How to properly scale GRACE estimates of the continental water storage variations?

Abstract

The estimation of terrestrial water storage variations at river basin scale is one of the most important applications of GRACE (Gravity and Climate Experiment) so far. Today, mature techniques are available to transform monthly GRACE gravity field models into mean water storage variations over a target area. Spatial filtering of GRACE is routinely used, and several isotropic or non-isotropic filters have been proposed in literature. Recently, more attention is paid to the problem of the bias, which is introduced by spatial filtering. The subject of this study is the amplitude and time behaviour of the bias for several target areas in Southern Africa. The regional hydrological model LEW is used to provide a time series of water storage variations inside and outside the target areas. This information is used to compute the bias. Correspondingly, GRACE estimates of water storage variations are corrected for the bias and compared with the LEW model output. The main conclusion of the study is that the bias caused by spatial smoothing results in monthly and annual amplitudes of mean water storage variations, which are too small. The bias-to-signal ratio is mainly determined by the filter correlation length. For the target areas in Southern Africa, a 1000 km correlation length for a Gaussian filter seems to be an appropriate choice. Then, the bias-to-signal ratio reaches values up to 50 % ; the monthly bias-to-signal ratio can even be larger. After bias correction, the differences in terms of annual amplitudes between GRACE and LEW take up values between 0 and 30 mm. The RMS difference of monthly amplitudes are reduced significantly and attain values up to 30 mm. The main conclusion of the study is that GRACE annual and monthly amplitudes of mean water storage variations over a target area have to be bias-corrected before GRACE is used to calibrate hydrological models.