A versatile net selectivity model, with application to Pacific salmon and freshwater
By Jeffrey F. Bromaghin
Fisheries Research, Volume 74, Issues 1-3, August 2005, Pages 157-168, ISSN 0165-7836, DOI: 10.1016/j.fishres.2005.03.004.
Gillnet catch data from the lower Yukon River, AK, collected from 1990 to 2003 in conjunction with a sonar study to estimate the abundance of migrating fish, were assembled. The full dataset contained 92,029 records with complete species and length information. A subset of data for the eight most prevalent groups of fish was selected for the estimation of net selectivity. The reduced dataset contained 89,984 records for Chinook salmon (Oncorhynchus tshawytscha), summer and fall runs of chum salmon (O. keta), coho salmon (O. kisutch), pink salmon (O. gorbuscha), humpback whitefish (Coregonus pidschian), broad whitefish (C. nasus), and various cisco (Coregonus) species. A Pearson function was used as a net selectivity model for all eight groups of fish, though a parameter was added to accommodate the catch of fish that are relatively large for a particular mesh. Because most of these relatively large fish were probably not gilled, but rather caught by body parts other than the operculum, the extra parameter can be thought of as a tangling parameter. The parameters of the modified Pearson model were estimated using maximum likelihood, and variances were estimated through bootstrapping. Gillnets were found to be most efficient, for all eight groups of fish, when fish length is approximately twice as great as the perimeter of a mesh, and the corresponding location parameter of the Pearson model was estimated with high precision. As the Pearson function has apparently not been used previously as a net selectivity model, its suitability was compared to the normal, lognormal, gamma, inverse Gaussian, and bi-normal functions, which are commonly employed as net selectivity models. Model fit was evaluated on the basis of the value of the likelihood function obtained, Akaike's Information Criterion and scaled deviance statistics, and plots of estimated models and scaled catch data. The Pearson model was found to be quite flexible, and fit the data as well as or better than the other models for all eight groups of fish considered. Other researchers may wish to consider its use with their data.
Keywords: Size selectivity; Unequal probability sampling; Gillnet; Catch-per-unit-effort; Maximum likelihood; SELECT