Urbanization, urban form, and PM2.5 concentration in China: a hybrid machine learning and semiparametric approach

Abstract

In the process of industrialization and urbanization, PM_2.5 poses significant risks to environment and public health. Understanding the key driving factors of PM_2.5 concentration is crucial, as this enables policymakers to develop targeted and effective control measures that protect both environment and public health. In this paper, we first identify the driving factors of PM_2.5 concentration from four aspects--urban environmental infrastructure, industrial structure, economic development, and urban form--in 286 Chinese cities surveyed from 2000 to 2018 by using the eXtreme Gradient Boosting (XGBoost) and Shapley Additive exPlanations (SHAP) methods. We then quantitatively investigate the mean and quantile effects of these driving factors on PM_2.5 concentration and estimate the interaction effect of population density, the most important driving factor, using a semiparametric varying coefficient model. In addition, we conduct a mediation effect analysis to show how population density affects PM_2.5 concentration indirectly via urban form. Our paper shows that: (1) Population density, the industrial nitrogen oxide discharge, and the proportion of service sector in gross domestic product (GDP) are identified as the three most important driving factors. (2) The effects of PM_2.5 driving factors are heterogeneous at different quantiles of PM_2.5 concentration distribution, and are significantly and nonlinearly affected by population density. And (3) Population density indirectly affects PM_2.5 concentration through accelerating the process of urbanization and changing urban form, where urban form is measured by urban expansion, urban compactness, and urban complexity.