Statistical Analysis And Correcting For Seasonal Weather Prediction Of DERF Model

In the past few dacades, the ability of numerical weather prediction has been improved obviously.However, seasonal weather prediction which is very important for disaster prevention and mitigation, still can not be put into service. It's necessary and useful to improve the performance of numerical forecast on this scale.Since statistical-dynamical prediction is proved by many studies as a simple and effective method for seasonal weather forecast, the output of Meteo France model of DEMETER project, is used to make a experment of statistical correction for model prediction on winter circulation in the extra-tropical of northern hemisphere.The innovative methods in this research are as follows:Unlike any previous researches, we focus on the Evaluating Scheme of model performance and the discussion of model correcting plans is put forward. Except for ACC(Anomaly Correlation Coefficient) which is commonly used, the ability of model forecast on NAO and anomalous components of seasonal mean circulation is also tested.The applicable fields for some general statistical-dynamical methods are discussed and blocking high index of Ural is predicted both by model forecast and corrected prediction.The major results and conclusions of this study are as follows:(1)First the ACC of model prediction is computered, then the model performance on anomalous components of seasonal mean circulation is tested by reducing the anomalous circulation of atmosphere to a feature line.(2)Here the EOFs are derived from the observation and will be used as invariant base modes to project the model prediction.Then the model forecast ability on these modes will be tested and prediction on'bad modes'will be corrected.(3)Three general schemes which are optimum subset regression (OSR) method, analogue method based on OSR result and analogue method based on previous indicators are used both in cross-validating and independent prediction to advance the model prediction on'bad modes'.Then the the applicable fields for these three statistical-dynamical methods are discussed.In this experiment, the OSR method failed, while the other two schemes show a possibility of improving the prediction techniques.(4)The original and corrected model predictions on Ural area are tested, then a new analogue correcting scheme for this area is tested in cross-validating. In addition, the Ural index of model prediction is calculated by two methods.Results show that the analogue correcting scheme for Ural area can advance numerical forecast remarkably.