Environ. Sci. Technol. 2006, 40, 6261-6268
Mercury in Soils, Lakes, and Fish in Voyageurs National Park (Minnesota): Importance of Atmospheric Deposition and Ecosystem Factors J . G . W I E N E R , * ,† B . C . K N I G H T S , ‡ M. B. SANDHEINRICH,† J. D. JEREMIASON,§ M. E. BRIGHAM,| D. R. ENGSTROM,⊥ L. G. WOODRUFF,| W. F. CANNON,# AND S. J. BALOGH4 University of WisconsinsLa Crosse, River Studies Center, La Crosse, Wisconsin 54601, U.S. Geological Survey, Upper Midwest Environmental Sciences Center, La Crosse, Wisconsin 54603, Chemistry Department, Gustavus Adolphus College, St. Peter, Minnesota 56082, U.S. Geological Survey, Water Science Center of Minnesota, Mounds View, Minnesota 55112, St. Croix Watershed Research Station, Marine on St. Croix, Minnesota 55047, U.S. Geological Survey, National Center, Reston, Virginia 20192, and Metropolitan Council Environmental Services, St. Paul, Minnesota 55106
Concentrations of methylmercury in game fish from many interior lakes in Voyageurs National Park (MN, U.S.A.) substantially exceed criteria for the protection of human health. We assessed the importance of atmospheric and geologic sources of mercury to interior lakes and watersheds within the Park and identified ecosystem factors associated with variation in methylmercury contamination of lacustrine food webs. Geologic sources of mercury were small, based on analyses of underlying bedrock and C-horizon soils, and nearly all mercury in the O- and A-horizon soils was derived from atmospheric deposition. Analyses of dated sediment cores from five lakes showed that most (63% ( 13%) of the mercury accumulated in lake sediments during the 1900s was from anthropogenic sources. Contamination of food webs was assessed by analysis of whole, 1-year-old yellow perch (Perca flavescens), a regionally important prey fish. The concentrations of total mercury in yellow perch and of methylmercury in lake water varied substantially among lakes, reflecting the influence of ecosystem processes and variables that affect the microbial production and abundance of methylmercury. Models developed with the information-theoretic approach (Akaike Information Criteria) identified lake water pH, dissolved sulfate, and total organic carbon (an indicator of wetland influence) as factors influencing methylmercury concentrations in lake water and fish. We conclude that nearly all of the mercury in fish in this seemingly pristine * Corresponding author phone: (608)785-6454; fax: (608)785-6959; e-mail:
[email protected]. † University of WisconsinsLa Crosse. ‡ U.S. Geological Survey, La Crosse. § Gustavus Adolphus College. | U.S. Geological Survey, Mounds View. ⊥ St. Croix Watershed Research Station. # U.S. Geological Survey, Reston. 4 Metropolitan Council Environmental Services. 10.1021/es060822h CCC: $33.50 Published on Web 09/06/2006
2006 American Chemical Society
landscape was derived from atmospheric deposition, that most of this bioaccumulated mercury was from anthropogenic sources, and that both watershed and lacustrine factors exert important controls on the bioaccumulation of methylmercury.
