Effect of a moderate-size reservoir on transport of trace elements in a watershed

ABSTRACT

Wildman RA, Forde NA. 2016. Effect of a moderate-size reservoir on transport of trace elements in a watershed. Lake Reserve Manage. 32:353–365.

We assessed the extent to which Grand Lake, Oklahoma (>30 m deep, >80 km long), retains Fe, Mn, P, As, Zn, Pb, and Cd. Filtered water samples and suspended sediment samples were collected upstream of, within, and downstream of the reservoir. We then estimated instantaneous, seasonal, elemental fluxes. In winter and spring, when storms brought high flows to the reservoir, Grand Lake modified flood water minimally. During these seasons, trace element distributions were determined by the passage of storm inflows through the reservoir. In summer, Fe, Mn, P, and As were enriched in anoxic bottom water and exported through the dam, which draws water from below the surface mixed layer. Concentrations of aqueous elements in the water column were lower following autumn overturn, perhaps due to precipitation of metal oxides and settling. Unlike Fe, Zn was retained in Grand Lake during all seasons. Concentrations of Cd and Pb in filtered samples were often below our detection limit. Logistic regression indicated that Zn predicted detectable Cd, and so Grand Lake probably sequesters Cd. Sequestration of Pb was unclear because detectable Pb was predicted by both Zn and Fe. This study shows that watershed hydrology determines the transport of trace elements through a reservoir during times of high flow but that vertical circulation and biogeochemistry dominate during summertime and autumn low flows. Understanding these mechanisms can aid reservoir managers who seek to reduce downstream loads of trace elements.

KEYWORDS:

• Anoxia
• flood
• phosphorus
• reservoir
• stratification
• trace metals
• watershed

Reservoirs of moderate or large size influence chemical transport in their watersheds. They are often located on large rivers to maximize inflow and thus often receive large loads of dissolved and particulate chemicals relative to lakes of similar size (Kalff 2002). As reservoirs trap suspended sediment, they trap a particulate chemical load (Horowitz et al. 2001, Lee et al. 2001, Wildman et al. 2011). Trace elements have been characterized in rivers (e.g., Horowitz et al. 2001, Müller et al. 2008), but less attention has been devoted to the modification of the downstream transport of such chemicals by reservoirs.

The hydrology of sizeable reservoirs and their major tributaries can affect chemical transport downstream in 2 important ways. First, increased seasonal flow in tributaries brings larger chemical loads to a reservoir (Horowitz 2008). Second, at low flow, the lacustrine zones of temperate-latitude reservoirs stratify in summer with the potential for hypolimnetic anoxia and subsequent reaeration during overturn or release from the dam (Kalff 2002). Hypolimnetic anoxia can lead to reductive dissolution of iron (Fe)-oxide minerals, which can be released downstream (Ashby et al. 2004). Elements that sorb to metal-oxides, such as arsenic (As) and phosphorus (P), can behave similar to Fe in reservoirs (e.g., Kneebone and Hering 2000). The simultaneous influences of watershed hydrology and biogeochemistry, which vary seasonally, prevent a simple understanding of the effect of a given reservoir on elemental transport in a watershed and thus complicate management efforts to reduce contamination by trace elements.

The purpose of this study was to explore the effect of Grand Lake, a reservoir in Oklahoma, on the transport of trace elements in the watershed of the Grand River. Grand Lake is particularly interesting because it lies at the upstream end of a chain of reservoirs and downstream of multiple sources of contamination. The transport of trace elements through this reservoir is important for management of water quality in its downstream watershed. We studied metals (lead [Pb], zinc [Zn], and cadmium [Cd]), a metalloid (As), and a nutrient (P) that represent most of the primary abiotic threats to water quality in this watershed. The Tar Creek Superfund Site, a historic mining district with abundant mine tailings rich in Pb, Zn, and Cd (Schaider et al. 2007, Andrews et al. 2009), is just upstream of Grand Lake. The Grand Lake watershed contains several industrial chicken farms that might release As to the environment (Hilleman 2007). The US Agency for Toxic Substances and Disease Registry (ATSDR) maintains a Substance Priority List that ranks contaminants by the threat they pose to human health. On it, As, Pb, Cd, and Zn are ranked 1, 2, 7, and 75, respectively (USATSDR 2016); As and Pb rank high on the list because As is a carcinogen and a neurotoxin in children, and Pb is a neurotoxin in adults and is especially potent in children (USATSDR 2007a, 2007b). Cd causes bone disease and kidney damage, and, in high doses, Zn causes gastrointestinal illness, anemia, and damage to the pancreas (USATSDR 2012, 2005). Grand Lake is eutrophic because of the agricultural land in its watershed (OWRB 2009), and so P is a nutrient of concern. The cycling of these and many other trace elements in lakes is strongly influenced by that of Fe and manganese (Mn; e.g., Kalff 2002), and so Fe and Mn were included in this study as well.

