雷达资料与中尺度数值预报的融合方法研究及其在临近预报中的应用

摘要

提通过融合多普勒天气雷达资料与中尺度数值预报产品，发展了一种便于临近预报业务使用的方法。该方法首先通过相关分析计算当前相同时刻雷达估测降水与中尺度数值预报的反射率因子估测降水之间的位置偏差，导出一个位移偏差矢量场；然后，利用人机交互的方式对矢量场进行分区，并对各分区的矢量场进行平滑处理，计算出各分区的平均位移偏差矢量；最后，采用最小二乘法对各分区连续多次的平均位移偏差矢量进行线性拟合，得到各分区平均位移偏差矢量随时间的变化特征，订正未来时段相应区域的数值预报反射率因子估测降水的位置偏差。利用该方法对2012和2013年夏季发生在重庆西部、四川东部的3次强降水天气过程进行临近预报试验并对预报结果进行了检验，结果表明：对0～2 h 的临近预报，融合预报效果总体上优于模式预报效果；另外，与雷达外推定量降水预报相比，0～1 h 雷达外推预报效果优于融合预报效果，1～2 h融合预报效果优于雷达外推预报效果。%A nowcasting method based on blending Doppler weather radar data and mesoscale numerical weather prediction (NWP)model products is presented.The method is as follows:Firstly,by using cor-relation analysis,position errors are calculated between radar precipitation estimate and precipitation esti-mated from reflectivity factor from the output of NWP model in this same time,and thus displacement de-viation vectors fields are obtained.Then,displacement deviation vector fields are partitioned with human-computer interaction and each deviation vector field gets smoothed,so the average displacement deviation vector of each partition is obtained.Finally,the trend variation characteristic of average displacement devi-ation vector of each partition with time is established by using least square method to linearly fit the con-tinuous time multiple average displacement deviation vectors for each partition,and according to the trend, spatial position deviation of precipitation estimated from reflectivity factor from the output of NWP model is corrected in the future periods.The method was once applied to three severe prceipitation cases in the summers of 2012 and 2013 that happened in the west of Chongqing and the east of Sichuan.The nowcast-ing verification results show that for the 0-2 h nowcasting,the performance of blending forecasts is gen-erally superior to model forecasts.Compared with quantitative precipitation forecast (QPF)of radar-based extrapolation,the performance of radar-based extrapolation QPF is superior to blending forecasts in the first hour but the performance of blending forecasts is superior to radar-based extrapolation QPF in the second hour.