China is facing serious air pollution problems especially that caused by high concentrations of ozone and fine particles. A key step to effectively control air pollution is the modeling and forecasting of air pollution. However, large uncertainties with complicated sources still exist in air pollution forecasting. The nonlinearity in chemical processes makes it difficult to identify those key uncertainty sources and carry out targeted constraints and corrections in the modeling study. Data assimilation method can combine modeling information with multi-source observations to improve the accuracy of air pollution simulation and forecast. In this paper, we briefly introduce model uncertainties, assimilation algorithms, and optimization of initial concentrations and emissions for air quality model in the field of air pollution data assimilation. Challenges and development trends in the study of atmospheric pollution data assimilation are also highlighted.