Estimation of Model Error Covariance Matrices of the Extended Kalman Filter for Validation of AMSR-E Soil Moisture Product

Xiwu Zhan, Paul R. Houser, Jeffrey P. Walker and Rolf H. Reichle

Poster Presentation

Due to imperfect instrument calibration and inversion algorithms, geophysical noise, representativeness error, communication breakdowns, and other sources, land surface soil moisture retrieved from satellite or aircraft remote sensing instruments contain uncertainties. As an effort of NASA’s AMSR-E Science and Validation Team, the extended Kalman filter (EKF) data assimilation scheme is implemented in NASA’s Land Data Assimilation System (LDAS) for the validation of the surface soil moisture product. A key step of this validation approach is the determination of the covariance matrices of model errors. In this study, a series of numerical experiments on the sensitivity of EKF output to different hypothetic model error covariance settings is carried out using the Mosaic model in the LDAS and the surface soil moisture data derived from the brightness temperature observations by the Tropical Rainfall Measurement Mission (TRMM) Microwave Imager (TMI) for the area of Southern Great Plains Hydrology Experiment 1999 in Okalahoma. By introducing various types of errors into the derived soil moisture data, the best model error covariance settings for correcting the introduced hypothetic observation errors are investigated. In addition to these stationary settings for model error statistics, parameterization schemes of nonstationary characteristics of model error covariance are also tested. Results of the numerical experiments and performance of EKF using the nonstationary model error covariance parameterizations for correcting the various types of observation error will be presented.