Introduction Methylmercury contamination has diminished the benefits produced by fishery resources in many North American waters (1). Mercury contamination of recreational fishes is widespread in the north-central United States and is the leading cause of surface water impairment in Minnesota, which has issued a statewide fish-consumption advisory for lakes since 1999 (2). Concentrations of methylmercury in game fish inhabiting interior lakes of Voyageurs National Park (Park) exceed criteria established by the state and the U.S. Environmental Protection Agency to protect the health of humans who eat noncommercial fish. Despite the Park’s semiremote location (north-central Minnesota) and seemingly pristine condition, northern pike (Esox lucius) from some lakes contain the highest concentrations of mercury reported in the state (Minnesota Fish Contaminant Database, Minnesota Pollution Control Agency, St. Paul, MN). Such observations have caused concern about potential ecological and health risks of exposure to methylmercury from consumption of fish from lakes in the Park. Moreover, concentrations in northern pike vary almost 10-fold among the small interior lakes in the Park, based on comparison of estimated standardized concentrations of mercury in edible fillets of 55-cm fish. The processes or factors causing this large variation in mercury levels in game fish in Park lakes have not been identified. Atmospheric deposition is widely considered to be the primary source of mercury accumulating as methylmercury in fish inhabiting lakes of the north-central United States (3-5). Yet it is highly improbable that the modern atmospheric loading of mercury to the lakes and their catchments varies sufficiently to account for the observed variation in fish-mercury levels among lakes, given the relatively small size of the Park, the lack of pronounced local sources of atmospheric mercury, and the spatial interspersion and close proximity of lakes with low and high mercury concentrations in fish. Monitoring of mercury in wet deposition by Glass and Sorensen (6) and the Mercury Deposition Network (MDN: http://nadp.sws.uiuc.edu/mdn/) has provided no indication of a significant spatial gradient in mercury deposition in the vicinity of the Park. We hypothesized that the variation in fish-mercury concentrations among interior lakes in the Park results from differences among lakes and watersheds in factors affecting the abundance of methylmercury. The characteristics of Park lakes and their watersheds vary substantially, and the transformations, cycling, and bioavailability of mercury can vary among lakes with different physicochemical and landscape features (7-9). Nearly all of the mercury in fish is methylmercury (10-13), a highly toxic compound that readily bioaccumulates and can biomagnify to high concentrations in aquatic food webs (1). The microbial methylation of inorganic Hg(II) is an important process affecting methylmercury concentrations in aquatic food webs supporting production of fish (1, 14, 15), and both internal and external sites of methylation can be important sources of methylmercury to lakes (9, 16). Given that wetlands are important sites of methylmercury production (16-18), we hypothesized VOL. 40, NO. 20, 2006 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
6261
TABLE 1. Range of Characteristics of the 17 Study Lakes and Their Watersheds in the Voyageurs National Park, MN variable
median
minimum
maximum
lake area (ha) mean depth (m) maximum depth (m) volume (106 m3) littoral area (%) watershed area (ha)a total wetland area (ha) connected wetland area (ha) pH (s.u.) TOC (mg/L)b Secchi depth (m) calcium (mg/L) dissolved sulfate (mg/L)
36.6 3.8 8.2 1.3 62 223.3 34 15.1 6.83 11.3 2.24 2.8 1.81
12.9 2.0 3.7 0.3 20 47.4 5.6 3.0 6.38 4.2 0.94 2.0 0.86
305 13 29 37.4 100 1735 447.3 293.1 7.59 17.3 4.95 7.8 3.56
a
Excludes lake area.
b
TOC ) total organic carbon.
that the spatial variation in methylmercury concentrations in lacustrine food webs in the Park was caused in part by variation in wetland influence among lakes. We here examine the relation of mercury in fish to characteristics of interior lakes and watersheds of the Park, with emphasis on factors that could influence the abundance of methylmercury and its entry into aquatic food webs. The objectives of this study were (1) to assess the importance of geologic and atmospheric sources of mercury to lakes and watersheds in the Park and (2) to identify ecosystem factors associated with variation in methylmercury concentrations in lacustrine food webs of the Park.
