Forests, Fires and Stochastic Modeling (06w5062)
John (Willard) Braun (University of Western Ontario)
Charmaine Dean (Western University)
Fangliang He (University of Alberta)
David Martell (University of Toronto)
Haiganoush Preisler (USDA Forest Service)
Stochastic models should prove to be very useful in attacking these kinds of problems. In recent years, point process intensity models have been successfully used in the study of earthquakes and volcanoes. Related intensity models have high potential in the forest fire context. For example, the times and locations of lightning strokes and fire ignitions can be viewed as a bivariate point process. A mutually exciting process is a possible model for this process where the lightning process drives the ignition process, but where the ignition process itself may have a self-exciting component representing the generation of jump-fires. The parameters of such models can be estimated efficiently using maximum likelihood; covariates such as fuel-type and moisture can be accounted for, and a form of residual analysis can be performed to assess the appropriateness of such models for given data. Interacting particle system models offer a way to stochastically model the spatio-temporal dynamics of a forest fire. Stochastic cellular automata may provide a way to simulate fire spread more effectively than current deterministic models. In particular, such models may yield a more natural approach to modelling jump-fires.
Mixed Markov Spatial Models:
Many forest ecology applications require methods for mixture models where the population consists of two or more subsets behaving differently and where spatial clustering is evident both in the placement of the population mixtures and their evolution over time. The context is quite broad but a specific example is the analysis of weevil attacks on pine trees in a stem-mapped plot where it is hypothesized that resistant trees are present which never succumb to attack. Non-resistant trees experience transitions from the attacked to non-attacked states each fall, for example, depending on spatially correlated random effects. The primary goal is the development of methods for inference for such spatially correlated mixture Markov models. In the specific forestry application, this is important for identifying resistant trees and their characteristics and how susceptibility to attack depends on the spatial environment and temporal effects. For modeling succession dynamics, stem mapping techniques, statistical point process modeling and computer simulation may be jointly utilized.
Aims and Scope:
The purpose of the proposed workshop is to engage forestry researchers and statisticians in the scientific enterprise of developing novel methods for addressing these problems. Specific problems connected to wildfires and their effects will be a focus, but the workshop is aiming to have a somewhat broader context; forest management and ecology issues will also be considered. Thus, experts in both forest fire research and forest ecology will form the team of participants.
It is envisioned that the proposed workshop would build upon collaborative initiatives already conceived in the forestry and statistical communities. We note that there has been a forestry session in recent Statistical Society of Canada annual meetings in which speakers from the forestry community have been invited to give talks to statisticians. The recent Joint Statistical Meetings in Toronto also had a session on wildfires organized by David Brillinger of UC-Berkeley and chaired by Ronald McRoberts of the USDA Forest Service. The forestry community has also made overtures to statisticians, inviting several to a Fire Science meeting at the University of Toronto in April, 2004 in order to discuss open problems and possible connections between statistics and forestry.