结合历史资料的数值天气预报误差订正

Error correction of numerical weather prediction with historical data

基于历史资料的模式距平积分预报订正(Anomaly Numerical-correction with Observations,ANO)方法,结合欧洲中期天气预报中心的ERA-interim再分析资料和0.1°×0.1°分辨率的中国地面自动站与CMORPH卫星反演降水资料融合逐时降水产品,对高分辨率非静力天气研究和预报(WRF)模式的数值预报结果进行订正试验,检验了ANO方法对灾害性天气、尤其是对持续性强降水预报的订正改进效果。对1983—2013年7月中旬四川地区订正前后数值预报结果与观测和再分析数据的比较表明,ANO方法不仅在环流场的预报订正试验中有较为显著的效果,对模式降水预报结果也有改进,能够有效提高模式对强降水的预报精度和评分,减小预报偏差。对2013年7月8—13日高分辨率预报结果的ANO订正试验发现,订正环流场各变量均有所改进,其中,位势高度距平相关系数平均提高了7.8%,均方根误差平均降低了55.7%,降水(特别是暴雨以上量级)的ETS评分和TS评分也有不同程度的提高,并得到多年独立样本的高分辨率数值预报订正结果的支持。

Using the rnethod Anomaly Numerical-correction with Observations （ANO） based on historical observation data and anomaly integration, we performed a numerical correction to the high-resolution simulations with the nonhydrostatic WRF3.5 model. The ERA-interim reanalysis and the 0.1°×0.1° rainfall data, which is a blending of the surface Automatic Meteorologi- cal Station （AWS） rainfall observation and retrieved precipitation from the CMORPH data, act as the historical observational series. The impact on numerical prediction of disastrous weather, especially persistent severe rainfall is evaluated. The results of the application of ANO in Sichuan severe rainfall case in July 2013 illustrate that the correction not only improves the atmos- pheric circulation sinmlation obviously, but also makes the modeled severe rainfall more realistic in comparison with the obser- vation. In addition to the amelioration of the circulation and precipitation forecasting, as well as ETS and TS score of rain, the decrease of precipitation bias is also observed. The correction to the high-resolution simulation of the case 8 - 13 July shows ob- vious improvement in the circulation quantities, with an increase of Anomaly Correlation Coefficient （ACC） of the geopotential height by 7.8% and decrease of Root Mean Square Error （RMSE） by 55.7% in average. Meanwhile, ETS and TS score of precipitation （especially torrential rain and above） are also well corrected. The conclusion is supported with the independent samples of high-resolution numerical results across several years.