Bayesian methods for marine mammal population assessment
Policy-makers increasingly need to use scientific data that are imprecise. This problem is particularly apparent for marine mammal management issues, where practical research constraints leave scientists and managers with the problem of drawing inference from sparse data. Effective use of such data therefore places great demands on our methods of data analysis and statistical inference. In this thesis I introduce novel Bayesian methods for the analysis of data on marine mammal abundance and trends. Bayesian methods are applied to a suite of case studies to inform current management issues of importance both in the UK and overseas. These include estimating the probability of density dependence in the growth of a killer whale (Orcinus orca) population inhabiting the inshore waters of Washington State; estimating the size of a widespread population of bottlenose dolphins (Tursiops truncatus) in the Bahamas; and assessing the population status and abundance trends of bottlenose dolphins within a newly designated Special Area of Conservation in the Moray Firth, NE Scotland. Each of these case studies uses model-based analysis of individual photo- identification data to make inference about unknown population parameters of interest. Specifically, Bayesian inference, based on "posterior" probability distributions and statements, is used to facilitate scientific reporting in the face of uncertainty about these unknowns. Additional issues addressed are the selection of alternative statistical models for inference based on posterior model probabilities; incorporating model selection uncertainty into inference through the estimation of model-averaged parameter estimates; and the use of random effects prior distributions to model the relatedness between unknown parameters and increase estimate precision. The application of these methods is accomplished through the use of Markov chain Monte Carlo sampling methods, which are implemented using the WinBUGS software.