Geostatistical methods comparison: an application on the concentration of PM2.5 in Lombardy

Livestock and fertilisers are widely acknowledged to be responsible for approximately 95% of all ammonia emissions, which undergo chemical reactions in the atmosphere to form fine particulate matter, or PM2.5 . The Lombardy region is characterised by high concentrations of PM2.5 and a high livestock density. The main objective of our study is to identify which model better explains the PM2.5 concentration in the Lombardy region at the local level. To achieve our objectives, we will utilise multiple modelling strategies, including a Hidden Dynamic Geostatistical Model (HDGM), a Generalized Additive Mixed Model (GAMM), and a Random Forest model with kriging applied to the residual term (RFRK). Furthermore, a comparison between each model will be performed to improve the interpretation of the models and a cross-validation scheme will be used to compare the prediction performance of the three different models.