A Bayesian Point Process Approach to Model Wildfire Patterns in Italy

We investigate the spatio-temporal behaviour of wildfires across Italy using event-level ignition records compiled within the national monitoring system managed by Forestry, Environmental and Agri-food Units Command. To capture the pronounced clustering of ignitions and their persistence over time, we model the ignition pattern through a spatio-temporal Log-Gaussian Cox process. Inference is car-ried out in a Bayesian framework using INLA via the inlabru R pack-age, enabling scalable estimation for complex latent structures. Resid-ual spatial variation is represented by Gaussian random fields imple-mented through the SPDE approach on a triangulated mesh over Italy, while penalized complexity priors provide principled regularization of model flexibility. The mo del incorporates selected environmental covari-ates assembled from heterogeneous sources, allowing us to quantify their association with ignition intensity at national scale. Overall, the pro-posed approach offers a unified framework to characterize wildfire igni-tion regimes in Italy and to support prevention and management strate-gies with uncertainty-aware evidence.