A land surface soil moisture data assimilation system based on the dual-UKF method and the community land model

Many studies have shown the deficiencies of the extended Kalman filter (EKF), even though it has become a standard technique used in nonlinear estimation. In the EKF method, the state distribution is propagated analytically through the first-order linearization of the nonlinear system, which can introduce large errors in variable estimation and may lead to suboptimal performance and sometimes divergence of the filter. The unscented Kalman filter (UKF) addresses these problems using a deterministic sampling approach to capture the posterior mean and covariance accurate to the third order for any nonlinearity, while the dual-UKF method uses two UKF filters (one for state variables and one for parameters, in contrast to only one filter in the usual UKF) to simultaneously optimize the model states and parameters using observational data. In this paper, we employ the dual-UKF method to account for the effects of land surface subgridscale heterogeneity and soil water thawing and freezing and implement it into the NCAR Community Land Model version 2.0 to build a data assimilation system for assimilating satellite observations of soil moisture. Experiments for two sites in north and south China show that this dual-UKF-based assimilation system outperforms the usual UKF- and EKF-based methods in reproducing the temporal evolution of daily soil moisture, especially under freezing conditions. Furthermore, the improvement also propagates, albeit to a lesser extent, to lower layers where observations are unavailable.