Inferring fisheries stock status from competing hypotheses

Abstract A suite of statistical catch-at-age models with time-varying natural mortality ( M ), selectivity and catchability were fitted to data for eastern Georges Bank Atlantic Cod ( Gadus morhua ) and Georges Bank Yellowtail Flounder ( Limanda ferruginea ) to address hypotheses regarding nonstationarities in these systems. Model averaging and ensemble methods were used to combine the results of converged models for each species. Model averaging was based on both predictive (i.e., cross-validation, AIC) and retrospective performance metrics. Models allowing large shifts in M provided the best fit to observed data for both species and were also regarded as the most plausible. For Atlantic Cod, variance in biomass estimates among the individual models was very high and multimodel estimates were sensitive to the relative weight given to competing models. In contrast, converged Yellowtail Flounder models produced broadly similar output and multimodel estimates were more robust to model weights. Our results suggest that the plausibility of competing models should be scrutinized, particularly when structural uncertainty is high.