GRACE Data as Non-Tidal Deformation in Reference Frame Determination:  From Position Timeseries Forecast to Frame Realization

Abstract

The terrestrial reference frame (TRF) is the foundation for all space-based, airborne, and ground-based Earth observations.  The accuracy of position data ultimately depends on the accuracy of TRF, which consists of precise Earth-centered coordinates of hundred of surface sites, many of which anchor the orbit estimates of the global navigation satellites.The non-tidal deformation (NTD) due to loading from atmosphere, ocean, and ground water tends to seasonally modulate the space geodetic position data used in determination of TRF.  Campaign to determine the International Terrestrial Reference Frame (ITRF) takes place approximately every 5 years (with a recent initiative to realize intra-campaign ITRF updates every year). Because of such latency in ITRF realizations, forecasts of the TRF positions are desired in a typical application, relying on the TRF velocity estimate to forecast the linear trajectory projection. Here, we examine how various models of NTD could affect the accuracy of such TRF position trajectory forecasts.In particular, we have seen that GRACE-based surface deformation data can be used as proxy of NTD in estimation of the site velocity.  At individual GNSS sites, use of the GRACE data were particularly effective in mitigating extremely large forecast errors at certain sites, out-performing the standard ITRF NTD estimation method (via a fit to a truncated Fourier series) by an average of 3 mm. Also, no degradation of the forecasts due to addition of the GRACE data was observed. In order to extend the use of the GRACE-based deformation data to a TRF realization, we need to address:  (1) the temporal coverage of the GRACE data, such as the gap between GRACE and GRACE-FO and pre-2003 period; (2) compatibility in site-to-site spatial correlations exhibited by GRACE versus TRF site position data.  We show that the spatio-temporal behavior of the GRACE deformation data can be "learned" empirically via multichannel singular spectrum analysis (MSSA) to fill the GRACE/GRACE-FO gap.  The MSSA GRACE data have improved the forecast accuracy in the gap where we have resorted to linear interpolation in the past, and the improved forecasts in the gap period were statistically indistinguishable from the period with actual GRACE data.  This encourages consideration for hindcasting the GRACE data into a much longer 1997-2003 period for the use in a full TRF realization.