Inferencing the GRACE/GRACE-Follow On Data Gap Using Bayesian Modeling

Over the past 19 years, the Gravity Recovery and Climate Experiment (GRACE) satellite mission and its Follow-On (GRACE-FO) deepened our understating of the Earth's mass change distribution. The 11-month gap between the two missions (07/2017-06/2018), in addition to missing data within each GRACE/FO mission, must be imputed to ensure continuous tracking of hydroclimate, ocean, and solid Earth systems and maximize the benefits of missions. Previous efforts to bridge the data gap on land relied on external information from land surface modeling via data assimilation or adopted statistical learning for the first mission using hydroclimate predictors to fill the gap. For the ice sheets, the ice mass balances models are often used. The objective of this research is to assess the value of Bayesian modeling for the data in the two missions by decomposing the geophysical signal into its temporal structures (e.g., secular trend, interannual variations, annual, semiannual, and residuals) and modeling each component using informed priors. The model parameters are treated as a degree of certainty, not as a fixed variable, allowing generation of sufficient sampling from the posterior distributions to cover much of the data and parameter space using a Markov chain Monte Carlo (MCMC) algorithm. Sampling over the missing data provides an approach to fill the data gap but results in high uncertainties. With ensuring the model convergence, the missed data are filled using the median of the posterior distributions, and the associated uncertainties are obtained at a 95% credible interval. We used the University of Texas at Austin, Center for Space Research (UT-CSR) mascon data to fill the gap and missing months in the two missions at a one-degree grid scale. Results show a good predictive performance with r-square > 0.6 over most of land and ocean. A lower r-square< 0.3 results are found in Sahara, southern Arabian Peninsula, north of Atlantic Ocean, and south of Indian and Pacific Oceans. Bayesian modeling is an economic data-driven approach to model GRACE missions gap and missing data.