Materials and Methods Study Area. Voyageurs National Park is an 886-km2 park along the Minnesota-Ontario border, east of International Falls, MN. The Park contains many small interior lakes and wetlands, principally on the Kabetogama Peninsula, which is surrounded by three larger lakes (Rainy, Kabetogama, and Namakan). Watersheds in the Park are largely forest-covered and characterized by thin soils and abundant outcrops of Precambrian bedrock. Annual precipitation averages about 68 cm, and lakes are typically ice covered for 5-6 months of the year. The 17 study lakes were dilute, low-alkalinity drainage lakes ranging in area from 12.9 to 305 ha (Table 1 and Supporting Information). The watersheds of the 17 lakes ranged from 47.4 to 1735 ha (excluding lake area). The area of connected wetlands, defined as wetlands adjoining the lakeshore or connected to the lake by a surface inflow, ranged from 3.0 to 293 ha, representing from 3.5% to 20% of total watershed area. Data on lake morphometry and landscape features were obtained from refs 19 and 5 and the National Park Service (International Falls, MN). Further information on the climate, geology, hydrology, lakes, and aquatic communities of the Park is available elsewhere (5, 20, 21). The wet deposition of mercury at a nearby MDN site (MN18) averaged 8.5 ( 0.9 (s.d.) µg m2 yr-1 during 19982004 (excluding 2002), similar to rates measured at other MDN sites in the region. Inputs of mercury to the forest canopy at similar settings within ∼140 km of the Park ranged from 290% (22) to 390% (23) of direct wet deposition. Applying these percentages to rates of wet deposition at site MN18, we estimate that total annual inputs (wet deposition plus forest-canopy deposition) to watersheds in the Park averaged roughly 25-33 µg m2 yr-1. Assessment of Geologic and Atmospheric Sources. The abundance of geologic mercury was assessed by analyses of bedrock and soil sampled in 2000 and 2001. The four principal rock types that underlay the study lakes and watersheds were sampled by taking composite chips or single grabs from 6262
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 40, NO. 20, 2006
outcrops at 12 sites. Soil was sampled from upland and lowland (but not wetland) sites in 13 of the 17 studied watersheds. Soils were sampled by horizon, with samples from the O, A, and C horizons collected from pits dug with a shovel and extended with a hand auger to the C horizon or a depth (often) limited by the presence of impenetrable rock. Samples from the C horizon typically were composed of weathered parent material, usually thin glacial deposits on bedrock. Samples of rock and soil (n ) 189) were prepared at the U.S. Geological Survey analytical laboratory (Denver, CO) and analyzed at XRAL (Don Mills, ON). Woodruff et al. (21) have described methods and presented complete geochemical data for all samples. Mercury in soils and rocks was determined with cold vapor atomic absorption spectrophotometry, with a detection limit of 0.02 µg Hg g-1 dry weight. The importance of atmospheric deposition as a source of total mercury was assessed by analyses of dated sediment cores (24). Cores about 1 m in length were collected from the profundal zone of five study lakes in July 1996 with a piston corer, sectioned on site, analyzed stratigraphically for total mercury, and dated by 210Pb, as described by Engstrom et al. (25). These five lakes were chosen for core analysis based on their relatively simple bathymetry, adequate depth (maximum depth ranged from 7 to 29 m), small catchment area, and minimal level of human disturbance. 210Pb was determined by alpha spectrometry with thermal distillation and isotope dilution (26), and dates and rates of sediment accumulation were calculated with the constant rate of supply model (27). Freeze-dried samples of sediment from 16 to 20 strata in each core were acid digested and analyzed for total mercury by cold vapor atomic fluorescence with single goldtrap amalgamation (28). One sample from each core was analyzed in triplicate, and one matrix spike and two reference materials were analyzed with each core. Assessment of Ecosystem Factors Related to Mercury Concentrations. This evaluation included determination of mercury in lake water and fish, geospatial characterization of watersheds, and information-theoretic analysis of mercury concentrations in relation to ecosystem factors. Mercury in Lake Water. Near-surface samples of unfiltered lake