Abstract:
Dealing with uncertainties in integrated age-structured assessment (ASA) models has become a central focus at all levels of fish stock assessment and management. My goal in this thesis was to uncover layers of uncertainty in ASA models. There were two major components to this approach: (1) dealing with uncertainty in datasets used in ASA models through examining effective sample size, and (2) dealing with uncertainty in ASA model structure through examination of effective number of parameters and simulations of school and stock structure. From the dataset uncertainty perspective, I investigated age and length composition datasets by first identifying possible sources of error and then by evaluating if it is feasible to include this error within ASA models. From the structural uncertainty perspective, I compared historical ASA models with spatially-explicit and metapopulation scale ASA models. Sources of uncertainty in age and length composition datasets include the spatial distribution of schools and age aggregation of fish within schools. To account partially for this error at the survey design level, the optimal approach is to sample from a greater number of schools, even if the sample size within any particular school is reduced. Also, it is possible to include this error within ASA models by parameterizing and estimating effective sample size with the Dirichlet distribution. Reduced uncertainty in parameters and management quantities resulted from spatially-explicit and metapopulation ASA models when compared to historical ASA model structures. Further, with possible climate change influences on fish populations use of spatially-explicit and metapopulation ASA models will allow stock assessment scientists to accurately and more precisely predict sustainable harvest levels.
