安徽省WRF模式短时强降水的预报检验

利用2005-2015年安徽省内1162个站点观测资料简要分析了短时强降水的时空分布特征,并利用中国气象局CLDAS(CMA Land Data Assimilation System)近实时降水资料检验2012-2015年安徽省WRF(Weather Research and Forecast)模式对短时强降水的预报性能,探讨不同空间插值方法、检验方法对预报效果的影响,以评估模式预报短时强降水的应用价值和使用注意事项。结果表明:短时强降水主要发生在大别山区和皖南山区;一年中发生次数呈单峰分布,集中于6-8月;日变化呈双峰状,强峰为北京时间下午15:00-19:00,弱峰为06:00-09:00,两个低谷分别为01:00、12:00前后。在两分类评分TS(Threat Score)检验中,各个季节评分均十分低,插值方法对TS评分影响不大。邻域法FSS评分(Fractions Skill Score)检验中,春季FSS评分低,最高仅可达15%,空间窗、时间窗、时间超前或滞后变化对FSS评分的影响不如夏季、秋季明显;夏季,不考虑时间窗时,单独的时间超前或滞后不能提高预报准确率;秋季,模式分别滞后1h或滞后2h预报结果优于同期预报,而超前1h或超前2h预报结果低于同期预报,表明秋季WRF模式对短时强降水的预报有一定滞后性。

Capability of Forecasting Short Term Precipitation Based on WRF in Anhui

Based on the hourly rainfall data of 1162 stations in Anhui from 2005 to 2015, the spatial and temporal distribution characteristics of short term heavy rains are analyzed. The results show that short term heavy rainfall happens mainly in Ta pieh Mountains and Wannan Mountains. The number of short term heavy rains changes in a single peak pattern annually, concentrated in June to August, and from October to next March, and short term heavy rainfall hardly happens. The diurnal change exhibits a double peak pattern: its stronger peak appearing from 15:00 to 19:00, and the weaker peak appearing from 06:00 to 09:00, while the two valleys around 01:00 and 12:00, respectively. Also, the capability of forecasting short term heavy rainfall based on the WRF model is estimated in different interpolation and assessment methods. The Threat Scores (TS), calculated in different interpolation methods, are all under 2%, and there is little difference among them. The Fractions Skill Score (FSS), a neighborhood based verification measure, is also used to assess the skill of forecast. The study shows that FSS varies with seasons, the space and time windows, and time bias. FSS in spring is the lowest, under 15%; it is more faintly affected by time bias, spatial and temporal neighborhoods than those in summer and autumn. In summer, the FSS curves, without temporal neighborhoods, are much similar, and obviously below those with temporal neighborhoods. In autumn, the model FSS with 1 to 2 hours later is higher than synchronous FSS, while FSS with 1 to 2 hours ahead is lower, which also happens in spring, but not so clear.