Developing a data-poor stock assessment model for New Zealand bycatch species: Estimating catchability and abundance under varying gear and population distribution scenarios

Abstract

This study investigated methods pertaining to the assessment of data-poor fishery stocks within New Zealand. Recent modelling approaches (termed ‘N-mixture’) have appeared in the ecological literature that enable simultaneous estimation of abundance and detectability (gear efficiency in a fisheries context) from count data corresponding to a population of interest. This approach forms part of the quantitative risk assessment framework known as Sustainability Assessment for Fishing Effects, which has been applied to a range of species in Australian fisheries. There is potential for application of such methods in New Zealand, where an estimated 80% of fish stocks lack assessment under the Quota Management System. Various models were developed as part of this study in order to obtain quantitative estimates relating to catchability and abundance. These involved a mixture of statistical distributions (log-normal, Poisson, and binomial) and were implemented in the R and JAGS software. A range of simulations were devised to produce artificial data sets for model validation and analysis purposes. Models were analysed using a Bayesian hierarchical approach in order to estimate the parameters of interest. The models under investigation were also applied to red gurnard (Chelidonichthys kumu) data from the GUR 1 stock management area in northeastern New Zealand, to provide a case study for real-world applicability. Model validation revealed that recovering estimates of gear efficiency and abundance was possible using the approach considered here. Analysis simulations provided information regarding the potential sources of bias, variability and incompatibility arising from assumptions in model specification and data generation. These findings were also considered in the context of application to red gurnard. The most important issues highlighted by this study include the role of over-dispersion on modelling and data requirements, handling zeroinflated data sets, and bias introduced from catch reporting procedures in commercial fisheries within New Zealand. The results of this study provide an important first step in assessing data-poor stocks, providing a cost-effective means of obtaining key fishing parameters without significant data requirements. Further research into this area will enable these findings to be applied alongside other methods in order to gauge the sustainability of fishing impacts on bycatch species.