Geographic Variation in Trace Element Composition of Juvenile Weakfish Scales

Abstract

We determined Mg: Ca, Mn: Ca, Sr: Ca, and Ba: Ca levels in scales from juvenile weakfish Cynoscion regalis from five estuarine locations along the Atlantic coast of the United States using laser ablation inductively coupled plasma mass spectrometry. Significant variability in the multivariate elemental signatures was found among sites within an estuary, among estuaries, and between collections from 1996 and 1997. Bootstrapped 95% confidence ellipses on mean scores from a canonical discriminant analysis found that most estuaries were significantly separated in discriminant space in both years. Linear discriminant function analyses (LDFA) were then used to classify individual weakfish to their natal estuary. In 1996, cross-validated classification accuracy ranged from 38% for Delaware Bay to 86% for Pamlico Sound, with an overall accuracy of 67%. Classification accuracy in 1997 ranged from 41% for Chesapeake Bay to 83% for Pamlico Sound, with an overall accuracy of 65%. The use of artificial neural networks (ANN) to classify individual fish increased overall accuracies, when compared to LDFA, by an average of 2% in 1996 and 10% in 1997. Interannual variability in the trace element signatures meant that fish could not be accurately classified to natal estuary based on ANNs parameterized by signatures collected from juvenile fish in a different year. Finally, the trace element levels in scales were significantly correlated with otolith concentrations from the same fish, suggesting that similar processes control both scale and otolith chemistry.