Tributary chloride loading into Lake Michigan

Abstract

Anthropogenic salt sources have contributed to rising salinities in the Laurentian Great Lakes. In Lake Michigan, chloride concentrations have risen from ~ 1–2 mg L⁻¹ in the 1800s to > 15 mg L⁻¹ in 2020. The watersheds of the approximately 300 tributaries of Lake Michigan vary in size and represent a wide range of land use, from undeveloped forested watersheds to urbanized and agricultural areas. The spatial variability in both size and land cover among Lake Michigan's tributaries contributes to enormous variation in chloride concentrations and loads. We performed a spatial assessment of Lake Michigan tributaries to calculate total annual salt loading, infer future conditions based on current patterns, evaluate the use of synoptic sampling, and identify watershed characteristics that drive high chloride concentrations. We found that the tributary load to Lake Michigan is 1.08 Tg yr⁻¹ of chloride, and that chloride concentrations in Lake Michigan will likely continue to slowly rise in the coming decades.

Scientific Significance Statement

The Laurentian Great Lakes hold 20% of the world's surface freshwater (Sterner 2021), which supports a regional economy with an estimated GDP of > $4 trillion (Campbell et al. 2015). Many of the manufacturing processes and millions of residents in the watershed rely on the freshwater from the Great Lakes, yet due to their vast volume of water, the threat of salinization is often overlooked.

Road salt and other anthropogenic salt sources such as industrial and water softener effluents have contributed to rising salinities in the Great Lakes. Chloride concentrations (an anion common to most salts) in the Great Lakes have risen over the last two centuries, from estimated concentrations ~ 1–2 mg L⁻¹ in the 1800s to present day concentrations > 10 mg L⁻¹ in Lake Michigan, Erie, and Ontario (Chapra et al. 2009). Lake Erie and Lake Ontario had large declines in chloride concentrations beginning in the 1980s following the introduction of the U.S. Clean Water Act (Chapra et al. 2009), whereas Lake Michigan steadily increased in chloride during this time.

For Lake Michigan, with a water volume of 4.918 trillion m³, it would take ~ 5 Tg (1 Tg = 1 million metric tons) of salt to raise the salinity by 1 mg L⁻¹. Based on the estimated annual loading of road salt in the Lake Michigan basin, this could be accomplished in 2–3 yr (Fig. 1a; Bock et al. 2018). From 1980 to 2020, chloride concentrations in Lake Michigan increased from ~ 9 to ~ 15 mg L⁻¹ (Fig. 1b). This annual increase of 0.125 mg L⁻¹ equates to ~ 0.625 Tg of additional chloride added to Lake Michigan per year. Considering chloride is 60% of the mass of salt (NaCl), this is equivalent to 1.05 Tg of salt—still less than the estimated annual road salt use in the basin (Fig. 1a). This discrepancy implies a portion of the salt used in the catchment may be stored, at least temporarily, on the terrestrial landscape and in groundwater (Dugan and Rock 2021; Green et al. 2021).

Fig 1

(a) Estimated road salt use in the Lake Michigan basin (derived from Bock et al. 2018). (b) Chloride concentrations in Lake Michigan collected by multiple agencies (U.S. Environmental Protection Agency [EPA] Great Lakes National Program Office, Illinois EPA, Wisconsin Department of Natural Resources, Little River Band of Ottawa Indians) across multiple sites. Data are from the U.S. Water Quality Portal (Dugan and Mooney 2021). Prior to plotting, the time series were detrended, and any residuals > 5 standard deviations from the mean were removed as outliers (0.7% of data). Transparency of points is to aid in visualization.

