Abstract: Facing the increasingly severe condition of air pollution, it is important to make rational and effective use of satellite data to improve the quality of numerical forecast of haze. In this study, the WRF-Chem (WRF coupled with Chemistry) model and GSI (Gridpoint Statistical Interpolation) 3D-VAR assimilation system are applied to assimilate MODIS and FY-3A/MERSI AOD (Aerosol optical depth) to simulate a haze case over China. Results indicate that assimilating AOD data can significantly improve the initial field, and MODIS and MERSI assimilation experiments have advantages in simulations of the center intensity and spatial distribution of AOD, respectively. AOD data assimilation has a positive effect on the subsequent forecast of PM_2.5 and PM₁₀, while the strength and distribution simulation of each individual variable is improved reasonably. From the perspective of statistical analysis, mean biases and root-mean-square errors of PM_2.5 and PM₁₀ both reduce dramatically, and the simulated mean values and median values of the two pollutants are closer to observations. Satellite AOD data assimilation has obvious regional and individual characteristics for the improvement of particulate matter prediction. MODIS and MERSI assimilation experiment showed advantages in the prediction of PM_2.5 and PM₁₀, respectively. In addition, the experiment results suggest that AOD data assimilation can significantly improve the forecast accuracy of haze weather. MODIS and MERSI assimilation experiments show some differences in the accuracy of PM_2.5 and PM₁₀ prediction. The assimilation of AOD data reduces the no-hitting rate and false-hitting rate of PM_2.5 and PM₁₀ forecast, and improves the TS score and ETS score. For distinctive satellite AOD data, the assimilation of AOD data has different effects on the intensity and distribution of AOD in the initial field, which also leads to different results in the numerical forecast. In this case, The MODIS AOD DA experiment proves to be advantageous in PM_2.5 forecast while MERSI is advantageous in PM₁₀ simulation. The results reveal that AOD satellite data assimilation has a positive effect on numerical prediction, which implies the broad prospect of AOD data assimilation in the application of air quality forecast.