Spatiotemporal wildfire modeling through point processes with moderate and extreme marks

Accurate spatiotemporal modeling of conditions leading to moderate and large wildfires provides better understanding mechanisms driving fire-prone ecosystems improves risk management. Here, we develop a joint model for the occurrence intensity wildfire size distribution, by combining extreme-value theory point processes within novel Bayesian hierarchical model, use it study daily summer data French Mediterranean basin during 1995–2018. The component models ignitions as log-Gaussian Cox process. Burnt areas are numerical marks attached points considered extreme if they exceed high threshold. is two-component mixture varying in space time that jointly fires. We capture nonlinear influence covariates (Fire Weather Index, forest cover) through component-specific smooth functions which may vary with season. propose estimating shared random effects between components reveal interpret common drivers different aspects activity. This increases parsimony reduces estimation uncertainty, giving predictions. Specific stratified subsampling zero counts implemented cope observation vectors. compare validate predictive scores visual diagnostics. Our methodology holistic approach explaining predicting activity associated uncertainties.