Article pubs.acs.org/est
Influence of Modern Stormwater Management Practices on Transport of Road Salt to Surface Waters Joel W. Snodgrass,*,†,‡ Joel Moore,† Steven M. Lev,† Ryan E. Casey,† David R. Ownby,† Robert F. Flora,† and Grant Izzo† †
Urban Environmental Biogeochemistry Laboratory, Towson University, Towson, Maryland 21252, United States Department of Fish and Wildlife Conservation, Virginia Tech, Blacksburg, Virginia 24061, United States
‡
S Supporting Information *
ABSTRACT: Application of road salts in regions with colder climates is leading to ground and surface water contamination. However, we know little about how modern stormwater management practices affect the movement of road salt through urban watersheds. We investigated groundwater contamination and transport of road salts at two stormwater ponds in Baltimore County, Maryland. In association with the ponds, we documented a plume of contaminated groundwater that resulted in Cl− loadings to the adjacent stream of 6574 to 40 008 kg Cl− per winter, depending on winter snowfall. We also monitored Na+ and Cl− ion concentrations and the temporal dynamics of conductivity at a range of stream sites in watersheds with and without stormwater management ponds. Streams draining watersheds with stormwater ponds had consistently higher conductivities and Cl− concentrations during base flow conditions and often exhibited greater peaks in Cl− and conductivity associated with winter storms and subsequent melting events, despite the degree of watershed development. Our results indicate that modern stormwater management practices are not protecting surface waters from road salt contamination and suggest they create contaminated plumes of groundwater that deliver Cl− and Na+ to streams throughout the year.
■
INTRODUCTION Use of NaCl (here after, road salt) as a deicing compound in the United States has grown since the 1960s, reaching 24.5 million metric tons of road salt in 2014.1 Since the 1980s sales of road salt has increased 40% faster than urban land cover, and Na+ and Cl− concentrations in streams have increased across the northern U.S., Canada, and Europe.2−5 The growth in Na+ and Cl− concentrations in streams over the last several decades has significant implications for aquatic ecosystem health and for drinking water quality. For example, Cl− concentrations as low as 33−108 mg L−1 are correlated with alterations in fish assemblages in Maryland streams6 and reduced survival of amphibians in wetland systems has been observed at 145 mg L−1 of Cl−,7 with reduced probabilities of occurrence of amphibians observed at even lower levels of Cl−.8 The EPA recommends Na+ concentrations in drinking water not exceed 20 mg L−1 for individuals on low-sodium diets.9 While the number of studies documenting road salt impacts on Na+ and Cl− concentrations in streams continues to grow, two key phenomena remain understudied: (1) processes that control the movement of dissolved road salt from impervious surfaces to streams; (2) temporal and spatial variations in Na+ and Cl− concentrations. With regards to road salt movement, it is clear that some fraction of dissolved road salt takes months to © XXXX American Chemical Society
years to travel from application sites to streams. Multiyear studies of water quality in streams in the northeastern U.S. consistently show increasing trends in Na + and Cl − concentrations during summer as well as winter months, and mass balance studies indicate that not all the road salt applied in a particular winter season is exported from watersheds within the same year.2,3,10 These two observations strongly suggest that some portion of dissolved road salt is entering groundwater where it slowly moves to streams. However, with few exceptions,11−13 previous work has studied road salt impacts on either streams or groundwater, but not both in the same watershed. Engineered structures such as stormwater management basins are likely to play a particularly important role in movement of road salt into groundwater. Stormwater management basins have become an increasingly common feature in (sub)urban developments with the goal of promoting infiltration and reducing runoff. Very little work has been done to document how stormwater management might affect Received: Revised: Accepted: Published: A
June 22, 2016 March 15, 2017 March 21, 2017 March 21, 2017 DOI: 10.1021/acs.est.6b03107 Environ. Sci. Technol. XXXX, XXX, XXX−XXX
Article
Environmental Science & Technology Na+ and Cl− concentrations in a connected groundwaterstream system.12 Here we investigated the effects of modern stormwater management practices, specifically the use of stormwater management ponds to detain storm runoff, on the movement of NaCl from road surfaces to streams. We focus on NaCl as most road salts used by state and local municipalities are primarily NaCl.14,15 We focus on two spatial scales to provide details of NaCl loading to streams while also investigating the generality of findings at a specific site to our broader study watershed. At the local scale, we focused on describing the NaCl groundwater plume associated with two stormwater ponds in the Red Run watershed of Baltimore County, MD, and the impact of the plume on NaCl loading to a small stream. At a larger spatial scale, we collected water samples at 24 sites throughout the Red Run watershed (19 km2) to document seasonal changes in road salt inputs to streams and compared those with and without stormwater management ponds in their respective watersheds. To compare the temporal dynamics of road salt inputs to streams at a finer scale, we also used continual monitoring of conductivity at three stream sites with and three stream sites without stormwater management ponds in their respective watersheds.