Because the limnology of a warm temperate reservoir varies seasonally, the ultimate independent variable (i.e., the presence of a reservoir in the watershed) was considered through 2 separate variables: inflow rate and state of vertical circulation. We conceptualized our dependent variable as the instantaneous mass flux of elements through the reservoir during different seasons. We compared this flux to the summed instantaneous mass loading of elements via tributaries entering the reservoir, which would be the mass flux downstream in the absence of the reservoir. Thus, this study sought to determine the effect of seasonal variation in flow and vertical circulation on the transport of a suite of trace elements through Grand Lake. Elements were studied in both aqueous and particulate form. We evaluated 4 hypotheses:

1.	During all seasons, the reservoir will retain particulate elements because of particle settling.
2.	Under vertically well-mixed conditions and base flow, transport of aqueous trace elements will be unaffected because biogeochemical conditions in Grand Lake resemble those of its tributaries.
3.	During stratification and base flow, the reservoir will export aqueous trace elements dissolved under anoxic conditions in the hypolimnion.
4.	During elevated inflows, fluxes of aqueous elements will be greater, and trends that occur at base flow will be magnified.

Study site

Grand Lake (sometimes called Grand Lake O’ the Cherokees) is a reservoir of moderate size (36 m maximum depth when not retaining floodwater, >80 km long) in northeast Oklahoma (Fig. 1). It is impounded by the Pensacola Dam, constructed to generate hydropower and to protect downstream areas from flooding (Hoagland 1986), has a volume of 1900–2000 gigaliters (GL; USACE 2015), and a watershed of 26,600 km². Less than 30% falls within another flood control project well upstream (USACE 2015), so most of the flow into Grand Lake is unregulated and, as a mainstem reservoir, likely receives a notable sediment load (Wetzel 2001, Kalff 2002). Unregulated inflows imply considerable variation in inflow rates during a year. The median of recorded daily mean inflows in the 9 years preceding this study was 94 m³/s, although drought conditions during our study (Dec 2010 through Nov 2011) resulted in a median inflow of 35 m³/s (Wildman 2016). Small storms in late winter resulted in inflows from 85 to 490 m³/s, and 2 large storms in spring each increased inflows to >550 m³/s for nearly a week (Wildman 2016). Despite these variations, the water level of Grand Lake generally varies <1 m throughout the year because, except when it retains and later releases floodwater, outflows generally match inflows. Consequently, the reservoir water balance was near neutral in December, increasing by 40 GL/day in February during storm inflows, decreasing by 16 GL/day in May as water was released following a large storm, and decreasing slightly in August and November as part of minor drawdowns of water level (Wildman 2016). Accordingly, water exchange was high in February and May and low in August and November (Wildman 2016).

Figure 1. Location of Grand Lake in Oklahoma and the United States. Arrows indicate direction of water flow from major tributaries the Neosho River (NR), the Spring River (SR), and the Elk River (ER) and minor tributaries Tar Creek (TC), Buffalo Creek (BC), and Honey Creek (HC) through the reservoir and into the Grand River (GR). Circles indicate locations where water and suspended sediment samples were collected, although samples were not collected at all locations on all sampling excursions. Gray-black shading indicates relative depth between the shoreline and the dam face, where water is usually 36 m deep. Map adapted from OWRB (2009).

Inflows to Grand Lake come primarily from the Neosho River in the northwest, Spring River in the north, and Elk River in the east, collectively contributing 75% of the water released from the dam during this study (Wildman 2016). The Neosho and Spring Rivers drain predominantly agricultural land, and the Elk River drains forested uplands and receives permitted discharge from poultry plants. In the northwest, Tar Creek drains the Tar Creek Superfund Site, and in the southeast, the Honey Creek watershed drains chicken farm runoff.

The confluence of the Neosho and Spring rivers, which has been submerged by Grand Lake, was the historical source of the Grand River, which now originates from Pensacola Dam. Penstocks of the dam, used for all dam releases except during floods, are 4 m tall screened openings centered ∼16 m below the water surface and located at the southwestern end of the dam. Downstream from the dam, water from Grand Lake flows through 2 additional reservoirs on the Grand River and 2 more on the Arkansas River before meeting the Mississippi River.

Grand Lake is a warm monomictic reservoir. During our study, it turned over in October, reached an isothermal minimum temperature of 5 C in winter, and exhibited stratification by early May. In midsummer, stratification was pronounced; the surface mixed layer was 8–11 m thick (Wildman 2016), extending to the bottom of the reservoir at distances >52 km upstream from the dam (i.e., the vicinity of the confluence of the Elk River arm with the thalweg; Wildman 2016). During this study in May, dissolved oxygen (DO) was >2.5 mg/L in the deepest water by the dam and 4.0–9.5 mg/L otherwise. In summer, the metalimnion and hypolimnion were anoxic, and the entire water column readily became oxic during autumn overturn (Wildman 2016).

