The Article Comparing Techniques for Air Pollution Prediction is Online!

We are thrilled to announce a groundbreaking scientific achievement from the AgrImOnIA project! Our latest deliverable, M3, unveils a comprehensive comparative analysis of cutting-edge predictive models designed to forecast air pollution concentrations in Lombardy, North Italy, spanning from 2016 to 2020. The article is published in the Environmental and Ecological Statistics and accessible in open access here.

In this study, researchers scrutinized three distinct models renowned for their predictive prowess: hidden dynamic geostatistical models (HDGM), generalised additive mixed models (GAMM), and random forest spatiotemporal kriging models (RFSTK). Despite their varying methodologies, all models showcased remarkable proficiency in capturing the intricate spatiotemporal patterns inherent in air pollution data.

Moreover, the research delves into station-specific analyses, shedding light on the nuanced performance of each model under localized conditions. Through parametric coefficient analysis and partial dependence plots, researchers uncovered consistent associations between predictor variables and pollution concentrations, offering invaluable insights into the underlying dynamics.

The significance of this study lies not only in its findings but also in its implications for future research and practical applications. By showcasing the complementary potential of both classical statistical approaches and modern machine learning techniques, this research underscores the efficacy of conventional methods in modeling correlated spatiotemporal data.