Operational Technology Research Of 6-15 Days Numerical Weather Prediction Based On Predictable Components

In the past few decades, the skill of numerical weather prediction has been improved significant. Useful skill in medium-range weather forecasts from present-day numerical weather prediction models typically extends to about 6 days. On the basis of the success of short-range weather forecast, the major numerical weather prediction centers around the world engage in technology development for improving the accuracy and timeliness of weather forecasts. However, for longer time scale, such as 6-15 days medium and extended range, the forecast quality is still poor. It is necessary to improve the quality of the forecast results. Medium and extended range forecast is very important for disaster prevention and mitigation. Therefore, further investigation on medium and extended range forecast has important for science and application issue.In this work, we attempt to establish a real-time medium and extended range forecasting system. For the present difficulties on medium and extended range forecast, some innovative methods and strategies are proposed. Based on an operational dynamical extended range forecasts (DERF) model of NCC/CMA, the 6-15 days operational prediction technology are studied. A real-time medium and extended range forecasting system has been established. The major results and conclusions of this study are summarized as follows:(1) The relationship of predictability between spatial and temporal scales is analyzed, in order to provide the basis for separating the predictable components and chaotic components from the perspective of time and spatial scales. The predictability of the spectral coefficients of spherical harmonics and empirical orthogonal functions (EOF) are examined by employing variance analysis. Through the comparison of the predictability of the spectral coefficients of spherical harmonics and empirical orthogonal functions, it is clear that EOF spectral component showing a better relationship with predictability, because it better reflects the link with the low-frequency flow.(2) Based on predictability theory, a method for separating the predictable components and chaotic components has been proposed. The predictable components are defined as that components which the error growth will not exceed a certain threshold, and they are the predictant of deterministic forecast. The definition of predictability is provided by the explained variance of predictable components. By reducing degree of freedom low dimensional climate attractor and combining the relation of predictability and the spatial and temporal scales, the computing method of predictable components have been proposed for application.(3) Regard the predictable components, which defined as the dynamical equation is not sensitive to initial conditions in forecasts, as the predictant of deterministic forecast, it is helpful to avoid the rapid internal error growth. A targeted numerical model for predictable components has been established based on an operational dynamical extended range forecasts (DERF) model of NCC/CMA. The model building processes are similar to the processing procedure of high-frequency gravity waves in primitive equation model, and avoid the re-establishment of predictable components equations and numerical model with inhibiting the development of chaotic components.(4) An analogue selection method under small degree of freedom is proposed, which is achieved by reducing degree of freedom on low dimensional climate attractor and predictable components. The regularity of the analogy of error is investigated. Analyses about the relationship of analogue degree between initial and error are studied. The results show that when the forecast have similar initial conditions, the forecast error has analogical characteristics. Compared with the system error, the error under analogical initial condition is more close to actual forecast error that is based on current initial value. Meanwhile, in the spatial structure, forecast error between analogical initial conditions has good consistency.(5) In this study, we tried to find a solution to the problem for two difficulties in analogue-dynamical method, especially with the emphasis in precondition of analogue-dynamical method. Because the analogue selection and error correction are only for predictable components, forecasting experiments indicate that this method can effectively improve the deterministic forecast skill of large-scale predictable components, and avoid the adverse effect of rapid error growth in small-scale components. Meanwhile, the analogue selection and error correction are for all the variables and levels of model, so that the improvement of forecast are for all the variables and levels of model too, which lay a foundation for its operational application.(6) An ensemble prediction method is proposed for predictable components and chaotic components. The predictable components and chaotic components, with different characteristics, should be forecast in different ensemble ways. For the predictable components, the forecast error has been reduced by accounting for uncertainties of model error; meanwhile, the forecast information of chaotic components are not from initial conditions because the chaotic components are so sensitive to given initial conditions, and because of model error, the initial perturbation method, which aim to generate an ensemble model forecast; have a debatable premise in common. So the probability distribution of the chaotic components forecast, is not based on an ensemble of initial conditions, but from the perspective of most likely climate probability distribution. It is valid to avoid the effect of model error. Based on the ensemble prediction method for predictable components and chaotic components, a 6-15 days medium and extended range forecasting system is established.(7) Some key problems for medium and extended range forecast system are discussed. The determination of predictable components and some respects in analogue-dynamical method with historical data to improve the dynamical model are discussed, respectively. Meanwhile, some studies of sensitive experiments on medium and extended range forecast system are investigated.This study is committed to improve the operational 6-15 days dynamical medium and extended range forecast level. A series of studies show that, comparing to the original operational forecast system, the forecast skills of developed medium and extended range forecast system are significantly improved. The results indicate that the new forecast system has potential operational application prospects, and our results provide support for further improvement and development of the real-time medium and extended range forecasting system.