Improving PM2.5 forecast during haze episodes over China based on a coupled 4D-LETKF and WRF-Chem system

Abstract

To improve the PM2.5 forecast during severe haze episodes, we developed a data assimilation system based on the four-dimensional local ensemble transform Kalman filter (4D-LETKF) and the WRF-Chem model to assimilate surface PM2.5 observations. The data assimilation system was successful in optimizing the initial PM2.5 mass concentrations. The root-mean-square error (RMSE) of the initial PM2.5 concentrations after assimilation decreased at 76.75% of the stations and the RMSE reduction exceeds 30% at 20.7% of the stations. The correlation coefficients for the PM2.5 analyses increased by more than 0.3 at 33% of the stations. The forecasts for the spatial distribution and evolution of the haze were improved remarkably after assimilation while the forecasts without assimilation usually significantly underestimated the PM2.5 mass concentrations during the severe haze episodes. The RMSE of the 24-h forecasts after assimilation can be reduced by 32.02% in the polluted regions. During haze episodes, the 48-h forecasts after assimilation can benefit from the assimilation to a similar extent with the 24-h forecasts. Both the forecast accuracy and the duration of assimilation benefits were improved remarkably which demonstrate the effectiveness of the 4D-LETKF-PM2.5 data assimilation system, and further experiments are to be conducted to improve its performance.