“Forecasting particulate matter concentrations by combining statistical models”

Abstract

Air pollutants have adverse effects on human health and play significant roles in urban planning and specifying air quality ranges. Atmospheric particulate matter is one of the criteria air pollutants that may have a dominant effect on air pollution. The present analysis was made using data collected over five years (2011–2015) in an urban area. Air pollutant concentrations and climate data are analyzed using four models: a multiple linear regression model, a principle component regression model, a logarithmic architecture, a principle component regression model with the variables that only have the highest factor loadings in each principal component. Thus, the proposed model combines both methods the multiple linear regression and the principal component analysis to obtain clearer and more reliable predictions. The prediction model’s accuracy has been verified operating several performance indicators, which revealed acceptable values, demonstrating that the proposed model can be used to predict pollutant concentrations. According to statistical indicators (RMSE, NMSE, CV, FB and IOA), the best prediction models were Model 3 for winter (0.06, 0.001, 0.03, −0.000, and 0.69), Model 4 for spring (0.08, 0.002, 0.04, −0.019, and 0.99), Model1 for summer (3.41, 0.005, 0.07, 0.000, and 0.98), and Model 2 for autumn (11.71, 0.018, 0.13, −0.000, and 0.57). In addition, Model2 generally gave appropirate values for all seasons and can be used as a common model. Finally, combined models based on principal component analysis and multiple linear regression outperformed models with only multiple linear regression in terms of error.