基于时效匹配法的融合降雨临近预报研究

基于天气雷达外推的降雨临近预报其有效预报期通常不超过3h,而数值天气预报的预热问题"Spin-up"往往导致短期内的预报能力较差。鉴于此,本研究通过"取长补短"设计了一种依据临界成功率指标(CSI)和均方根误差(RMSE)的时效匹配方法,对两者的降雨预报进行融合,并选取了位于大清河水系的阜平、紫荆关流域的4场时空分布均匀程度各异的典型降雨过程进行试验应用。研究基于像素追踪的PBN外推临近预报和数据同化的WRF模式预报,通过时效匹配法实现两者的融合,融合方案为:1 h以内(不含1 h)采用PBN外推预报,1～3 h采用PBN与WRF的融合预报结果,3 h以上(不含3 h)则全部使用数值天气预报。通过对4场典型降雨过程预报效果对比发现:(1)随预见期延长,PBN外推预报权重逐渐下降,WRF模式预报权重逐渐增加;(2)融合权重在降雨初期、末期的变化并不显著,但在降雨发展过程中融合交叉点出现明显变化;(3)融合预报的效果优于两种单独预报,时效匹配法在时空分布均匀的降雨场次表现最好,其次为时间不均匀空间均匀场次,在时空不均匀场次表现较差。

Blending rainfall nowcasting based on the time-effectiveness matching method

In order to overcome the deficiency that the effective forecast period of weather radar based extrapolation is less than 3 hours and the short-time forecast limitation of the NWP model due to its "spin up" problem,the time-effectiveness matching method is designed for rainfall Nowcasting blending,which regards the critical success index (CSI) and the root mean square error (RMSE) as evaluation indices. Four typical storm events with different rainfall distribution evenness in space and time are selected to study from the Fuping and Zijingguan watersheds in the Daqinghe river basin. Firstly,the 0-3h radar-based extrapolation is carried out by using the pixel-based Nowcasting (PBN) algorithm. Meanwhile the 0-6h NWP forecasting is completed by using the WRF model with the assimilation of GTS and radar reflectivity observations. The blending method is then designed as:the PBN extrapolation is used when the forecast lead time is within 1h (excluding 1h), the PBN and WRF blending results are blended and adopted for the 1h-3h lead times,and the WRF prediction is used with the lead time beyond 3h (excluding 3h). The results show that the blend weight of PBN decreases and that of WRF decreases with the increase of the forecast lead time. The weights do not have obvious changes in the start and end periods of the storm processes,but changes quickly with the blending intersections occurring earlier as the storms develop. Given the results and discussions of the four typical storm events, the present blending method can be an effective tool to improve the Nowcasting accuracy, and it performs the best for storms with evenly spatio-temporal rainfall distributions,and the second best for those with even rainfall in space but uneven rainfall in time, whereas the worst for uneven rainfall in both space and time.