Bayesian Hierarchical Modelling of Spruce Budworm Development

Abstract

The management of destructive forest pests such as the spruce budworm relies on accurate modelling of their development. Predicting the timing of specific events in the life cycle is crucial for pest control tactics and for modelling the landscape-scale dispersal of the insect. This thesis implements a Bayesian hierarchical thermal response model for the larval stages of the spruce budworm. The model was fitted to data collected from a laboratory rearing experiment on wild spruce budworm colonies collected from locations across Canada and on a fully lab-reared colony. The results were compared across developmental stages and geographic origins. The Bayesian model was implemented with the non-linear, temperature-dependent development rate curve outlined in Schoolfield et al. 1981 and the framework in Régnière et al. 2012 for individual variation and interval censored data. Posterior samples were obtained and a quadratic relationship was observed between developmental stage and an intercept parameter of the development curve. A second model was fitted to the data incorporating this structure. Distributions of development rate estimates at each rearing temperature were obtained from each posterior sample and it was observed that the lab-reared colony developed more quickly than the wild colonies. In future work, the posterior samples can be used to generate simulated populations for prediction, with uncertainty fully propagated throughout.