Abstract
Biogeochemical conditions and landscape can have strong influences on mercury bioaccumulation in fish, but these effects across regional scales and between sites with and without point sources of contamination are not well understood. Normal means clustering, a type of unsupervised machine learning, was used to analyze relationships between forage fish (Fundulus heteroclitus and Menidia menidia) mercury (Hg) concentrations and sediment and water column Hg and methylmercury (MeHg) concentrations, ancillary variables, and land classifications within the sub-watershed. The analysis utilized data from 38 sites in 8 estuarine systems in the Northeast US, collected over five years. A large range of mercury concentrations and land use proportions were observed across sites. The cluster correlations indicated that for Fundulus, benthic and pelagic Hg and MeHg concentrations were most related to tissue concentrations, while Menidia Hg was most related to water column MeHg, reflecting differing feeding modes between the species. For both species, dissolved MeHg was most related to tissue concentrations, with sediment Hg concentrations influential at contaminated sites. The models considering only uncontaminated sites showed reduced influence of bulk sediment MeHg for both species, but Fundulus retained sediment drivers at some sites, with dissolved MeHg still highly correlated for both. Dissolved organic carbon (DOC), chlorophyll, land use, and other ancillary variables were of lesser importance in driving bioaccumulation, though DOC was strongly related within some clusters, likely in relation to dissolved Hg. Land use, though not of primary importance, showed relationships opposite to those observed in freshwater, with development positively correlated and forests and agriculture negatively correlated with tissue concentrations across clusters and species. Clusters were composed of sites from geographically distinct systems, indicating the greater importance of small scale drivers of MeHg formation and uptake into the food web over system or region-wide influences.