Methods

Water samples were collected in December 2010 and February–March, May, August, and November 2011. They were collected from 5 (Dec, Nov) or 6 (Feb–Aug) locations spaced at approximately equal distances across the thalweg of Grand Lake, the tributaries described earlier, and the Grand River <500 m below the dam (Fig. 1). Based on circulation parameters measured concurrently, 1–4 depths at each site were sampled; one sample was always collected from the surface mixed layer, and additional samples were collected from below the surface mixed layer and from water ≤1 m from the sediment–water interface. Samples were pumped through silicone tubing precleaned with dilute nitric acid and rinsed thoroughly with distilled deionized water. Before sample collection at a given site and depth, 3 tube volumes of reservoir water were passed through the tubing and discarded. Reservoir water samples were filtered through preweighed 0.45 μm polyethersulfone membranes, either within 24 h in a class-100 clean bench (Dec–May, when DO was >4.3 mg/L in all samples) or immediately inline (Aug–Nov). After filtering, samples were acidified with ultrapure nitric acid. Field blanks and occasional field duplicates verified the low background concentrations of analytes and reproducibility of field protocols.

Elements in the aqueous phase were quantified by inductively coupled plasma (ICP) mass spectrometry (Mn, Pb, Zn, Cd, As) and ICP optical emission spectrometry (Fe, P). Measurements were calibrated with standard solutions made by diluting commercially available stock solutions. Detection limits were 0.1 μg/L for Mn, Pb, Zn, Cd, and As and 1 μg/L for Fe and P. During a given analysis, no samples were analyzed until the instruments were tuned within operating specifications, method blanks returned concentrations below detection limits (BDL), and a linear calibration curve was produced.

Elements associated with suspended sediment were quantified after extraction from filter membranes, which were first oven-dried (at 65 C until mass was constant) and weighed to determine sediment mass. Membranes were subsequently microwave-digested in concentrated ultrapure nitric acid in Teflon digestion vessels. Digestates were diluted 1:10 with distilled, deionized water and analyzed like the water samples described earlier. Further 100- or 1000-fold dilutions with 5% ultrapure nitric acid were usually necessary to bring samples within the range of concentrations of the calibration curve. Digestion efficiency was verified by digesting NIST 2709 and 2711 certified reference materials several times along with batches of samples. Filter blanks and digestion vessel blanks verified that techniques were trace-metal clean. Detection limits of these measurements varied between samples because, although the ICP detection limits were constant, the digestate volumes, sediment masses on filters, and volumes of water filtered were not. The interquartile range (IQR) of detection limits was 0.2–1.7 mg/kg, and the median was 0.5 mg/kg.

Elemental fluxes were calculated for each sampling excursion by multiplying the average flow of the days when sampling occurred by the concentrations of elements in samples collected from tributaries and the dam tailrace and then subtracting the tailrace values from the sum of the tributary values. When concentrations were BDL, half the detection limit was used for this calculation. Volumetric flows of the Grand, Neosho, Spring, and Elk rivers and Honey and Buffalo creeks were downloaded from the US Geological Survey (USGS 2013). The detection limits for the elemental measurements were carried through this calculation to determine detection limits for calculated fluxes. These detection limits varied due to varying river flow rates.

Relationships between trace elements and qualitative predictor variables (e.g., depth, location, season) were explored using principal component analysis (PCA). This statistical technique expresses the variance of a many-dimensional dataset not on axes that correspond to individual variables but on new orthogonal axes called principal components (PCs) aligned with successively decreasing fractions of the variance of the dataset (e.g., Shine et al. 1995). This process allows visualizations of groupings of both variables and samples based on the concentrations of the trace elements measured in this study. In filtered samples, when some elements were too often BDL to permit useful inclusion in PCA, relationships with other elements were instead explored with multivariable logistic regression. These analyses treated trace elements with many concentrations available above detection limits as continuous predictor variables for low-concentration elements evaluated as bimodally distributed (i.e., above or below the detection limit) response variables.

Contour plots of elemental concentrations in Grand Lake were created from measurements at discrete depths and distances from the dam using Tecplot 360 (Tecplot, Inc.; Bellevue, WA). Vertical interpolation based on circulation parameters was required before allowing Tecplot to interpolate longitudinally between sampling locations. All available data were used to create contour plots.

Results

Concentrations of trace elements

The range of Fe in all filtered samples (Fe_f and similar) was BDL–220 μg/L with a median of 15 μg/L and an IQR of BDL–33 μg/L (Fig. 2). Concentrations in tributaries were not significantly different from those in the reservoir (t-test, P > 0.1) with one exception. In August, concentrations in anoxic summertime bottom water sometimes exceeded 100 μg/L, causing the reservoir water to be significantly higher in Fe_f than the tributaries (t-test, P < 0.01). Concentrations >50 μg/L also occurred in the large inflow captured in the February–March sampling, indicating that during different seasons Fe enters the reservoir in the aqueous phase from both internal loading and the watershed upstream. Downstream concentrations of Fe_f were similar to or lower than those in the reservoir. No other meaningful spatial trends were observed (Supplemental Fig. S1). Oxic waters were not devoid of Fe_f, and although concentrations were BDL in winter, they were 10–35 μg/L in February–March and May and 7–54 μg/L in November.

