The effects of applying mis-specified age- and size-structured models

- An operating model based on an age- and size-structured population dynamics model that included growth platoons is developed.
- Assessment methods based on age-, size-, and age- and size-structured population dynamics models are evaluated.
- Simpler models of those considered lead to least-worst performance.

Stock assessments estimate biomass and fishing mortality in absolute terms and relative to reference points. Most of the world's stock assessments are based on age- or size-structured population dynamics models, with few stock assessments directly accounting for both age and size dynamics. However, the life history parameters of fish and invertebrate populations are often functions of both age and size. An operating model that is based on an age- and size-structured population dynamics model is used to evaluate the performance of assessment methods based on age-, size- and age- and size-structured population dynamics models. Variants of the operating model and the assessment methods, which include 'platoons' to better represent individual variation in growth, are considered to explore the impact of this source of uncertainty on the performance of stock assessment methods. Simulation experiments based conceptually on the assessment for Pacific cod in the Eastern Bering Sea are used to explore the consequences of applying assessment methods that are mis-specified in terms of the population dynamics model and the way variation in growth is modelled. The age-structured assessment methods perform poorest, most likely because they mis-represent how mortality, due to fishing, is removed from the population in the operating model, but also because of mis-specification of the growth curve. The assessment methods with platoons can outperform those that ignore platoon structure when this is present, but their performance when there are no platoons is such that, overall, simpler assessment methods based on size-structured, or age- and size-structured population dynamics models, appear best. Conducting assessments using multiple model types and selecting the best model based on the lowest AIC was, however, the best approach overall.