The importance of adjusting contaminant concentrations using environmental data: A retrospective study of 25 years...

To improve the statistical power of detecting changes in contaminant concentrations over time, it is critical to reduce both the within- and between-year variability by adjusting the data for relevant confounding variables. In this study, we present a method for handling multiple confounding variables in contaminant monitoring. We evaluate the highly variable temporal trends of Polycyclic Aromatic Hydrocarbons (PAHs) in blue mussels from the central Baltic Sea during the period 1987-2016 (data from 25 years during this period) using various regression analyses. As potential explanatory variables related to PAH exposure, we use mussel size and retrospective analyses of mussel δ¹⁵N and δ¹³C (representing large scale biogeochemical changes as a result of e.g. eutrophication and terrestrial inputs). Environmental data from concurrent monitoring programmes (seasonal data on Chlorophyll-a, salinity and temperature in the water column, bioturbation of sediment dwelling fauna) were included as variables related to feeding conditions. The concentrations of high-molecular-weight and low-molecular-weight PAHs in blue mussel were statistically linked to different combinations of environmental variables. Adjustment using these predictors decreased the coefficient of variation in all 15 PAHs tested and improved the statistical power to detect changes. Moreover, the adjustment also resulted in a significant downward trend for fluoranthene that could not be detected initially. For another PAH, benzo(g,h,i)perylene, adjustment which reduced variation resulted in the loss of an apparent downward trend over time. Hence, our study highlights the importance of using auxilliary data to reduce variability caused by environmental factors with general effects on physiology when assessing contaminant time trends. Furthermore, it illustrates the importance of extensive and well designed monitoring programmes to provide relevant data.