A fast forecasting method for PM2.5 concentrations based on footprint modeling and emission optimization

Abstract

We propose a method for fast forecasting of PM2.5 concentrations in the North China Plain based on footprint (source–receptor relationship) modeling and emission inventory inversion. A backward Lagrangian stochastic particle dispersion model was employed to derive the footprint, using meteorological fields and boundary layer parameters provided by the WRF model. An analytical Bayesian inversion model was used to optimize existing emission inventories using long-term, multi-site PM2.5 monitoring data. The fast simulation of PM2.5 concentrations was based on the optimized inventory and the footprint results. Two-year simulations were carried out for six cities (Baoding, Beijing, Dezhou, Shijiazhuang, Tianjin, and Tangshan), with model establishment and emission inversion in the first year (2015) and test forecasting in the second year (2016). Promising simulation results were obtained even when using the primary emission inventory. For all six cities, the fractions of simulations of measurements within a factor of two ranged from 0.49 to 0.68, and the correlation coefficients ranged from 0.40 to 0.56 in 2015. The model also well reproduced temporal variations in PM2.5 concentrations in Beijing during severe haze episodes in the winter of 2015. Great improvement was achieved for the simulations by using the optimized emission inventory. The proportion of samples that met the PM model criteria increased from 88% to 97%, and the proportion that achieved the modeling goal increased from 25% to 44%. This method maintained its high forecasting skill in 2016, with 92% and 46% of samples meeting the PM model criteria and achieving the modeling goal, respectively. However, the corresponding values were86% and 39% if emission optimization was not applied.