Applying autoregressive models in analysis of GRACE-Mascon time-series

Abstract

This study discusses how to model the noise in a Gravity Recovery and Climate Experiment (GRACE)-Mascon derived Equivalent Water Thicknesses (EWT) time-series. GRACE has provided unique information for monitoring variations in EWT of continents in regional or basin scale since 2002. To analyze a GRACE EWT time-series, a standard harmonic regression model is used, but usually assuming white noise-only stochastic model. However, like almost all kinds of geodetic time-series, it has been shown that the GRACE EWT time-series contains temporal correlations causing colored noise in the data. As well known in geodetic modelling studies, neglecting these correlations leads to underestimating the uncertainties, and so misinterpreting the significancy of the parameter estimates such as trend rate, amplitudes of signals etc. In this study, autoregressive noise modeling, which has some advantageous compared to the approaches and methods frequently applied in geodetic studies, is considered for GRACE EWT time series. For this aim, three important basins, namely the Yangtze, Murray-Darling and Amazon basins have been examined. Among some applied autoregressive models, the ARMA(1,1) model is obtained as the best-fitting noise model for analyzing the EWT changes in each basin. The obtained results are discussed in terms of forecasting, significancy and consistency with GRACE-FO mission.