A framework for conducting, evaluating, and communicating probabilistic wildland fire prediction

As part of wildland fire management practitioners use deterministic fire behavior models to anticipate how the fire environment will affect future fire growth. Additional current tools (e.g. FSPro) aim to estimate the uncertainty in those predictions by calculating a range of possible outcomes based on historical weather variability at weather stations in the fire area or by varying inputs across the natural range of all possible inputs.We will develop demonstrate and evaluate an improved framework for forecasting fire growth and effects with probabilistic predictions to at least four incidents where the variability and spread in an ensemble of simulations of each incident (built in Phase I on uncertainty in two key inputs: current forecasted weather and fuel loading) provide additional information requested by managers. The additional intelligence would additionally predict the most likely outcome our confidence in the prediction and the potential for low probability but high impact outliers. This probabilistic fire prediction methodology - which can be applied to a wide range of fire modeling tools - requires new approaches to forecast evaluation/validation and communication of uncertainty (both its calculation and graphical representation) to a wide range of stakeholders. Letters of support come from four stakeholders. The outcome will advance scientific research fire management and generate a foundation for business application development. Outcomes could feed into a variety of professional services at local and strategic scales across public and private sectors.