Estimating urban PM2.5 concentration: An analysis on the nonlinear effects of explanatory variables based on gradient boosted regression tree

Abstract

It is important for air pollution study and control to understand the relationships between the concentrations of fine particulate matter and the influencing factors. This study builds a PM2.5 concentration estimation model with high accuracy considering major factors including meteorological conditions, traffic conditions, and urban spatial layouts based on Gradient Boosted Regression Tree approach. Based on the proposed model, the relative importance of the explanatory variables is presented. The meteorological variables, traffic condition variables and urban spatial layout variables have the predictive powers of 64.6%, 21.2% and 14.2% respectively. The nonlinear relationships between each variable and urban PM2.5 concentration are visualized by partial dependency plots and discussed in detail. The results show that in order not to exceed the limitation of regional PM2.5 concentration, the local government can lower city's temperature, improve urban ventilation and precipitation, alleviate traffic congestions, balance the road construction and PM2.5 generation, increase urban green area and concentrate Points of Interests.