Application Researth On The Ensemble Kalman Filter (EnKF) With A Medium-Range Numerical Weather Prediction (NWP) Spectral Model At A T106L19 Resolution

One of the important advantages of the Ensemble Kalman Filter is to analysis the data by using the flow-dependant background error covariance. Meantime, as a development of the standard Kalman Filter, Ensemble Kalman Filter can be applied in an nonlinear system by breaking through the assumption of original linear system to study the numerous and complicated synoptic situation.Medium-Range Numerical Weather Forecast is always attracts peoples' attention. This paper mainly to study the applicability and feasibility of ensemble Kalman Filter, and to develop an effective and feasible Ensemble Kalman Filter data assimilation system with a global spectral model at a Medium-Range resolution.Firstly, we studied the assimilated ability of the Ensemble Square Root Filter in some different situations which is the different sizes of the observations, the different distribution of the observations and the different error distribution of the observations with a simple model---Lorenz96. As a result, we find that if the DOF of the observations is identical to that of the model, the filter will be more stable. And if the observation is less, covariance localization is vital to the stability of the filter. The analysis errors will be larger in the area with less observations. Filter will be more dependent on the spread of the ensemble members in the Non-Gaussian observation error situation.An Ensemble Square Root Filter data assimilation system with a Medium-Range Numerical Weather Prediction (NWP) Spectral Model at a T106L19 Resolution is developed. Here, we give some results of the sensitivity tests by OSSEs. The results show that the 40 ensemble members are enough to achieve the requirements of analysis error and the localization scale has relation to the ensemble size. Ensemble Kalman Filter gives a coordinated initial filed, so the forecast model doesn't need to use Normal Mode Initialization (NMI) in the current tests. Adopting the Vertical Error Correlation (VEC) will improve the assimilation result, especially reduce the temperature errors in the low levels of the model. Regular covariance inflation avoids the filter divergence to a certain degree. Introducing moist observations will be benefit to not only the moist field but only dynamic field, thermal field and quality field. And the 'uniq' scheme will be more suit to the data assimilation system.The result of OSSEs which simulated the real observation environment shows that the analysis errors and prediction error of the 500hPa height field mainly located near the trough-line and ridge line, and the prediction errors has a trend from fast to slow and then to fast with the growth of time. Southern Hemisphere doesn't have predictability. The result of the real observations data assimilation tests shows that the analysis errors and prediction errors of 500hPa height field in tropical and subtropical is smaller, and in the Mid-High latitude of northern hemisphere the errors is larger. The result of OI is opposite to that of Ensemble Kalman Filter.