Figure 2. Distribution of trace element concentrations in filtered water samples. Data are depicted along several one-dimensional number lines sorted by sampling location and excursion. Circles denote samples collected upstream of Grand Lake in tributaries. Diamonds denote samples collected <8 m deep in the water column (the bottom of the summertime surface mixed layer). Triangles denote samples collected ≥8 m deep. Squares denote samples collected downstream of Pensacola Dam. Symbols appear at the median values of each subset of data, and error bars extend to the 5^th and 95^th percentiles of each subset of data. The numbers above the panel showing iron (Fe) concentrations indicate the number (n) of observations in each subset of samples, and these apply to the corresponding number lines for all panels other than Fe. Labels at the bottom of each panel identify sampling excursions.

We observed similar trends but different magnitudes between Mn_f, P_f, As_f, and Fe_f. Anoxic August bottom water contained Mn concentrations of 1–4 mg/L, and the IQR of Mn for all filtered samples was 2–57 μg/L (Fig. 2). Hypolimnetic anoxia in summer led to concentrations of Mn_f 1–3 orders of magnitude greater than in tributaries and downstream during summer. Concentrations of Mn_f in the reservoir fell between August and November to 2–35 μg/L, and downstream from the dam, releases were 93 and 22 μg/L in summer and autumn, respectively. The IQR of P_f concentrations was 22–140 μg/L (Fig. 2), significantly higher in the reservoir than in tributaries across all sampling excursions (t-test, P < 0.05), especially in summertime bottom water (t-test, P < 0.0005). Concentrations of As_f were <5 μg/L (Fig. 2), with anoxic hypolimnetic waters 4.7 μg/L and oxic water 1–2 μg/L and invariant. For each of Fe_f, Mn_f, As_f, and P_f, concentrations of samples from anoxic summertime bottom water were significantly higher than those collected from other reservoir water (t-test, P < 0.05). With the exception of variation with depth as described here, spatial variation of Mn_f, As_f, and P_f was minor (Supplemental Fig. S2–4).

Elements prevalent at the Tar Creek Superfund Site upstream behaved differently from Fe_f, Mn_f, P_f, and As_f. Anoxic water did not show an enrichment of Zn_f (Fig. 2). In Tar Creek, Zn_f easily exceeded 1000 μg/L and was >100 μg/L in the main tributaries of Grand Lake and in the upper part of the reservoir. Concentrations of Zn_f were consistently and significantly higher (t-test, P < 0.05) in the tributaries feeding the northern end of Grand Lake than in our sampling location in the reservoir >30 km downstream from those tributary sampling location (99 vs. 24 μg/L in Dec, 157 vs. 51 μg/L in Feb, 42 vs. 8 μg/L in May, and 5 vs. 0.9 μg/L in Aug). Concentrations in the rest of the reservoir were <15 μg/L, with no clear spatial pattern (Supplemental Fig. S5). The highest concentrations of Pb_f and Cd_f occurred in Tar Creek in February, 1.9 μg/L and 1.7 μg/L, respectively, but were otherwise usually BDL in filtered samples.

The IQR of Fe in all suspended sediment (Fe_ss and similar) was 7800–25,000 mg/kg, with extremes an order of magnitude lower and higher, respectively (Fig. 3). Maximal values occurred in bottom water during autumn and downstream in February–March; however, across all sampling excursions, deep water samples were not significantly higher in Fe_ss than other samples from the reservoir (t-test, P > 0.1; Fig. 3). Otherwise, concentrations of Fe_ss in the reservoir and below the dam were not significantly different from those in the tributaries (t-test, P > 0.1), which did not vary appreciably throughout the year (Supplemental Fig. S6).

Figure 3. Distribution of trace element concentrations in suspended sediment samples. Because the detection limit varied between samples, only the approximate maximum detection limit is denoted where appropriate. Other details are as in Figure 2.

The IQRs of Mn_ss and P_ss were 350–3300 mg/kg and 1100–2800 mg/kg, respectively (Fig. 3), with spatiotemporal trends closely resembling those of Fe_ss (Supplemental Fig. S7 and S8). Concentrations of As_ss were generally <50 mg/kg except for elevated values (100–8900 mg/kg) during winter (Fig. 3). Other than high values in February–March 1–2 orders of magnitude higher than those measured in other months, As_ss in tributaries did not vary across locations or time (Supplemental Fig. S9). Like Fe_ss, concentrations of Zn_ss, Pb_ss, and Cd_ss were greater in some samples collected in autumn in deep water (Fig. 3; Supplemental Fig. S10 and S11).

Flux calculations and data analysis

Loading of trace elements via tributaries to Grand Lake varied consistently with inflow volumes (Table 1). In filtered water samples, mass fluxes of all elements were 1–3 orders of magnitude greater during high flows in February–March than in December, August, or November (Table 1A). These trends were broadly similar in suspended sediment samples with one exception: for As, the February–March flux was 3–6 orders of magnitude larger than that of other months (Table 1B). Fluxes of Fe, Mn, Zn, P, and As were significantly higher than those of Pb and Cd, and fluxes of Mn were significantly higher than those of As (Kruskal–Wallis test, H < 0.01).

