A novel spatiotemporal prediction approach to fill air pollution data gaps using mobile sensors, machine learning and citizen science techniques

Abstract

Particulate Matter (PM) air pollution poses significant threats to public health. We introduce a novel machine learning methodology to predict PM2.5 levels at 30 m long segments along the roads and at a temporal scale of 10 seconds. A hybrid dataset was curated from an intensive PM campaign in Selly Oak, Birmingham, UK, utilizing citizen scientists and low-cost instruments strategically placed in static and mobile settings. Spatially resolved proxy variables, meteorological parameters, and PM properties were integrated, enabling a fine-grained analysis of PM2.5. Calibration involved three approaches: Standard Random Forest Regression, Sensor Transferability and Road Transferability Evaluations. This methodology significantly increased spatial resolution beyond what is possible with regulatory monitoring, thereby improving exposure assessments. The findings underscore the importance of machine learning approaches and citizen science in advancing our understanding of PM pollution, with a small number of participants significantly enhancing local air quality assessment for thousands of residents.