Time-correlated noise signatures in gravity records

Nowadays it is widely acknowledged, that all geodetic records of geophysical processes exhibit temporal correlation, which should not be ignored when parameters of interest (e.g. seasonal signals or trend) are to be determined. Gravity changes recorded by superconducting gravimeters (SG) should represent real gravity variations for all frequencies higher than 10^-7 Hz. However, they should be firstly corrected for long-term drift. The value of this drift can be obtained by subtracting the absolute gravity measurements (AG) from the SG values or by analytical determination. In this research we solve a problem of drift removal twofold: by applying 6-degree polynomial and wavelet decomposition, to absorb as much power in low frequency as possible. All seasonal signals that still remained in data (from annual till sub-diurnal) were removed with Least Squares Estimation. What is left in the residuals (stochastic part) is being considered as a “noise”. Unlike GPS, the stochastic part of SG time series does not follow power-law processes in the whole frequency domain. In this research we tested a set of autoregressive-moving-average (ARMA) models, which describe a stationary stochastic process as two polynomials: auto-regression (AR) and moving average (MA). Besides, the Generalized Gauss-Markov (GGM) model was applied. Both were considered as a combination with white noise. The optimal model was selected upon the values of Maximum Likelihood Function (MLF) as well as Akaike (AIC) and Bayesian (BIC) Information Criteria. Since superconducting gravimeters are sensitive to both deformation and Newtonian processes, the environmental effects should be clearly noticeable in the Power Spectral Density charts. To investigate the way in which these effects significantly modify the PSDs we used atmospheric and hydrologic (continental hydrosphere and ocean non-tidal) loading effects calculated in the EOST Loading Service and SG data from 20 globally distributed stations stored in the GFZ’s Information System and Data Center. The obtained optimal stochastic model can finally be used to infer realistic uncertainties on the instrumental drift for instance.