Published in Air Quality, Atmosphere & Health, the article “Spatial-temporal prediction of ambient nitrogen dioxide and ozone levels over Italy using a Random Forest model for population exposure assessment” presents findings used to support national exposure assessment and environmental epidemiology studies planned within the BEEP project (Big Data in Environmental and Occupational Epidemiology, www.progettobeep.it). The study confirms the potential of machine learning methods for accurately predicting atmospheric pollutant levels at high spatial resolution.
This work employs an innovative approach that integrates results from a chemical transport model (FARM) with machine learning (ML) techniques to produce daily concentration fields of NO2 and O3 at high spatial resolution (1 km) across Italy. The Random Forest model (ML-RF) utilized various space-time predictors for the period 2013-2015, including results from FARM simulations and other parameters such as population, land use, and road networks. The good performance of the FARM model for NO2 and O3 was further enhanced by the ML “Random Forest” model, which demonstrated reduced underestimation of NO2 (with fractional bias results close to zero). Additionally, this approach highlighted significant spatial gradients for these pollutants.