Novel Data-driven High-Frequency Mass Change Models from GRACE orbit residuals

Michal Cuadrat-Grzybowski , Pieter N. A. M. VIsser, Frederik Jacobs, Joao G. Teixeira da Encarnacao

We present a fully data-driven framework for transforming residual K-band range-rate (KBRR) data from GRACE into 5-day mass change models expressed in Equivalent Water Height (EWH). The approach first derives residual range and Line-of-Sight Gravity Differences (LGDs) from monthly post-fit residual range-rates and combines them with 5-day post-fit residuals. A hybrid formulation, merging LGD- and range-based terms, is then applied to minimise regression errors associated with LGD side-lobe patterns. The resulting hybrid grids are empirically linked to EWH through a spatially resolved linear regression, yielding 5-day estimates. The method relies solely on standard monthly GRACE processing products, requiring no additional dynamic modelling, Kalman filtering, or regularisation. Relative to the gridded monthly Level-2 EWH uncertainties, the 5-day solutions exhibit a moderate uncertainty increase of typically 13--20\% over land and only 8--12\% over the oceans, substantially smaller than expected from a direct unconstrained inversion at equivalent temporal resolution. The physical relevance of the recovered sub-monthly signals is demonstrated through two independent case studies, capturing (i) the 2007 monsoon flooding event in Bangladesh and (ii) the January 2013 Cyclone Oswald event. The accompanying uncertainty framework provides spatially and temporally resolved error estimates, enabling quantitative interpretation of the 5-day fields and facilitating their use in downstream geophysical and data-assimilation applications.