Evaluation of Phosphate and E. coli Attenuation in a Natural Wetland Receiving Drainage from an Urbanized Catchment

Abstract

A natural wetland receiving drainage from a 24-ha urbanized catchment in the Falls Lake Watershed of North Carolina was evaluated to determine if it was providing ecosystem services with regards to phosphate and Escherichia coli (E. coli) attenuation. Inflow and outflow characteristics including nutrient and bacteria concentrations along with physicochemical properties (discharge, pH, oxidation reduction potential, temperature, and specific conductance) were assessed approximately monthly for over 2 years. The median exports of phosphate (0.03 mg/s) and E. coli (5807 MPN/s) leaving the wetland were 85% and 57% lower, respectively, relative to inflow loadings, and the differences were statistically significant (p < 0.05). Hydraulic head readings from three piezometers installed at different depths revealed the wetland was a recharge area. Phosphate and E. coli concentrations were significantly greater in the shallowest piezometer relative to the deepest one, suggesting treatment occurred during infiltration. However, severe erosion of the outlets is threatening the stability of the wetland. Upstream drainageway modifications were implemented to slow runoff, and septic system repairs and maintenance activities were implemented to improve water quality reaching the wetland and Lick Creek. However, more work will be needed to conserve the ecosystem services provided by the wetland.

1. Introduction

Excess nutrient and bacteria concentrations are commonly cited reasons for the impairment of water resources [1,2,3,4]. For example, approximately 46% of the river and stream miles and 21% of lakes assessed in the US were listed as being in poor condition because of excess phosphorus [1]. Elevated concentrations of fecal bacteria were reported in 23% of rivers and stream miles assessed, and pathogens were identified as a major pollutant of coastal waters [1]. The primary productivity of freshwater streams and lakes is often limited by the supply of reactive phosphorus (phosphate) [5]. When phosphate is added to water resources, it may stimulate blooms of algae and other primary producers, thus providing a food source for higher organisms [6]. However, too much phosphorus can lead to excessive growth of harmful algal blooms and cyanobacteria, some of which produce toxins that are dangerous for humans, wildlife, and aquatic organisms [2]. For example, microcystin, a toxin produced by cyanobacteria has been detected in 39% of lakes sampled in the US [1]. Algae blooms may cover the surface of the waters, preventing light from reaching submerged aquatic vegetation (SAV) and causing die off of SAV [7]. When SAV and algae die, waters can become depleted of dissolved oxygen during decomposition, resulting in a poor habitat for aquatic life and causing water use impairment. Some of the major sources of excess phosphorus loading to surface waters include agricultural runoff [8,9], livestock waste [10], urban runoff [11], municipal sewer discharges [12,13], septic systems [12,14], and wildlife waste [15]. Efforts to reduce the loading of nutrients from these sources is important for improving water quality and ecosystem health [2].

Fecal indicator bacteria such as E. coli are used to assess the potential public health risks of waters used for recreation and water supply. While most are not harmful themselves, research has shown that when concentrations of E. coli exceed certain thresholds, the likelihood of negative health outcomes increases due to the presence of actual pathogens [16]. For recreational waters that are routinely tested, the geometric mean of E. coli should be below 126 colony forming units per 100 mL, and no more than 10% of samples should exceed 410 cfu/100 mL. E. coli in surface waters may originate from many of the same sources as phosphorus including wastes from humans [17,18], pets [19], and wildlife [20]. Fecal bacteria can be delivered to surface waters via urban runoff that flushes waste deposited onto impervious surfaces [21], agricultural runoff from livestock waste deposited or applied to fields [22], discharges of municipal sewer effluent [23], and malfunctioning septic systems [17]. Reducing the delivery of E. coli from watersheds influenced by anthropogenic activities (e.g., agriculture, urbanization, wastewater discharges) to receiving waters is important for public and environmental health.

Wetlands are transitional zones between terrestrial and aquatic environments and often provide valuable ecosystem services including pollutant attenuation and reduction in runoff. For example, studies have shown that natural wetlands can reduce exports of phosphorus via slowing runoff/flow, thus allowing for sedimentation of particulate-bound phosphorus and plant uptake of phosphate [24]. Phosphorus can also sorb to soil and be assimilated by microorganisms [25] and/or combine with soluble cations such as iron, aluminum, and/or calcium to form minerals that precipitate out of solution [26] and become buried in wetlands. In a review of 57 wetlands, it was reported that 84% were sinks for phosphorus [27]; thus, most were providing ecosystem services regarding nutrient retention. In a different review of studies that included 203 wetlands [28], a median TP reduction of 46% was reported as water moved through the wetlands. Some of the factors that may influence the phosphorus removal efficiencies of wetlands include hydraulic retention time, which is a result of inflow loading rate (flow rate and phosphorus concentration) and wetland area [29], and temperature [30]. Longer retention times encourage sedimentation and allow for more opportunity for immobilization of phosphorus via plant and microbial uptake. Plant density is typically greater during warmer periods, also providing enhanced uptake and sedimentation [30].

