The Research Of Extracting Predictable Components And Forecasting Techniques In Extended-range Numerical Weather Prediction

It is of great importance for the extended-range forecasting in the coming10-30days to prevent and relieve meteorological disasters which have important impacts on national economy. The extended-range forecasting is still on its infancy stage all over the world, and the corresponding operational prediction needs further improvement. At present, it is slow for the improvement of the extended-range forecast. The causes are the limit of atmospheric predictability.Based on the operational dynamical extended range forecast model of National Climate Center of China Meteorological Administration and the princeple of extracting predictable components, a method of extracting predictable components from the complex atmospheric model is proposed in the view of the error growth of the numerical model variables.The model for the coming10-30days is established based on the predictable components. The results of the hindcasting for the model are good.Created forecasting component model to the existing operational forecast model, without having to rewrite the program to describe the equations of atmospheric dynamics, and better portability.It is promising for the model to apply to the operational forecasting. The results and conclusion are as following:(1) A valid method is proposed to extract predictable components in the complex numerical model.The growth of the forecasting error for different components is studied by establishing the nonlinear mapping between the initial field and the predict field and employing tangent linear approximation for the operator of the error growth. The predictable components are defined as the ones whose growth errors are relatively slow, the remaining components as the non-predictable components. The predictability is defined from the explained variance the predictable components occupation. Using Lorenz model and its tangent and adjoint model, the evolution of the error growth operator is studied, and the impact of the forecasting error in different directions on the limit time of the forecast.(2) A fast non-adjoint algorithem is proposed to extract predictable components based on the CNOP.A reasonable objective function between the initial field and the predict field is established, and an optimal algorithem is adopted in every step to extract the base of the predictable components in the process of the model integral. The algorithem avoids the huge work to write the adjoint model and simplifies the calculation. The validity of the algorithem is proved by the Lorenz model. It is found that the forecasting results of the predictable model have a longer predictable time than the original model through the forecasting experiments.(3) A predictable component model is established based on the T63L63model.The predictable components is extracted from the complex model using the fast non-adjoint algorithm. The scheme of how to establishment the predictable components model is illustrated. The forecasting results in different height levels and regions by historical case hindcasting are compared. It is proved that the predictable components model can improve the forecasting level of the extended-range forecasting in the coming10-30days.(4) The analogue correction technique based on historical data information to improve the predictable components model is developed.The historical data is divided into predictable and unpredictable components on the base of the predictable components. For the predictable components in the model, the analogue field can be found in the historical data. The process decreases the dimensions of the variables in the judgement of the analogue field, removes the fast components'influence which is sensitive to the initial values, and provides useful information in the model integral. The experiment based on the predictable components and analogue correlation in the integral process to the predictable components proves that the technique can effective improve the model forecasting level of T63L16.(5) The ensemble forecasting is employed to the chaotic components based on probability method.The predictable components of the historical analogue field is integrated in the predictable model. The integrated results subtract by the real time data is what we call chaotic components under the ananlogue initial value conditions.Using these chaotic components in the historical data,the probability distribution of the chaotic components is reasonable estimated in the process of prediction.