Data assimilation of GRACE terrestrial water storage information: solution assessment, error characterization and algorithmic procedure

Monday, 15 December 2014: 3:25 PM
Carly Sakumura and Srinivas V Bettadpur, Univ Texas - Ctr Space Rsch, Austin, TX, United States
The NASA/DLR GRACE mission has delivered over a dozen years of time-variable gravitational information that has seen widespread use in the study of processes in hydrology, oceanography, the cryosphere, and is particularly critical to inform, improve, and validate computational models of the Earth system. Assimilation of GRACE products into land surface models has been investigated using the monthly smoothed and destriped gridded-mass (Level-3) datasets. However, the coarse spatial and temporal resolution of these fields limits their use for data assimilation. This study offers advances in (1) improving the signal content retained from the GRACE satellite observations via the use of sliding window solutions and (2) improvements of the error characterization and data assimilation procedure of GRACE TWS information into the Community Land Model (CLM).

Presently, approximately thirty equally weighted days of data are used to estimate each gravity field. However, through differential weighting of the daily GRACE files used in each solution it is possible to optimize the desired signal characteristics while fulfilling the observability requirements. This differential weighting scheme creates a pseudo-daily “sliding window” product and increases the frequency retention by boosting the low frequency gain and providing gravity field solutions on shorter time scales. The effectiveness of the modified solutions is assessed through assimilation of both the sliding window and traditional monthly solutions into CLM using the Data Assimilation Research Testbed (DART). The error characterization of the GRACE observations is expanded within the DA framework to show and accurately reflect the spatial and temporal error patterns present in the data. Furthermore, the transformation between model and observation space and the disaggregation of the update calculated by the Ensemble Kalman Filter to the finer resolution CLM state variable space bears further examination. The addition of improved frequency retention, error characterization and model state-observation transformation algorithmics offers increased effectiveness and new methods of data extraction from time-variable gravity datasets from the GRACE and future geodetic observing missions.