Total Variation Regularized GRACE(‐FO) Inversion

Gravity estimation from satellite‐satellite tracking missions such as GRACE(‐FO) is an ill‐posed inverse problem. The conventional approach to regularized inversion of GRACE(‐FO) measurements uses ‐Tikhonov regularization with a heuristic constraint matrix derived based on knowledge of spatiotemporal distribution of the signal/noise. We present Total Variation (TV) regularization as an alternative approach, that is, effective for GRACE(‐FO) inversion without the need for heuristic constraint matrices. We use an iterative regularization strategy to limit the signal loss due to regularization bias. The resulting solutions compare well with the ‐Tikhonov mascons from UTCSR/JPL estimated with more elaborate constraint design. The TV‐derived trends over Greenland agree with the UTCSR mascon trends to within 4 Gt/yr and with the JPL trends to within 11 Gt/yr. The TV solution timeseries shows good agreement with the UTCSR/JPL mascons at large hydrological basins with less than 2% difference in the annual amplitude recovered over the Amazon and the Ganges basins. The TV solutions are derived without the use of explicit spatial constraints, and require no operator‐knowledge of the distribution of the signal/noise features in the data, and reflect the inherent information content within the observations.