Assessing the Impacts of Birmingham’s Clean Air Zone on Air Quality: Estimates from a Machine Learning and Synthetic Control Approach

Abstract

AbstractWe apply a two-step data driven approach to determine the causal impact of the clean air zone (CAZ) policy on air quality in Birmingham, UK. Levels of NO2, NOx and PM2.5 before and after CAZ implementation were collected from automatic air quality monitoring sites both within and outside the CAZ. We apply a unique combination of two recent methods: (1) a random forest machine learning method to strip out the effects of meteorological conditions on air pollution levels, and then (2) the Augmented Synthetic Control Method (ASCM) on the de-weathered air pollution data to isolate the causal effect of the CAZ. We find that, during the first year following the formal policy implementation, the CAZ led to significant but modest reductions of NO2 and NOX levels measured at the roadside within (up to 3.4% and 5.4% of NO2 and NOX, respectively) and outside (up to 6.6% and 11.9%) the zone, with no detectable changes at the urban background site outside the CAZ. No significant impacts of the CAZ were found on concentrations of fine particulates (PM2.5). Our analysis demonstrates the short-term effectiveness of CAZ in reducing concentrations of NO2 and NOX.