Abstract
Recruitment success of groundfish in the Gulf of Alaska has been linked to large-scale environmental drivers in the North Pacific. Stock assessment models can include environmental drivers and produce population forecasts to understand the effects of these drivers on population dynamics and performance of fishery management. However, it is not clear how to select among stock assessment models that differ in the way they include environmental drivers of recruitment, and how harvest control rules will perform given different model configurations and future climate uncertainty. I used simulation testing to determine the robustness of model selection tools (Akaike's Information Criterion,Mohn's retrospective statistic, and hold-out validation) to choose among a set of stock assessment models for a Gulf of Alaska flatfish-like species. I configured four operating models in Stock Synthesis that contained alternative assumptions about how an environmental index affects recruitment. I then fit a set of estimation models with different assumptions about an environmental driver of recruitment to test model selection tools. I also forecasted estimation models with correctly specified and mis-specified environmental drivers of recruitment under two climate emissions scenarios with F[subscript]MSY and F[subscript]40% harvest control rules. Akaike's Information Criterion, Mohn's retrospective statistic, and hold-out validation were not robust tools for selecting correctly specified estimation models fit to data from operating models. Using models chosen from model selection tools led to slight negative bias in estimates of yield at the biomass target. Incorporating information about environmental drivers of recruitment within stock assessment leads to different population dynamics and management advice. Forecasting these models under future climate change scenarios using catch advice from mis-specified models led to the population crashing in some instances and overfishing in long-term forecasts. The F40% harvest control rule led to overfishing and collapse less frequently than the FMSY harvest control rule for these models. If short-term or long-term management objectives are to manage sustainable fisheries where there is a well-known environmental driver of recruitment, stock assessments should include this information and robust harvest control rules should be used.