Evaluation of a Statistical Model-Based Prediction of Mercury Concentrations in Ontario Sport Fish
MetadataShow full item record
Since the mid-1970s, the Ontario (Canada) Ministry of Environment (OMOE) has been collecting data on fish tissue mercury (Hg) contamination in provincial waterbodies. By 2004, approximately 160,000 fish from 86 species at over 1,600 sites were tested for Hg. This large database is primarily used to issue advisories for safe human fish consumption via publication of the biennial Guide to Eating Ontario Sport Fish. Analysis to uncover spatio-temporal trends while maximising the use of data points is complicated by the application of a non-random heterogeneous sampling design. The National Descriptive Model for Mercury in Fish (NDMMF) developed by the United States Geological Survey (USGS) is a statistical model of Hg concentrations that can potentially mitigate these challenges by separating the spatiotemporal variability of fish-[Hg] sampling while considering the effects of species, size, and fish sample portion type. However, the NDMMF has not been fully exploited, likely due to lack of rigorous evaluation. We conduct the first detailed investigation on the ability of the NDMMF to reproduce the observed fish-[Hg] in coolwater walleye (Sander vitreous) and warm-water yellow perch (Perca flavescens). Approximately two-thirds of both walleye and yellow perch [Hg]-length relationships could be accurately predicted using the NDMMF. For these cases, a majority (>85%) of the estimates are within the same consumption advisory categories as the interpolated [Hg] value based on the observed data, using an average-length fish. For the remaining incidences with significantly different NDMMF fish [Hg]-length relationships compared to those from the observed data, the NDMMF notably yields similar results, with a majority (>75%) of [Hg] estimates still falling within the same consumption advisory categories. For the small fraction of incidences with inaccurate advisory categorization, the instances of conservative over-prediction (<18%) would be of little human health concern as these would recommend fewer meals than otherwise suggested using observed data. For the few instances when [Hg] is under-predicted (<11%), the nature of the human health concern would be relatively minor because the advisory classification is almost never (<1%) more than one category less restrictive.