Table 1. Water flux (GL/day) and mass flux of elements in (Table 1A) filtered samples (kg/d) and (Table 1B) suspended sediment samples (kg/d).

Table 1A
Excursion	Water^a	Fe	Mn	Pb	Zn	Cd	As	P	DL^b
Summed loading of all tributaries
Dec	3.1	52	54	BDL	230	0.39	2.1	250	0.31
Feb–Mar	58	4400	1700	8.2	1600	BDL	66	14,000	5.8
May	16	320	290	BDL	520	BDL	62	1600	1.6
Aug	2.4	8.7	12	BDL	6.3	BDL	4.4	90	0.24
Nov	2.2	3.1	2.3	BDL	5.2	BDL	0.16	12	0.019
Export via Pensacola Dam
Dec	3.0	BDL	17	BDL	3.3	BDL	3.7	240
Feb–Mar	18	200	120	BDL	13	BDL	25	3800
May	32	1100	165	1.6	120	1.6	30	3800
Aug	11	5.4	1000	0.54	14	0.54	9.3	320
Nov	8.3	130	180	0.42	9.2	0.42	8.4	4.1
Net flux through Grand Lake^c
Dec	0.12	50	36	BDL	220	0.39	−1.6	5.4
Feb–Mar	40	4200	1600	7.3	1600	BDL	41	11,000
May	−16	−800	130	−0.81	400	−0.81	32	−2200
Aug	−8.4	3.3	−990	−0.42	−7.8	−0.42	−4.9	−230
Nov	−6.1	−130	−180	−0.41	−4.0	−0.41	−8.2	8.1
Table 1B
Excursion	Water^a	Fe	Mn	Pb	Zn	Cd	As	P	DL^b
Summed loading of all tributaries
Dec	3.1	1200	150	1.3	240	0.23	1.1	200	0.024
Feb–Mar	58	190,000	1900	770	21,000	43	16,000	11,000	1.1
May	16	8900	450	33	1500	3.2	2.3	620	0.092
Aug	2.4	560	110	2.2	47	0.20	0.25	42	0.0096
Nov	2.2	160	11	0.50	9.3	0.047	0.065	11	0.002
Export via Pensacola Dam
Dec	3.0	310	8	1.7	120	0.40	0.94	71
Feb–Mar	18	13,000	790	13	500	0.18	600	530
May	32	28,000	1000	45	720	1.0	8.8	1900
Aug	11	1600	120	1.6	30	BDL	0.62	120
Nov	8.3	1500	540	1.9	25	0.13	1.5	140
Net flux through Grand Lake^c
Dec	0.12	860	140	−0.37	120	−0.16	0.12	130
Feb–Mar	40	180,000	1100	753	20,000	43	15,000	10,000
May	−16	−19,000	−590	−12	790	2.1	−6.4	−1200
Aug	−8.4	−1100	−6.7	0.57	17	0.17	−0.37	−77
Nov	−6.1	−1300	−530	−1.4	−16	−0.082	−1.4	−130

^aNet water flow through Grand Lake taken from Wildman (2016).

^bDetection limit for Mn, Pb, Zn, Cd, and As; for Fe and P, DL is 10 times higher than the value shown; DL applies to both loading and export values.

^cLoading via tributaries minus export out the dam was calculated before rounding to the appropriate number of significant figures.

The aqueous and suspended sediment phases each dominated transport of trace elements through the reservoir during different seasons and for different elements (Table 1). Large positive fluxes (i.e., net loadings) of most elements measured in this study were observed in both phases in February–March, when loading via tributaries exceeded loss from the reservoir via dam releases. Net loadings to the reservoir in February were 1–2 orders of magnitude lower for Fe_f, Pb_f, Zn_f, Cd_f, and As_f than for these elements in suspended sediment. At other times, flux of aqueous elements was of the same order of magnitude as the flux of elements associated with suspended sediment, and sometimes it was less than an order of magnitude larger. In May, when the dam was releasing water from the downstream end of Grand Lake to accommodate large inflows at the upstream end, a net export of Fe_f, Fe_ss, Mn_ss, P_f, and P_ss was observed; however, Zn_f, Zn_ss, and Mn_f were loaded to the reservoir in May. In August and November, most elements were exported from the reservoir, although fluxes were smaller than in months of high flow because water flux was small. In August, flux calculations indicated a comparatively small loading of Fe_f to the reservoir and a much larger export of Fe_ss.

Variance in the PCA performed on filtered samples (PCA_f) was broadly distributed, with just 38% described by the first PC (PC1_f and similar), 21% by PC2_f, and 16% by each of PC3_f and PC4_f. A biplot showing the relationship of variables and samples to these axes (termed “loadings” and “scores,” respectively; see Shine et al. 1995) is characterized by high values on PC1_f for all elements included in this analysis and a range of values on PC2_f with Fe_f and As_f together and opposite from Zn_f, and also with Mn_f and P_f near each other (Fig. 4). This finding indicates general covariance of Fe_f and As_f across all the samples of the dataset, covariance of Mn_f and P_f across all samples in the dataset, and noncovariance of Zn_f with Fe_f and As_f. Because of numerous concentrations BDL, Cd_f and Pb_f were excluded from the PCA_f.

