Prediction of PM 2.5 Concentrations Using Principal Component Analysis and Artificial Neural Network Techniques: A Case Study: Urmia,Iran

Abstract

Forecasting PM_2.5 concentration in ambient air quality is of great concern to urban management administrative due to its harmful health consequences and interference with the safe and comfortable use of the environment. In this study, the prediction of PM_2.5 concentration using three artificial neural network models, including multi-layer perceptron (MLP), radial basis function (RBF), and generalized regression neural network was investigated. The effect of principal component analysis (PCA) technique on improving the results was studied as well. Urmia City, Iran, was selected as the case study. Air pollution parameters, that is, NO₂, CO, PM₁₀, and PM_2.5 were obtained from Urmia air quality monitoring station No. 3 and meteorological data, including temperature, relative humidity, and wind speed, were collected from the Urmia airport synoptic station. Three scenarios of input data were proposed to address the effect of time lag. According to the results, the highest correlation coefficient (R²) and the minimum values of mean squared error and mean absolute error parameters were obtained from RBF network with input data scenario No. 2 (including the data of 1 and 2 days before the forecasting). PCA application not only reduced the number of input data in MLP network but also increased the correlation coefficient between real data and the predicted one by 3.8%. Due to the fact that NO₂, as the major source of nitrate aerosols, has low retention time in the atmosphere (1–2 days), and considering the significant relationship between PM_2.5 and NO₂ concentrations in Urmia city, it can be concluded that the data of 3 days before the forecasting day might not contribute meaningfully in PM_2.5 prediction.