The booming of the economy in Hong Kong (HK) and the Pearl River Delta Economic Zone (PRD EZ) results in severe air pollution problems. Efficient data driven-based policy-making is of great importance in coping with the challenges. This thesis, with its focus on air quality modelling, applies the sensitivity analysis methodology to evaluate the effectiveness of integrated emission control policies and cross-boundary sectoral contributions to provide a holistic emission control potentials. A novel three-dimensional variational (3D-Var) coupled modelling system is set to fusing hourly surface observations to investigate the Spatio-temporal variations of air pollution. This Ph.D. defense will illustrate three parts. Firstly, a new framework of refining production-based emission inventory is proposed to explore the yearlong Spatio-temporal variation of PM_2.5 and O₃ in HK and PRD EZ. Results of Spatio-temporal analysis confirm the model capable of reproducing the two-pollutant concentrations in most cities for most months. Secondly, hypothetical dynamical scenario studies are carried out to evaluate China’s emission reduction policies in the 13^th Five-Year-Plan and short-term measures in HK. An integrated emission control potential map covering various emission sectors gradually extending to the whole of China is present for the first time. Calculating the health co-benefits of O₃ formation confirms the effectiveness of the control policies. The central government adopts the feasibility of tightening specific pollutants in HK using the scenario study from 2020 to 2025. Thirdly, a 3D-Var coupled data assimilation system illustrating detailed Spatio-temporal variations of pollutant concentration aims to improve the boundary conditions for the nested model running-mode. The impact of the data assimilation frequency, multiple pollutant assimilations, and the assimilated analysis and forecast benefits are discussed. One key finding is that assimilating PM_2.5 on the outermost domain brings substantial improvement of the PM_2.5 model simulations for the innermost domain of which is a potential substitute method comparing with the previous domain-wide data fusion algorithms. The coupled data assimilation system could be a useful tool for improving air quality forecasting and diagnosing model deficiencies. Lastly, potential future research directions will be discussed, as well.