Fecal bacteria may also be removed in wetlands via several processes including but not limited to sedimentation, filtration, and sorption [23], disinfection via exposure to ultraviolet rays [31] and predation [32]. Bacteria may attach to sediment via the electrostatic attraction created because of differences in the surface charges of soil and bacteria. As water travels from the inlet towards the outlet of a wetland, sediment and the attached bacteria may fall out of suspension as the velocity of water decreases due to friction with wetland vegetation, thus reducing the export of both sediment and bacteria [31]. Bacteria in the wetland influent may be exposed to predators such as micro-zooplankton that graze on and thus reduce concentrations of bacteria as water passes into the wetland [23,32]. In wetlands that have areas open to sunlight, the mortality of indicator bacteria may occur due to ultraviolet radiation exposure [31]. Like with the removal of phosphorus, the hydraulic retention time has been reported as a significant factor regarding bacteria removal [23]. Wetlands may be constructed in upland areas to mimic processes in natural wetland and to remove pollutants including nutrients and E. coli from wastewater [24,25,31,33], urban stormwater [24,34,35,36], and agricultural drainage [30,37].

The primary goal of this study was to determine if a natural wetland receiving drainage from an urbanized catchment was providing ecosystems services with regards to phosphorus and E. coli attenuation. A secondary goal was to implement measures that may enhance the pollutant removal efficiency of the wetland and improve the quality of water entering Lick Creek and ultimately Falls Lake. Specific objectives included: (1) measure inflow and outflow from the wetland approximately monthly over a 2-year period; (2) collect water samples at the inlets and outlets of the wetland for phosphorus and E. coli concentration analyses and physicochemical characterization; (3) determine influent and effluent loadings of phosphorus and E. coli and determine removal efficiencies of the wetland; (4) implement best management practices to slow runoff entering the wetland and improve the performance of malfunctioning and improperly maintained septic systems in the watershed.

2. Materials and Methods

2.1. Study Site

The study site was a natural riparian wetland adjacent to Lick Creek, in Durham, North Carolina, USA (Supplemental Figure S1). Lick Creek is a tributary of Falls Lake, which serves as the main water supply for the greater Raleigh area and provides recreational opportunities including boating, fishing, and swimming. Lick Creek and Falls Lake are impaired for aquatic habitat and water quality, respectively, and efforts to reduce nutrient, sediment, and bacteria loadings to these waters are mandated by state regulatory agencies [38]. Also suggested is the conservation of riparian wetlands lining Lick Creek, such as the one that is the focus of this research. The wetland evaluated in this study received drainage from a 24-ha catchment (Figure 1) with a land use that was predominantly residential development. The homes (n = 44) in the watershed use onsite wastewater systems (OWSs) for wastewater treatment and dispersal. Soils in the watershed include the White Store, Mayodan, Creedmore, and Chewacla series, most of which have expansive clay mineralogy with low permeability in the subsoil and shallow (<1 m below surface) water tables [39], which may reduce the functionality of the OWSs. Annual mean rainfall in the area is approximately 122 cm, with monthly averages ranging from 8.6 cm in November to 11.9 cm in March [40]. The mean annual daily high temperature is 21 °C, and the mean annual daily low is 8.9 °C. The summer months of July and August have the greatest mean high temperatures of 31.7 °C and 30.6 °C, respectively. The winter months of January and February have the coldest mean low temperatures of −2.2 °C and −1.7 °C, respectively. Inflow to the wetland occurs via two small streams, labeled inlet 1 and inlet 2 (Figure 1). The stream channels dissipate within the wetland and water spreads across the ~1.5 ha surface (Figure 2). The overstory comprises various tree species including Pinus taeda, Liquidambar styraciflua, and Acer rubrum, while the understory includes Saururus cernuus. Chewacla and Wehadkee soils, which frequently flood and typically have seasonal water tables within 0.3 m of the surface, comprise approximately 75% of the soils within the wetland [39]. A levee at the top of the bank of Lick Creek partially impounds water within the wetland. However, in 3 distinct locations, the levee was breached, allowing for surface water outflow to Lick Creek (Supplemental Figure S3). Lick Creek is incised, and thus there was a decrease in elevation of more than 6 m from the levee to the stream bed of Lick Creek (Figure 3). The extreme drop off was contributing to erosion and landward migration of the 3 breach locations labeled outlets 1, 2, and 3 (Supplemental Figure S1). Because of the severe erosion observed, it was unknown if the wetland was providing ecosystems services with regards to nutrient and bacteria attenuation.

