Outlier identification and correction for GRACE aggregated data

Gravity measurements within the Gravity Recovery and Climate Experiment (GRACE) provide a direct measure of monthly changes in mass over the Earth’s land masses. As such changes in mass mainly correspond to water storage changes, these measurements allow to close the continental water balance on large spatial scales and on a monthly time scale within the respective error bounds. When quantifying uncertainties, positive and negative peaks are detected in GRACE aggregated monthly time series (from different data providers) that do not correspond to hydrological or hydro-meteorological signals. These peaks must be interpreted as outliers, which carry the danger of signal degradation. In this paper an algorithm is developed to identify outliers and replace them with hydrologically plausible values. The algorithm is based on a statistical approach in which hydrological and hydro-meteorological signals are used to control the algorithm. The procedure of outlier detection is verified by evaluating catchment based aggregated GRACE monthly signals with ground truth from hydrology and hydro-meteorological signals. The results show improvement in the correlation of GRACE versus hydrometeorological and hydrological signals in most catchments. Also, the noise level is significantly reduced over 255 largest catchments.