Figure 4. Biplot of principal component analysis of elemental concentrations in filtered samples. Loadings are depicted as open double circles and correspond to upper and right-hand axes. Scores are other symbols and correspond to lower and left-hand axes. Scores are represented as: (filled diamonds) upstream samples from February-March and May (i.e., months of high flow); (open diamonds) other upstream samples; (filled squares) summertime surface mixed layer; (open squares) other lake samples from <8 m deep; (filled triangles) summertime metalimnion and bottom water; (open triangles) other lake samples from ≥8 m deep; and (filled circles) downstream samples. PC1 and PC2 described 38% and 21% of the variance in the dataset, respectively.

The plot of PC_f scores showed that most samples collected in river inflows during months of high flow plotted with high values on PC1_f and PC2_f, whereas many samples collected from depth in summer plotted with high values on PC1_f and comparatively low values on PC2_f (Fig. 4). These 2 groupings, the most pronounced of the dataset, corroborate other observations in this study by showing that, across 5 variables, the highest concentrations in this dataset occurred during large inflows or anoxic bottom water. Comparing these PC scores to the PC loadings of the variables indicates that the river inflows were comparatively low in Fe_f and As_f and high in Zn_f, whereas the opposite was true for anoxic bottom water.

Multivariate logistic regression identified the logarithm of concentrations of Zn_f as the only significant predictor for Cd_f (Fig. 5); the relationship was robust, with a McFadden's pseudo R² of 0.69. Conversely, Pb_f was predicted by the logarithms of concentrations of Fe_f and Zn_f in a multivariate logistic regression that resulted in a McFadden's pseudo R² of only 0.39.

Figure 5. Plot of logistic function relating occurrence of detectable Cd_f to logarithm of Zn_f concentration by showing observed values (open diamonds, left vertical axis), the percentage of Cd observations that exceeded the detection limit for a given 0.5-log-unit window on the horizontal axis (gray circles, right vertical axis), and the logistic function fitting the observed data (dotted line, left vertical axis).

In suspended sediment samples, PC1–3 explained 42, 25, and 15% of the variance of the dataset, respectively. No clear trends could be discerned from the biplots resulting from this PCA (data not shown).

Discussion

The concentrations of Fe_ss, Mn_ss, Zn_ss, Pb_ss, and Cd_ss measured in this study were generally comparable to previous measurements of streambed sediment from the Neosho River, the Spring River, and Tar Creek (Andrews et al. 2009). However, concentrations of Fe_ss in all downstream samples and in bottom water samples in November exceeded the median of 12,000 mg/kg measured by Andrews et al. (2009) in tributaries. One sample in particular, 1 of 2 collected downstream of the dam in February–March, exceeded 250,000 mg/kg, a concentration suggesting that some particles exported downstream during elevated dam releases can be substantially enriched in Fe_ss. A previous study reported fluvial sediment samples collected from many river and stream sites in the United States (Horowitz and Stevens 2008). Median values of samples from rivers that drained either >50% agricultural land or >50% rangeland (Mn_ss, P_ss, and As_ss near 800, 100, and 7 mg/kg, respectively) fell between the 25^th and 50^th percentiles of concentrations of samples measured in our study. In our study, both Zn_ss and Cd_ss were consistently 10–100 times higher and Pb_ss was ∼10 times higher than median values of 90, 0.4, and 20 mg/kg, respectively, found by Horowitz and Stevens (2008). These comparisons indicate that the suspended sediment of Grand Lake is enriched in metals derived from upstream mining and can also be enriched in other elements under certain limnologic conditions, described later. The elevated concentrations of As_ss across Grand Lake in February–March matched high concentrations in tributaries during that time and occurred after the water column had been well-mixed and oxygenated for several weeks. Therefore, they were probably caused by variations in watershed hydrology (perhaps related to storm runoff and high river flows), not lacustrine biogeochemistry.

Compared to previous measurements of water samples collected in the Neosho River, the Spring River, and Tar Creek (Andrews et al. 2009), Fe_f, Mn_f, Zn_f, Pb_f, and Cd_f data from this study fall in the same ranges with 2 exceptions. First, redox-active metals (i.e., Fe_f and Mn_f) were higher in floodwater and in the anoxic bottom water sampled in this study (median Fe_f measured by Andrews et al. [2009] in the Neosho and Spring Rivers was <20 μg/L; median Mn ranged between 10 and 80 μg/L). Second, Andrews et al. (2009) measured much higher concentrations of Zn_f, Pb_f, and Cd_f in Tar Creek (1000–10,000, 0.1–1, and 3–90 μg/L, respectively) than we measured in Grand Lake. We are aware of no previous measurements of these elements in the Elk River or Honey Creek or of profiles of aqueous metals in Grand Lake.

