Joint modeling and prediction of massive spatio-temporal wildfire count and burnt area data with the INLA-SPDE approach

This paper describes the methodology used by team RedSea in data competition organized for EVA 2021 conference. We develop a novel two-part model to jointly describe wildfire count and burnt area provided organizers with covariates. Our proposed relies on integrated nested Laplace approximation combined stochastic partial differential equation (INLA-SPDE) approach. In first part, binary non-stationary spatio-temporal is underlying process that determines whether or not there at specific time location. second we consider hurdle log-Gaussian Cox (hurdle-LGCP) positive data, i.e., an LGCP shifted data. Dependence between captured shared random effect. modeling approach performs well terms of prediction score criterion chosen organizers. Moreover, our results show surface pressure most influential driver occurrence wildfire, whilst net solar radiation are key drivers large numbers wildfires, temperature evaporation areas.