Abstract:
This thesis presents and extends the 2014 Bayesian length-structured stock assessment model that was used to assess the blackfoot paua stock in Quota Management Area, PAU 5A. The performance of ADMB, the software used in the 2014 assessment, was compared to that of Stan and TMB, which are two new statistical programming languages. Shortcomings with the model for paua growth were identified and new formulations were presented to address these. Several model assumptions were scrutinized using sensitivity analyses. These assumptions included the period seasonal recruitment variation was estimated, the assumption that the stock were in an virgin state prior to 1965 and the assumption that the Beverton-Holt steepness parameter, H, was equal to a fixed value of 0.75. Finally, attempts were made to accurately estimate H using both vague and informative prior distributions. The latter were constructed by carrying out a meta-analysis on the PAU 5B and 5D stocks. Overall Stan had the best performance and proved to be the easiest of the three languages to code the model. There were insufficient data to accurately estimate the recruitment deviations any earlier than the period currently used (1986), the model was insensitive to violating the assumption of an equilibrium state prior to 1965 and a number of reference points were extremely sensitive to varying H. There was evidence that there were insufficient data to accurately estimate H when assuming vague priors. There were also indications that a steepness of 0.75 may have been conservative and that recruitment is primarily driven by environmental-based variables. Alternatively, when informative priors were used, the estimation of H was primarily driven by the prior, which was found to be strongly influenced by the steepness values within the stock assessments used in the meta-analysis. These results should help improve the current New Zealand paua stock assessment models. They are also further reaching, particularly the newly presented growth model, which could be adopted for any species where length-structured assessments are used.