Abstract:

By using the Butterworth band-pass filter to process the daily gridded data of multi-factors’ fields of NCEP/NCAR about the three levels (200 hPa, 500 hPa and 700 hPa) from 1980 to 2014, and a 30-60 days low frequency oscillation was kept. By means of period analysis, scoring double standard screening and optimal subset regression methods, global data about low frequency elements of 150 days before forecasting days at 3 levels were fitted and extended to 30 days from June to August during 2011-2014. On this basis, the 10-30 days precipitation forecast  was made and tested in 44 stations in mid-west region of Northwest China from June to August during 2011-2012 by using gradually-filtering analog method. The results are as follows: (1) The fitting accuracy of low frequency elements fields gradually reduced with the increase of forecast time length, and then it tended to be stable. Furthermore, each low frequency element at 3 levels passed the test of 0.05 significant level over the most key region. (2) When the number of similar rainfall cases was more than or equal to 3, there was a rainfall in forecast, and the corresponding Cs score was the highest. The average prediction accuracy of 44 stations was 0.585, and the Cs probability values was higher than the 30-year climatic mean (1980-2010) in general. Therefore, this method has a good application prospect for summer precipitation forecast in mid-west region of Northwest China.

Key words: low-frequency signal, period analysis, gradually-filtering analog method, extended range forecast

摘要：

1980—2014年的NCEP/NCAR 3个层次(200 hPa、500 hPa和700 hPa)多个要素场逐日格点资料进行Butterworth带通滤波，保留周期为30～60 d的低频信号，通过周期分析、双评分标准筛选和最优子集回归方法对2011—2014年全球6—8月各要素场预报日前150 d的低频数据进行逐日外延(至未来30 d)。在此基础上，通过逐步相似过滤方法对西北地区中西部44个台站2011—2012年夏季降水进行10～30 d的预报及效果检验。结果表明：(1)对低频要素场的拟合准确度随预报时效增长逐步降低，后期趋于稳定；且各层次各低频要素相应的大部分关键区通过0.05显著性水平检验；(2)当相似个例中降水个例数≥3时预报当日有降水，其对应的Cs评分最高，44个台站的Cs评分均值为0.585，且Cs评分整体高于气候概率值。因此，利用该预报方法进行降水预报是可行的，可用于实际业务中。

关键词: 低频信号, 周期分析, 逐步相似过滤, 延伸期预报