A data-driven firebrand generation model for application in wildland fire simulation platforms

This study develops a firebrand generation model, designed for integration into wildland fire simulation platforms, for fire spotting simulation. Firebrands are burning fuel fragments released from an active fire line that can ignite new fires upon landing. Known as “fire spotting,” this process often becomes the dominant mode of wildfire spread, underscoring its importance in wildland fire simulations. Modeling firebrand generation is the first step to achieve realistic fire spotting simulations. However, the literature lacks firebrand generation models suitable for integration into wildland fire simulation platforms. While various experimental studies have attempted to quantify firebrand generation, they each have limited focus and are not readily generalizable into a firebrand generation model. In this study, we have compiled and integrated the existing experimental data in the literature to develop a data-driven firebrand generation model. The model comprises three components that estimate firebrand yield, mass distribution, and mass versus projected area. The inputs include fuel type, the mass of fuel consumed, fuel moisture content, and wind speed, and outputs include firebrand realizations (samples) with assigned mass and projected area. Validation studies show good agreement between estimated and observed distributions of firebrand mass and projected area, with Chi-Square test failing to reject the null hypothesis between the observed and predicted firebrand mass and area distributions. Additionally, in a validation case, the model achieves average 7% error in predicted mean of the firebrand projected area. The resulting model is computationally efficient, relies on readily available input parameters within the existing wildland fire models, and provides reasonable predictions, making it a suitable model for implementation in current wildland fire simulation platforms.