相似集合预报方法在北京区域地面气温和风速预报中的应用

相似集合是近年来提出的一种基于相似理论、大数据挖掘和集合预报思路的统计释用方法。文中首先介绍了相似集合的基本原理，并应用该方法对北京快速更新循环数值预报系统（BJ-RUC）v3.0预报地面要素开展了订正释用试验。结果表明，相似集合订正后，在0—36 h预报时段内，10 m风速的均方根误差降低44%，2 m气温的均方根误差降低22%，均方根误差均显著减小。对比测站预报误差的水平分布，相似集合方法的应用对于提升非城区站点的10 m风速预报、复杂地形区域的2 m气温预报具有更为明显的效果。相同预报因子的相似集合和支持向量机方法对模式10 m风速和2 m气温预报均具有显著且相似的订正效果，但相似集合方法具有计算资源需求较少、不需要大量人工干预的优势。相似集合方法形成的集合较好地模拟了模式平均误差的增长情况，集合离散度与集合平均均方根误差表现出理想的统计一致性，即相似集合方法在形成确定性预报的同时，还能够提供预报要素的不确定性或概率信息。因此，相似集合方法在模式预报订正及释用方面具有广阔的应用前景。 

Application of analog ensemble method to surface temperature and wind speed prediction in Beijing area

Analog Ensemble (AnEn) is a statistical interpretation method that is based on similarity theory, big data mining and ensemble forecasting. The basic principle of this method is introduced first. It is then applied to revise the ground elements predicted by BJ-RUCv3.0. The results show that the root-mean-square-errors (RMSEs) of 10 m wind speed and 2 m temperature are significantly decreased during the forecast lead times of 0-36 h after using AnEn. The RMSE of 10 m wind speed is decreased by 44%,and the RMSE of 2 m temperature is decreased by 22%. Comparing horizontal distribution of prediction errors at stations, it is found that the application of the AnEn method has more obvious effects for 10 m wind speed prediction at non-urban stations and 2 m temperature prediction at complex terrain area stations. AnEn and Support Vector Machines (SVM) with the same predictors have significant and similar effects on Numerical Weather Prediction (NWP) model predictions of 10 m wind speed and 2 m temperature. However, the AnEn method has the advantage of needing less computing resource and less manual intervention. The pattern of average error growth can be well simulated by the AnEn method, and the RMSE of AnEn mean and average ensemble spread show ideal statistical consistency. It not only yields deterministic predictions, but also provides uncertainty or probability information for the prediction factors. Therefore, the AnEn method will have a broad application prospect in NWP model interpretation prediction.