mostly in the western areas. The estimated true leakage represents up to 30% of total TWS signal and varies from single cm (e.g., Douro, Ebro, Narva) to even 20 cm (e.g., Dniestr, Northern Dvina, Tigris-Euphrates).