Spatiotemporal Bayesian Model of PM Levels in South Korea

Abstract

This paper determines the factors which significantly affect pollution in The Republic of Korea. Using data from the Korean Statistical Society, a Bayesian spatio-temporal model is fit to identify the relationship between PM 2.5 and predictor variables such as temperature, road length, traffic density, etc. Spike and slab priors are implemented to identify the subset of significant predictor variables for the final model. The model is then used to predict PM 2.5 for the 16 provinces of South Korea and compared against a baseline model which does not take spatial or temporal dependence into account. We find that the spatio-temporal model outperforms the non spatio-temporal model when using the mean squared error for model comparison.