The Western United States Large Forest-Fire Stochastic Simulator (WULFFSS) 1.0: A monthly gridded forest-fire model using interpretable statistics

Abstract. We developed WULFFSS, a new stochastic monthly gridded forest-fire model for the western United States (US). Operating at 12-km resolution, WULFFSS calculates monthly probabilities of forest fires ≥100 ha and area burned per fire. The model is forced by variables related to vegetation, topographic, anthropogenic, and climate factors, organized into three indices representing spatial, annual-cycle, and lower frequency temporal domains. These indices can interact, so variables promoting fire in one domain amplify fire-promoting effects in another. Fire probability and size models use multiple logistic and linear regression, respectively, and can be easily updated as new data or ideas emerge. During its training period of 1985–2024, WULFFSS captures 72 % and 83 % of observed interannual variability in western US forest-fire frequency and area, respectively. It reproduces regional differences in seasonal timing, frequencies, and sizes of fires, and performs well in cross-validation exercises that test the model’s accuracy in years or regions not considered during model training. While lacking fine-scale fire dynamics, WULFFSS’ use of classic statistics promotes interpretability and efficient ensemble generation. Designed to run within a vegetation ecosystem model, bidirectional feedbacks between vegetation and fire can identify how ecosystem changes have altered or will alter fire-climate relationships across the western US. The model's predictive power should improve with increasingly accurate and extensive observational data, and its approach can be extended to other regions. Here we provide a thorough description of the WULFFSS model, including the motivation underlying its development, caveats to our approach, and areas for future improvement.