A step-wise decorrelation approach for improving time-variable gravity fields from GRACE-FO data

Abstract

The time-variable gravity field solutions from the Gravity Recovery and Climate Experiment (GRACE) and GRACE Follow-On (GRACE-FO) mission are generally contaminated by the correlation errors, specifically parameter correlation (strong parameter coupling) and observation noise correlation (colored instead of white noise). In this context, we propose a decorrelation approach to pursue an improved time-variable gravity solution following a step-wise processing. The step-wise decorrelation approach comprises three steps, standard, parameter decorrelation, and observation noise decorrelation processes. First, the standard process serves to establish a reliable signal reference for the following. Then, the parameter decorrelation is implemented through the separate estimation of orbit and gravity field parameters. Finally, to achieve the goal of observation noise decorrelation, the post-fit residuals obtained from the result of parameter decorrelation are used to estimate a colored noise model, which is considered to determine the final gravity field model, specifically termed the step-wise decorrelation solution. The basic idea is that under the regularization constraints of separate estimation for dynamic parameters, the reduced dynamic parameter space allows certain low-frequency perturbative errors to emerge in post-fit residuals, enabling comprehensive characterization of observation noise correlation. Using this step-wise decorrelation approach, we process monthly GRACE-FO gravity field time series and evaluate the performance of them from the aspects of signal and noise. Spectral and spatial domain analyses of noise levels confirm significant noise suppression in the final solution. For instance, it achieves ~ 66% noise reduction in term of mean geoid height difference beyond 80 degrees relative to the standard solution. The basin-scale signal assessment for the step-wise decorrelation solution demonstrates unbiased estimation of both full signals (notably trends and seasonal components) and high-frequency intra-annual variations (up to 6 cycle per year) and the accurate extraction of regional episodic events, showing high consistency with the latest Science Data System (SDS) products. Beyond GRACE-FO, the universality test confirms the applicability of the approach to the data processing scenario of GRACE, with potential extensions to next-generation gravity missions.