Istituto Superiore per la Protezione
e la Ricerca Ambientale

Cerca

New high-resolution maps of air pollution in Italy, created by integrating statistical models and machine learning

A study published in the journal Atmospheric Environment compares two advanced approaches for estimating concentrations of atmospheric particulate matter (PM10, PM2.5) and nitrogen dioxide (NO2) across the entire national territory, with a spatial resolution of 1 km × 1 km. The study was conducted by researchers from ISPRA, the Department of Epidemiology of the Lazio Regional Health Service, ARPAE Emilia-Romagna, and CNR-ISAC.

The research, conducted as part of the project “Outdoor Air and Health: An Integrated Atlas to Support Decision-Making and Research”—funded under the National Complementary Investment Plan (PNC)—compares a Bayesian statistical model (INLA-SPDE) with a Random Forest machine learning algorithm. This approach integrates data from air quality monitoring networks with meteorological information, territorial data, and simulations from the national atmospheric modeling system, kAIROS.

The results demonstrate that both approaches can accurately reproduce the spatial and temporal distribution of pollutants. Combining the two methods into an ensemble model further enhances performance, enabling the creation of detailed maps of population exposure to air pollutants.

The resulting maps highlight the areas of the country most affected by air pollution—notably the Po Valley and major urban centers—providing a valuable tool for exposure assessment, epidemiological research, and support for air quality policies as mandated by the new European Directive 2024/2881.

References

Reatini M.A., Nobile F., Stafoggia M., Stortini M., Amorati R., Veratti G., Cattani G., Comparison of Bayesian hierarchical modelling and Random Forest algorithm for spatiotemporal prediction of PM10, PM2.5 and NO2 across Italy, Atmospheric Environment, 389 (2027), 122456. DOI: 10.1016/j.atmosenv.2026.122456