Fine particulate matter and ozone variability with regional and local meteorology in Beijing, China

Abstract

Fine particulate matter (PM2.5) and surface ozone air pollution have severe health consequences. Since China implemented the nationwide Air Pollution Prevention and Control Action Plan (APPCAP) in 2013, annual average PM2.5 concentrations have decreased while O3 concentration have increased in Beijing. It is unclear, however, to the extent that short-term meteorological variability has influenced these trends. To separate the influence of meteorology from long-term emissions-induced changes, we quantify the portion of the long-term PM2.5 and O3 concentration signals associated with short-term meteorological variability. Building off recent literature highlighting the regional nature of pollutant variability, we incorporate both locally observed meteorology and average regional reanalysis. We isolated sub-seasonal variability using the Kolmogorov-Zurbenko filter. Next, we trained and compared the utility of three statistical models of varying complexity (a linear model, a general additive model with splines, and a random forest) in their ability to predict out-of-sample daily concentrations. The random forest model yields the best predictive capability in holdout tests and predicts changing meteorological contributions to PM2.5 and ozone concurrent with changing concentrations across the study period. Results show an overall decrease in meteorology induced PM2.5 concentration variability at daily scales since 2013, but variability from meteorology has increased relative to mean PM2.5 concentrations. In contrast, O3 shows the opposite trend, with increasing meteorology-induced variability (meteorology-induced variability normalized by mean observed O3 has decreased). Our results show the importance of including regional meteorology in explaining local PM2.5 and O3 variability and quantify links between long-term policy implementation and short-term meteorological contributions to pollutant concentrations.