低纬度地区局地暴雨的神经网络预报方法研究

THE RESEARCH ON METHODS OF FORECASTING LOCAL RAINSTORMS IN LOW LATITUDE AREA USING THE NEURAL NETWORK

以南宁市所辖8个站暴雨集中的6—8月逐日降水量作为预报对象,采用人工神经网络方法进行了新的数值预报产品释用预报方法研究。通过运用动力相似法,结合日本降水预报模式对未来暴雨发生的可能性进行判别,然后通过对欧洲中期数值预报中心预报场进行滑动分区车氏展开计算,求出与降水量序列相关较好的预报因子,并对这些因子进行自然正交分解,有效浓缩数值预报产品因子,建立了南宁市逐日暴雨的神经网络释用预报模型。利用该预报模型,对2006年6—8月的逐日暴雨预报试验结果表明,该预报模型对南宁市的暴雨强降水具有很好的预报能力。

The daily rainfall from June to August, a time of concentrated torrential rain, taken from eight stations of Nanning, are used as the object of forecast in our study of new methods in interpreting numerical prediction products employing artificial neural networks. In conjunction with a Japanese rainfall forecast model, the technique of dynamic similarity is used to distinguish future possibilities for rainstorms to take place. Then, the Chebyshev sliding-window expansion is applied to the forecast field by the European Center for Medium-range Weather Forecast （ECMWF） for forecast factors best correlated with the series of rainfall. The factors are then studied with empirical orthogonal function （EOF） to reduce efficiently factors for numerical prediction products and to set up a forecast model for day-to-day torrential rains interpreting the neural network. As shown from the model results of the forecast experiment for June to August 2006, it is suggested that the model does well in forecasting torrential rains over the Nanning area.