Bayesian statistics for fishery stock assessment and management

This work is about the use of Bayesian statistics in fishery stock assessment and management. Multidimensional posterior distributions replace classical parameter estimation in surplus-production and delay-difference models. The maximization of expected utilities replaces the estimation of optimal policies. Adaptive importance sampling is used to obtain approximations for posterior distributions. The importance function is given as a finite mixture of heavy-tailed Student distributions. The performance of the method is tested in five case-studies, two of which use data simulation. Real data refer to Skeena river salmon (Oncorhynchus nerka), Orange Roughy (Hoplostethus atlanticus) and Pacific cod (Gadus macrocephalus). The results show that the technique successfully approximates posterior distributions in higher dimensions even if such distributions are multimodal. When comparing models in terms of their performance as management tools, simpler and less realistic models can do better than more sophisticated alternatives. The Bayesian approach also sheds new light on the controversy about the Orange Roughy fishery.Science, Faculty ofStatistics, Department ofGraduat