Revisiting the data-driven deviation approach for leakage correction in GRACE data processing

The mass change maps derived from GRACE spherical harmonic data come with inherent striping noise, which needs to be filtered during post-processing. Filtering of data leads to signal attenuation and leakage. Several leakage correction methods are available to restore the filtered signal, majorly based on hydrological models or data-driven approaches. In the data-driven correction approach, the primary focus is to account for attenuation and leakage without the aid of external hydrological models. The data-driven methods either use a scaling approach in which the filtered values are scaled after applying a leakage correction, or they use a deviation approach in which the filtered values are corrected by a deviation term and a leakage term. This study concerns the derivation of the deviation term. In deriving the deviation term, the choice of the extent of the catchment average and the deviation field has an impact on the final correction terms. If the deviation term is taken as a global field, then a scaling term appears in addition to the deviation term. Our preliminary investigation shows that there are significant differences between deviation-only and deviation-cum-scaling approaches. We perform our investigation over the GRDC basins as well as the HydroSHEDS level 5 dataset.