2.2. Physicochemical Monitoring

Inflow and outflow measurements were recorded approximately monthly between January 2018 and June 2019, then from August 2020 to April 2021. The gap in data collection was related to funding availability. Stream velocity (m/s) was measured using a Global Water FP111 Flow Probe (Global Water, Gold River, CA, USA). The width (m) and average depth (m) of each stream/outlet was determined using a tape measure and grade rod. Discharge (m³/s converted to L/s) was calculated as the product of mean stream depth, stream width, and velocity. Total wetland inflow was the sum of discharge from inlets 1 and 2, while total outflow was the sum of discharge from outlets 1, 2, and 3. A YSI 556 multimeter (YSI Environmental, Yellow Springs, OH, USA) was used to determine the pH, temperature, specific conductance (SC), and oxidation reduction potential (ORP) of water entering and exiting the wetland. The probes were calibrated prior to each site visit. A stilling well was installed in the stream channel at inlet 2 during August of 2020. The well was secured by driving a stake into the stream bed close to a tree and using zip ties and tape to secure the well. A Solinst 3001 Levelogger Junior (Solinst Canada Ltd., Georgetown, ON, Canada) water level logger was programmed to record every 20 min and lowered to the bottom of the well. A Solinst 3001 Barologger (Solinst Canada Ltd., Georgetown, ON, Canada) was tied to a tree near inlet 2 and programmed to record air pressure every 20 min. The air pressure readings were used to compensate for changes in pressure recorded by the water level loggers that were not associated with water depths. The high-frequency water level data were compared to the monthly manual stage readings to determine how close the mean and median values were between the data sets and provide perspective on the overall hydrologic dynamics. High frequency water level data were recorded between 17 August 2020 and 29 June 2021.

Three piezometers were installed at different depths (0.6 m, 1.6 m, 3 m) within the wetland (Supplementary Materials) to evaluate hydraulic head differences and determine if recharge or discharge conditions were present. The piezometers were constructed by coupling 5 cm diameter schedule 40 PVC casing to well screen. Augers were used to create boreholes for the 3 piezometers. The piezometers were installed in the boreholes and sand was poured in the annular space until the entire well screen was covered. A bentonite and sand mixture was used to seal the remaining annular space to the ground surface. Piezometer casings extended above the surface more than 0.45 m to ensure that surface water did not overtop the casings during extreme rainfall events. A laser level was used to determine the relative elevation to the top of each piezometer casing. Depth to water in each piezometer was measured using a Solinst Model 107 temperature level and conductivity meter (Solinst Canada Ltd., Georgetown, ON, Canada). The depth to water was subtracted from the relative elevation of the casing top to determine relative elevation of the water surface in each piezometer. When the relative elevation of the water in the shallower piezometers exceeded water elevations in the deeper piezometer, then recharge conditions were noted. Discharge conditions were present when the relative elevation of water in the deeper piezometer exceeded the shallower piezometers. Groundwater samples were collected from the piezometers using disposable bailers (Supplementary Materials) after the depth to water was recorded. Each piezometer was purged using a new, separate bailer and allowed to recharge (typically for 10 min) prior to sample collection. Groundwater samples were transferred from the bailers to sample bottles and to the YSI 556 multimeter calibration cup for physicochemical analyses (pH, ORP, SC, temperature). Groundwater measurements and samples were collected during 10 different months spanning September 2018 to June 2019.

2.3. Water Analyses

Samples for phosphorus analyses were vacuum-filtered using 1.5- and 0.7-micron microfiber glass filters to remove suspended solids. Material captured by the filter was used to analyze for particulate phosphorus while the filtrate was analyzed for total dissolved phosphorus and phosphate using a SmartChem 170 (KPM Analytics, Westborough, MA, USA) color spectrometer. Total phosphorus was calculated as the sum of particulate and dissolved phosphorus. Phosphate was analyzed during every sample collection event when flow was measurable (n = 22), while particulate phosphorus and total phosphorus were analyzed during the last 17 sample events. Water samples were analyzed for E. coli using IDEXX Colilert media with IDEXX Quanti-Tray 2000 methods (IDEXX Laboratories, Inc., Westbrook, ME, USA). After preparation, samples were incubated at 35 °C for 24 h and viewed with a blacklight for enumeration.

Qualitative Taqman™ polymerase chain reaction (PCR) was used to determine if human wastewater from the 44 septic systems in the watershed was a source of bacteria in the drainageways discharging into the wetland. This procedure used fluorescent dye to amplify the DNA. The tests were completed using the Qiagen™ (Hilden, Germany) and UNEX protocols to extract DNA from the samples. Analyses of the samples was performed with a Lightcycler^® 480 II instrument (Roche, Mannheim, Germany) and Superscript III One-Step qRT-PCR kit (Invitrogen, Carlsbad, CA, USA). The equipment utilized the following program: 50 °C for 2 min at a ramp rate of 4.4 °C, 95 °C for 10 min at a ramp rate of 4.4 °C, 45 cycles at 95 °C for 15 s at a ramp rate of 4.4 °C, 60 °C for 1 min at a ramp rate of 2.2 °C. Primers and probes for bacterial general indicator Bacteriodales detection were BacGen Rev 5′-CACGCTACTTGGCTGGTTCAG-3′, BacGen Fwd 5′-CTGAGAGGAAGGTCCCCCAC-3′, BacGenLNA probe 5′-/56-FAM/AG+CA+GT+G+A+G+GAATATT/3IABkFQ/-3′. Primers and probes for bacterial human indicator detection were BacHum Rev 5′-CGCTACACCACGAATTCCG-3′, BacHum Fwd 5′-CGCGGTAATACGGAGGATCC-3′, BacHum LNA Probe 5′-/56-FAM/AAGTTT+GC+GG+C+T+CAAC/3IABkFQ/-3′ (IDTDNA, Coralville, IA, USA). Similar protocols have been used in prior studies [21,22] to help identify bacteria sources. Stream samples from inlets 1 and 2 were collected during late summer (September) and early winter (December) of the project period. The samples were initially compared to the general indicator Bacteriodales. If the general indicator Bacteriodales was detected, then the sample was run against the human-specific Bacteroides fragilis (ATCC Manassas, VA, USA). A wastewater sample from a septic tank served as the positive human control, and a pet (dog) waste sample was the negative control.

