Improvement of PM2.5 forecast over China by the joint adjustment of initial conditions and emissions with the NLS-4DVar method

Abstract

Particulate pollution is a serious environmental problem that affects regional air quality and the global climate. We developed an advanced joint chemical data assimilation system to improve atmospheric aerosol forecasting based on the nonlinear least squares four-dimensional variational (NLS-4DVar) method. The chemical initial conditions (ICs) and emission fluxes were optimized jointly every 24 h in the system by assimilating multi-time observations. The NLS-4DVar approach allows us to perform joint assimilation with a large state vector benefited from its high computational efficiency. Observed hourly surface fine particulate matter (PM2.5) concentrations were assimilated into the Weather Research and Forecasting model coupled with the Community Multiscale Air Quality (WRF-CMAQ) model with 32-km spatial resolution. According to the results of sensitivity experiments, simultaneous adjustment of ICs and emissions brings more significant improvement of PM2.5 48-h forecasting than optimizing only the ICs. The impact of joint assimilation on PM2.5 mass concentration forecasting over China from 10 to November 24, 2018 was evaluated. In the Yangtze River Delta region, joint assimilation reduced the root mean square error by 48.4% for estimations of the initial concentration fields, 21.9% for 24-h forecasts, and 13.4% for 48-h forecasts. We also assessed the differences between optimized and prior emissions. These results indicate that the joint data assimilation system can effectively reduce the uncertainty in PM2.5 predictions during pollution episodes by simultaneously optimizing ICs mass concentrations and emissions.