In this study, flux of Fe occurred mostly in particulate form, whereas in most seasons >50% of the flux of P and As occurred in filtered samples. In February and August, the flux of Mn_f was greater than that of Mn_ss, and the reverse was true otherwise. Flux of Zn species were similarly inconsistent across sampling excursions. Although the majority of mass flux can frequently be attributed to the aqueous phase in some elements in this study, this does not imply that the solid phase does not dominate the yearly mass balance of all elements, as has been observed in some major rivers (Horowitz et al. 2001), because at least 12 hydrologically spaced measurements are needed to accurately determine a yearly flux (Horowitz 2008). Despite this limitation, our flux calculations provide insight into the effect of hydrology on the transport of elements through Grand Lake because they describe a range of inflow rates that can be taken as representative of a range of hydrologic events.

The rising limb of a storm inflow (observed in February–March) loaded trace elements into Grand Lake, which can be explained by the imbalance of inflows rich in elements and outflows of lower concentration from a long reservoir that is not longitudinally mixed. During the falling limb of a storm inflow (observed in May), several elements in both the aqueous phase and suspended sediment were exported out of the reservoir along with the excess water retained during the flood. Both of these periods of high inflow were also characterized by little difference between the trace element chemistry of the reservoir and its tributaries, which we attribute to low water residence time, low primary production, weak or nonexistent stratification, and an oxic water column that made biogeochemical modification of trace elements unlikely. These observations resemble those from another well-flushed reservoir basin in which water chemistry matched that of the inflowing river (Bellanger et al. 2004). Our data from Grand Lake suggest that, during stormy seasons, managers of Grand Lake can expect minimal modification of floodwater by the reservoir and spatial distributions of trace elements determined mostly by the passage of floodwater through the system. During this study, the volume of inflows during winter and spring was >230% of the volume of the reservoir, and each of 2 large storms may have flushed the thalweg completely in as little as 1 week (Wildman 2016).

When elevated inflows from storms subsided, summer stratification led to hypolimnetic anoxia (Wildman 2016). Subsequent elevated concentrations of Fe_f and Mn_f in bottom water samples suggest reductive dissolution of metal oxides in sediment and settling particles. Elevated concentrations of P_f and As_f suggest desorption of these elements from metal oxides. Grand Lake exported Fe, Mn, and P from its anoxic summertime hypolimnion, with Mn and Fe measured downstream predominantly in filtered and particulate samples, respectively. However, Fe may be leaving the dam in the aqueous phase and precipitating immediately upon entering oxic tailwater due to the rapid oxidation kinetics of Fe²⁺ (Morgan 2005). This event likely explains the low concentrations of Fe_f below the dam that led to an apparent net loading of Fe_f to the reservoir in August. The export of Fe, Mn, P, and As in August, despite the low volume of water exchanged (Wildman 2016), indicates that reductive dissolution of particles in anoxic bottom water represents a source of dissolved trace elements to the river below the dam. The shift of control of trace element cycling from inflow hydrology to lacustrine biogeochemistry when inflow rates were low is consistent with research elsewhere (Bellanger et al. 2004).

Much lower concentrations of Fe_f, Mn_f, P_f, and As_f in the water column shortly after autumn overturn suggest oxidative precipitation of Fe and Mn. The concentrations of Fe_f and Mn_f in November that were higher than those of the preceding December suggest that, in this natural setting, oxidation kinetics might have been slower than in laboratory settings (e.g., Morgan 2005) and that oxidative precipitation may not have been complete at the time of our sampling. This interpretation is supported by the low flow through the reservoir during and between our summer and autumn sampling excursions, indicating that water exchange was too low to explain the decrease in the concentrations of trace elements in filtered samples (Wildman 2016). Scavenging of P_f and As_f onto newly precipitated particles within the reservoir may explain the decrease in these elements in filtered samples from August to November. These interpretations are consistent with our observation of some elevated concentrations of Fe_ss, Mn_ss, P_ss, and As_ss in bottom waters during this season. During a vertically mixed period of low flow (observed in our December excursion), the reservoir closely resembled its tributaries for all elements in the aqueous phase and in suspended sediment, except it passed P downstream in both phases.

The broad distribution of variance in both PCAs suggests that elements in this study respond to a range of relatively balanced influences. Although these PCAs alone do not indicate clear trends, the loadings of variables on PC2_f combine with spatial distributions and mass flux calculations to indicate that, although As_f and Fe_f covary in this system, Fe_f has minimal influence over Zn_f. This observation resembles those made in the anoxic hypolimnion of a eutrophic lake where many previous studies have occurred (Achterberg et al. 1997). In Grand Lake, Zn_f decreased as water entered the reservoir in all seasons and did not respond to biogeochemical depletion of DO. Verification of this sequestration through analysis of sediment samples was beyond the scope of this study, but sediment cores collected by Andrews et al. (2009) from the upper region of Grand Lake contained Zn concentrations an order of magnitude higher than in the Neosho River upstream of its confluence with Tar Creek.