2.4. Pollutant Loading and Treatment Efficiency

Flow (L/s) was multiplied by the concentrations of phosphorus (mg/L) and E. coli (MPN/100 mL) at the two inlet and three outlet locations. Total inflow loading was the sum of inlet 1 and 2 loads, while total exported loads were the sum from outlets 1, 2, and 3. Treatment efficiency of the wetland was calculated using Equation (1).

$$Treatment Efficiency=\frac{Inflow Loads-Outflow Loads}{Inflow Loads }\times 100\%$$

(1)

2.5. Statistical Analyses

Anderson–Darling tests were performed using Minitab 20 statistical software (Minitab, LLC, State College, PA, USA) on water quality data to determine if they demonstrated a normal distribution. The data that were not normally distributed were log10 transformed, and normality tests were run on the transformed data. T-tests were used to determine if differences in inflow and outflow characteristics were significantly different at p < 0.05 for data that exhibited normal or log10 normal distributions. Mann–Whitney nonparametric tests were used to determine significance for data that were not normal or log₁₀ normally distributed. Concentrations of E. coli exiting the wetland were compared to US EPA [16] recreational water quality criteria (geometric mean should not exceed 126 cfu/100 mL; no more than 10% of samples should exceed 410 cfu/100 mL). It was also reported that the frequency of outflow exports was less than that of inflow loads. Spearman’s rank (nonparametric data) or Pearson (data with normal distributions) correlations were performed on some data to determine if statistically significant relationships were observed.

2.6. Drainageway Modifications and Onsite System Improvements

A walking survey of the watershed was conducted to assess the condition of the two main drainageways and their tributaries that provided inflow to the wetland, and to determine if there were signs of OWS malfunction (e.g., effluent surfacing, straight pipes, wastewater backup in home). The morphology of the drainageway channels was observed and in locations where erosion or evidence of poor bank and/or bed integrity were observed, the addresses were recorded. Discussions with the property owners were held to determine their perspective on the stability and functioning of the drainageways in their yards and whether they would be supportive of mitigation efforts. Most but not all property owners who were experiencing issues with drainage and erosion were cooperative and wanted to participate. For most of the identified sites, the banks of the drainageways were tapered to a less steep slope (3:1 horizontal/vertical ratio or greater) by removing the soil/sediment using an excavator (Figure 4). The soil was removed from the site via dump trucks or spread onto uplands and covered with seed and straw. Geotextile fabric was anchored into the “new” bank and rip rap or rock was placed on the fabric. The capacity of the drainageways was increased to accommodate more stormflow and the banks were stabilized by placement of the fabric and rip rap. For smaller drainageways and ditches, rock was also placed onto the stream bed (Supplementary Materials). Overall, 230 m of drainageway was stabilized and an estimated 115 m³ of loose soil/sediment was removed from the banks and beds. Rock check dams were used in the channels to help reduce peak flows, drainageway erosion, and ultimately the erosion of the wetland. These practices were expected to help the stability of the wetland by reducing the volume and energy of water entering the wetland from the watershed during and shortly after rainstorms.

Discussions with homeowners in the study area revealed that many had not routinely or ever had their septic tank pumped to remove the solids, and during the walking survey, four different properties were observed that had septic effluent surfacing in their yard. With funds from the grant, septage from septic tanks at 15 different properties (34% of homes) in the watershed was pumped (Figure 5). The drainfields at 3 properties with OWS malfunctions were repaired or replaced, and the cracked lid of one septic tank was replaced to prevent infiltrating rainwater from hydraulically overloading the drainfield. New drainfields were installed at 2 sites (Figure 6). At the third site, a tree root had completely clogged the drainfield trench. The trench was uncovered until the location of clog was found (Supplementary Materials) and effluent from the tank was piped to a different portion of the drainfield that had not been used due to the clog.

3. Results and Discussion

3.1. Flow Dynamics

The total inflow to the wetland varied during the study period and ranged from no measurable flow (0 L/s) during field sampling events in July 2018 and June 2019 to 65.3 L/s and 22.6 L/s during January of 2018 and February of 2019, respectively (Figure 7). Total outflow leaving the wetland also varied from no flow on dates in July 2018, June 2019, and September 2020 to 31.2 L/s (January 2018) and 21.4 L/s (February 2019) when discharge was greatest. Overall, colder months had higher flow than warmer months. For example, during the fall and winter months (October–March), the median inflow (4.7 L/s) and outflow (3.5 L/s) were greater relative to the median inflow (0.85 L/s) and outflow (0.16 L/s) during the warmer spring and summer months (April–September) (Figure 7). Differences between inflow during cold and warm months (p = 0.004) and outflow during cold and warm months (p = 0.007) were statistically significant. Rainfall was evenly distributed throughout the seasons (Supplemental), and differences in monthly precipitation were not significantly different (p = 0.464) when comparing warm (mean 11.8 cm/month) and cool (mean 10.0 cm/month) periods. Evapotranspiration rates are typically higher during the “growing season” spanning the spring and summer months (April–September) due to longer daylight hours, warmer temperatures, and active plant uptake and transpiration of water, thus resulting in lower discharge [36]. Differences in evapotranspiration may explain the higher inflow and outflows during cooler periods as the mean temperatures during fall and winter months (8.9 °C) were significantly (p < 0.001) lower relative to summer and spring (22.2 °C), thus resulting in less evaporation (Supplementary Materials). Transpiration was likely lower during the cooler months when the deciduous trees and shrubs lose their leaves and there is a reduction in the uptake of water.

