风暴尺度集合预报中的混合初始扰动方法及其在北京2012年“7.21”暴雨预报中的应用

风暴尺度集合预报系统（Storm-Scale Ensemble Forecast system，简称SSEFs）中集合成员之间发散度不足一直都是研究的难点。本文尝试了将Barnes空间滤波融入到集合转换卡尔曼滤波（ETKF）更新预报系统中的混合初值扰动法。该方案将ETKF方法的小尺度信息与来自于侧边界条件扰动的大尺度信息相结合，缓解了扰动在侧边界不匹配的问题。通过2012年北京“7.21”暴雨并使用邻位方法对比分析了不同初值扰动方案在不同时间尺度与空间尺度上的特征，在此基础上进一步探讨了构造混合初始扰动法的可行性。结果表明:ETKF试验所构造的初始扰动无法与侧边界条件扰动相匹配，混合后的初始扰动可以有效缓解SSEFs中由于初始扰动与侧边界扰动不匹配产生的虚假波动，其中大尺度信息保留较多的混合试验（ETKF80）和动力降尺度方案（Down）在减少虚假波动方面的效果最优；从集合离散度来看，在前期暖区降水阶段ETKF的离散度在小尺度上最大，随着锋面降水的开始，Down的离散度逐渐超过ETKF，而使用各滤波波段构造的混合试验同时具备ETKF与Down二者的特征。选择合理的滤波波段可以获得最为合理的离散度表现（ETKF180），说明仅考虑侧边界匹配（Down和ETKF80）并不能获得最合理的集合离散度，应综合考虑其他因素。从降水概率预报结果来看，选取合适的滤波波段所构造的混合扰动试验同样获得了较好的效果。

A Blending Method for Storm-Scale Ensemble Forecast and Its Application to Beijing Extreme Precipitation Event on July 21, 2012

In order to overcome the under-dispersive problem in the storm-scale ensemble forecast system (SSEFs), a new blending method to generate initial perturbation is designed and tested for the WRF SSEFs. This new scheme is based on the combination of the ETKF (Ensemble Transform Kalman Filter) and the Barnes filter for scale decomposition. This scheme is applied to the simulation of the Beijing extreme precipitation event on July 21, 2012and the neighborhood methods is employed to verify the performance of this new scheme. Results indicate that the blending method can effectively solve the scale mismatch problem in the lateral boundary in storm-scale ensemble prediction system, in which ETKF80 (with wavelength scale of 180 km) and Down (Dynamical downscaling) show the best overall performance. The Dispersion Fractions Skill Score (DFSS) shows that the ETKF has a larger spread in small scales during the period of warm area precipitation while Down produces a larger spread at large scales during period of frontal precipitation. The experiments with initial perturbations generated by the blending method take advantage of both the ETKF and Down. The ETKF180 (with wavelength scale of 180 km) generates the most reasonable ensemble spreads. Results also indicate that in order to get better ensemble spread in SSEFs, not only the lateral scale mismatch but also some other elements (such as the interaction between different scales of initial perturbation) should be considered. The blending method ETKF180 also improves the precipitation probability forecast.