长三角地区冬季霾污染日数的季节预测  

Seasonal prediction of winter haze days in the Yangtze River Delta

作　　者：董莹 尹志聪[1,2,3] 段明铿 DONG Ying;YIN Zhicong;DUAN Mingkeng(Collaborative Innovation Center on Forecast and Evaluation of Meteorological Disasters(CIC-FEMD)/Key Laboratory of Meteorological Disaster,Ministry of Education(KLME)/School of Atmospheric Sciences,Nanjing University of Information Science and Technology,Nanjing 210044,China;Southern Marine Science and Engineering Guangdong Laboratory,Zhuhai 519080,China;Nansen-Zhu International Research Centre,Institute of Atmospheric Physics,Chinese Academy of Sciences,Beijing 100029,China)

机构地区：[1]南京信息工程大学气象灾害教育部重点实验室/气象灾害预报预警与评估协同创新中心/大气科学学院,江苏南京210044 [2]南方海洋科学与工程广东省实验室,广东珠海519080 [3]中国科学院大气物理研究所竺可桢-南森国际研究中心,北京100029

出　　处：《大气科学学报》2021年第2期290-301,共12页Transactions of Atmospheric Sciences

基　　金：国家自然科学基金资助项目(42088101,91744311);江苏省大学生创新创业训练计划项目(201910300043Z)。

摘　　要：将霾日数年际增量作为预测对象、前期外强迫因子作为自变量,分别运用多元线性回归方法和广义相加模型建立长三角地区冬季霾日数预测模型。综合考察“去一法”交叉验证和循环独立样本实验的结果,选出适用于各个模型较优的建模方法,并对比长三角地区冬季霾日数预测模型(MODEL1)和长三角地区冬季霾日数分月预测模型(MODEL2)。MODEL1、MODEL2的均方根误差(解释方差)分别为2.69(80.01%)、2.76(79.04%),两类模型均能成功捕捉霾日数的年际-年代际趋势和极值。MODEL2预测的霾日数距平同号率(97.3%)优于MODEL1(86.49%),具有良好的距平符号捕捉能力。MODEL1采用11月之前的外强迫因子,可提前一个季度预测冬季霾日数;MODEL2采用更新的外强迫因子,可不断预测每月霾污染状况。通过两类模型组合使用,可更准确预测长三角地区冬季霾日数,为霾污染治理提供可靠的科技支撑。In recent years with the development of social economy,the Yangtze River Delta has experienced serious haze pollution,which has brought great harm to traffic safety,ecosystem and human health.Taking the interannual increment of haze days as the prediction object and the external forcing factors in the early stage as the independent variables,the prediction models of winter haze days in the Yangtze River Delta are established by using the multiple linear regression method and the generalized additive model.By comprehensively investigating the results of“one-year-out”cross validation and cyclic independent sample prediction test,this paper selects the optimal modeling method applicable to each model and compares the Yangtze River Delta winter haze days prediction model(MODEL1)and the Yangtze River Delta haze days prediction model in different winter months(MODEL2).Root mean square errors(explained variances)of MODEL1 and MODEL2 are 2.69(80.01%)and 2.76(79.04%),respectively.Both models can successfully capture the interannual and interdecadal trends and extreme values of haze days.The percentage of the same sign(meaning mathematical signs of fitted and observed haze days anomalies are same)predicted by MODEL2(97.3%)is better than that predicted by MODEL1(86.49%),showing that MODEL2 has better ability to capture the anomalous signs.By selecting external forcing factors before November,MODEL1 can predict winter haze days one quarter in advance.MODEL2 can constantly predict monthly haze pollution by selecting newer external forcing factors.By combining the two models,it can more accurately predict winter haze days in the Yangtze River Delta,which provides reliable scientific and technological support for the haze pollution control.

关 键 词：霾污染 年际增量 多元线性回归 广义相加模型 短期气候预测 

分 类 号：X831[环境科学与工程—环境工程]