Bayesian Network Relative Risk Models (BN-RRM) were developed to assess recent (2005-2020) risk of mercury (Hg) exposure to the freshwater ecosystems of Great Slave Lake (GSL) and the Mackenzie River Basin (MRB) in the Canadian Northwest Territories. Risk is defined as the probability of a specified adverse outcome
here the adverse outcome was the probability of environmental Hg concentrations exceeding the Hg regulatory guidelines (thresholds values) established to protect the health of humans and aquatic biota. Environmental models and Hg monitoring studies were organized into a probabilistic (Bayesian network) model which considered six Hg input pathways, including atmospheric Hg deposition, Hg release from permafrost thaw, terrestrial to aquatic Hg transfer via soil erosion, and the proximity to mining, fossil fuel developments, and retrogressive permafrost thaw slumps (RPTS). Sensitivity analysis was used to assess spatial trends in influence of the sources to Hg concentrations in freshwater and in the tissue of five keystone fish species (lake whitefish, lake trout, northern pike, walleye, and burbot) which are essential for the health and food security of the people in the MRB. The risk to the health of keystone fish species, defined by toxicological dose-response curves, was generally low but greatest in GSL where fish size, mine proximity, and soil erosion were identified to be important explanatory variables. These BN-RRMs provide a probabilistic framework to integrate advances in Hg cycling modeling, identify gaps in Hg monitoring efforts, and calculate risk to environmental endpoints under alternative scenarios of mitigation measures. For example, the models predicted that the successful implementation of the Minamata Treaty, corresponding to 35%-60% reduction in atmospheric Hg deposition, would translate to a ∼1.2-fold reduction in fish Hg concentrations. In this way, these models can form the basis for a decision-support tool for comparing and ranking risk-reduction initiatives.