气溶胶粒径吸湿增长多因素影响的GAM模型

GAM Model for Multi-factorial Effects on Aerosol Particle Size Hygroscopic Growth

该文利用成都市2017年10-12月浊度计、黑碳仪和GRIMM180环境颗粒物分析仪的逐时观测数据,以及该时段同时次大气能见度(V)、相对湿度(RH)和二氧化氮(NO2)监测资料,通过免疫进化算法反演气溶胶粒径吸湿增长因子(Gf),分析了以RH为单变量Gf(RH)参数化方案的模拟偏差,并基于广义可加模型系统探讨了Gf对多因素变化的响应关系。结果表明:(1)Gf(RH)参数化方案在低相对湿度条件下(RH<85%)均能很好地模拟Gf随相对湿度的变化特征,但在高相对湿度条件下(RH>85%)的拟合值则会出现较大的偏差。(2)利用多元递归法和方差膨胀因子(VIF)相结合构建了RH、C_(BC)、C_(BC)/C_(PM1)、C_(PM1)/C_(PM2.5)和C_(PM2.5)/C_(PM1)0的解释变量集(C_(BC)、C_(PM1)、C_(PM2.5)、C_(PM1)0分别为BC、PM1、PM2.5、PM10的质量浓度),其中RH和C_(BC)对Gf影响较大,对应单变量GAM模型中的调整判定系数(R2)分别为0.78和0.17;C_(BC)/C_(PM1)、C_(PM1)/C_(PM2.5)和C_(PM2.5)/C_(PM1)0对Gf影响相对较小,对应单变量GAM模型中的调整判定系数(R2)分别为0.08、0.12和0.03。(3)以RH、C_(BC)、C_(BC)/C_(PM1)、C_(PM1)/C_(PM2.5)和C_(PM2.5)/C_(PM1)0为自变量构建了Gf多因素影响的GAM模型,对应的调整判定系数(R2)和压轴回归决定系数(R2)分别为0.868和0.872,降低了在高相对湿度条件下(RH>85%)的拟合偏差,显著提升了霾天气条件下气溶胶粒径吸湿增长因子的模拟效果。

The simulation deviation of Gf(RH)parameterization scheme with RH as the univariate variable was analyzed on the basis of hourly observational data of Chengdu City by means of nephelometer,aethalometer and GRIMM180 ambient particles monitor during the period of October-December,2017,and the simultaneous data of atmospheric visibility(V),relative humidity(RH)and nitrogen dioxide(NO2),aerosol particle size hygroscopic growth factor(Gf)retrieved by an immune evolutionary algorithm;moreover,the response of Gf to multifactor variation was systematically explored by the generalized additive model(GAM).Accordingly,the results suggested the Gf(RH)parameterization scheme could well simulate the variation feature of Gf in case of low RH(when RH<85%),while under the high relative humidity condition(RH>85%),the fitted values would considerably deviate from observed ones.The explanatory variables of RH,Csc,Cc/CpM,Cpm/C_(PM2.5)and C_(PM2.5)/CpMio were proposed by using the combination of multivariate recursion and VIF method,among them RH and Csc exerted considerable effects on Gf,the adjusted coefficients of determination(R^(2))in the corresponding univariate GAM model being 0.78 and 0.17,respectively;while Csc/C_(PM1),Cpm/C_(PM2.5)and C_(PM2.5)/Cpmo had relatively small effects on Gf,the adjusted coefficients of determination(R^(2))in the corresponding univariate GAM model were 0.08,0.12 and 0.03,respec-tively;and the GAM model with RH,C_(BC),Csc/C_(PM1),Cpm/C_(PM2.5)and CpM2.s/CpMio as independent variables was constructed for the multi-factorial effects of Gf,and correspondingly,the adjustment determination coefficient(R^(2))and the determination coefficient of reduced major axis regression were respectively 0.868 and 0.872,thus cutting down the fitting deviation under high relative humidity conditions(RH>85%),and significantly enhancing the simulation effect of aerosol particle size hygroscopic growth factor under smog weather condition.