An ensemble-Kalman filter-based dual assimilation of thermal-IR and passive microwave retrievals of soil moisture

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Tuesday, 19 January 2010: 9:15 AM
B304 (GWCC)
Christopher R. Hain, Univ. of Alabama, Huntsville, AL; and W. T. Crow, M. C. Anderson, and J. R. Mecikalski

Recent work in hydrologic data assimilation has shown promise with respect to the improvement of predicted hydrologic variables such as soil moisture (Crow et al. 2005; Reichle et al. 2007; Kumar et al. 2008). The ensemble Kalman filter (EnKF) is the most widely used framework in the area of hydrologic data assimilation, as its allows for the nonlinear propagation of background model error. Previous studies have focused on the assimilation of synthetic soil moisture observations or retrievals of surface soil moisture from passive microwave (PM) sensors such as AMSR-E (Reichel et al. 2002; De Lannoy et al. 2007; Crow and Wood 2003; Reichle and Koster 2005; Reichle et al. 2007).

Hain et al. (2009) demonstrated the feasibility and relative skill of retrieving a composite surface and root-zone soil moisture signal from a thermal-IR framework using surface flux estimates from the Atmosphere-Land Exchange Inversion model (ALEXI). Retrievals of soil moisture from ALEXI provide information relative to the vegetation cover of the pixel; over dense vegetation a root-zone signal is provided, while over bare soil a surface soil moisture signal is provided.

In this study, retrievals of ALEXI (TIR) and AMSR-E (PM) soil moisture are assimilated with the Land Information System (LIS) and the NOAH LSM. A series of data assimilation experiments are completed with the following configuration, (a) no assimilation, (b) only ALEXI soil moisture, (c) only AMSR-E soil moisture, and (d) ALEXI and AMSR-E soil moisture. The first set of experiments is completed by varying the prescribed retrieval and background model error, along with the analysis of innovation statistics, to solve for the “optimal” retrieval and background error covariance for each configuration. The assimilation configurations are completed again using the “optimal” retrieval and background error covariance and validated against in-situ soil moisture observations from the Oklahoma Mesonet and SCAN sites across the central and eastern United States. Finally, the relative skill of each assimilation configuration is quantified through an extensive data-denial experiment, where the LSM is forced with an inferior precipitation dataset (in this case, the TRMM 3B42RT precipitation dataset). The ability of each assimilation configuration to correct for precipitation is quantified through the comparison of the results with a single simulation over the same domain with a high-quality (NLDAS) precipitation dataset.