基于数值预报与雷达外推预报融合方法的降水回波短临预报试验研究

Experimental study on short-term and impending prediction of precipitation echo based on blending method of numerical prediction and radar extrapolation prediction

采用傅里叶-梅林变换、多尺度光流法及威布尔分布,对湖北省2020年6—7月4次降水过程雷达回波进行多尺度预报试验及其相位和强度校正。在此基础上,利用双曲正切函数对校正后的模式降水回波预报和雷达回波外推临近预报进行融合。最后,基于预报技巧评分和平均绝对误差及命中率等指标对不同时效、尺度及回波阈值的预报结果进行定量分析。结果表明:(1)融合后的0~3 h降水回波预报在范围和位置上均较模式预报和雷达外推预报改进明显,尤其对强回波预报有明显优势,对对流预报有积极作用;0.01°×0.01°尺度的0~1 h预报效果明显好于其他尺度及预报时效。(2)武汉RUC模式预报效果最差,平均绝对误差(mean absolute error,MAE)最大为6.1~8.2 dBZ,而融合预报效果最好,MAE最小为4.7~6.5 dBZ。0.01°×0.01°尺度下融合预报的命中率(probability of detection,POD)随回波阈值和预报时效增加而降低,而其他尺度下20 dBZ回波阈值的平均POD最大、MAE最小,平均POD(MAE)均高于(低于)其他2种预报。总体来看,融合预报明显优于单一预报,对改进0~3 h定量降水预报有一定参考。

The Fourier Merlin transformation, multi-scale optical flow method and Weibull distribution were used to carry out multiscale prediction experiments of radar echo on four precipitation cases from June to July 2020 in Hubei Province, and the phase and intensity correction of model products were realized. On this basis, the corrected radar echo prediction of numerical model and radar echo extrapolation prediction were blended by using hyperbolic tangent function. Finally, the prediction effect of blending method with different prediction time, scales and thresholds of echo intensity was quantitatively analyzed by means of prediction skill score, mean absolute error(MAE) and probability of detection(POD). The results are as follows:(1) Compared with model prediction and radar extrapolation, the 0-3 hours precipitation echo predicted by the blending method improved obviously in range and location, the advantage of blending prediction was obvious, especially to strong echo, and it had a positive influence on prediction of convection. The 0-1 hour prediction effect of precipitation echo with 0.01°× 0.01° spatial scale was obviously better than those with other scales and prediction time.(2) MAE of Wuhan RUC model prediction was the largest with a range of 6.1-8.2 dBZ, while for blending forecast it was the smallest with a range of 4.7-6.5 dBZ. POD of blending prediction for 0.01°× 0.01° scale decreased with increase of echo threshold and prediction time, while the average POD was the maximum and MAE was the minimum for precipitation echo with 20 dBZ threshold at other scales, the average POD(MAE) of blending prediction was higher(lower) than other two kinds of prediction. On the whole, the blending prediction was obviously superior to single prediction, and it can provide reference for improvement of 0-3 hours quantitative precipitation forecast.