High salt loading to freshwater environments in the Great Lakes region and northeastern United States have been tied to urban land use and human-dominated landscapes (Dugan et al. 2020; Moore et al. 2020; Stets et al. 2020). Tributaries of the Great Lakes are primary sources of terrestrial-derived pollutants and they serve as a direct connection between the land and nearshore zones (Robertson and Saad 2011; Makarewicz et al. 2012). The watersheds of the approximately 300 tributaries of Lake Michigan vary in size from 2 km² to over 15,000 km² and represent a wide range of land cover and land use—from undeveloped watersheds dominated by wetlands to highly urbanized and farmed areas. The spatial variability in both size and land cover among Lake Michigan's tributaries can lead to enormous variation in pollutant concentrations, loads, and areal yields (Mooney et al. 2020). In the 1970s, the annual chloride load to Lake Michigan was estimated as 0.9 Tg, broken down into tributary loading at 0.598 Tg, direct loading at 0.220 Tg, and atmospheric loading at 0.083 Tg (Sonzogni et al. 1978, 1983). As the human population in the Great Lakes basin continues to rise, the risk of salinization persists. Here, we perform a spatial assessment of salt loading from tributaries into Lake Michigan to calculate total annual salt loading, infer future conditions based on current patterns, evaluate the use of synoptic sampling, and identify watershed characteristics that drive high salt concentrations.

Methods

Objective 1: Calculate tributary chloride load into Lake Michigan

Synoptic sampling is a highly efficient sampling strategy for assessing water quality across large geographic distances, especially in areas with no long-term monitoring (Mantzouki et al. 2018). In 2018, we sampled 235 of Lake Michigan's ~ 300 tributaries between 10 and 15 July (Mooney et al. 2021). All sampling took place under low- to moderate-flow conditions when streams were near or below their 30^th percentile of discharge for 2018. Surface water from each tributary was collected at the road crossing nearest to the mouth via bridge sampling methods (U.S. Environmental Protection Agency [EPA] 2013). Collected water samples were immediately filtered through a 0.45 μm glass fiber filter and stored on ice until they were frozen within 10 h of collection. Water samples remained frozen until chloride analysis was performed.

Water samples were analyzed via ion chromatography using a Thermo Scientific Dionex ICS-2100 equipped with an IonPac AS11 analytical column and an AG11-HC column guard. The ion chromatograph was calibrated with six NaCl standards ranging from 1 to 500 mg L⁻¹ and run with a 30 mM NaOH eluent solution. Twenty-five randomly selected sites were run in duplicate. The sample from Susan Creek, MI, appeared anomalous, with chloride concentrations of 611 and 645 mg L⁻¹ (run in duplicate). This sample was removed from all analyses, leaving 234 tributaries.

To estimate chloride load, concentrations needed to be paired with discharge estimates. Since only a small percentage of Lake Michigan tributaries are gaged, discharge was estimated for each tributary by the discharge-area ratio method as a 5-d mean around the sampling date as well as annual mean and median flow (see Supporting Information and Mooney et al. 2020). Based on the analysis of high-frequency specific conductance data from two Lake Michigan tributaries, the likelihood of a grab sample being representative of the 5-d mean concentration is > 80% (see Supporting Information). To scale our synoptic sampling to an annual load, we multiplied concentration by the annual median discharge. Median discharge was chosen over mean discharge as it is less representative of high flows, which tend to dilute concentrations. This snapshot approach of scaling to the annual load does not account for seasonal fluctuations in chloride concentrations nor complexities arising from dilution during storm events (see the Objective 3 section).

The Great Lakes Aquatic Habitat Framework (GLAHF) database (Wang et al. 2015; Forsyth et al. 2016) documents 293 noninterfluve catchments in the Lake Mighigan basin. Our sampled watersheds (n = 234) accounts for 98.97% of the total basin area, leaving 1.03% of the basin not accounted for. Watershed area (km²) and road densities (km road km⁻²) for individual tributary watersheds were calculated using the GLAHF database. Mean watershed imperviousness was determined using the 2011 National Land Cover Database (NLCD).

Objective 2: Infer future conditions

We used a simple mass balance model from Sonzogni et al. (1983) to interpret our load calculations vs. historical data and long-term projections:

(1)

where the change in chloride concentration dc/dt in the lake can be solved analytically as:

(2)

where C_[t] is the chloride concentration (mg L⁻¹ or g m³) at time t, W is the total chloride load (g), Q is the discharge out of the lake (m³ yr⁻¹), and V is the lake volume (m³). Residence time is equal to V/Q.

Objective 3: Evaluate the use of synoptic sampling

We compared our 2018 observations with publicly available historical data (Dugan and Mooney 2021). Chloride data from Lake Michigan and Lake Michigan tributaries were downloaded from the U.S. Water Quality Portal (WQP; Read et al. 2017) using the dataRetrieval R package (De Cicco et al. 2018; R Core Team 2020). Data were not restricted by sampling date, and sites were spatially matched to tributary watersheds and Lake Michigan using GLAHF shapefiles with the st_intersects function in the sf R package (Pebesma 2018).

