Classical individual assignments versus mixture modeling to estimate stock proportions in Atlantic salmon (Salmo salar) catches from DNA microsatellite data

Mixture modeling is shown to outperform classical individual assignments for both estimating stock composition and identifying individuals' sources in a case study of an eight-locus DNA microsatellite database from 26 Atlantic salmon (Salmo salar) stocks of the Baltic Sea. Performance of the estimation methods was compared using self-assignment tests applied to each of the baseline samples and using independent repeat samples from two of the baseline stocks. The different theoretical underpinnings, hypothesis testing versus decision theory, of the methods explain their estimation capacities. In addition, actual catch samples from three northern Baltic Sea sites in 2000 were analysed by mixture modeling, and estimated compositions were consistent with previous knowledge. Baltic main basin and Gulf of Finland stocks were each minor components (<1% at any site), and three groups of Gulf of Bothnia stocks, wild (36%–43% among sites), Finnish hatchery (15%–49%), and Swedish hatchery (11%–41%), were each important with the two hatchery contributions trending geographically.