Multi-horizon forecasting of average daily PM2.5 concentrations: A case study of air pollution in Singapore

Abstract

Understanding the spatial and temporal dynamics of surface concentrations of particulate matter (PM2.5 and PM10) is essential in air quality modelling and climate research. Forecasting of air pollutants is necessary to understand the variation of pollutants and to plan and implement air pollution control measures. This work proposes a methodology for forecasting PM2.5 concentration using various meteorological parameters over multiple time horizons. The proposed deep learning-based models forecast the daily average values concentration for PM2.5 in Singapore. The performance metrics indicate the efficacy of the proposed model in forecasting PM2.5 concentrations over different horizons. The analysis shows that PM2.5 concentrations are best forecasted for the 3-days ahead scenario, with RMSE and MAPE being 3.767 µg/m3 and 6.82%, respectively.