Objective 4: Identify watershed characteristics that drive high salt concentrations

A quantile regression forest (QRF) was used to model the relationship between observed chloride concentrations in the 234 tributaries and watershed characteristics (land use percentages, population and road density, watershed area, and stream order). QRF methods are identical to Dugan et al. (2020) following Meinshausen (2006). A QRF method was chosen to accommodate correlated predictor variables and nonlinear responses, and was implemented using the Ranger R package (Wright and Ziegler 2017).

Results

Chloride concentrations recorded across the 234 Lake Michigan tributaries ranged from < 1 to 265 mg L⁻¹ in the summer of 2018. Across all of the tributaries, the mean chloride concentration weighted by watershed area was 25.4 and 27.2 mg L⁻¹ when weighted by median discharge. The highest concentrations were generally found along the southern Wisconsin and Indiana shorelines (Fig. 2a).

Fig 2

(a) Chloride concentrations in 234 Lake Michigan tributaries collected between 10 and 15 July 2018. (b) Chloride concentrations from stream sites in the Lake Michigan basin collected across multiple years by multiple agencies. Data are from the U.S. Water Quality Portal (www.waterqualitydata.us).

Our estimate of 2018 median discharge from tributaries into Lake Michigan is approximately 1256 m³ s⁻¹ (Mooney et al. 2020), which closely matches the 2000–2019 average of 1245 m³ s⁻¹ from the NOAA-GLERL monthly hydrometeorological database (Hunter et al. 2015). Applying an area-weighted chloride concentration of 25.4 mg L⁻¹ would equate to a tributary load of 1.079 Tg yr⁻¹. Five tributaries, the Grand, St. Joseph, Fox, Kalamazoo, and Milwaukee Rivers, account for 71% of the total load. This is roughly twice the 0.598 Tg estimate by Sonzogni et al. (1983). Given the size of Lake Michigan, rainfall is also a large contributor to the hydrologic budget. The annual atmospheric chloride load, based on a wet deposition concentration of 0.5 mg L⁻¹ (NADP 2018) and an annual rainfall of 0.917 m onto the surface of Lake Michigan (Hunter et al. 2015), is an additional 0.025 Tg of chloride. Based on a steady state model of loading into Lake Michigan (no change in annual load or hydrology), a tributary load of 1.10 Tg would eventually result in an in-lake concentration of 24.3 mg L⁻¹ (Fig. S4). This trajectory is slightly lower than historical chloride concentrations suggest (Chapra et al. 2009), which might indicate our tributary load underestimates the total load into Lake Michigan and the total load is closer to ~ 1.2 Tg yr⁻¹ (Fig. S4).

We compared our synoptic sampling to 51,126 WQP chloride observations from rivers and streams in the Lake Michigan basin. Dates of observations ranged from 1952 to 2021, with 90% of observations collected after 1990. The highest concentrations, ranging from 100s to 44,000 mg L⁻¹, are associated with an outflow of the Milwaukee General Mitchell International Airport (Corsi et al. 2001). Other sites with high chloride are seen in the watersheds of the Fox, Milwaukee, Root, and St. Joseph rivers (Fig. 2b). Across the entire WQP dataset (not spatially weighted), the mean chloride concentration was 111 mg L⁻¹ and the median was 48 mg L⁻¹. In the winter months (January–March), mean and median chloride increased to 231 and 56 mg L⁻¹, respectively. Tributary outflow chloride concentrations collected in 2018 were within the same range as observations from individual watersheds (Fig. 3a). When WQP data were subset to locations within 1 km of the tributary outflow, median concentrations were highly correlated to our 2018 observations (r² = 0.91; Fig. 3b).

