| A statistical, age-structured, multispecies modeling framework that integrates data from fisheries, research surveys, and predator diets to assess the status and productivity of harvested fish was developed. Parameters could be estimated based either on including the diet data and model components for predation in the likelihood, or ignoring these data and model components. When the data sources from diets were included, the form of the predator feeding response as well as the parameter values associated with specific forms was estimated and the best model(s) for an individual data set selected using AIC. The framework was applied to data for three fish species on the Aleutian Shelf, Alaska; walleye pollock, Atka mackerel, and Pacific cod. The estimates for quantities of management and ecological interest changed depending on whether or not predation was explicity included in the models, and on the relative weighting of the different sources of data based on the values specified for the effective sample sizes and variances. Simulation testing indicated the models that explicitly included predation performed no better than the models that ignored predation at estimating the quantities all models could estimate, and that additional quantities estimable by the models including predation were often poorly estimated. The effects of uncertainty regarding predator ration were evaluated by sampling theoretical distributions for the assimilation efficiencies and von Bertalanffy growth parameters. Ration uncertainty for each species individually affected the estimates for quantities such as the spawning biomasses for other species. The use of multiple structural models, data weightings, and data sampling is recommended to explore the uncertainty beyond the perceived uncertainty associated with only a single set of model assumptions when estimating quantities of management and ecological interest. |
