Assesing AI algorithms for predictive modeling of spatiotemporal PM₁₀ air pollution

Abstract

Without a doubt, air pollution is one of the most serious issues confronting our world today, which presents significant health and environmental risks, exacerbating respiratory ailments and contributing to climate change. Air pollutants’ spatial and temporal variability is the basis for effective air quality management, necessitating more accurate predictive models. The study aims to assess particulate matter of a diameter smaller than 10 μm (PM₁₀) forecasts using the European Union’s Space Copernicus program mission of monitoring the atmosphere and tracking air pollutants, the Sentinel-5 Precursor satellite (5P) TROPOspheric Monitoring Instrument (TROPOMI), coupled with meteorological variables and observations from air quality monitoring stations. Root mean square error (RMSE) and mean absolute error (MAE) measure the model’s accuracy. The study integrated machine learning algorithms and diverse datasets to enable precise spatial modelling of PM₁₀ concentrations using a geographic information system (GIS). The results obtained peak accuracy during the heating season validation yielded an RMSE of 4.52 μg/m³, MSE of 20.44 (μg/m³)², and MAE of 3.30 μg/m³, while testing resulted in an RMSE of 4.38 μg/m³, MSE of 19.21 (μg/m³)², and MAE of 3.19 μg/m³.