Analysis of GRACE uncertainties by hydrological and hydro-meteorological observations

The uncertainty in GRACE gravimetric data emanates from a superposition of errors from different sources, whose separation is difficult. In situ hydrological and hydro-meteorological observations, under certain conditions, facilitate the identification of different uncertainty types, and the quantification of their contribution to the overall error budget. In this paper, monthly GRACE solutions, hydrological and hydro-meteorological data, and their respective combinations are assessed to ascertain their uncertainties. Unphysical outliers in GRACE data are identified, removed, and replaced, which leads to a significant increase (up to 70%) in correlation with hydrological and hydro-meteorological data, and also a sizeable reduction (up to 40%) in noise level. On further statistical analysis, physical signals that neither have hydrological nor hydro-meteorological character are identified in the GRACE data at the residual level. These signals have a long-range correlation, which is untypical of hydrological/hydro-meteorological signals, and also have a strong latitudinal dependence. In keeping with their character of long-range correlations, these signals are collectively termed as correlated error and removed. This further increases the correlation of GRACE data with hydrological and hydro-meteorological data by up to 30%, and also reduces the noise level by up to 50%. Due to its latitudinal dependence, the reduction of correlated error is of specific importance to applications of GRACE in the boreal and equatorial regions with low signal amplitudes. (C) 2012 Elsevier Ltd. All rights reserved.