基于GAM模型分析影响因素交互作用对PM_(2.5)浓度变化的影响

Interactive Effects of the Influencing Factors on the Changes of PM_(2.5) Concentration Based on GAM Model

对南京市2013~2015年PM2.5及影响因素的时间变化序列,运用广义可加模型(GAM)分析影响因素交互作用对PM2.5浓度变化的影响.结果表明,PM2.5及影响因素都基本服从正态分布类型,影响因素间具较强相关性,其中气温、气压和水汽压间具有显著相关性.PM2.5浓度变化的单因素GAM模型中,所有影响因素均通过显著性检验,其中SO_2、CO、NO_2等影响因素的模型拟合度较优,方程解释度较高;PM2.5浓度变化的多因素GAM模型中SO_2、CO、NO_2、O_3、平均降雨量(PRE)、平均风速(WIND)和相对湿度(RHU)等影响因素对PM2.5浓度变化解释率为73.9%,对其变化具有显著性影响;通过多因素对PM2.5浓度变化影响效应的诊断分析,得到SO_2、NO_2和WIND与PM2.5浓度变化呈线性关系,CO、O_3、PRE和RHU与PM2.5浓度变化呈非线性关系;在影响因素交互作用对PM2.5浓度变化影响的GAM模型中,SO_2与CO、PRE、RHU间交互作用,CO与NO_2、O_3、PRE、WIND、RHU间交互作用,以及NO_2与WIND、PRE、RHU间交互作用,都在P<0.01(或P<0.05)水平下显著影响PM2.5浓度变化;大气污染物SO_2、CO及NO_2分别与气象等其它因素的交互作用对PM2.5浓度变化产生最主要影响作用;通过对影响因素交互作用GAM模型可视化三维图分析,定量研究了影响因素交互作用对PM2.5浓度变化的影响特征.结论表明,运用GAM模型,能够定量化分析影响因素交互作用对PM2.5浓度变化的影响,研究方法具有一定创新性,对PM2.5浓度污染与控制研究具有重要意义.

In this paper,the generalized additive model（ GAM） was introduced to analyze the interactive effects of the influencing factors on the change of PM2. 5concentration during 2013-2015 in Nanjing city. The results showed as follows： PM2. 5and its influencing factors appeared to follow normal distribution. There were strong correlations among the influencing factors,especially among the temperature（ TEM）,pressure（ PRS） and water vapor pressure（ VAP）. For the single influencing factor GAM models of PM2. 5concentration,all influencing factors passed the significance test. Moreover,the equation fitting degrees of SO2,CO,and NO2 were much higher. In the multiple influencing factors GAM models of PM2. 5concentration,the contribution of the SO2,CO,NO2,O3,precipitation（ PRE）,wind and relative humidity（ RHU） to the change of PM2. 5concentration was 73. 9% with significant impacts on the change of PM2. 5concentration. Based on the diagnostic analysis of the effect of multi factors on the change of PM2. 5concentration,there were linear relationship between PM2. 5and SO2,NO2 and wind,and non-linear relationship between PM2. 5and CO,O3,PRE and RHU. The GAM models,which considered the interaction of SO2 respectively with CO,PRE and RHU,the interaction of CO respectively with NO2,O3,PRE,Wind and RHU,and the interaction of NO2 respectively with Wind、PRE and RHU,all passed the significance test（ P〈0. 01 or P〈0. 05）. The interaction of SO2,CO and NO2 respectively with other factors such as meteorologicalfactors had the most important influence on the change of PM2. 5concentration. At last,through the visualized three-dimensional map of the GAM models considering the interaction of the influencing factors on the PM2. 5concentration,the interactive effects of the influencing factors on PM2. 5concentration were quantitatively modeled. Our results demonstrated that GAM could be used to quantitatively analyze the interactive effect of the influencing factors on the change of PM2. 5concentration. Therefore,the research method is innovative and important for PM2. 5pollution and control.