Strategy And Methodology Of Dynamical Analogue Prediction

Based on a comprehensive review for numerical weather prediction and short-term climate prediction, it is pointed out in the present dissertation that the development of prediction strategy and methodology is an important approach to increase prediction performance besides improving observation and developing model. It is further emphasized the important values in the effective combination of statistical and dynamical methods by utilizing historical analogue information in dynamical prediction. So, on the basis of previous researches, some innovative strategies and methods, which are associated with numerical forecast correction, ensemble forecast, model post-processing technique, and predictability, et al., are put forward under the new framework of dynamical analogue prediction (DAP) by introducing analogy idea. Many exciting results related theoretical analyses and numerical experiments are achieved, which indicate the wide perspective of studies and applications on these strategies and methods. The major conclusions of this study may be summarized as follow:(1) The relationship of "analogue prediction—low-frequency flow regimes (LFFRs)—predictability" is physically examined. Both numerical experiments and diagnostic analyses show that the behaviors of forecast and its error are characterized by analogy both in theoretical and real atmospheric model at different timescale. By comparing statistical and dynamical analogue prediction, three classes of analogy problems and the physical basis of DAP are first raised, and the analogy principle of dynamical prediction are put forward.(2) The inverse problem that information of historical analogue data is utilized to estimate model errors, and the new idea on prediction of prediction errors, are raised. Based on the analogy principle of dynamical prediction, an analogue correction method of errors (ACE) is put forward and developed, which can identify specific historical data for the solution of the inverse problem in terms of the particularity of current prediction. Furthermore, the theoretical superiority of DAP equation to either analogue deviation equation or analogue correction equation of errors is found out. The new concepts such as period of analogue updating (PAU) and interval of correcting errors (ICE) are also raised.(3) Global forecasting experiments on Lorenz model show that, PAU, ICE, and interval of observation samples have significant impact on analogue correction of errors, as well as analogue number, systematical and random observation errors. It is found that the simple linear estimation method can well replace the hyperplane approximation method in actual application of ACE. Forecasting experiments on T63 model exhibit evident improvement of prediction skill and a new errors diagnosis procedure by rerunning is developed. Actual predictive efficiency may be influenced by the analogy problem of sea surface temperature.(4) Focused on finite persistence of initial analogue, a new method of DAP based on multi-reference-state updating is put forward by introducing information of multi historical analogues. In this method, not only analogue-dynamical model and ACE may be regarded as dynamical prediction kernel, but also the optimal forecast vectorcan be estimated from those forecasts based on different reference states. The considerable predictive skill is shown both in the experiment results of Lorenz model and T63 model.(5) On the basis of analogue correction of errors, a model error parameterization (MEP) is developed, and a new method of analogue-dynamical ensemble forecast (ADEF) is further raised by employing MEP. The new technique associating with weighted ensemble which is developed from simple arithmetical mean of multi results of DAP, and in-process ensemble by introducing analogue updating, is also put forward. An ADEF experimental system by applying this new ensemble method to analogue-dynamical extended-range forecasting is primarily established and developed, by which experiments of real-time quasi-operational monthly forecast show exciting results.(6) The theoretical problems and errors estimate methods on which the final ACE is applied to monthly and seasonal short-term climate prediction are deep discussed, which documents the feasibility of estimating current forecast errors from historical forecast errors based on analogues. The potential capability of final ACE applied to actual prediction is successful validated in a predictability study and it is shown that this post-processing method has the well performance of improving predictive skill and reproducing prediction variance in the forecast experiments of both monthly mean circulation and summer circulation and rainfall.(7) Some key problems in selecting analogues such as defining analogue indices, selecting analogue variables, determining analogue metric and analogue region range, et al., are deep examined. A two-level method for selecting analogue and an integrated analogue index are raised in ocean-atmospheric system, and related problems of evolution analogue and tendency analogue are discussed, and a spectrum-coefficient-based analogue index is developed. The impacts of analogue metric on prediction with different timescales and the influences of spatial degree of freedom on selecting analogue and predicting problem are discussed, respectively. It is also found that there exists a significant relationship of anti-correlation between analogue quality and predictive skill on analogue correction of errors.(8) The EOF dataset with a few degrees of freedom is primarily established. In order to investigate the close linkage among analogue prediction, the LFFRs and predictability, the classification studies of summer LFFRs in northern hemispheric extratropics are performed, as well as the large-scale low-frequency rainfall regimes in summertime over China which is the important component of LFFRs. All these theoretical and diagnostic studies are helpful to improve the relative process of selecting analogue.