was measured using a SonTek Flowtrack Acoustic Doppler Velocity meter. Local Scale Study. We used a series of piezometers placed in two stormwater ponds, and across the floodplain between the ponds and stream (SI Figure S1), to delineate the groundwater plume of Cl− created by the ponds and investigate its temporal dynamics. The site is underlain by a poorly drained, acidic (pH ≤ 5.3), base-poor (ultisol) soil with relatively low cation exchange capacity (∼10 mequiv/100g).18 The ponds were constructed as “dry” ponds that held surface water for several days following precipitation events and groundwater samples were collected during dry periods for this study. The elevation across the study site drops from approximately 169 MAMSL within the ponds to 162 MAMSL at wells adjacent to the stream (SI Figure S1). The sampling depth below the surface for the stormwater pond piezometers was approximately 15 cm and for the floodplain piezometers approximately 90 cm. We used grab samples, discharge measurements, and continuous monitoring of stream conductivities and water depths to estimate loading of Cl− resulting from road salt travel from the two stormwater ponds through the groundwater plume to the stream. We collected water samples from all piezometers and from the stream on four occasions from June 2009 through May 2010. We placed continually recording conductivity and water level probes in the stream above and below the input of the salt plume created by the ponds; the probes were separated by 15 m of stream channel. On 24 occasions we made field measurements of discharge and water depth at the upstream and downstream gauges for the estimation of rating curves. We collected water samples (n = 29) spanning the range of conductivity measures to estimate relationships between conductivity and Cl− concentrations for the upstream and downstream set of probes. Large-Scale Study. We sampled water at 24 sites in the Red Run watershed every 1 to 2 months across the transition from winter to summer of 2011 (SI Figure S1). We selected sites to represent three stormwater management conditions. Sites that had no other sample sites upstream (i.e., were assumed to be independent of other sites) and that had only one type of stormwater management in their respective watersheds were classified as managed (n = 7) if they had stormwater ponds that received runoff from impervious surfaces or unmanaged (n = 4) if they did not have stormwater ponds. The remaining sites were located below other sample sites, often drained land where runoff was only partially intercepted by stormwater ponds, and were classified as combined. All sites were sampled on each sampling date with the exception of January, when four combined sites could not be sampled due to high discharge and anchor ice along shorelines. We also placed data loggers in three streams draining watersheds with (sites 17, 20, and TR) and without (sites 12, 23, and 24; SI Figure S1) stormwater management ponds to monitor conductivity. We chose watersheds with a range of impervious surface cover, from 1.2% to 22.6% in watersheds without stormwater ponds and from 9.8% to 21.0% in watersheds with stormwater ponds. Five of the sites were located in the Red Run watershed and corresponded with sites included in the survey of streamwater chemistry. We also included a site on a small tributary to Towson Run (39.3956N, 76.6215W) to document conditions at a site with a greater degree of urbanization that included stormwater management
■
MATERIALS AND METHODS General Approach. We conducted the majority of the study in the Red Run watershed of Baltimore County, Maryland (Supporting Information (SI) Figure S1). Crystalline bedrock underlies the Red Run watershed with schist comprising the majority of the bedrock with some ultramafic and small amounts of marble.16 The Red Run watershed experienced two phases of development over the past 60 years. Before approximately 1985 a number of single-family residential neighborhoods were developed along with relatively small commercial areas within a rural/agricultural land use matrix. Before 1986 stormwater management efforts were minimal and less than ten ponds of any type were found in the watershed.17 After 1986 and the designation of the watershed as an area of smart growth by Baltimore County, there was an upsurge in development of lands for single and multifamily residential and commercial use under best stormwater management practices, including stormwater management ponds and maintenance of riparian buffer strips. By 2007 the number of stormwater ponds in the watershed exceeded 200.17 We focused on Cl− ions as a tracer of road salt movement through both ground and surface water because the main source of Cl− in the Red Run watershed is road salts and Cl− is the more labile ion in NaCl salts.8,12 To investigate the degree to which salt ions moved through groundwater versus entered surface waters through surface runoff, we analyzed the relationship between molar concentrations of Na+ and Cl−. As Na+ has a higher affinity for sites on soil particles, a Na+:Cl− ratio less than one (i.e., a slope less than one when molar Na+ is regressed against molar Cl− ion concentrations) indicated a greater relative degree of travel via groundwater; the lower the ratio, the greater the relative degree of travel through groundwater. All water samples were diluted to below 1000 μS cm−1 and then filtered (0.45 μm PTFE syringe filter) before being analyzed by ion chromatography (IC; Dionex ICS-5000) for Na+ and Cl− concentrations. Each IC run included blanks, which did not significantly exceed detection limits. Discharge B
