Efficacy of using stable isotopes coupled with chemometrics to differentiate the production method and geographical origin of farmed salmonids.

The δ2H and δ18O of 105 salmonids cultured in freshwater and seawater and from different regions were combined with linear discriminant analysis (LDA), k-nearest neighbor (KNN), and random forest (RF) to create discrimination models. To assess the stability of the discrimination models, seasonal variation in δ2H and δ18O in salmonids cultured in different systems was studied. δ2H and δ18O were significantly different between salmonids cultured in freshwater and seawater and from different geographical origins. δ2H and δ18O of salmonids cultured in an open system were vulnerable to seasonal effects. The KNN model had 100% accuracy for identifying the production methods of salmonids and was less affected by seasonal variation. The RF model had the highest accuracy for identifying the geographical origins of salmonids with an accuracy of over 80%. Thus, δ2H and δ18O were more effective for identifying the production methods of salmonids than their geographical origins.