| Summary: | Overfishing and environmental changes impose high risks on the wellbeing of the world's fish stocks. It is commonly acknowledged that fisheries management should be risk averse, following the principles of the precautionary approach, but unfortunately the statistical stock assessment methods often lack the ability to estimate the uncertainties related to their results. Further challenges arise from the fact that stocks which are in the most desperate need of a stock assessment are often data poor and resources to gather new data from them are scarce. Bayesian statistical inference can be utilized to conduct stock assessments cost-efficiently since these methods provide a formal way to combine information from various sources including databases, literature and expert knowledge. Bayesian inference is essentially a learning process where existing information is combined into a prior distribution, which is further updated with the most recent data. The result, a posterior distribution, expresses the best available knowledge about the phenomenon, including the related uncertainty. Furthermore, Bayesian hierarchical models enable learning between similar units, for example, stocks of the same or related species. This thesis consists of four studies that use Bayesian hierarchical models to improve knowledge in the fisheries stock assessment. Correlations between biological parameters, arising from different life history strategies among species, are utilized in paper [I] so that a data rich set of length-weight parameters can reduce the uncertainty of length-fecundity parameters for a data poor species. In paper [II], similarities between stocks of Atlantic salmon are used to estimate the stock specific key parameters of eggs-to-smolts relationship. A predictive distribution of this key parameter is also estimated, and could be used as an informative prior in a subsequent study of another Atlantic salmon stock. In addition, a hierarchical model is built to study structural uncertainty, estimating posterior ...
|
|---|