Bayesian spatial modelling of satellite-derived wildfire counts across Italian municipalities

Abstract This study investigates the spatial distribution of wildfire counts for Italian municipalities, focusing on some challenges inherent to modelling spatially aggregated areal count data. Leveraging high-resolution satellite-derived fire data aggregated to administrative units, we model spatial dependence and heterogeneity using the Integrated Nested Laplace Approximation (INLA) framework. Based on a single-season case study, the analysis addresses some key modelling issues, including model selection for hierarchical structures through Leave-Group-Out Cross-Validation (LGOCV) and the mitigation of spatial confounding. The results underscore the importance of municipal-level characteristics, such as land use, demographic trends, and socioeconomic conditions, in shaping wildfire patterns across the country.