A Study Of Error Schemes In Estimating Soil Moisture With Ensmble Square Root Filter

Soil moisture is a key variable in the terrestrial water cycle and land-atmosphere interactions and plays an important role in the areas of weather, climate, hydrology, ecology and agriculture. However,ground-point observation is very accurate, but very limited in space; Remote sensing methods can get a wide range of surface soil moisture, but can not provide the root-zone soil moisture; Although land surface model (LSM) could simulate global soil moisture, the simulated soil moisture is not accurate because of inaccuracy of model physics, input parameters、atmospheric forcing and initial state variables. Data assimilation method, which is based on observation and model predicted information errors, can integrate observational information in the dynamic process of the numerical model, and provide better soil moisture profile estimates. However, assimilation method, especially the ensemble based assimilation method is dependent on description of the background and observation uncertain information. Therefore, handling and describing the various types of errors reasonably in the process of assimilation become a hot issue. Focusing on error handling methods in soil moisture assimilation, the thesis includes the following three parts:Firstly in charpter3, unknown model errors are regarded as a whole by using an additive time-correlated error sheme.The performance of this sheme is investigated, in which the ensemble square root filter (EnSRF) is used to assimilate surface soil moisture for the estimation of soil moisture profile. The results show that adding time-correlated errors to ensemble members can result in the increase of ensemble spread and then prevent ensemble convergence. The accuracy of soil moisture estimates is obviously improved.Secondly in chapter4, the performance of three inflation methods (i.e., constant multiplicative covariance inflation method, relaxation-to-prior perturbations(RTPP) method and relaxation-to-prior spread(RTPS) method) is evaluated in the context of using EnSRF to assimilate surface soil moisture for estimation of soil moisture profile. As relaxation-to-prior perturbations and relaxation-to-prior spread method inflate more where observations are dense, which are more reasonable compared to constant multiplication covariance inflation method,the estimated soil moisture is obviously closer to the true value. Compared to relaxation-to-prior perturbations method, the relaxation-to-prior spread method performs somewhat better and the corresponding estimated soil moisture is closer to the true value, although the relaxation-to-prior perturbations method produces more balanced analyses.In charpter5, the feasibility of simultaneously estimating the soil moisture profile and correcting the soil parameters with EnSRF is investigated, in the presence of parameter errors. The experimental results show that correcting soil parameters is very important to the estimation of soil moisture profile; uncorrected parameters will seriously influence the updating of model states. However, the data assimilation scheme of simultaneously updating the model states and model parameters is comparatively sensitive to parameters’initial guess; only appropriately selecting the initial guess can make sure that the parameters and model states are successful estimated at the same time.Finally, in chapter6, a detailed summary is presented.