Improved Retrieval and Forecast of Surface Air Quality based on Source Analysis, Numerical Modeling and Hybrid Statistical Dynamic Model

Local air quality conditions depend on local and regional climatic conditions, source contribution, emission strength of individual pollutant precursors, and the trajectory of air masses, with effects being more conspicuous in urbanized areas with complex terrain pattern, thus induces a need to develop a reliable data-analytic model to capture the realistic associations between emission changes, meteorological conditions and surrounding chemistry processes of the investigated spatial region. Using available emission database, modeled meteorological outputs and raw pollutant attributes of Hong Kong, a new Hybrid Statistical-Dynamic Model was developed by taking advantages of both statistical and deterministic features, and was applied into retrieving historical pollution profile and forecasting next-day surface pollutant concentrations.

By considering the contribution ratio of local to regional pollutant and emission sources, influence of background source and regional meteorological conditions, the categorization of outputs from a coupled regional meteorological and chemistry model was performed, and was adopted to parametrize the equations of a statistical Generalized Additive Model (GAM) within the framework. The established model was shown capable in retrieving temporal patterns of short-term PM_2.5, PM₁₀ and NO₂ concentrations in Hong Kong, but suffered from underestimation during pollution episodes. Thus, bias-adjustment techniques like Hybrid Forecast and Kalman Filter were applied into complementing such statistical deficiency with the aid of observational datasets, and the effectiveness of these techniques were validated by categorical assessments and common statistical metrics. The development of this hybrid model opens new windows in projecting scenario-based pollutant changes, and providing project-based opportunities for acquiring reasonable pollution forecasts.