Item

Abstract

For a reliable quantification of terrestrial water storage (TWS) variations, it is important to provide functional and stochastic models that best explain the data. Although spatial covariance models of GRACE/GRACE-FO TWS data are well developed, temporal covariances are often left unexplained. However, this results in unrealistic and too optimistic parameter uncertainties, i.e. those of TWS trends. By assessing the post-fit residuals of the global gravis TWS dataset, we show that the standard TWS time series model should account for the existence of temporal correlations that are present in the data. For a major part of TWS data, power-law noise models explain temporal covariances. We further show that the magnitude of estimated uncertainties of the trend function strongly depend on the intensity of time correlations. By means of a clustering analysis it is demonstrated, that certain regions (e.g., river basis) can be characterized and separated by common-mode TWS signals, which supports the explanation of regional hydrological effects. We highlight how these signals can be parameterized in the functional model, and how optimal model combinations can be chosen by means of information criteria. To the end, we apply the same methodology to the JPL mascon RL06 dataset, and explore differences in noise and signals w.r.t. the gravis dataset. We conclude that – to accurately quantify hydrological variations – the TWS uncertainty budged should account for temporal correlations present in the data. Until not all TWS signal constituents are explained and validated, an empirical noise calibration method provides a realistic explanation of TWS data uncertainties.