Comparing Accuracy of Stock Status Estimates Among Integrated Models Using Functional-Form and Random-Walk Catchability

Many factors can cause catchability for a fishery-dependent index of abundance to change over time, and these changes can bias stock assessment estimates of historical abundance.  Time-varying catchability can be included in stock assessment many different ways, including by estimating a random-walk process error, or by using either single- or multi-species data to inform an assumed functional relationship between catchability and other derived variables such as stock abundance.  Although there are many proposed ways to account for time-varying catchability, few studies have analyzed the accuracy of stock status and benchmark estimates arising from these different methods.

In this study, I simulate age-structured data with density-dependent catchability, and apply multiple time-varying catchability methods to evaluate their relative accuracy in estimating stock status and management benchmarks.  Methods including: (1) assuming constant catchability; (2) discarding fishery-dependent indices; (3) estimating a random-walk process error in catchability; (4) using data from the focal stock to estimate density-dependent catchability; and (5) using a Bayesian prior from auxiliary stocks to inform the density-dependent catchability parameter.  When age composition data is abundant, I find that the random-walk method performs better than other methods.  However, when age composition data is unavailable and survey data are uninformative, single- and multi-species density-dependence models perform better than random walk and constant catchability models.  Finally, I find that the relative performance of single- and multi-species models is dependent upon data availability and the underlying catchability model.