Acquiring observation error covariance information for land data assimilation systems

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Tuesday, 19 January 2010: 11:30 AM
B304 (GWCC)
Wade T. Crow, USDA/ARS, Beltsville, MD

Recent work has presented the initial application of adaptive filtering techniques to land surface data assimilation systems. Such techniques are motivated by our current lack of knowledge concerning the structure of large-scale error in either land surface modeling output or remotely-sensed estimates of land surface water and energy balance variables. Most current adaptive techniques are based on classical whitening approaches in which a lack of temporal auto-correlation within filtering innovations is assumed to be a necessary and sufficient condition for optimal filter performance. However, the application of these approaches to the assimilation of remotely-sensed surface soil moisture has uncovered two serious problems. First, the iterative application of whitening approaches to land surface models leads to extremely slow convergence on optimal error parameters and is therefore not appropriate for satellite data sets with limited temporal heritage. Second, errors in available remotely-sensed soil moisture data sets are commonly too heavily auto-correlated for whitening approaches to function effectively. This presentation will illustrate these problems and develop an alternative methodology which circumvents both limitations. The approach is based on the application of a so-called “triple collocation” technique to independently estimate observational errors in remotely-sensed surface soil moisture. Such approaches estimate error magnitudes in a given geophysical variable by averaging across variations within three independently-obtained estimates of the variable. Here errors in surface soil moisture retrievals obtained from the Advanced Microwave Scanning Radiometer (AMSRE-E) are estimated via triple-collocation and used to constrain optimal modeling errors (by tuning modeling errors until normalized filtering innovations are variance unity). Real data validation results within several heavily-instrumented ground test sites reveal that the procedure leads to faster convergence to optimized error parameters and significantly enhances surface soil moisture estimates relative to existing adaptive filtering approaches.