A Markov Chain Monte Carlo Model to Forecast Wildfire Suppression Costs

A Markov Chain Monte Carlo (MCMC) model provides a probabilistic forecast of project cost related to wildfire suppression in consideration of SPC fire outlook area and risk level. The model is constructed using 6-year distributions of the number of wildfires per SPC fire weather outlook unit area and burn area per wildfire, each for which a Weibull curve is fit. Based on 10,000 random simulations, probability density functions (PDFs) are generated that provide a probabilistic forecast of total burn area for an SPC outlook of a given areal extent. A bivariate spline surface is fit to a 2D histogram of observed values of burned area and associated project cost to formulate a statistically-viable project cost for each of the 10,000 simulations, resulting in a PDF of forecast project cost for wildfires of a specific SPC outlook. An MCMC model is trained separately for the three SPC outlook risk categories (elevated, critical, and extremely critical). Using one-year of test data, preliminary model results verify well against observed project costs per SPC outlook.