Performance of operational fire spread models in California

Background Wildfire simulators allow estimating fire spread and behaviour in complex environments, supporting planning and analysis of incidents in real time. However, uncertainty derived from input data quality and model inherent inaccuracies may undermine the utility of such predictions. Aims We assessed the performance of fire spread models for initial attack incidents used in California through the analysis of the rate of spread (ROS) of 1853 wildfires. Methods We retrieved observed fire growth from the FireGuard (FG) database, ran an automatic simulation with Wildfire Analyst Enterprise and assessed the accuracy of the simulations by comparing observed and predicted ROS with well-known error and bias metrics, analysing the main factors influencing accuracy. Key results The model errors and biases were reasonable for simulations performed automatically. We identified environmental variables that may bias ROS predictions, especially in timber areas where some fuel models underestimated ROS. Conclusions The fire spread models’ performance for California is in line with studies developed in other regions and the models are accurate enough to be used in real time to assess initial attack fires. Implications This work allows users to better understand the performance of fire spread models in operational environments and opens new research lines to further improve the performance of current operational models.