Accounting for Spatial Correlation Errors in the Assimilation of GRACE into Hydrological Models through localization

M. Khaki, M. Schumacher, E. Forootan, M. Kuhn, J.L. Awange, A.I.J.M. van Dijk

Research output: Contribution to journalArticle (Academic Journal)peer-review

29 Citations (Scopus)
203 Downloads (Pure)


Abstract Assimilation of terrestrial water storage (TWS) information from the Gravity Recovery And Climate Experiment (GRACE) satellite mission can provide significant improvements in hydrological modeling. However, the rather coarse spatial resolution of GRACE TWS and its spatially correlated errors pose considerable challenges for achieving realistic assimilation results. Consequently, successful data assimilation depends on rigorous modelling of the full error covariance matrix of the GRACE TWS estimates, as well as realistic error behavior for hydrological model simulations. In this study, we assess the application of local analysis (LA) to maximize the contribution of GRACE TWS in hydrological data assimilation. For this, we assimilate GRACE TWS into the World-Wide Water Resources Assessment system (W3RA) over the Australian continent while applying LA and accounting for existing spatial correlations using the full error covariance matrix. GRACE TWS data is applied with different spatial resolutions including 1° to 5° grids, as well as basin averages. The ensemble-based sequential filtering technique of the Square Root Analysis (SQRA) is applied to assimilate TWS data into W3RA. For each spatial scale, the performance of the data assimilation is assessed through comparison with independent in-situ ground water and soil moisture observations. Overall, the results demonstrate that LA is able to stabilize the inversion process (within the implementation of the SQRA filter) leading to less errors for all spatial scales considered with an average RMSE improvement of 54% (e.g., 52.23 mm down to 26.80 mm) for all the cases with respect to groundwater in-situ measurements. Validating the assimilated results with groundwater observations indicates that LA leads to 13% better (in terms of RMSE) assimilation results compared to the cases with Gaussian errors assumptions. This highlights the great potential of LA and the use of the full error covariance matrix of GRACE TWS estimates for improved data assimilation results.
Original languageEnglish
JournalAdvances in Water Resources
Early online date1 Aug 2017
Publication statusE-pub ahead of print - 1 Aug 2017


  • Data assimilation
  • Localization
  • Hydrological model

Fingerprint Dive into the research topics of 'Accounting for Spatial Correlation Errors in the Assimilation of GRACE into Hydrological Models through localization'. Together they form a unique fingerprint.

Cite this