Establishing the Framework for Land Data Assimilation of GRACE Terrestrial Water Storage Information

Abstract:

Assimilation of terrestrial water storage (TWS) data from the Gravity Recovery and Climate Experiment (GRACE) mission into current land surface models can correct model deficiencies due to errors in the model structure, atmospheric forcing datasets, parameters, etc. However, the assimilation process is complicated by spatial and temporal resolution discrepancies between the model and observational datasets, characterization of the error in each, and limitations of the algorithms used to calculate and update the model state. This study aims to establish a framework for hydrological data assimilation with GRACE. GRACE is an independent and accurate but coarse resolution terrestrial water storage dataset. While the models cannot attain the accuracy of GRACE, they are effective in interpretation and downscaling of the product and providing continuity over space and time. Accurate assimilation of GRACE TWS into LSMs thus will take the best characteristics of each and create a combined product that outperforms each individual source.

More specifically, this framework entails a comprehensive analysis of the deficiencies and potential improvements of the satellite data products, the assimilation procedures and error characterization, and assimilation effectiveness. A daily sliding window mascon GRACE TWS product, presented previously, was developed to reduce smoothing in time and space during assimilation into the Community Land Model 4.0. The Ensemble Kalman Filter assimilation algorithms are analyzed and adapted for use with the new products, GRACE error information, and model characteristics. Additional assimilation tools such as Gaspari-Cohn localization and ensemble inflation are implemented and tuned for the model and observation properties. Analysis of the observational data, model data, sensitivity and effectiveness of the assimilation routines, and the assimilated results is done through regional comparison with independent in-situ datasets and external model results.