Verification of improved analog ensemble methods for forecasting extreme temperature and wind speed in Beijing

Upon the current requirement of the extreme weather forecast and service, we developed two improved prediction schemes (that is, schemeⅠ and scheme Ⅱ) based on the ECMWF-IFS model (EC model) and the model output statistics (MOS) method on the basis of the Anolog Ensemble (AnEn) method. First, taking the EC model forecasts from 2016 to 2018 and their corresponding observations as the training dataset, the overall performance of schemeⅠ, scheme Ⅱ, and AnEn for the extreme temperature and wind speed in Beijingfrom January 1 to December 31 in 2019 is tested and evaluated against the observations at 364 stations. The results show that the prediction accuracy of schemeⅠ and scheme Ⅱ is better than that of AnEn for both extreme temperature (T) and wind speed (V_M), particularly for scheme Ⅱ. Second, according to the 2nd and 98th percentiles, the thresholds of extreme low temperature (T_m) and extreme high temperature (T_M) at the different stations in Beijing are -22.3 ℃ and 38.8 ℃, respectively. The overall prediction results of schemeⅠ and scheme Ⅱ for T in this region show that the two schemes are significantly improved compared to AnEn, and their mean absolute errors (EMA) are reduced by 11.90% and 21.43%, respectively. Similarly, according to the 98th percentile, the V_M threshold of each station in Beijing is set at 20.3 m·s^-1, and the EMA of V_M forecast with schemeⅠ and scheme Ⅱ is reduced by 23.08% and 26.52%, respectively, compared with AnEn. Finally, the prediction results of T_m, T_M and V_M at each station in Beijing show that schemeⅠ and scheme Ⅱ have improved in T and V_M on the basis of AnEn, and more than 94% of stations show that scheme Ⅱ has better performance. In addition, the spatial distributions of prediction accuracy of T and V_M show that the two improved schemes have better performance on the prediction of T and V_M in the mountainous areas than in the plain areas.