DOI: 10.1021/acs.est.6b03107 Environ. Sci. Technol. XXXX, XXX, XXX−XXX
Article
Environmental Science & Technology
dependent variable. We also included models with the additive effects of independent variables, a model with just molar Cl−, and a null model with just an intercept term. For groundwater samples we did not include a model with the main effects of month only because under all conditions, we expected some degree of relationship between molar Na+ and molar Cl−. To compare temporal dynamics of conductivity among sites where conductivity probes were deployed, we constructed cumulative frequency distributions for each site and calculated the 10th, 50th, and 90th percentiles. We calculated bootstrapped 95% confidence intervals for each percentile for each site.
ponds. We deployed data loggers in fall of 2011 and operated them almost continually for two years. During winter months, when we expected spikes in conductivity associated with the application of road salts, we recorded conductivity every 30 min. During summer months, when spikes in conductivity were not expected, we recorded conductivity hourly to conserve battery life. We used ArcGIS and the 2006 National Land Cover data set for impervious surface19 to estimate impervious surface cover for all watersheds corresponding to each sample site. We used a GPS unit to locate sample sites and imported them into ArcGIS for determination of watershed boundaries. We digitized watersheds based on digital elevation models, 2008 aerial photography, and field validation of stormwater flow pathways. For each watershed, we estimated impervious surface cover as a weighted average where weights were the proportion of each watershed falling into each percentage category from 0 to 100% impervious cover. Modeling and Statistical Analyses. To estimate Cl− loading due to the salt plume, we converted conductivities to estimates of Cl− concentration and then calculated the difference between estimated export of Cl− at the upstream and downstream set of probes. Because of intermittent flow conditions from approximately March until October in both study years, we limited our analysis of loading to the late fall and winter months of 2008−2009 (October 1, 2008 through February 28, 2009) and 2009−2010 (November 1, 2009 through March 31, 2010). These periods included a winter with below average snowfall (2008−2009; 27 cm) and a winter with one of the largest snowfall totals on record (2009−2010; 195 cm; SI Figure S2) as reported at Baltimore-Washington International Airport (National Environmental Information Service, National Oceanographic and Atmospheric Administration). To estimate input due to groundwater recharge to the stream, we separated groundwater and surface runoff loading using a recursive digital filter method.20 To predict Cl− concentrations from field conductivity measurements and estimate discharge, we fit piecewise regression models with two line segments. Further information on ground and surface water separation and piecewise regression models can be found in the SI. For the large-scale study of streams we analyzed relationships between impervious surface cover, stormwater management, and streamwater Cl− concentrations on a monthly basis. We used an information-theoretic approach21 to compare five models for each month sampled. The full model included stormwater management (managed, unmanaged, or combined), percent impervious surface, and their interaction. We also included an additive model that did not include the interaction term; models that included the main effects of stormwater management or impervious surface cover separately; and a null model that included only a single intercept. We calculated Akaike information criterion with a correction for small sample size (AICc) and model weights following Burnham and Anderson.21 All models were fit using the glm function of the R program.22 To investigate relationships between molar Na+ and Cl−, we also used an information-theoretic approach. For water samples from pond and floodplain piezometers our full model included molar Na+ as the dependent variable and molar Cl−, sampling month, and their interaction as dependent variables. For streamwater samples we again fit separate models for each month of sampling and included stormwater management as a
■
RESULTS Local Scale Study. Across all water samples, concentrations of Cl− were highest in groundwater beneath the ponds (x̅ = 3828 mg L−1; SD = 4493), reaching a maximum in February 2010 of 18,540 mg L−1 in one well (February x̅ = 7422 mg L−1; SD = 6061; Figure 1). Concentrations of Cl− in groundwater
Figure 1. Changes in Cl− ion concentrations over the winter of 2009− 2010 for water samples from piezometers located within stormwater management ponds (Pond) and across the adjoining stream floodplain (Floodplain) contaminated with road salts in the Red Run watershed, Baltimore County, Maryland. Median (heavy line), 25th to 75th percentiles (red box), and ranges of larger values (whiskers and points). The bottom panel is stream concentrations adjacent to the floodplain as measured from water samples collected on the same day that wells were sampled.