water were taken in May, July, and September 2001 and May and July 2002. Ultraclean sampling procedures were applied during handling of samples. Methods for sampling and analysis of lake water for total mercury and methylmercury have been described (29). Concentrations of methylmercury in lake water were below the limit of detection (0.04 ng L-1) in 2 of the 17 lakes during five sampling events and in 4 of the lakes during one sampling event. In these cases, we used 0.02 ng L-1, half of the detection limit, as the estimated methylmercury concentration in statistical applications (30). Mercury in Fish. The concentration of total mercury in whole, 1-year-old yellow perch (Perca flavescens) is a useful indicator of methylmercury concentrations in lacustrine food webs (31). During their first year, yellow perch feed largely on zooplankton and small benthic invertebrates (32, 33). It can, therefore, be reasonably assumed that variation in concentrations of mercury in 1-year-old yellow perch among interior lakes in the Park is due to ecosystem processes and factors influencing the abundance of methylmercury in lakes, rather than to substantial differences in the trophic position. Widespread and abundant in lakes of the Park (5) and the Upper Midwest (34), the yellow perch is a preferred prey of many piscivoressincluding northern pike, walleye (Sander vitreus), and common loons (Gavia immer)sand an important link in the food-web transfer of methylmercury (31, 35, 36). One-year-old yellow perch were sampled a few days after ice melt in May 2000, 2001, and 2002 with small-mesh trap
nets, seines, and small electroshocker fished in littoral habitat. Each fish was measured (total length to 1 mm), weighed (to 0.01 g), and stored at e-30 °C in a labeled Ziploc bag until lyophilization at e-50 °C to a constant dry weight. The age of each fish was determined by examination of three or more scales taken from the area of insertion of the left pectoral fin (37). We analyzed a total of 612 yellow perch from the 17 lakes, with the sample size for individual lakes ranging from 5 to 61 fish (median, 39); these fish had hatched the previous spring and resided in a lake for about 1 year. Lyophilized perch obtained in 2000 were digested whole. Lyophilized samples of whole perch obtained in 2001 and 2002 were pulverized and homogenized with a mortar and pestle; a 50-mg sample of the homogenate from each fish was acid digested and analyzed by flow injection cold-vapor atomic absorption spectrophotometry with a Perkin-Elmer FIMS 100 (13). Data are reported as whole-body concentrations (dry weight) of total mercury. Method precision (relative standard deviation) from triplicate analyses of homogenized fish averaged 6.4%. Mean recovery of mercury from fish samples spiked before digestion was 97%. Mean measured concentrations of total mercury in five standard reference materials were within or very near the certified ranges, which ranged from 57.4-64.6 to 850-1050 ng g-1 dry weight (Supporting Information). Concentrations in all fish exceeded our calculated method detection limits. Geospatial Analysis and Wetland Indicators. Wetlands were classified and delineated from vegetative cover maps developed from aerial photographs taken in fall 1995 and fall 1996 (19). Watersheds were delineated with a 30-m Digital Elevation Model. Streamflow paths were assessed with the Hydro extension for ArcView (Minnesota Department of Natural Resources, Management Information Services Bureau, St. Paul, MN), to distinguish between wetland areas that drained directly to the study lakes (here termed “connected wetlands”) and wetlands that lacked a surface water connection to lakes. Wetlands are significant sources of organic matter for lakes, and the proportion of a catchment that is wetland strongly influences the yield of organic matter and organic carbon from a watershed (38-40). We calculated mean concentrations of total organic carbon (TOC) for each lake from data reported by Goldstein et al. (29), who analyzed TOC in near-surface samples of lake water taken in May, July, and September 2001. The validity of TOC as an indicator of wetland influence was evaluated by linear regression of TOC against connected wetland area in the watershed. Information-Theoretic Modeling. Linear models were constructed with predicted variables pertaining to mercury and predictor variables pertaining to ecosystem characteristics. In accordance with the