A spatio-temporal model for the analysis and prediction of fine particulate matter concentration in Beijing

Effective air quality management and forecasting in Beijing is urgently needed as the region suffers from the worst air pollution in any standards. However, the statistical mechanism of the PM(2.5)formation with respect to various factors is underexplored in this region and China in general. Through an elaborate application with refinement of a spatio-temporal model with varying coefficients to the dynamics of PM(2.5)around Beijing based on a large dataset, we provide a comprehensive interpretation for the dynamics of PM(2.5)concentration with respect to its gaseous precursors, meteorological conditions and geographical variables. Furthermore, we conduct multistep temporal forecasts on a rolling basis for both the PM(2.5)concentration and the pollution levels. With the help of the expectation-maximization algorithm, the proposed models estimated for eight seasons from March 2015 to February 2017 around Beijing provide satisfactory in-sample fits and generate more accurate out-of-sample forecasts, compared with Finazzi and Fasso's original model as well as other alternative models. Valuable insights in tackling the excessive air pollution in Beijing are suggested from the comprehensive application of our model.