Based on the improved three-dimensional variational assimilation system, the assimilation of ground fine particulate matter (PM2.5) and satellite aerosol optical depth (AOD) is conducted and the effectof analysis field assimilation on the PM2.5 forecast improvement is evaluated. A continuous pollution process is selected in this study, and the ground PM2.5 and AOD observation data are assimilated individually and simultaneously. The results show that compared with PM2.5 individual assimilation, AOD assimilation alone is more effective in improving the accuracy of the AOD analysis field, but the accuracy of the PM2.5 analysis field is reduced obviously. While simultaneous assimilation of PM2.5 and AOD makes the simulation of aerosol optical-physical properties achieve the best overall effect. Moreover, assimilation tests can effectively reduce the missing report rate. For the case of mild and moderate pollution, the choice of assimilation PM2.5 or AOD does not have a significant impact on the forecast. However, in the case of severe pollution, the comprehensive forecast of PM2.5 is the best when the ground PM2.5 and the whole layer AOD are assimilated simultaneously.