information-theoretic approach, we applied judgment based on scientific understanding of factors and processes controlling the abundance of methylmercury in selecting predictor variables. The methylmercury concentration in lacustrine food webs is influenced by an array of physicochemical variables that affect the mercurymethylation potential of the lakes and their watersheds. To limit the number of predictor variables in our models, we included only key variables known to influence the methylation of mercury or the abundance of methylmercury, and we excluded redundant variables identified by correlation and principal component analyses. Four key physicochemical variables related to the mercury methylation potential of ecosystems were selected for evaluation as predictor variables in models: lakewater pH (41-43), dissolved sulfate (15, 4446), percent littoral area (8), and TOC in lake water, an indicator of wetland influence (16-18). Values for predictor variables were obtained as follows. Lakewater pH was the mean, calculated as the -log of mean
H+ activity, from field (2000, 2001, 2002) and laboratory (2002) pH measurements, with two to (usually) four measurements per lake. Dissolved sulfate was the mean concentration in near-surface samples of lake water taken in May 2000 and 2002. Percent littoral area was from Kallemeyn et al. (5). Total organic carbon was the mean of three near-surface samples taken from each lake in May, July, and September 2001 (29). Predicted variables included the mean concentration of methylmercury in unfiltered lake water (near-surface samples taken in May 2001 and May 2002, soon after ice melt when lakes were weakly stratified) and the mean concentration (dry weight) of total mercury in whole, 1-year-old yellow perch (sampled in May 2000, 2001, and 2002). We assume from prior analyses (13) that the mercury in yellow perch was present almost entirely as methylmercury. Models were compared with information-theoretic criteria (47, 48) to identify the ecosystem variables that were most closely related to the variation in mercury concentrations in water and in 1-year-old yellow perch from the 17 lakes. With this method, the best model is considered to be that with the minimum Akaike Information Criterion (AICC) corrected for a large number of parameters relative to sample size
AICC ) -2log e(L(θˆ )) + 2K + 2K(K + 1)/n - K - 1
(1)
where K denotes the number of estimable parameters in the model, L(θ) is the maximized log-likelihood, and n is the sample size (48). The other models in the set were assessed relative to the best model with ∆AICCi
∆AICCi ) AICCi - min AICC
(2)
and by Akaike weights (wi) R
wi ) exp(- ∆AICCi/2)/
∑ exp(- ∆AIC
Cr/2)
(3)
r)1
where ∆AICCi, AICCi, and wi pertain to the ith model. Models with ∆AICCi e 2 have substantial support, whereas models with ∆AICCi g 10 have essentially no support (48). Akaike weights (wi) are approximations of the probability that a particular model is the best model in an entire set of models. We used small numbers of independent variables and avoided including redundant variables to reduce the chance of overspecification in the models compared. The Proc Mixed procedure in SAS (SAS Institute Inc., Cary, NC) was used to calculate AICC values for all models, and the SAS Proc Reg procedure was used to calculate adjusted coefficients of determination (r2) and the probability of the F-statistic for each linear model. A Type I error of 0.05 was used to judge statistical significance.
Results and Discussion Assessment of Geologic and Atmospheric Sources. Analyses of soil, bedrock, and sediment cores indicate that atmospheric deposition was the dominant source of mercury to the Park. Bedrock and C-horizon soils contained much lower concentrations of total mercury than did soils of the O and A horizons (Supporting Information), indicating that geologic sources within the Park have contributed little mercury to the studied watersheds. Concentrations of total mercury in bedrock were at or below the detection limit of 20 ng g-1 dry weight. In the 13 sampled watersheds, the grand mean concentration of total mercury in C-horizon soils was 49 ng g-1 dry weight (range of means 20-93 ng g-1), much less than that in soils from the O horizon (grand mean 324 ng g-1, range 246-448 ng g-1) and the A horizon (grand mean 126 ng g-1, range 58-256 ng g-1). Concentrations of total mercury in A-horizon soils were positively correlated with the organic VOL. 40, NO. 20, 2006 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
