三类强对流天气临近预报的模糊检验试验与对比

Fuzzy Verification Test and Comparison of Three Types of Severe Convective Weather Nowcasting

强对流天气具有尺度小、演变快的特点,为了满足强对流预报检验、评价的需求,本文引入了模糊检验方法,该方法通过在空间等属性上进行尺度变换处理,可获得预报在不同空间尺度上的评价信息。以中国气象局SWAN等短临预报业务系统提供的1 h回波外推预报为例,对三种类型强对流天气系统进行了模糊检验试验对比,并据此构造了三种理想强对流天气模型,进一步研究了各种模糊检验方法的特性,发现:相对于"点对点"的传统检验方法,模糊检验能够在不同尺度和评价策略上给出有关预报的更多信息,给予预报更加全面和客观的评价;针对不同的评价策略,同一个预报的最优尺度是有差异的;不同的模糊检验方法各有特点,适用范围也有差异;相对于传统检验方法,模糊检验方法的应用范围更广,尤其是当预报偏差达到一定程度时,多种模糊检验方法仍然能够给出有参考意义的评分。综合来看,对于高阈值、小尺度特征的强对流事件,低判别标准的最小比例法、模糊逻辑法和多事件列联表等检验方法更有应用价值。

Severe convective weather is hard to forecast because of the character of small scale and rapid de- velopment. Fuzzy verification methods can get evaluation information at different spatial scales by using a spatial window or neighborhood surrounding the forecast and/or observed points, and are introduced into the verification of severe convective weather in this article. Focusing on the three types of severe convec- tive weather, some operational nowcasting products like as one-hour reflectivity extrapolation product of the Chinese Meteorological Administration （CMA） SWAN （Severe Weather Analysis and Noweasting） system, are verified using the fuzzy method. Then, three ideal severe convective weather models are also built and verified to give a further study on the above-mentioned methods. The results show that compared to traditional metrics with the stagey of ＂point to point＂, fuzzy verification can glean additional information in different scales and evaluation strategies, evaluating forecasts more comprehensively and objectively. Based on different evaluation strategies, one forecast has different optimal scales and each fuzzy verification method has its own feature and application. When forecast has large bias, fuzzy verification methods canstill give effective or ＂useful＂ scores while traditional metric can only give poor scores. For the severe con vective events with characteristics of high thresholds and small scales, the fuzzy verification methods inclu ding minimum coverage with low fraction, fuzzy logic and multi-event contingency table show more poten tial value than the traditional ones.