Estimation of age composition from length data by posterior probabilities based on a previous growth curve: application to Sebastes schlegelii

Abstract

We developed a system to estimate the age composition of a fish population (Sebastes schlegelii) from length data by considering fish growth, length variation, proportion of age classes, and sexual dimorphism. Reasonable interpolations allowed age composition to be estimated when length data and age–length relationships were measured in different seasons. A growth curve was fitted to the mean length growth using a maximum likelihood method with an assumption of a normal distribution in length variation. Posterior probabilities were constructed with normal distributions according to Bayes’ theorem, and age composition and its confidence limits were reasonably estimated from the posterior probabilities and by bootstrap resampling. The influence of annual fluctuations of population properties was assessed by cross-validation, which was improved by updating the prior probabilities. While the new system was more robust than the age–length key for small numbers of aging data, it was impossible to improve the system by focusing on the length data alone because the correlation between the estimation error and the likelihood calculated from the length data alone was weak.