beneath the ponds were also relatively high during June 2009 (x̅ = 4639 mg L−1; SD = 4174) and were lowest during November 2009 (x̅ = 798 mg L−1; SD = 746). In contrast, concentrations of Cl− in the shallow groundwater of the floodplain (421 mg L−1; SD = 717) were over 9 times lower than concentrations in shallow groundwater below the ponds and increased from June 2009 through May 2010 (Figure 1). Separate plumes of salt contaminated groundwater originating from each of the upslope stormwater ponds were evident on all sampling occasions (Figure 2). Mean concentrations for the entire sample period were low on the C
DOI: 10.1021/acs.est.6b03107 Environ. Sci. Technol. XXXX, XXX, XXX−XXX
Article
Environmental Science & Technology
Figure 2. Interpolation of the groundwater plume of Cl− ions (mg L−1) associated with two stormwater management ponds in the Red Run watershed, Baltimore County, MD. The ponds are located in the bottom left corner and the stream in the upper right corner of the maps. The stream flows from west to east. Piezometers are indicated as black points.
northwest side of the well field (x̅ = 12 mg L−1; SD 8), which suggests minimal salt impact, but relatively high in the body of the plumes directly downgradient of the ponds (x̅ = 1391 mg L−1; SD = 1096). The seasonal pattern for stream Cl− concentrations in grab samples was similar to groundwater, with lower concentrations in June and November 2009 and higher concentrations in February and May 2010 (Figure 1). Ratios of Na+:Cl− in waters beneath the stormwater ponds were near 1, suggesting NaCl inputs were the primary source of Cl− to groundwater beneath the ponds (Figure 3). The slope of the line relating molar Na+ to molar Cl− for water samples from wells within the ponds was slightly less than 1 (95% CI = 0.869−0.963) and the most supported model included only molar Cl− (SI Table S2). The average Na+ contribution to the total cation charge was 88.9% while Mg2+ and Ca2+ contributed 10.9%. In contrast, Na+:Cl− ratios in shallow groundwater of the floodplain plume were often much less than 1; smaller ratios tended to be associated with higher Cl− concentrations and relationships varied among sampling dates (Figure 3). The most supported model of Na+ molar concentrations was the full model including molar Cl−, sample date, and their interaction (SI Table S2). The model suggested that the slopes of the lines relating molar Na+ to molar Cl− during June and November were greater than slopes during February and May (SI Table S3). In floodplain groundwater, the average Na+ contribution to the total cation charge dropped to 75.9% while Mg2+ and Ca2+ contributed 22.9%. During late fall and early winter, generally before January of both study years, increases in discharge associated with rain events produced a sharp decrease in conductivity and Cl− concentrations in stream waters (Figure 4 and SI Figure S5). In contrast, during January and February of both years very small increases in discharge associated with frozen precipitation events produced extremely large increases in conductivity and Cl− concentrations, approaching 1000 mg L−1 and 1400 mg L−1 at the downstream conductivity probe during events in the
Figure 3. Relationships between molar Na+ and molar Cl− ion concentrations for water samples collected from piezometers within stormwater management ponds and across the adjoining stream floodplain contaminated with road salts in the Red Run watershed, Baltimore County, Maryland. Dotted lines are the 1:1 ratio and red lines are fitted linear regressions.
