A Bayesian LSTM model to evaluate the effects of air pollution control regulations in Beijing, China

Abstract

Rapid socio-economic development and urbanization have resulted in serious deterioration in air-quality in many world cities, including Beijing, China. This study attempts to examine the effectiveness of air pollution control regulations implemented in Beijing during 2008–2019 through a data-driven regulatory intervention analysis. Our proposed Bayesian deep learning model utilizes proxy data including Aerosol Optical Depth (AOD) and meteorology as well as socio-economic data, while accounting for confounding effects via propensity score estimation. Our results show that air pollution control regulatory measures implemented in China and Beijing during 2008–2019 reduced PM2.5 pollution in Beijing by 11 % on average. After the introduction of Action Plan for Clean Air in China and Beijing in late 2013, as compared to the hypothetical PM2.5 concentration (without any regulatory interventions), the estimated PM2.5 reduction increased dramatically from 15 % in 2015 to 44 % in 2018. Our results suggest that Beijing’s air quality has improved gradually over the past decade, though the annual PM2.5 pollution still exceeds the WHO threshold. In this regard, the air pollution control regulations introduced in Beijing and China tend to become more effective after 2015, suggesting a 2-year time lag before the stringent air pollution control regulations starting from 2013 takes any strong positive effects. Moreover, as compared to the air pollution control regulations introduced before 2013, newly introduced policy-making governance, which couples the policy-makings of the local jurisdictions with that of the central government, and the new policy measures that tackle the vested interests of the local stakeholders in Beijing and its nearby cities, alongside with the stringent local and national air pollution control regulations and plans, should help reduce air pollution and promote healthy living in Beijing over the longer term.