Abstract:
|
Level set dynamics have proven to be an effective and flexible technique for tracking the spread of wildland fire fronts. With a specified function that gives the rate of advance of a fire front, the method can produce realistic topological spread of the fire (e.g., merging fronts, absorbing islands). The limiting component of level-set dynamics is the specification of this spread function (velocity). Here, we investigate a Bayesian implementation that learns the spread function for level set dynamics given covariates and random effects. There are several numerical challenges to this implementation. The approach is demonstrated on simulated data and observations of large wildland fires in the Western US.
|