winters of 2008−2009 and 2009−2010, respectively. During the winter of 2008−2009, we estimated an input of 6574 kg of Cl− to the stream from the salt contamination plume. Of this load, 60% occurred during base flow conditions as defined by the recursive digital filter (SI Figure S6). In contrast, loading D
DOI: 10.1021/acs.est.6b03107 Environ. Sci. Technol. XXXX, XXX, XXX−XXX
Article
Environmental Science & Technology
and summer, differences in Cl− concentrations among sample sites were primarily related to stormwater management type within watersheds, with the most supported models for April, June, and August containing only stormwater management and other models in the model sets having evidence ratios of greater than two (SI Table S4). During April, June, and August relationships between Cl− and impervious surface cover remained weak (r < 0.41). Similar to Cl− concentrations in stream waters, molar ratios of Na+:Cl− varied seasonally. During January ratios were slightly above 1 when molar concentrations of Cl− were above approximately 5 mmol L−1 (SI Figure S7). The full model provided the best fit during January (SI Table S5), although a model containing only Cl− accounted for greater than 99% of the variation in Na+ concentrations and had a slope slightly greater than one (95% CI = 1.03−1.10). Because a much greater variation in ratios during February, particularly several occasions when molar Na+ concentrations greatly exceeded molar Cl− concentrations (SI Figure S7), there was a much weaker relationship between the two ions (r2 = 0.24) and the slope of the line relating the two ions did not differ from 1 (95% CI = 0.45−3.72). By March a strong relationship between molar Cl− and Na+ had returned (r2 = 0.99; SI Figure S7) and the slope of the line relating the two was only slightly less than 1 (95% CI = 0.90−1.00). By late spring and early summer, ratios of Na+:Cl− had dropped below 1 for most sites and the degree of drop increased with Cl− concentrations (SI Figure S7; Table S5). For the months of April, June, and August, relationships between ions were strong (April: r2 = 0.95; June: r2 = 0.87; August: r2 = 0.82) and the slopes of lines relating ion concentrations were well below 1 (April: 95% CI = 0.49−0.60; June: 95% CI = 0.55−0.76; August: 95% CI = 0.42−0.62). In general, higher base flow conductivities and larger spikes in conductivity associated with storm events resulted in greater 50th and 90th percentiles of conductivity for sites with stormwater ponds in their respected watersheds (Figure 6). In all cases there was no overlap in the 95% confidence intervals for the 50th and 90th percentiles of conductivity between sites with and without stormwater management ponds in their respective watersheds (Figure 6; SI Table S6). The 90th percentile for sites where stormwater was managed ranged from 447 to 1773 μS cm−1, while the same range for sites where there was no stormwater management was 258−278 μS cm−1. The highest conductivities were from the most highly urbanized site with stormwater management.
Figure 4. Discharge at the downstream sensor and estimated Cl− ion concentrations at the upstream and downstream sensor during the 2008−2009 winter. Data are means for 24 h periods to facilitate plotting.
was much greater in the winter of 2009−2010 when the total input to the stream was estimated at 40,008 kg of Cl− and a smaller proportion of the input occurred during base flow conditions (36%). Large Scale Survey. Overall, there was a decline in Cl− concentrations from winter months through spring and into summer for all stream sites, but declines were greater for sites draining watersheds without stormwater ponds (Figure 5).
Figure 5. Mean Cl− ion concentrations for streamwater collected at sites with (managed), without (unmanaged), and with a combination (combined) of watershed area draining into stormwater management ponds. The inset graphs show Cl− ion concentrations as a function of watershed impervious surface for a month when the relationship was relatively strong (January) and a month when the relationship was weak (August). Error bars are ±1 SE.
