One of the major sources of uncertainty in mass change estimates from level 02 grace products comes from the signal degradation due to filtering of noisy gravity field products. Filtering suppresses noise but also changes the signal via attenuation and leakage. Therefore, many methods have been devised to tackle the unavoidable signal loss due to filtering. However, most of these methods lack mathematical analysis that is essential for understanding the cause and effect of filtering. Furthermore, they use hydrological models to compute correction terms, such as leakage, bias, or scale factor, for repairing the damage due to filtering. Recently, a data-driven method was proposed for improving the filtered grace products, which was shown to be superior to three widely used model-based methods. However, the method works efficiently only for catchments above a minimum size. This limitation is due to the usage of a uniform layer approximation for deriving a scale factor, which is used to counter the attenuation of the catchment-confined signal. In this contribution, we avoid this approximation, and therefore the usage of scale factor, which lifts the limitation and provides a better mathematical relation. The new data-driven method is able to restore the signal loss due to filtering independent of the catchment size. We validate the method in a realistic grace-type closed-loop simulation environment and compare it with other popular approaches. We show that for 22 out of 32 catchments (small to large size and located in different climatic zones) the improved data-driven method outperforms other methods.