A Bayesian assessment of the mercury and PCB temporal trends in lake trout (Salvelinus namaycush) and walleye (Sander vitreus) from lake Ontario, Ontario, Canada

• We examine the temporal trends of PCBs and THg in fish from Lake Ontario.
• The PCB levels have steadily decreased over the past four decades.
• Mean annual mercury levels are characterized by a “wax and wane” pattern.
• Dynamic linear modeling offers a robust hindcasting tool with a flexible structure.
• Specific fish species may not act as bio-indicators for all contaminants.

Polychlorinated biphenyls (PCBs) and total mercury (THg) are two of the most prevalent contaminants, resulting in restrictive advisories on consuming fish from the Laurentian Great Lakes. The goal of this study is to examine the temporal trends of the two contaminants in walleye (Sander vitreus) and lake trout (Salvelinus namaycush) for Lake Ontario. We employed Bayesian inference techniques to parameterize three different strategies of time series analysis: dynamic linear, exponential decay, and mixed-order modeling. Our analysis sheds light on the role of different covariates (length, lipid content) that can potentially hamper the detection of the actual temporal patterns of fish contaminants. Both PCBs and mercury demonstrate decreasing temporal trends in lake trout males and females. Decreasing PCB trends are evident in walleye, but the mean annual mercury levels are characterized by a “wax and wane” pattern, suggesting that specific fish species may not act as bio-indicators for all contaminants. This finding may be attributed to the shifts in energy trophodynamics along with the food web alterations induced from the introduction of non-native species, the intricate nature of the prey–predator interactions, the periodicities of climate factors, and the year-to-year variability of the potentially significant fluxes from atmosphere or sediments. Finally, a meaningful risk assessment exercise will be to elucidate the role of within-lake fish contaminant variability and evaluate the potential bias introduced when drawing inference from pooled datasets.