Fig 3

(a) Comparison of 2018 tributary observations (yellow points with brown outline) with boxplots of all long-term observational data from 35 watersheds with > 20 sampling events. Dark brown boxes represent the 25^th to 75^th percentile, with whiskers extending to the 5^th and 95^th percentile. Wheat colored vertical lines extend the whiskers to the 1 and 99^th percentiles. Narrow gray vertical lines represent all watersheds in the WQP data. Yellow points without an outline show remaining watersheds sampled in 2018. (b) Median chloride concentrations from sampling locations < 1 km from tributary sampling locations (# watersheds = 43) vs. 2018 tributary observations. Data points are colored by % urban land use in the watershed.

Eight of the largest watersheds all show relatively little seasonality in chloride concentrations across sampling locations in the watersheds (Fig. 4), with the lowest concentrations in April and the highest concentrations in the fall (Fig. 4). In these watersheds, summer concentrations are fairly representative of annual concentrations, which supports our use of synoptic sampling. In smaller urban watersheds, chloride concentrations peak in the winter months (Fig. 5). This seasonal cycle of high winter chloride driven by road salt runoff can be seen in the Milwaukee River, Oak Creek, and the Calumet River (Fig. 5).

Fig 4

Monthly boxplots of chloride concentrations across sampling locations in watersheds of the Lake Michigan basin that are less than a given distance (5–30 km) from the tributary outlet. Watersheds are ordered in descending order by area. Boxes represent the 25^th to 75^th percentiles, with whiskers extending to the 5^th and 95^th percentiles. Numbers of observations in each boxplot are presented in the center panels. All data are from the U.S. Water Quality Portal (www.waterqualitydata.us). Watershed (dark green) and 2018 sampling location (point) are shown on right panel. Point colors correspond to the data shown in Fig. 2a.

Fig 5

Monthly boxplots of chloride concentrations across sampling locations in urban watersheds of the Lake Michigan basin. Boxes represent the 25^th to 75^th percentiles, with whiskers extending to the 5^th and 95^th percentiles. Numbers of observations in each boxplot are presented in the rightmost panels. All data are from the U.S. Water Quality Portal (www.waterqualitydata.us). Watershed (dark green) and 2018 sampling location (point) are shown on right panel. Point colors correspond to the data shown in Fig. 2a.

The QRF revealed that the top predictor of chloride concentration was percent of impervious surface in the watershed (Fig. S3), but different forms of anthropogenic development contribute to higher chloride concentrations. Feature contribution plots reveal that while predictors such as mean imperviousness and percent urban land use scale linearly with chloride concentration (Fig. S3c,e), percent agriculture has a positive influence on chloride concentration only when agricultural land use is > ~ 60% (Fig. S3g).

Discussion

The total chloride load from the 234 Lake Michigan tributaries sampled in the summer of 2018 was 1.079 Tg yr⁻¹. Assuming the watershed is at steady state with respect to tributary and atmospheric chloride loading, the chloride concentration of Lake Michigan is expected to rise from current concentrations of 15 to 24 mg L⁻¹ over the next two centuries (Fig. S4).

Our use of synoptic sampling to assess the current status of Lake Michigan tributaries is supported by historical chloride observations, which show that large watersheds with < 20% urban land use have small seasonal shifts in chloride concentrations (Fig. 4). Five watersheds, the Grand, St. Joseph, Fox, Kalamazoo, and Milwaukee, account for 71% of the total chloride load. In these watersheds, the lowest concentrations are typically recorded in April, which aligns with spring runoff. This observation supports previous research documenting dilution regimes during high spring discharge due to chloride limitation (Kelly et al. 2019; Rock 2021). The chloride peak in September likely reflects a time lag between groundwater recharge dynamics and base flow contributions (Perera et al. 2013). Shallow groundwater throughout the Great Lakes basin has been shown to be contaminated with anthropogenic chloride (Green et al. 2021).

