Numerical Model-Based Artificial Neural Network Model and Its Application for Quantifying Impact Factors of Urban Air Quality

Knowledge of the relationship between air quality and impact factors is very important for air pollution control and urban environment management. Relationships between winter air pollutant concentrations and local meteorological parameters, synoptic-scale circulations and precipitation were investigated based on observed pollutant concentrations, high-resolution meteorological data from the Weather Research and Forecast model and gridded reanalysis data. Artificial neural network (ANN) model was developed using a combination of numerical model derived meteorological variables and variables indicating emission and circulation type variations for estimating daily SO2, NO2, and PM10 concentrations over urban Lanzhou, Northwestern China. Results indicated that the developed ANN model can satisfactorily reproduce the pollution level and their day-to-day variations, with correlation coefficients between the modeled and the observed daily SO2, NO2, and PM10 ranging from 0.71 to 0.83. The effect of four factors, i.e., synoptic-scale circulation type, local meteorological condition, pollutant emission variation, and wet removal process, on the day-to-day variations of SO2, NO2, and PM10 was quantified for winters of 2002–2007. Overall, local meteorological condition is the main factor causing the day-to-day variations of pollutant concentrations, followed by synoptic-scale circulation type, emission variation, and wet removal process. With limited data, this work provides a simple and effective method to identify the main factors causing air pollution, which could be widely used in other urban areas and regions for urban planning or air quality management purposes.