基于GAM的北京PM_(2.5)浓度变化的影响因素研究

为辨识与测度不同影响因子对PM_(2.5)浓度变化的作用机理,以2015年北京PM_(2.5)浓度时间演变模式为基础,建立PM_(2.5)与各大气污染物(PM_(10)、SO_2、NO_2、CO、O_3)及气象因素(日均温度、风力、风向)的GAM模型,探索不同因素对PM_(2.5)浓度变化的影响作用。结果显示:(1)北京PM_(2.5)浓度具有夏秋季低、春冬季高的时间分布特点;(2)2015年北京PM_(2.5)浓度变化与PM_(10)、SO_2、NO_2、CO大体呈线性正相关,且正相关程度由强到弱为:COPM_(10)SO_2NO_2,而与O_3、温度和风因子的关系更为复杂;(3)GAM模型的拟合优度R~2为0.725,线性回归模型的拟合优度R~2为0.519,相比较,GAM模型对PM_(2.5)浓度变化的解释度提高了20.6%。研究表明,GAM模型对于建立PM_(2.5)浓度变化与影响因素间综合性复杂关系更灵活、更可靠,优于线性回归模型。

Exploring the factors influencing PM2.5 concentration change in Beijing based on Generalized Additive Model

Based on the mechanism of PM_2.5 concentration change in Beijing in 2015, the generalized additive model (GMA) of PM_2.5 and atmospheric pollutants (PM₁₀, SO₂,NO₂,CO,O₃), as well as the meteorological factors (daily mean temperature, wind scale, wind direction) are established to explore the impact of different factors on PM_2.5 concentration changes. The results showe that: (1) Beijing PM_2.5 concentration has the characteristics of low distribution of summer and autumn, however higher in spring and winter; (2) The PM_2.5 concentration in Beijing is linearly positively correlated with PM₁₀, SO₂, NO₂, CO. The positive correlation is from strong to weak: CO> PM₁₀> SO₂> NO₂, and the relationship with O₃, temperature and wind factor is more complicated. (3) The goodness R² of the GAM model is 0.725, and the goodness of the linear regression model is 0.519. Compared with the GAM model, the explanatory degree of the PM_2.5 concentration increased by 20.6%. The results show that the GAM model is more flexible and reliable than the linear regression model in establishing the complex relationship between PM_2.5 concentration and influencing factors.