Study On The Different Land Data Assimilation Algorithms

In recent years, soil moisture has gained much attention as a very important variable in the land surface process researches. It is essential to correctly estimate and forecast soil moisture profile in numerous studies of atmosphere, hydrology, agriculture and climate. In this thesis, different methods are applied to estimate soil moisture profile, including the direct interpolation, adjoint-based four-dimension variational (4D-Var) method, an ensemble square root filter and a hybrid four-dimension variational method.Chapter 1 describes the important role of soil moisture, presents the problems, and then proposes the main aim and scope of this research.Chapter 2 summarizes three kinds of basic methods in estimating soil moisture, and also the variational data assimilation method, ensemble Kalman filter and the comparative study between them are fully explained.Chapter 3 presents how to set up an adjoint-based 4D-Var assimilation system to retrieve the soil moisture profiles by assimilating the surface soil moisture measurements. Firstly, we derive the soil moisture forecast difference equations and associated tangent and adjoint models. Secondly, we verify the tangent model, adjoint model and the gradient calculated by adjoint model. Finally, we do some synthetic tests to study the capability of the adjoint-based 4D-Var method in estimating the soil moisture profiles.Chapter 4 studies the capability of ensemble square root filter (EnSRF) to estimate soil moisture profiles by assimilating surface soil moisture observations. Firstly, we study how the ensemble number, surface soil moisture observational error and model error impact on the results. Secondly, we investigate some moments of the ensemble member pdfs, i.e. ensemble spread, skewness and kurtosis, to interpret the ensemble before every analysis step. And then, we do some tests to find the optimal covariance inflation factor which minimizes the error in the EnSRF. Finally, EnSRF is compared to direct insertion method. In Chapter 5 the adjoint-based 4D-Var method and EnSRF are compared. Some synthetic tests are done to study the impacts of surface soil moisture observational errors, model errors and observation frequencies on the estimation.In Chapter 6, we combine adjoint-based 4D-Var method with EnSRF method according to the result of Chapter 5 and propose a hybrid 4D-Var method to estimate soil moisture profiles. We do some synthetic tests to investigate the capability of hybrid variational method to retrieve the soil moisture profiles by assimilating the surface soil moisture measurements. The results indicate that the hybrid variational method can correctly estimate soil moistures in the deep layers.In Chapter 7 we summarize all contents in the thesis and predict the possible use of the methods in this thesis.