Kalman filtered daily GRACE gravity field solutions in near real-time - first steps

Until now, monthly GRACE gravity field models have been available with a time delay of about two months, which just allows for the ‘confirmation after occurrence’ and to assess the severity of a hydrological extreme event. As part of the EGSIEM (European Gravity Service for Improved Emergency Management) project, a technology demonstrator for a near real-time (NRT) gravity field service will be established, with the goal of reducing this latency to a maximum of five days and to provide daily gravity field models. These rapid gravity time series will enable us to monitor the global water storage in NRT and to observe floods and droughts as they occur. This information will help rapid mapping providers to react earlier to such hydrological events, resulting in improved emergency and assistance efforts. As basis of the daily gravity field recovery serves a Kalman filter framework, in which GRACE data is combined with prior information to increase the redundancy of the gravity field estimates. The Institute of Geodesy at Graz University of Technology (TU Graz) routinely processes daily gravity field time series in post-processing as part of static gravity field releases. In preparation of the operational phase of the service, several aspects of this processing chain have been inspected in order to improve the gravity field solutions and move towards NRT.

This contribution summarizes the results of the first project phase at TU Graz on the basis of a new daily gravity field time series. We present an improved stochastic state-space model, derived from empirical covariance estimates of geophysical model output and evaluate its performance using in-situ measurements as well as GRACE range-rate residuals. To satisfy the maximum time delay of five days, forward-backward smoothing as is used in the standard post-processing chain, has been replaced by an on-line smoothing algorithm. The effect of this novel state estimator is investigated by comparison of the computed time series with post-processing and forward-only filtered solutions. From the results of this evaluation we determine the error levels which can be expected from the near real-time gravity field estimates.