Consequences of Alternative Catchability Models for a Stock Assessment of American Eels

Assumptions about catchability can have large effects on estimates from stock assessment models.  Our objective was to compare effects of different assumptions about time varying catchability on an assessment of American eels (Anguilla rostrata) in the Potomac River.  We developed a series of stock assessment models that differed in their assumptions about fishery catchability and the standard deviation (SD) of the indices of abundance.  The assessment models were sex- and age-structured and were fitted to a fishery independent recruitment index and a sex- and age-aggregated catch per unit effort (CPUE) index from the commercial pot fishery. We examined seven models of time-varying catchability for fishery CPUE (constant, white noise (two levels of assumed variance), random walk (two levels of assumed variance), density dependent, and effort dependent) and three levels of assumed SD of errors for the recruitment index (0.2, 0.3, and 0.4).  Estimated average abundance differed substantially among the models with range of 4.5-21.9 million eels.  Most models were hypersensitive to relatively small changes in the assumed SD of the recruitment index.  Notable exceptions were the density dependent catchability model and the white noise catchability model with a process error log-scale SD of 1.0, which had estimates that differed by less than 10% across the levels of recruitment index SD.  We suggest that a lack of sensitivity to assumed SDs of indices may be a useful method for selecting among competing assessment models with different assumptions about catchability, but more study is required.