Essential features of the next-generation integrated fisheries stock assessment package: A perspective

Integrated analysis (or integrated population modelling) methods have become the preferred approach for conducting stock assessments, and providing the basis for management advice for fish and invertebrate stocks since the publication of a seminal paper by Fournier and Archibald in 1982. Methods to assess fish stocks based on single-species, single-area, age-structured models are now standard, with the major debates associated with these models related to data choice, model configuration assumptions, and data weighting. However, the current generation of stock assessment packages is not addressing all of the needs of stock assessment analysts and managers. A major challenge for any next-generation stock assessment package is the set of extensions needed to assess stocks that do not satisfy the ‘well-mixed single-stock’ paradigm. In addition, the next-generation stock assessment package needs to: (a) be able to capture age and size/stage dynamics simultaneously yet computationally efficiently, (b) scale from data-rich to data-poor, (c) include some multi-species capability, and (d) more appropriately deal with temporal variation (e.g., random effects and state-space models). In relation to data, there is a need to ensure that the next-generation stock assessment package better handles tagging data (age-size/stage models may help in this regard), in particular, to be able to use close-kin mark-recapture data. Efficient methods are needed to share parameter priors among stocks (satisfying the promise of the ‘Robin Hood’ paradigm). The next-generation stock assessment package needs to have associated appropriate training programs and documentation. Adoption of such a package will be facilitated by a data entry system that is well-documented, does not require specification of inputs that will not be used in an application, has an expert system to configure default settings based on best practices, and has associated code to automatically produce diagnostic statistics. Some technical challenges that have plagued stock assessment for decades warrant continued attention (at the theoretical and applied level) such as automatic data weighting and tuning, how to handle spatial and stock structure, improved coding to facilitate application of state-of-the-art methods for quantifying uncertainty, and adoption of true state-space formulations to allow more parameters to be treated as random effects. Future needs for features cannot be anticipated, so the key design cons