Bayesian hierarchical surplus production model of the common whelk Buccinum undatum in Icelandic waters

Abstract

Whelk fisheries globally suffer a from a lack of data and dearth of stock assessments, even though they are growing and can contribute important income locally in some regions. Hierarchical Bayesian methods are especially suitable for estimating biological parameters and reference points for data-poor stocks, as they can borrow strength from stocks with more information. Here we analyze the common whelk (Buccinum undatum) fishery in Iceland, which was initiated in late 1996 and is restricted to Breiðafjörður in Western Iceland. Whelk may be vulnerable to overexploitation due to low dispersal, differences in growth and maturation on small spatial scales, and limited regulation. A population dynamics model for whelk was used to test whether inclusion of spatial variation improved model fit, and demonstrates where more data are needed to improve stock assessment. We used CPUE data as an index of biomass and catches to fit a Bayesian hierarchical Schaefer surplus production model that included six sub-regions, and compared this model with all models that included successively less hierarchical structure. We also consolidated environmental and whelk demographic data from Iceland, including growth, maturity, and mortality rates, to form prior probability distributions for population growth and carrying capacity parameters. To determine whether spatial catch distribution would aid sustainability, we compared results when catches were evenly distributed over sub-regions to those based on status-quo catch distributions. The best model chosen included spatial variation in population dynamics, but more contrast in the data or additional abundance data are needed to be useful for stock assessment. As whelk is traditionally a data-poor fishery, our results yield hyperparameter distributions that can be used as prior information for assessments of whelk elsewhere, as well as sensitivity analyses that show how influential prior distributions can be on model selection and results.