厄尔尼诺/拉尼娜事件对区域气温的影响与预测——以沈阳地区为例  

The effect and forecaster of the El Ni?o/La Ni?a events on climate in Shenyang

作　　者：严韬 徐明洁 葛非凡 蒋跃林[1] 温日红[3] 程志庆[1] 吴文革[4] YAN Tao;XU Mingjie;GE Feifan;JIANG Yuelin;WEN Rihong;CHENG Zhiqing;WU Wenge(School of Resources and Environment,Anhui Agricultural University,Hefei 230036;College of Agriculture,Shenyang Agricultural University,Shenyang 110866;Institute of Atmospheric Environment,China Meteorological Administration,Shenyang 110166;Anhui Academy of Agricultural Sciences,Hefei 230031)

机构地区：[1]安徽农业大学资源与环境学院,合肥230036 [2]沈阳农业大学农学院,沈阳110866 [3]中国气象局沈阳大气环境研究所,沈阳110166 [4]安徽省农业科学院,合肥230031

出　　处：《安徽农业大学学报》2019年第1期57-64,共8页Journal of Anhui Agricultural University

基　　金：国家重点研发计划项目课题(粮食作物生产灾害防控与产后安全绿色储藏技术集成2018YFD300905)资助

摘　　要：利用1961—2015年国家气象信息中心沈阳站的日平均气温资料、美国国家海洋和大气管理局(National Oceanic and Atmospheric Administration,NOAA)提供的多变量ENSO指数(multivariate ENSO index,MEI)资料等,在分析沈阳地区气温月际变化的基础上,结合厄尔尼诺/拉尼娜事件对其影响特征,利用线性倾向估计和非线性自回归(nonlinear auto regressive models with exogenous inputs,NARX)神经网络模型分别对沈阳地区2011—2015年的气温进行预测。结果表明,1961—2015年共计660个月中,沈阳地区11月—3月气温的变异系数在20%以上,远大于其他月份。1961—2015年的厄尔尼诺/拉尼娜事件往往在秋冬季达到最大强度,或为导致沈阳地区11月—3月气温变异增强的原因之一。厄尔尼诺事件结束之后的春季,沈阳地区气温偏低的概率逾70%。沈阳地区气温随MEI变化的线性倾向值为0.98,决定系数为0.98且通过了0.01的可信度检验。利用MEI对沈阳地区的气温进行同期和时滞预测,NARX的预测结果均优于一元线性回归模型。当气温滞后MEI16个月时,两者的相关系数达到最大且通过了0.01的显著性检验,此时回归模型预测的相关系数为0.59,较同期预测提升了79%;NARX预测的均方误差(mean-square error,MSE)为0.49,较同期预测降低了36%,相关系数为0.86,较同期预测提升了8%。Based on the analysis of inter-monthly temperature variation in Shenyang,using the daily mean temperature data from National Meteorological Information Center during 1961-2015,the multivariate ENSO index(MEI)data from NOAA,and combined with the influence of El Nino/La Nina events on its characteristics,we have predicted the temperature from 2011 to 2015 in Shenyang by linear propensity and nonlinear autoregressive(NARX)neural network model.The results showed that in the 660-months from 1961 to 2015,the variation coefficient of temperature from November to March was more than 20%,which was much higher than that in other months.The El Nino/la Nina events in 1961-2015 tended to reach their maximum intensity in autumn and winter.That may one of the reasons for the increased variation of temperature from November to March.In the spring after the El Nino event,the probability of lower temperatures was more than 70%.The linear trend value of the temperature change with MEI was 0.98,and the determination coefficient was 0.98 and passed the credibility test of 0.01.Synchronous and delay predicting of temperature were carried out by MEI.The prediction results of NARX were better than the one-dimensional linear regression model.When the temperature delay MEI for 16 months,the correlation coefficient between them reached the maximum and passed the significance test of 0.01.At this time,the correlation coefficient predicted by the regression model was 0.59,which was 79%higher than the corresponding prediction.The mean square error(MSE)predicted by NARX was 0.49,36%lower than the corresponding prediction,and the correlation coefficient was 0.86,8%higher than the corresponding prediction.

关 键 词：多变量ENSO指数(MEI) 非线性自回归模型(NARX) 动态神经网络 短期气候预测 

分 类 号：P467[天文地球—大气科学及气象学]