PM_2.5 concentrations estimation using machine learning methods with combination of MAIAC - MODIS AOD product - a case study in western Iran

The harmful effects of ambient air pollution on human health have been consistently documented by many epidemiological studies around the world, and it is estimated that at least seven million deaths worldwide each year are caused by the effects of air pollution. Harmful airborne particles are identified by the Particulate Matter (PM) parameter, which is a term that used for solid and liquid particles in varying size, shape, composition and with different sources, suspended in the air. The aim of this study is to build PM_2.5 concentrations estimation model with meteorological data such as PBLH, and a combination of two AOD’s data retrieved from MODIS satellite (MAIAC - MODIS AOD product) using machine learning methods. The study area is in the Western part of Iran, where dust storms as one of the most important sources of air pollutants increasing sharply in recent decades, and this increase has caused numerous health and environmental problems. The data period is four years from 1 January 2018 to 31 December 2021, and three machine learning methods, LightGBM, MLP, and Random Forest algorithms were used. For the three typical machine learning methods, the RF model presents the best result by obtaining the lowest RMSE (30.1 μg/ m³) and MAE (25.0 μg/ m³) values in combination with the highest R² (0.64) value for daily predictions.