Nowcasting Burn Probability for the Contiguous United States

The probability of wildfire occurrence has been successfully predicted on coarse spatial scales (~0.5°) based on empirical modelling of burned area. This method has limited scope for predicting the wildfire hazard at local scales and in the near-term, both necessary for wildfire management. A high-resolution, practically applicable hazard model can be built through quantifying the probability of fire as a function of site-specific present and antecedent climate and vegetation variables. Here we apply the known Poisson and Pareto distributions of wildfire occurrence and fire size in a two-step hazard model, where the probability of a location being affected by wildfire is approximated using multiple climate and vegetation parameters. In addition, we examine other predictor variables that have been used for modelling fire at coarse resolution, e.g. road density, to determine at what spatial scale they lose predictive power. The study focuses exclusively on the contiguous United States, due to its comparatively long and high-resolution record of wildfire events.