资源环境科技发展态势分析平台

To improve the PM_2.5 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 PM_2.5 observations. The data assimilation system was successful in optimizing the initial PM_2.5 mass concentrations. The root-mean-square error (RMSE) of the initial PM_2.5 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 PM_2.5 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 PM_2.5 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-PM_2.5 data assimilation system, and further experiments are to be conducted to improve its performance.