GFS集合降雨预报的校正后处理研究

Study on statistical post-processing of GFS ensemble precipitation forecasts

耦合降水集合预报信息与水文模型是水文预报发展的一个重要方向。然而由于大气运行初始条件及模式的不确定性数值降雨预报不可避免地存在误差。基于全球集合预报系统(GFS)提供的1～8d预见期的降雨集合预报数据,研究了基于扩展型Logistic算法和异方差扩展型Logistic算法发展的5个统计后处理模型对淮河流域息县子流域GFS预报降雨的校正效果。结果表明,5个模型对GFS预报降雨均具有较好的校正效果,但随着预见期的增长,各个模型的校正能力呈衰减趋势。总体而言,相较于基于扩展型Logistic算法的3个模型,基于异方差扩展型Logistic算法的2个模型具有更优的校正能力。

Coupling ensemble precipitation prediction with hydrological models is an important development direction of hydrological forecasting.However,due to the uncertainty of the initial atmospheric conditions and model physics,numerical precipitation forecasts inevitably have errors.In this study,based on GFS ensemble precipitation reforecasts with a 1-8-day lead time,we analyzed five statistical post-processing models that were developed based on extended logistic regression(ELR)and heteroscedastic extended logistic regression(HELR)algorithms,and compared their correction effects on the GFS ensemble precipitation forecasts in Xixian sub-basin of Huai river basin.The results indicated that these five models all made significant improvements to the GFS raw forecast;but with the extension of the lead time,their correction effects tended to attenuate.In general,the two HELR-based models had better performance compared to the other three ELR-based models.