Validation of a Predictive Model for Fish Tissue Mercury Concentrations

In an effort to reconcile the extensive variability in fish tissue mercury (Hg) in North Carolina, we previously created a predictive model using data collected from 1990 to 2006. Our model identified four factors—fish trophic status, fish species, ecoregion, and water pH—that explained 81% of the variation in fish tissue Hg. Herein, we evaluate the performance of this model by using two independently collected data sets describing fish tissue Hg in North Carolina and Virginia. Our model explained 75% of the variation in independently collected tissue Hg data from North Carolina and 68% of the variation in data collected from the same ecoregion types in Virginia. Although the relationships were relatively strong, the model generally underestimated observed tissue Hg. Residual analysis indicated that at least some of the model bias was due to error in predictions for sites near (within 10 km of) coal‐fired power plants. The model was not significantly biased when sites near power plants were removed from the independent data set, suggesting that the model was best at predicting mean tissue Hg in adult fish collected from water bodies greater than 10 km from power plants; such sites constitute the majority of the landscape. Fortuitously, fish at sites near power plants have significantly less tissue Hg than fish at sites far from power plants and are therefore of less concern to public health officials. The validated model can serve as a fish tissue Hg screening tool for use by state and federal agencies because it was created to predict Hg levels in numerous species and systems, and it only requires data that are readily available or relatively simple and inexpensive to measure. Identification of specific systems that may have highly contaminated fish and would not otherwise be sampled due to limited resources will help to focus sampling and enhance efforts to protect fish consumers.