光流法在乌鲁木齐机场多普勒雷达回波外推中的应用

Application of Optical Flow Methods in Doppler Radar Echo Extrapolation at Urumqi Airport

摘要: 本文采用Georgy Ayzel等提出的4种光流法计算模型，使用乌鲁木齐机场多普勒雷达组合反射率回波分别对一次强降水和两次雷雨大风天气的回波进行未来2小时逐6分钟的临近外推实验，并对外推预报的回波进行主观、客观评估，结果如下：(1) 利用光流法对雷达组合反射率图像进行临近外推是可行的，稠密光流算法较稀疏光流算法在雷达回波的临近外推上更具优势。(2) 4种光流模型对大尺度系统约束下演变缓慢回波的外推效果较中小尺度对流系统更优，对稠密光流算法而言，中小尺度对流系统的回波外推时间越短，准确性越高，60 min后的外推预报不具有很好参考价值，而对大尺度系统约束下缓慢演变的回波其外推预报至2小时仍具有较高的准确性，具有较好的参考价值。(3) 光流法无法对回波的生消、强弱变化进行很好的外推预报，因此当回波在机场或关注区域附近生成时，光流法外推可能是失败的，对此类系统的外推预报还需要如数值预报等手段进行辅助。(4) 对于大尺度条件约束下的强降水回波，考虑旋转效应的DenseRotation模型比其它模型有更好的外推效果，而对于类似中小尺度系统等快速演变的回波，不考虑旋转效应的Dense模型具有更好的外推效果。

Abstract: This study utilizes four optical flow calculation models proposed by Georgy Ayzel et al. to conduct nowcasting experiments on composite reflectivity echoes from Doppler radar at Urumqi Airport. Experiments were performed for one heavy precipitation event and two thunderstorm high-wind events, generating 6-minute interval extrapolations for a lead time of up to 2 hours. Subjective and objective evaluations of the extrapolated echoes yielded the following conclusions: (1) Optical flow methods are feasible for short-term extrapolation of composite reflectivity images, with dense optical flow algorithms demonstrating superior performance over sparse optical flow algorithms. (2) All four models exhibit better extrapolation accuracy for slowly evolving echoes under large-scale weather systems compared to meso- and micro-scale convective systems. For dense optical flow algorithms, shorter extrapolation times (within 60 minutes) provide higher accuracy for convective systems, whereas extrapolations beyond 60 minutes lack reliability. In contrast, extrapolations for large-scale systems maintain high accuracy even at 2 hours. (3) Optical flow methods are unable to effectively predict the initiation, dissipation, or intensity changes of echoes. Consequently, when echoes develop near airports or areas of interest, optical flow extrapolation may fail, requiring supplementary methods such as numerical weather prediction. (4) For heavy precipitation echoes under large-scale conditions, the DenseRotation model, which accounts for rotational effects, outperforms other models. For rapidly evolving systems such as meso- and micro-scale convection, the standard Dense model (without rotational effects) achieves better extrapolation results.