Our analysis has four limitations that may lead to underestimation of chloride loads. First, we did not sample all tributary inflows due to logistical constraints, and left roughly 1% of noninterfluve basin area unaccounted for (see the Methods section). Second, in the watersheds with > 20% urban land use (Fig. 5), chloride concentrations peak in February. This is consistent with road salt runoff with periods of low discharge (Corsi et al. 2015), and summer synoptic sampling is likely underestimating annual loads from these small urban tributaries. The largest of these highly urbanized tributaries is the Milwaukee River, which accounts for approximately 8% of the total tributary load (5^th largest contributor overall). If our measurement of 72 mg L⁻¹ is not representative of annual conditions, and the true annual mean of the Milwaukee River is 120 mg L⁻¹, this would raise the total tributary flow-weighted concentration from 27.2 to 28.7 mg L⁻¹, and the total tributary load to 1.138 Tg. Third, concentrations of chloride may have changed over time. From the available time series of chloride concentrations, this does not seem to be the case in the last 10–20 yr (Figs. S1, S2). Lastly, we did not account for direct overland or groundwater discharges, nor direct inputs to the lakes (storm sewers, wastewater outflows). In Lake Michigan, direct groundwater inputs are estimated to be only ~ 2–2.5% (Kornelsen and Coulibaly 2014) of total inflows. However, in urban areas, or areas that have an industrial legacy of brine pollution, such as Manistee, MI, concentrations in these discharges could be quite high (Cherkauer et al. 1992; Rediske et al. 2001). Sonzogni et al. (1983) calculated direct loads to Lake Michigan of 0.220 Tg in the 1970s, with half attributed to direct municipal effluent and half to industrial loads. It is unknown how these have changed, although it is likely that industrial loads have decreased. For instance, the 1975 load for the Manistee River was 0.16 Tg (95% of which was attributed to point sources; Sonzogni et al. 1978). Most of the plants in Manistee (“The Salt City of the Inland Seas”) have closed, and our calculated load for the Manistee River in 2018 was 0.04 Tg. Chloride concentrations around municipal outflows, such as the Milwaukee Metropolitan Sewerage District outflows, show periodic spikes in chloride, and reinforce the importance in monitoring point source discharges.

Even from just one snapshot sampling of Lake Michigan tributaries we found that impervious surface was a strong predictor of chloride concentration (Fig. S3). Of the 15 largest watersheds in the Lake Michigan basin, only 4, the Menominee (which delineates the border of Wisconsin and the Upper Peninsula of Michigan), Manistique, Peshtigo, and Oconto Rivers, have median chloride concentrations ≤ 10 mg L⁻¹. These watersheds are all ≤ 5% urban. A strong correlation between surface water chloride concentrations and road density, road proximity, impervious surface has been noted through the U.S. Midwest and Northeast (Kelting et al. 2012; Dugan et al. 2017; Moore et al. 2020), and is often considered a proxy for road-salt application, which is the predominant source of anthropogenic salt at a large scale (Overbo et al. 2021). We also see an indication that agriculture contributes to chloride concentration, which likely reflects the use of potassium chloride as a synthetic fertilizer (USDA 2019).

Our findings highlight the complexity in managing the Lake Michigan basin for chloride pollution. While large watersheds are the largest contributor of chloride to the lake, their concentrations are far below EPA toxicity thresholds (230 mg L⁻¹) and their riverine habitat is more protected than smaller urban watersheds where winter concentrations routinely surpass 230 mg L⁻¹. This creates a conflict between management priorities of protecting lake vs. tributary habitat. In addition, success of management actions in watersheds with high contribution of chloride-laden groundwater may not immediately be apparent due to decadal time lags between groundwater fluxes responding to surface changes (Green et al. 2021). Our chloride inventory of Lake Michigan tributaries suggests that the chloride concentration in the lake is in disequilibrium with chloride loading from its basin, and will continue to slowly rise in the coming decades. The rise from ~ 15 to ~ 24 mg L⁻¹ may be ecologically significant, as research has shown negative impacts to biotic communities well below water quality guidelines (Arnott et al. 2020). Lake Ontario currently has a chloride concentration ~ 23–24 mg L⁻¹, and could be used as a reference for future conditions in Lake Michigan.

Acknowledgments

Funding for this work was provided by a Department of Interior Northeast Climate Adaptation Science Center graduate fellowship to RJM. The project described in this publication was supported by Grant or Cooperative Agreement No. G12AC00001 from the United States Geological Survey. Its contents are solely the responsibility of the authors and do not necessarily represent the views of the Northeast Climate Adaptation Science Center or the USGS. This manuscript is submitted for publication with the understanding that the United States Government is authorized to reproduce and distribute reprints for Governmental purposes. We thank Will Rosenthal for assistance collecting tributary water samples, and the reviewers and editors who provided thoughtful feedback throughout the publication process.