Exploring SO₂ air pollution in Plovdiv through multivariate adaptive regression splines: A case study

Abstract This case study delves into the assessment of Sulfur dioxide (SO 2 ) air pollution in Plovdiv by employing Multivariate Adaptive Regression Splines (MARS) to model and understand the factors influencing daily SO 2 levels. By analyzing a dataset characterized by an average SO 2 pollution level of 0.43ppm, this study highlights the potency of MARS in capturing the non-linear relationships and complex interactions between SO 2 concentrations and measured meteorological and atmospheric time series in the form of quantitative and categorical variables. In particular, an increase in model performance is demonstrated by performing a modified Yeo-Johnson transformation on pollutant data and constructing additional predictors such as lag variables, date variable and dummies. A significant achievement of this investigation is the attainment of a coefficient of determination (R 2 ) exceeding 0.91. This high level of accuracy highlights the efficiency of MARS as a flexible and reliable machine learning tool thus emphasizing its potential in contributing to the urban air quality management toolbox.