Simulation of the December 2021 Marshall fire with a hybrid stochastic Lagrangian-cellular automata model

A stochastic model based on a combination of the cellular automata approach for forest fires and a random walk for firebrands and hot gases has been further developed and used to simulate the Marshall fire, Colorado, December 2021. Typical heat release profiles for burning wooden houses from the literature were used to distill information on the burning duration and ignition delay time needed to model the hours-long firebrand emission from wooden buildings in the Marshall area during this fire. In addition to information on vegetation and housing structures from in-person inspection, satellite images were used to estimate various model parameters. The results give reasonable predictions for the extent of the fire and its time evolution. A parametric analysis further highlighted the sensitivity of predictions to the parameters used in the model and suggested areas for improvement. The very low computational cost of the model, ease of operation, and acceptable accuracy suggest that the proposed framework can be used for operational decision-making and damage assessment.

Automated summary

A stochastic model combining cellular automata approach and random walk method was developed to simulate the Marshall fire in Colorado in December 2021. Information on burning duration and ignition delay time for firebrand emission was distilled from literature profiles of burning wooden houses. Satellite images and in-person inspection provided information on vegetation, housing structures, and model parameters. The model predicted the extent and time evolution of the fire reasonably well, and a parametric analysis identified sensitivity and areas for improvement. The low computational cost and ease of operation make the proposed framework suitable for operational decision-making and damage assessment.