Overall, during 69% of the discrete sampling events, outflow from the wetland was lower than inflow. The median inflow (2.46 L/s) was significantly (p = 0.014) higher relative to the median outflow (1.07 L/s). Prior research [41] has shown that forested lands, especially with a mixture of large (>10 m height) and small (3–5 m height) trees can transpire significant volumes of both soil water and groundwater, thus influencing the hydrological cycle. Transpiration rates exceeding 2.4 mm/day have been reported in several studies of forested watersheds in New York, USA [41], Argentina [42], and Georgia, USA [43]. Similar evapotranspiration processes and rates may also have been active within the wetland assessed in this study, given the prevalence of large trees and understory vegetation, thus influencing outflow volumes. The high-frequency water level logger data obtained near inlet 2 showed that the median stage was 15.6 cm and thus within a few centimeters of the median of the monthly discrete stage readings (18.3 cm) (Figure 8). Regression analyses revealed a statistically significant relationship (p < 0.001) between stream stage and discharge (r² = 0.73; discharge = 0.4328 stage − 7.508). When the stage at inlet 2 exceeded 24 cm, then outflow from the wetland was equal to or greater than inflow. The high frequency logger data showed that between August 2020 and late June 2021, the stage at inlet 2 exceeded 24 cm in 28% of the readings or 2113 h (88 days). The highest median monthly stage was observed during February 2021 (31.7 cm) and the lowest was observed during August 2020 (3.6 cm) (Figure 8). During the winter months of January, February, and March, the median monthly stage readings were above 24 cm, suggesting most of the inflow passed directly across the wetland. However, between August 2020 and December 2020, and from April 2021 to the end of the monitoring period in late June of 2021, the median monthly stage was under 24 cm, suggesting that during most of that time, the wetland outflow was lower relative to inflow.

During each groundwater sampling event, the hydraulic head for the deepest piezometer (total depth 3 m) was lower relative to the shallow (0.6 m) and intermediate-depth (1.6 m) piezometers, indicating that the wetland was a recharge area [44]. The median hydraulic head readings for the shallow (4.9 cm below surface) and intermediate (5.2 cm below surface) piezometers were significantly different (p < 0.05) relative to the deep piezometer (17.4 cm below the surface) (Figure 9). Inflow to the wetland was pooling on the surface and gradually infiltrating and percolating through soil. When inflow exceeded the infiltration and surface storage capacities, runoff occurred, and water discharged to Lick Creek via the three outlets. The high-frequency stage readings suggest that outflow was reduced relative to inflow during 72% of the logger recording period and during these times, infiltration occurred.

3.2. Physicochemical Characteristics

The mean physicochemical properties of inflow including specific conductance (411.3 to 484.5 µS/cm), temperature (14.9 to 15 °C), and oxidation reduction potential (100.2 to 105.7) were higher relative to outflow properties (specific conductance: 273.3 to 346.5 µS/cm; temperature: 12.4 to 14.9 °C; oxidation reduction potential: 53.6 to 75.8) (

Table 1. Mean (top) and standard deviation (bottom) of physicochemical properties of inflow and outflow from the wetland and groundwater sampled from the shallow, intermediate, and deep piezometers.

Location	Turbidity (NFU)	pH	Specific Conductance (µS/cm)	Temperature (°C)	Oxidation Reduction Potential (mV)
Inlet 1	21.8	7.0	411.3	15	105.7
Inlet 2	25.2	6.8	484.5	14.9	100.2
Outlet 1	188.2	6.9	273.3	13.6	53.6
Outlet 2	74.2	6.8	333.9	12.4	75.8
Outlet 3	59.0	7.1	346.5	14.9	75.4
Shallow		6.6	260.6	13.3	26.8
Intermediate		6.5	244.0	13.7	−12.6
Deep		6.5	512.7	13.2	−48.7
Location	Turbidity (NFU)	pH	Specific Conductance (µS/cm)	Temperature (°C)	Oxidation Reduction Potential (mV)
Inlet 1	26.8	0.5	272.2	5.7	104.6
Inlet 2	25.6	0.6	247.9	5.9	114.2
Outlet 1	300.6	0.4	82.5	5.8	78.1
Outlet 2	214	0.4	160.7	4.5	117.1
Outlet 3	67.3	0.5	116.6	5.3	119.4
Shallow		0.3	50.2	5.8	72.4
Intermediate		0.5	39.8	5.3	63.8
Deep		0.6	119.1	3.9	108.2

