基于低频振荡信号的中国南方冬半年持续性低温指数延伸期预报试验

Extended range forecast experiments of persistent winter low temperature indexes based on intra-seasonal oscillation over southern China

利用1961—2009年36°N以南、108°E以东中国大陆191个站点逐日最低气温和NCEP/NCAR再分析日平均格点资料,研究与区域持续性低温事件有关的大气低频振荡信号,寻找可以在一定程度上表征不同类型区域持续性低温事件的指数,并尝试结合DERF2.0系统的预报产品进行持续性低温指数的延伸期预报试验。结果表明:(1)在研究范围内的区域持续性低温事件可以归纳为江北型、江南型和全区域型3类,其中江北型和江南型事件的环流背景差异体现在异常环流中心的纬度位置上,而全区域型事件属于增强型的江北型事件;(2)江北型和江南型区域平均最低气温时间序列的10—30 d低频分量的位相和强度变化与区域持续性低温事件的发生有显著关系,可以作为表征区域持续性低温事件指数和预报量;(3)100°—120°E范围内850 hPa温度场距平的经验正交函数分解前两个主模态具有显著的10—30 d变化周期,并且其空间结构分别与江北型和江南型事件的典型环流特征一致,前两个主模态时间系数能够作为持续性低温指数的预报因子;(4)检验结果表明,DERF2.0系统对上述预报因子有一定的预报能力。在延伸期预报时效内,利用统计学和动力学相结合的方法制作的持续性低温指数的预报效果好于模式直接预报的2 m气温,该预报方法有助于提升区域持续性低温事件的延伸预报能力。

On the basis of daily NCEP/NCAR reanalysis product and observational data for 1961 - 2009, this study investigates the low frequency oscillation signals of regional persistent low temperature events （RPLTEs） to the south of 36°N in China and identifies indexes that can be used to characterize RPLTEs. These indexes are then used as predictands in extended range fore- cast experiments based on the DERF2. 0 hindcasts. Results show that the RPI.TEs can be classified into three types, i. e. North of Yangtze River, South of Yangtze River, and the entire region. The types of North of and South of Yangtze River have their own key common circulation features that are distinguished by latitudes of anomalous circulation centers and characterized by low frequency wave trains propagating from northwest to southeast in Asia. 10 - 30 d low frequency components of the daily minimum temperature series of North of Yangtze River （T1） and South of Yangtze River （T2） are defined as the persistent low temperature indexes （RPI.TIs）. The phase and amplitude of the RPLTIs have a close relationship with the RPLTEs and are used as the predictands in extended range forecast experiments. EOF1 of the 850 hPa temperature anomalies between 100°- 120°E coincides with the low-frequency mode of T1 while EOF2 coincides with that of 7：e. Projection of daily data onto the pair of leading EOFs of 850 hPa temperature anomalies yields time series of principal components that can serve as an effective filter for low frequency oscillation without the need for bandpass filtering and makes the time series of the two principal components effective predictors for real-time application. DERF2.0 hindcasts and stepwise regression statistical method are employed to ex- plore extended range forecast （ERF） of RPLTIs. The forecast skill of this statistical dynamical prediction for 2-m temperature is better than that of DERF2.0 （direct model output） in real-time experiments.