|Institution:||University of Washington|
|Keywords:||Bootstrap; Delta method; Fisheries; MCMC; Stock assessment; Uncertainty; Statistics; Natural resource management; fisheries|
|Full text PDF:||http://hdl.handle.net/1773/35582|
Uncertainty is an integral part of fisheries stock assessment. Successful resource management requires scientific analysis to evaluate the uncertainty about the status of each stock and related quantities of interest. A failure to incorporate uncertainty into management advice increases the risk of suboptimal yields and can lead to a fishery collapse. In practice, it is not always clear which features of stock assessment data make them informative or uninformative, and it is also unclear how well different statistical methods are likely to perform when evaluating uncertainty. This study uses simulation analysis to measure the performance of alternative methods, based on a large number of simulated datasets where the underlying true values are known. The methods are then applied to data from an actual fishery, and the overall inference takes into account the performance of the methods in the simulations. The results show that the historical levels of stock size and harvest rate greatly affect how informative the data are about the current stock status. The key parameters natural mortality M and stock-recruitment steepness h pose challenges when it comes to statistical estimation, and long-term management advice is likely to depend strongly on the estimated or assumed values of M and h. The most informative fishing history is one where the data include years of high and low stock size, which is informative about h, as well as high and low harvest rates, which is informative about M. The results also indicate that confidence intervals describing the uncertainty about the stock status and other quantities of interest are likely to be too narrow in general. Benchmark analysis indicates that the delta method, Markov chain Monte Carlo (MCMC), and profile likelihood approaches are likely to perform better than the bootstrap for quantifying uncertainty. A bias correction algorithm for the bootstrap improved its performance, but not enough to match the performance of the other methods. Additional approaches to evaluate the estimation uncertainty include retrospective analysis and bivariate confidence regions for the current stock status. The use of harvest control rules to incorporate uncertainty into management advice is also discussed. The main value of this study is to present a comprehensive overview and evaluation of methods to analyze uncertainty. The study concludes with a checklist of recommendations for confronting uncertainty in stock assessment. Advisors/Committee Members: Hilborn, Ray (advisor).