). The turbidity of water entering the wetland (21.8 to 25.2 NFU) was typically lower relative to water exiting the wetland (59.0 to 188.2 NFU) (Table 1). The mean pH of water entering (6.8 to 7.0) and exiting the wetland (6.8 to 7.1) was similar (Table 1). Groundwater from the different piezometers had variable mean physicochemical property values including the lowest mean oxidation reduction potential (−48.7 to 26.8 mV), the lowest specific conductance (244 µS/cm), the lowest pH (6.5 to 6.6), and mean temperatures (13.2 to 13.7 °C) that were lower relative to inflow and outflow (Table 1). Thus, as wetland inflow infiltrated the soil, it recharged and mixed with groundwater and the resulting outflow from the wetland had specific conductance and oxidation reduction potential values that were significantly (p < 0.05) lower than inflow. Outflow was significantly (p = 0.005) more turbid relative to inflow, likely from the erosion of the outlets that discharge to Lick Creek (Supplemental Figure S3).

3.3. E. coli Concentrations and Loading

The median flow-weighted concentration of E. coli entering the wetland (1078 MPN/100 mL) was not significantly different (p = 0.493) relative to the median outflow concentration (649 MPN/100 mL). In just over half (52%) of the sampling events, inflow concentrations were greater than or equal to outflow concentrations (Figure 10). Concentrations of E. coli entering and exiting the wetland were greater than the US EPA (2012) statistical threshold value of 410 MPN/100 mL in 76% of sampling events, thus exceeding the 10% frequency established as the benchmark. Furthermore, the geometric mean of E. coli in wetland inflow (858 MPN/100 mL) and outflow (719 MPN/100 mL) was greater than the geometric mean standard of 126 MPN/100 mL established by the EPA [16]. Concentrations of E. coli observed in wetland inflow and outflow were within the range of concentrations reported in prior studies conducted in central and eastern North Carolina where streams draining watersheds with residential development were sampled. For example, a study [45] was conducted on two small (201 to 440 ha) watersheds in Pitt County, NC, and during baseflow conditions, the streams had geometric mean concentrations of E. coli of 2296 and 389 MPN/100 mL, respectively [45]. In another study [46] of a different watershed in Pitt County, Town Creek had a geometric mean of 126 MPN/100 mL E. coli during baseflow and 639 MPN/100 mL during stormflow. Geometric mean concentrations of E. coli ranging from 89 to 792 MPN/100 mL in nine watersheds in Durham, NC, were reported [47], with eight of the nine watersheds exceeding the EPA geometric mean standard of 126 MPN/100 mL [16]. All of these studies show that stream discharge from urbanized watersheds in eastern North Carolina often contain concentrations of E. coli that are elevated relative to EPA suggested metrics.

Influent loading of E. coli (13,661 MPN/s) was significantly greater (p = 0.006) relative to outflow exports of E. coli (5807 MPN/s) equating to a median treatment efficiency of 57% (Figure 11). These results are similar to the treatment efficiency (59%) reported [34] for a wetland constructed in Greenville, NC to treat stormwater runoff and within the range of treatment efficiencies (26% to 98%) reported in a stormwater wetland in Sydney Australia [48] and for two stormwater wetlands in Charlotte, NC (33% to 96%) [49]. In 68% of the sampling events, exports of E. coli were lower relative to imports. These data suggest that processes in the wetland such as evapotranspiration that resulted in lower outflow relative to inflow volumes were important for the reduction in E. coli loadings to Lick Creek. There were eight sampling events when export loadings of E. coli exceeded influent loadings, and 75% of these occasions occurred during the cool season (December to March) when evapotranspiration was expected to be lowest and stream stage was highest. The outflow exports of E. coli were significantly correlated (r = 0.727; p < 0.001) to the inflow loadings of E. coli; thus, pro-active measures to reduce runoff volumes and contributions of E. coli from the watershed (to the wetland) may help improve the water quality entering and exiting the wetland.

Groundwater sampled from the deep piezometer in the wetland had a median concentration of E. coli (1 MPN/100 mL) that was significantly (p = 0.011) lower relative to groundwater sampled from the shallow piezometer (816 MPN/100 mL). Therefore, the attenuation of E. coli via filtration, sorption, predation, and/or die off may have resulted in lower concentrations observed in the deeper groundwater. Reducing runoff volumes entering the wetland should allow for increased infiltration rates and filtration of E. coli and less overflow and bypass of fecal indicator bacteria removal mechanisms. Longer residence times in the wetland may also allow for increased treatment via settling and disinfection via exposure to UV [49]. Microbial source tracking analyses showed that human sources of bacteria were present in stream water entering the wetland. This was expected given the identification of malfunctioning septic systems observed during the walking survey at the start of the project. While it is likely that animal sources of bacteria were also present in the streams, having qualitative affirmation that wastewater from septic systems was influencing water quality with regards to fecal bacteria helped to steer mitigation efforts. Repairs to four septic systems and maintenance activities such as pumping the septic tanks and cleaning effluent filters were enacted to improve wastewater treatment and ultimately, the quality of water draining to the wetland. This will be discussed in more detail in a later section. The median area normalized export of E. coli leaving the wetland (242 MPN/s/ha) was within the range of watershed exports reported in studies conducted in North Carolina including Town Creek (44 MPN/s/ha to 1840 MPN/s/ha) [46] and Fork Swamp (196 MPN/s to 36,824 MPN/s) [47], which were also watersheds with predominantly residential development land use in Eastern North Carolina.

