Widespread occurrence of emerging contaminants in Great Lakes tributaries led to the development and publication of a vulnerability index (VI) to assess the potential exposure of aquatic communities to chemicals of emerging concern (CEC) in the Great Lakes basin. The robust nature of the VI was tested to evaluate the underlying statistical model and expand the spatial domain of the index. Data collected at 131 new sampling sites (Test 1) and published data from independent studies (Test 2) were used to test the model predictions. Test 1 water and sediment samples were analyzed for the same classes of CEC chemicals and compared to the predictions for the original VI. Concentrations and numbers of unique CECs detected in water and sediment samples were similar between the original data and the two test datasets, although CECs tended to have higher detection frequencies in the original dataset compared to the Test 1 and Test 2 datasets. For example, 69 CECs were detected in ≥30% of water samples in the original dataset compared with 17 CECs in the Test 1 data and 59 in the Test 2 data. Predicted vulnerability for test sites agreed with actual vulnerability 64% of the time for water and 71% of the time for sediment. Agreement percentage results were greater when individual sites were grouped by river, with 82% agreement between predictions and actual vulnerability for water and 78% agreement for sediment. For the entire dataset, the VI ranks correlated with an independent estimate of potential biological impact. Agreement percentage was the greatest for low or high vulnerability index values but highly variable for sites that are classified as having medium vulnerability. Despite the underlying variability, there is a significant correlation (R2 = 0.26; p < 0.01) between the VI ranking of tributaries and the independent ranking of potential negative biological impact.
Keywords: Bioeffects; Boosted regression tree (BRT) models; CECs; Contaminant occurrence; Great Lakes; Vulnerability index.
Published by Elsevier B.V.