基于BMA方法的地面气温的10～15 d延伸期概率预报研究

利用TIGGE资料提供的欧洲中期天气预报中心(ECMWF)、美国国家环境预报中心(NCEP)、英国气象局(UKMO)三个预报中心2013年6月1日至8月31日的地面2 m气温10~15 d预报资料,对延伸期地面气温进行贝叶斯模式平均(Bayesian Model Averaging,BMA)预报试验。结果表明,BMA方法的预报效果随训练期长度而改变,训练期长度为30 d时预报效果最优。BMA方法可提供全概率密度函数,定量描述预报不确定性的大小,且陆地上预报不确定性大于海洋上的预报不确定性,高纬度地区预报不确定性大于低纬度地区的预报不确定性。利用CRPS评分对BMA概率预报技巧进行评估,发现预报技巧随预报时效的延长降低,且预报技巧在海洋上优于陆地、低纬度地区优于高纬度地区。此外,3 d、5 d和7 d滑动平均的预报值反映某些天气过程的平均要素预报,对于提高10~15 d延伸期概率预报技巧有一定效果,且滑动天数越长,预报效果越好。

Extended range probabilistic forecast of surface air temperature using Bayesian model averaging

In this study, based on the 10-15 day extended range ensemble forecasts of European Centre for Medium-Range Weather Forecasts(ECMWF), National Centers for Environmental Prediction (NCEP)and United Kingdom Met Office(UKMO)in the TIGGE dataset, the probabilistic forecasts of surface air temperature during the period from 1 June to 31 August 2013 were conducted using BMA(Bayesian Model Averaging). The results showed that the forecasting skill changed with the length of the training period, reaching its optimal value when the length of the training period was 30 days. BMA could provide full PDF(Probability Density Function), and quantitatively describe the forecast variance and uncertainty. The uncertainty and error on the land(higher latitude)were larger than those on the sea(lower latitude). Moving average methods improved the forecast skill of surface air temperature, and the longer the moving average period was, the better of the forecast performance would be.