3.4. Phosphorus Concentrations and Loading

The median flow weighted concentration of TP at the inlet (0.11 mg/L) was not significantly different (p = 0.233) to the outflow concentration (0.09 mg/L) (Figure 12). In 62.5% of the sampling events, concentrations of TP for inflow exceeded those for outflow (Figure 12). The highest concentration of TP in wetland inflow occurred during May 2019 (0.87 mg/L) and on this occasion, particulate phosphorus accounted for 98% of the total phosphorus. On average, particulate phosphorus comprised 57% of the total phosphorus for wetland inflow; thus, the inflow during early May 2019 may have been influenced by a recent erosive activity. Outflow from the wetland contained an almost equal percentage of particulate (50.3%) and dissolved (49.7%) phosphorus. Median concentrations of TP in wetland inflow and outflow observed in this study were within the range of concentrations (0.08–0.13 mg/L) reported [14] for six streams monitored in Eastern NC that drained watersheds with residential development. The median phosphate concentration in wetland inflow (0.09 mg/L) was significantly greater (p = 0.01) relative to the median outflow concentration (0.03 mg/) and during just over 70% of sampling events, the outflow concentration was lower relative to inflow (Figure 13). These data suggest that the wetland was efficient at removing dissolved, reactive forms of phosphorus. Prior studies have shown that phosphate may be removed in soil via various processes including adsorption and mineral precipitation, resulting in a reduction of greater than 70% in concentration prior to infiltrating water reaching the water table [26,50]. The median concentration of phosphate in wetland inflow (0.09 mg/L) was not significantly different (p = 0.218) relative to groundwater sampled from the shallow (0.05 mg/L) piezometer. However, significant differences were observed when comparing the median concentration of phosphate in wetland inflow to groundwater sampled from the intermediate piezometer (median 0.04 mg/L; p = 0.041) and deep (median 0.001 mg/L; p < 0.001) piezometers. Hydraulic head data revealed the wetland to be a recharge area; thus, as water infiltrated and percolated through the wetland soil, phosphate removal mechanisms including adsorption, mineral precipitation and/or immobilization via plant uptake may have resulted in a reduction in phosphate concentrations. Based on the mean pH and redox potential readings from groundwater within the wetland (Table 1), the mineral vivianite may precipitate [12,51,52] and become buried, thus reducing exports of phosphorus. Immobilization of phosphorus via uptake from riparian vegetation may also have been significant. For example, in a review of multiple studies [53], the total uptake of phosphorus by riparian vegetation was reported to be between 4.5 and 12.6 kg/ha/yr and the woody storage of phosphorus ranged from 1.7 to 6.9 kg/ha/yr. A review of reference [54] showed that the mean removal of TP for floodplain wetlands was 21 kg/ha/yr and was associated with plant uptake, sorption, mineral precipitation and burial.

The median loading of TP at the inlet (0.41 mg/s) was significantly greater (p = 0.011) relative to the median outflow loading of TP (0.17 mg/s) and resulted in a TP treatment efficiency of 59% (Figure 14). The efficiency reported for the natural wetland in this study is similar to efficiencies reported for five forested wetlands receiving wastewater treatment plant effluent in Louisiana (>60% when inflow loading was <10 g/m2/yr) [29], a constructed stormwater wetland in Greenville, NC (63%) [34], and the mean (>66%) of multiple wetlands constructed as a polishing step for wastewater generated from the hospitality industry [33]. In 81% of the sampling events, inflow loading was greater than outflow exports of TP (Figure 14). The effectiveness of the wetland in reducing loads of TP was influenced by hydrology. More specifically, there was a significant inverse correlation between TP treatment efficiency of the wetland and inflow discharge (r = −0.656; p = 0.006) and influent TP loading (r = −0.521; p = 0.031). Therefore, during periods of high inflow and TP loading rates, TP removal in the wetland was lower relative to periods of lower flow and lower TP influent loading. The median inflow loading of phosphate (0.20 mg/s) was significantly (p = 0.010) greater relative to the median outflow (0.03 mg/s), and in 18 of 22 sampling events when flow was measurable, inflow loadings exceeded outflow exports (Figure 15).