Previous research at the Tar Creek Superfund Site can provide insight into a possible mechanism of Zn_f sequestration in the upper region of Grand Lake. In Tar Creek, some Zn(aq) precipitates as Zn-carbonate or sorbs to Fe-oxides (Bostick et al. 2001, Schaider et al. 2014) while the remaining Zn(aq) serves as the probable source of the high concentrations of Zn_f observed in this study. The lack of association between Zn_f and Fe_f in Grand Lake implies that sorption of excess Zn_f to Fe-oxide particles from the Neosho and Spring rivers is an unlikely mechanism for Zn_f sequestration. However, attenuation of Zn_f by precipitation as carbonate minerals is plausible because the Spring River is likely rich in carbonate (suggested by abundant limestone in its watershed and elevated pH relative to the Neosho River; Andrews et al. 2009) and because precipitation of Zn-carbonate particles is thermodynamically likely in Tar Creek waters (Schaider et al. 2014). Further research would be required to investigate this potential mechanism of Zn_f sequestration but would be relevant for reservoir management by demonstrating the transferability of observations at Grand Lake to other reservoirs downstream of mining areas.

Sequestration of Zn by Grand Lake implies that reservoir managers in the Grand River Basin can rely on Grand Lake to limit the downstream transport of mine waste across a range of hydrologic conditions. The logistic regression model showing that Zn_f predicts the presence of detectable Cd_f indicates that Grand Lake probably also sequesters Cd_f. This prediction is particularly advantageous for reservoir managers because it implies that broad trends of Cd_f can be understood indirectly by measuring Zn_f, which requires less sensitive instrumentation to characterize in this system because of its higher concentrations. The behavior of Pb_f is less clear, in part because the best model that predicted Pb_f used both Fe_f and Zn_f as predictors and because that model was not robust.

Future work pertaining to trace element flux through Grand Lake might focus on the spatial distribution of elements in the reservoir sediment. Between 55 and 65 km from the dam (i.e., the vicinity of the confluence of the Elk River arm with the thalweg), the summertime surface mixed layer reached the reservoir bed and most sedimentation occurred (Wildman 2016), indicating that this region is probably the transition zone of the reservoir (Wetzel 2001). Because fine sediment contains elevated concentrations of trace elements (Horowitz and Elrick 1987, Wildman et al. 2011), future work could test the hypothesis that the sediment of the transition zone is enriched in trace elements. During floods, however, sedimentation can occur at Pensacola dam (Wildman 2016), and, consequently, future research could also explore the hypothesis that much of the thalweg sediment is enriched in trace elements relative to background.

Conclusions

During our study period, the flux of Fe, Mn, P, and As through Grand Lake was controlled by hydrology when inflows were low and high (i.e., winter and spring, respectively) and the water column was vertically mixed. During these times, spatial distribution of trace elements in the reservoir was without meaningful trends, the reservoir chemically resembled its tributaries, and it did little to alter the transport of aqueous trace elements downstream. These findings support the second and fourth hypotheses of this study. When flows were minimal and the water column was stratified, elements that entered the aqueous phase in the anoxic hypolimnion were exported from the reservoir, which supports the third hypothesis of this study. Additionally, these trends were not observed for Zn, which was loaded to the reservoir in all seasons. Concentrations of Pb and Cd were consistently low, but Cd_f concentrations above the detection limit were predicted by Zn_f concentrations. This finding implies that the frequency of large inflows and the volume of summertime hypolimnetic water released from Pensacola Dam will be primary drivers of the yearly flux of several metals through Grand Lake, and that Grand Lake sequesters Zn, preventing its transport downstream. The presence of Grand Lake in the watershed of the Grand River does not always imply that all elements associated with suspended sediment are trapped in this reservoir, refuting the first hypothesis of this study. Instead, some elements are passed downstream under some conditions, indicating that Grand Lake is not large enough for all particles entering from upstream to settle.

Acknowledgments

This study was enabled by the generous logistical, laboratory, and personnel support of the Grand River Dam Authority, specifically Darrell Townsend, Sam Ziara, Jacklyn Jaggars, Scott Cox, and Sean Allred. Steve Nikolai, Andy Dzialowski (Oklahoma State University), and Lance Phillips (Oklahoma Water Resources Board) shared important ancillary data. Mollie Thurman and Emily Estes (Harvard) provided essential sampling support, and Nick Lupoli, Chitra Amarasiriwardena, and Zhao Dong (Harvard) provided analytical support. Rebecca Jim, Earl Hatley (L.E.A.D. Agency), Laurel Schaider (Harvard), and Marjorie Wonham (Quest) provided key background information and helpful discussion. Jim Shine (Harvard) provided essential support in all aspects of this project. The final manuscript benefitted from 3 anonymous reviews and the work of Associate Editor R. Thomas James.

Funding

This project was funded by a French Environmental Fellowship granted to Wildman through the Harvard University Center for the Environment and by a gift from the Akatsuka Orchid Company, Ltd.