Auxiliary diagnostic analyses used to detect model misspecification and highlight potential solutions in stock assessments: application to yellowfin tuna in the eastern Pacific Ocean

Integrated models (IMs) for stock assessment are simultaneously fit to diverse data sets to estimate parameters related to biological and fishery processes. Modelmisspecificationmay appear as contradictory signals in the data about these processes andmay bias the estimate of quantities of interest. Auxiliary diagnostic analysesmay be used to detectmodelmisspecification and highlight potential solutions, but no set of good practices on what to use exist yet. In this study, we illustrate how to use auxiliary diagnostic analyses not only to identify modelmisspecification, but also to understand what data components provided information about abundance. The diagnostic tools included likelihood component profiles on the scaling parameter, age-structured productionmodels, catch-curve analyses, and two novel analyses: empirical selectivity andmonthly depletion models. While the likelihood profile indicated model misspecification, subsequent analyses were required to indicate the causes as unmodelled changes in selectivity and spatial structure of the population. The consistency between the catch-curve models, the monthly depletion models and the IM information on abundance comes from a strong signal shared by several purse-seine fisheries data sets: the length composition data informs absolute abundance while the indices of abundance constrain the trend in relative abundance.

Automated summary

The study demonstrates the use of auxiliary diagnostic analyses to identify and understand model misspecification, with various tools revealing consistency in information about population abundance from different data sets. The analyses highlight the importance of considering various components of data in stock assessment models, to better estimate biological and fishery processes.