3.5. Mitigation Efforts

The erosion observed along the three wetland outlets (Supplemental Figure S3) was likely being influenced by processes upstream and downstream from the wetland. More specifically, incision and lowering of the stream bed of Lick Creek created a steep “drop off” of water exiting the wetland and nick points were created. The nick points migrated inland towards the wetland after large rain events. Nick point retreat has been documented in many studies where abrupt changes in channel elevation were observed [55,56,57]. Erosion of stream beds and banks are symptoms of “urban stream syndrome” and have been reported in numerous studies [58,59,60]. Another factor contributing to erosion of the wetland was increased stormwater runoff from the watershed due to residential development and additional impervious surface. Drainage from a housing development in the northwestern portion of the watershed was directed to the wetland, thus increasing flow during storms. The combined effects of increased runoff from the watershed entering the wetland and the incision of Lick Creek contribute to erosion along the three wetland outlets. While it was beyond the scope and funding of this research to design and implement stabilization practices for Lick Creek, funding was secured to slow runoff from the drainage area (watershed) upslope from the wetland. As shown in Figure 4 and Supplemental Materials, streams/drainageways (~230 m length) were modified to increase the storage capacity of the channels and stabilize the banks and beds. Check dams were implemented within the larger channels to slow runoff and reduce the likelihood of erosion. Most of the drainageway modifications were completed towards the end of the project period due to delays related to the COVID-19 pandemic and the health status of various property owners living along the degraded drainageways; thus, a pre- and post-implementation comparison was not made. However, prior research [61] has shown that check dams can reduce runoff volumes by 17% and increase hydraulic residence times by 29% or more. Other research [62] reported that implementation of check dams in drainageways helped to reduce the total volume and dynamic flows and the stormflow transport of pollutants, such as fecal indicator bacteria [21].

Septic tank effluent typically has concentrations of E. coli and phosphorus that exceed 42,000 cfu/100 mL [63] and 5 mg/L [12,26,64], respectively. In properly maintained and functioning septic systems, phosphorus and E. coli in septic tank effluent are removed in the soil beneath drainfield trenches via processes including sorption, mineral precipitation and immobilization for phosphorus, and filtration, predation, sorption, and die off for E. coli [65]. Malfunctioning septic systems may contribute to elevated concentrations of E. coli and phosphorus in adjacent water resources [12,17,45,47,50,66]. Repairing a malfunctioning septic system may improve nearby water quality. We originally planned to monitor water quality in the streams before and after implementation of the septic system repairs and maintenance activities to assess the effectiveness of the improvements. However, like with the drainageway modifications, most of the septic repairs were completed towards the end of the project period due to limited communication with property owners during the COVID-19 pandemic and delays in obtaining permits for the repairs. Prior work has shown that repairing septic systems can reduce nutrient and bacteria loading to water resources. For example, a reduction in fecal indicator bacteria in groundwater near a septic system after the drainfield was replaced was reported in eastern North Carolina [66]. Research [67] has suggested that repairing septic drainfields that are hydraulically malfunctioning may result in an annual reduction in phosphorus loading to waterways by an estimated 1.2 kg per home. During this project, septic tanks from 15 different properties were pumped. Discussions with the homeowners regarding the thickness of the solids layers and the need for routine maintenance such as pumping the tanks was communicated. Waiting too long to pump a septic tank may increase the likelihood of hydraulic malfunctions. For example, excessive buildup of solids reduces the hydraulic residence time of wastewater in the tank, thus reducing settling/sedimentation and increasing the likelihood of clogging the effluent filter or pipe. Another benefit of pumping septage from a typical 3780-liter-capacity septic tank is that it may result in the removal of up to 1 kg of phosphorus [64] from the tank and the watershed.

4. Conclusions

The primary goal of this study was to determine if a natural wetland (1.5 ha surface area) receiving drainage from a small watershed (24 ha) was providing ecosystem services with regards to phosphorus and E. coli removal. The median loadings of the total phosphorus and E. coli exiting the wetland were 59% and 57% lower, respectively, relative to the inflow loadings. The wetland was providing pollutant removal services, especially during the warm, growing season when outflow was lower relative to inflow. The nutrient and bacteria treatment efficiency of the wetland decreased during the winter and during high-flow periods associated with heavy rain. Efforts to slow runoff entering the wetland during storms such as the enlargement and stabilization of stream banks and the installation of check dams were implemented in the tributary streams. These practices have been shown in other work [61] to reduce outflow by 17%, which may result in load reduction improvements of 0.03 mg/s of total phosphorus and 988 MPN/s of E. coli for this watershed. Despite the efforts to better manage runoff in the catchment, ecosystem services provided by the wetland may be at risk due to erosion of the wetland outlets which are migrating. More work including structural stabilization of the wetland outlets along with the implementation of additional stormwater control measures in the catchment and within the tributary streams is required to prevent erosion of the wetland.

Several malfunctioning septic systems identified in the catchment were contributing to the elevated concentrations of E. coli and phosphorus entering the wetland. Four septic systems were repaired, and 15 septic tanks were pumped to help improve the performance of the systems. These efforts likely resulted in a one-time removal of 15 kg of phosphorus from the catchment (septage pumping) [64] and an annual reduction of 4.8 kg in phosphorus associated with the repairs of the malfunctioning septic systems [67]. Projects like this require gaining the trust of the communities and identifying where the practices may provide the most influence. Sometimes it takes years to establish relationships with property owners, especially if they are concerned that they could be held financially responsible for fixing malfunctioning septic systems or degraded drainageways. Securing funds that cover the costs of the septic repairs and drainageway modifications and establishing relationships with property owners are vital to the success of mitigation projects. Wetland conservation and restoration efforts can provide valuable water quality benefits in this and other urbanizing watersheds that receive nonpoint source pollutants. Monitoring water quality and flow in watersheds before and after the implementation of stormwater control measures and septic system improvements is suggested to improve our understanding of the effectiveness of these efforts.