基于偏最小二乘回归方法的西北区寒潮延伸期预报

Extended-range forecast of cold wave in Northwest China based on partial least squares regression method

利用2000—2013年西北地区376个国家站逐日最低气温及NCEP FNL再分析资料,基于西北区站点降温幅度和欧亚大陆500 hPa低频高度场,使用偏最小二乘回归方法建立西北区站点寒潮延伸期预测模型,分别以2017年1月的一个寒潮日和非寒潮日为例进行预测试验,并对2014—2017年冬半年(共728 d)西北地区的寒潮进行回报试验。结果表明,该模型能够提前10~30 d较好预测寒潮日与非寒潮日,对寒潮日提前10~30 d预测的降温幅度≥8℃的站数平均为32.9个,命中率最高可达67%,而对非寒潮日提前10~30 d预测的降温幅度≥8℃的站数平均为2.7个。对2014—2017年冬半年西北地区延伸期寒潮日的预测结果显示,该模型提前10~30 d预测的平均站点命中率为11.5%,空报率为70.3%。规定降温幅度≥8℃的站数超过15个时为一个寒潮日,则平均CS评分为0.1,预测的寒潮日与实际寒潮日对应最好。该模型可用于西北区地区延伸期寒潮的预测,具有一定的参考价值。

Based on the daily minimum temperature data of 376 national stations in Northwest China and NCEP FNL reanalysis data from 2000 to 2013,the temperature drop amplitude of the stations and the 500 hPa low-frequency height field in Eurasia were analyzed.An extended-range forecast model of cold wave at stations in Northwest China was established by the partial least squares regression method,and the cold wave day and the non-cold wave day in January 2017 were predicted as an example.The cold wave forecast experiment was carried out using the data in Northwest China in winter half year from 2014 to 2017(728 d in total).Results show that the model can well predict cold wave days and non-cold wave days 10—30 d in advance.The average number of stations with temperature drop≥8℃predicted 10—30 d in advance for cold wave days is 32.9,with the highest hit rate of 67%,while the average number for non-cold wave days is 2.7.The forecast results of cold wave days in Northwest China during the extended-range of winter half year from 2014 to 2017 show that,the average station hit rate predicted by the model 10—30 d in advance is 11.5%,and the empty rate is 70.3%.It is a cold wave day,When the number of stations with temperature drop≥8℃exceeds 15.The average CS score is 0.1,and the predicted cold wave day corresponds to the actual cold wave day best.Therefore,the model can be used to predict the cold wave during the extended-range in Northwest China,and has a certain reference value.