6263
FIGURE 1. Concentrations (dashed lines) and rates of accumulation (solid lines) of total mercury (HgT) in sediment cores from profundal zones of five lakes in Voyageurs National Park, MN. carbon content of soil (r2 ) 0.82). The forest floor and A-horizon soils were a significant sink for atmospheric mercury in these and other forested watersheds (22, 49), retaining mercury deposited onto the forest canopy and later delivered to the forest floor in litterfall and throughfall. In sediment cores, concentrations of total mercury increased from a low background in preindustrial sediments deposited before 1860 to maximum values in the late 1900s (Figure 1). Preindustrial concentrations ranged from 47 to 167 ng g-1 dry weight, whereas modern (1990s) concentrations ranged from 102 to 364 ng g-1. Concentrations were lowest in the cores with the highest sedimentation rates, indicating that the sediment matrix acts as a diluent to the mercury inputs. Enrichment ratios (modern: preindustrial concentrations) varied little among cores (mean ( SD: 2.1 ( 0.1). Cores from the five lakes exhibited similar temporal patterns in rates of total-mercury accumulation, increasing above stable, preindustrial background rates in the early to mid 1800s (Figure 1). Annual accumulation rates for the five cores ranged from 2.4 to 21.2 µg m-2 in preindustrial times and from 13.5 to 42.4 µg m-2 yr-1 in the 1990s. Differences in total-mercury accumulation among the cores was partly a function of sediment focusingsthe preferential deposition of fine-grained sediments to the core sitesas indicated by the higher 210Pb flux in the cores with the highest mercury accumulation. An approximate correction for sediment focusing, based on the ratio of core-site 210Pb flux to the regional atmospheric flux (∼0.5 pCi cm-2 yr-1), yielded a much narrower range of mercury accumulation rates (7.818.6 µg m-2 yr-1). Moreover, much of the remaining variance in mercury accumulation can be explained by differences in delivery of mercury in catchment runoff. Both catchment size (catchment area: lake surface area) and the percentage of wetland in the catchment explained a significant portion of the variation in the focus-corrected total-mercury fluxes (r2 ) 0.78 and 0.89, respectively). The important point here is that mercury fluxes in single sediment cores are strongly influenced by sedimentation patterns within the lake and by watershed characteristics that control catchment runoff and are not a simple measure of atmospheric mercury deposition rates per se (50). Flux ratios comparing modern (1990s) to preindustrial (before 1860) accumulation of total mercury provide a more comparable measure of change in mercury deposition among lakessassuming no change in watershed loading or in-lake retention of mercury during the period of record (50). These ratios ranged from 2.0 in Shoepack Lake to 5.7 in Little Trout Lake, with a mean of 3.25 ( 1.5 for the five lakes. These 6264
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 40, NO. 20, 2006
values are similar to the range of flux ratios reported for cores from many other lakes in northern Minnesota (4, 25, 51) and other parts of North America (52-55). Hence, the trends in mercury accumulation in cores from the Park are consistentsin both timing and magnitude of increaseswith the broad regional pattern of mercury loading from atmospheric deposition. Moreover, most of the mercury deposited in the lakes since preindustrial times was from anthropogenic emissions, with the fraction of mercury derived from anthropogenic sources (HgAnthr) estimated with the equation 1996
∑ (Hg - Hg ) i
HgAnthr )
p
1900
(4)
1996
∑ Hg
i
1900
where Hgi is the mercury inventory for time interval i, and Hgp is the inventory for the same interval assuming a preindustrial (pre-1860) rate of mercury accumulation. With this approach, we estimated that 63 ( 13% (mean ( 1 SD) of the mercury accumulated in the sediment cores during 1900-1996 (the year of core collection) was from anthropogenic sources and that 66 ( 12% of the mercury in the uppermost stratumsreflecting mercury accumulation during 1990 to 1996swas anthropogenic. Mercury in Lake Water. Mean concentrations of total mercury in unfiltered lake water, sampled in May 2001 and May 2002, ranged from 0.45 ng L-1 in Mukooda Lake to 3.3 ng L-1 in Net Lake. Corresponding mean concentrations of methylmercury ranged from