| Abstract | The mountain pine beetle is forest insect that undergoes intermittent population eruptions. At epidemic levels, it must kill its host tree to reproduce. Currently, an outbreak in British Columbia and Alberta, Canada covers over 14 million hectares of mature pine forests, exerting landscape-level mortality and carbon impacts on the order of megatonnes. We consider two different ways of modeling mountain pine beetle outbreaks across space and over time in British Columbia, Canada. One approach is a spatial-temporal autologistic regression model in the framework of Markov random fields. The other approach is a generalized linear mixed model with spatial-temporal random effects. We devise computationally feasible algorithms for Bayesian inference in both approaches. An example using real data of a mountain pine beetle outbreak is provided for illustration. |
|---|