■
DISCUSSION We found that routing of runoff contaminated with road salts to stormwater ponds resulted in plumes of highly contaminated groundwater moving from ponds to streams. The groundwater beneath both stormwater ponds was extremely high in Cl− and represents a large source of Cl− and Na+ to the downgradient groundwater and stream. For the month with the highest average Cl−, groundwater beneath the stormwater ponds had Cl− concentrations that were 27% of seawater −19 000 mg L−123 − and the stormwater pond well with the highest Cl− approached seawater concentrations. Median Cl− concentrations were 102 to 103 times higher below the stormwater ponds, greater than 10 times higher in the floodplain, and 1−7 times higher in the stream than Cl− concentrations in uncontaminated groundwater from the Loch Raven Schist (similar bedrock to the study area) in Baltimore County, Maryland (8.1 mg L−1)24 and median Cl− concentrations of Maryland
During January, when Cl− concentrations in streams were relatively high (x̅ = 407 mg L−1; SD = 324; Figure 5), the model predicting Cl− concentrations and including only percent impervious surface was the most supported model, with other models in the model set having evidence ratios greater than eight (SI Table S4). During January the relationship between percent impervious surface cover and Cl− concentrations was positive and strong (r = 0.77; inset A, Figure 5). However, by February and March the most supported models included both impervious surface cover and stormwater management (SI Table S4) and relationships between Cl− concentrations and impervious surface cover were weaker (February: r = 0.55; March: r = 0.34). By late spring E
DOI: 10.1021/acs.est.6b03107 Environ. Sci. Technol. XXXX, XXX, XXX−XXX
Article
Environmental Science & Technology
concentrations were three times higher in streams draining watersheds with stormwater ponds than those without (Figure 5). Although the degree of watershed urbanization (as measured by impervious surface area) is responsible for the majority of the variation in Cl− concentrations among streams in winter months, during summer months Cl− concentrations were more related to stormwater management (SI Table S4). The 50th percentile of conductivity measurements were on average almost three times higher in streams draining watersheds with stormwater ponds (x̅ = 632 μS cm−1) when compared to streams draining watersheds without stormwater ponds (x̅ = 215 μS cm−1, Figure 6). Additionally, an unmanaged site (12) with 22.6% impervious cover had a 50th percentile conductivity of 221 μS cm−1, whereas a managed site (TR) with 27% impervious surface had a 50th percentile conductivity of 1174 μS cm−1 (Figure 6). It is interesting to note that the latter watershed has the longest history of urbanization and road salt application among the watersheds we studied. The range of Na+:Cl− ratios in groundwater beneath the stormwater management ponds (accounting for an analytical uncertainty of ∼10%) matches seawater (0.85)23 and road salt used by the Maryland State Highway authority (0.92).26 The match in Na+:Cl− ratios demonstrates that this groundwater has little net chemical interaction with the stormwater management pond sediments. While it is possible that input to streams from throughflow may have a different Na+:Cl− ratio than surface runoff or groundwater, the ∼1:1 Na+:Cl− ratios observed in January, February, and March for the large scale survey (SI Figure S7) indicate that throughflow from storms and snowmelt either has similar Na+:Cl− ratios to surface runoff or does not have a significant effect on stream chemistry. In contrast, the low Na+:Cl− ratios in floodplain groundwater result from chemical interaction with the floodplain materials with a reduced proportion of the cation charge contributed by Na+ in the floodplain groundwater (75.9%) as compared to groundwater below the stormwater pond (88.9%). When aqueous Na+ is high, road salt-driven cation exchange reactions have been documented to reduce the Na+:Cl− ratio in streams.26−29 Formation of chlorinated organic compounds has also been shown to be an important process that may result in retention of Cl− in soils.30,31 However, several factors suggest that formation and storage of chlorinated organic compounds is relatively minor compared to storage of Na+ through cation exchange reactions. A review of Cl− budgets for small, forested watersheds, which retain a much greater fraction of organic compounds than urbanized watersheds,32 found a strong relationship that did not differ from 1:1 between input and output of Cl−.33 We documented significantly elevated Cl− concentrations in the groundwater (421 mg L−1 as compared to 3 mg L−1 in groundwater from a forested watershed ∼15 km to northeast34) and Na+:Cl− ratios much less than one. Finally, laboratory experiments suggest a relatively rapid shift from storage to export when Cl− is added to soil columns.35 Taken together our data suggest that Na+ and Cl− moves to streams via at least two routes: (1) quick flow paths to the stream where Na+ and Cl− arrive within weeks to months of application in January to March as indicated by Na+:Cl− ratios of approximately 1 and (2) slower flow paths where Na+ and Cl− have longer residence times in groundwater and interact more extensively with floodplain sediments as indicated by Na+:Cl− ratios of much less than 1 in April, June, and August. The relative role of these paths in transporting Na+ and Cl− to
Figure 6. Cumulative frequency distributions of specific conductivity for three streams with stormwater ponds in their respective watersheds (sites TR, 17, and 20) and three streams with no stormwater ponds in their respective watersheds (sites 12, 23, and 24). The 10th, 50th (median), and 90th percentiles are marked on each graph and the maximum value observed is provided. Data were recorded on a half hour or hourly basis from fall of 2011 through the fall of 2013; see SI Table S6 for sample sizes.
watersheds with
