我国冬季持续低温事件预报模型的建立

利用1979—2010年我国冬季522个站点的日平均温度资料,使用基于尺度分离的多元线性回归方法建立预报模型,对我国冬季的持续低温事件的预报做了一定的尝试。将温度按照一定的方法进行尺度分离,分解出年际、月、季节内尺度温度,对各个部分分别进行拟合再相加。挑选特定海区的海温、北极海冰作为年际尺度温度预测因子,多个低频气象场作为季节内温度预测因子,利用1979—2003年资料拟合出各因子回归系数。用重构的冬季温度挑选出持续低温事件并与实际的事件进行对比检验,计算出此方法的TS评分为0.57,再利用2008年初我国南方大范围的持续低温事件检验拟合结果。检验结果表明基于时间尺度分离的多元线性回归方法在持续低温事件的预报中有一定的参考意义。

Construction of winter persistent cold events prediction model in China

By using winter daily temperature data from 522 stations during 1979-2010, with Multiple Linear Regression(MLR) method, a prediction model was constructed. Winter temperatures are divided into three parts:annual, monthly and intraseasonal. Sea surface temperature in some certain areas and Arctic sea ice are selected as the prediction factors of annual temperature and some low frequency filtered meteorological fields are for intraseasonal temperature. Regression coefficients are calculated by using temperature data from 1979 to 2003, which were used to reconstruct winter temperatures. Persistent cold events (PCEs) picked out by regressed temperatures are compared with those by real temperatures. TS score for the method is 0.57. The catastrophic freezing rain and sleet in early 2008 is further used to test the result. The result shows that MLR based on time-scale division is to some degree useful for PCEs prediction.