Introduction

Quantifying sediment fluxes through rivers and streams is essential to addressing a variety of environmental, economic, and aesthetic issues. Suspended sediment—directly or indirectly—causes limited productivity, reductions in populations, and changes in community composition of aquatic organisms (Henley et al. 2000; Bilotta and Brazier 2008; Jones et al. 2012); facilitates the flux of adsorbed nutrients, metals, and other chemicals through stream systems (Taylor and Owens 2009); fills in reservoirs (Madadi 2022); and reduces public perception of water quality and recreational suitability (Angradi et al. 2018). Several of these issues have occurred in Enka Lake, a privately-owned lake in Western North Carolina, United States. Turbid water, nutrients, E. coli, and sediment buildup from contributing tributaries have affected the recreational and aesthetic value of the lake to the community and its residents. A dredging program was carried out in 2016, and a sediment management study was commissioned in 2017 to consider future dredging frequencies, engineered forebays to capture sediment before reaching the lake, and stormwater best management practices (BMPs) to address the source of sediments in the watershed. This study was initiated to help evaluate these sediment management options by measuring streamflow and sediment loads from the two main tributaries into Enka Lake.

Stream sediment fluxes can be estimated using sediment rating curves (SRCs), which describe empirical relationships between stream discharge (Q, m3/s) and total suspended solids (TSS, mg/L) (Asselman 2000; Horowitz 2003; Vercruysse et al. 2017; Efthimiou 2019). In a common application, continuous stream stage measurements are obtained using a stage recorder and converted to discharge using a stream rating curve (stage-discharge relationship) (Turnipseed and Sauer 2010). TSS is determined through manual sampling and analysis of discrete samples or by measuring a surrogate (e.g., turbidity) and converting to TSS using a site-specific conversion factor or calibration curve (Horowitz 2003). SRCs have been used to quantify suspended sediment in streams all over the world (e.g., Horowitz et al. 2008; Hicks et al. 2011; Duvert et al. 2011; Harrington and Harrington 2013; Dominic et al. 2015; Leta et al. 2023). This paper describes an evaluation of sediment fluxes into Enka Lake and the management implications for improving water quality.

Site description

Enka Lake is a 25-ha (62-acre) reservoir in the Southern Appalachians of Western North Carolina, United States (Fig. 1). Lake level is controlled by a dam, which was built in 1933 to supply water for a nearby industrial facility. In 2001, the lake and approximately 400 hectares (1000 acres) were sold and developed into the Biltmore Lake residential community; the lake is now used primarily for recreation, with a beach, clubhouse, boathouse, and a walking/biking path.

Fig. 1
figure 1

Enka Lake watershed map, showing monitoring locations and land use

Full size image

The lake’s watershed encompasses 15.38 km2, of which 12.4 km2 (81%) is part of the Bill Moore Creek (BMC) and Wise Branch (WB) subwatersheds south of the lake (Fig. 1). The BMC watershed is primarily forested and includes part of Pisgah National Forest; the remaining land uses include low-density residential development and limited pastureland (Table 1). The WB subwatershed is much smaller than the BMC subwatershed, but has a higher percentage of agricultural and residential land use. Water flowing through both BMC and WB is very clear (transparent) during baseflow conditions but becomes turbid during and following large precipitation events. As stormflow enters the lake, coarse sediments are deposited at the inlets, while a cloudy front of finer sediments can be observed moving across the lake and toward the dam. Approximately 1300 m3 (1700 cubic yards) of sediment were dredged from three lake inlets in 2016, of which over 900 m3 (1200 cubic yards) were removed from the southern-most inlet fed by BMC and WB. The dredging project required lowering lake levels, stranding the beach, dock, and boathouse. Residents raised questions about the expected frequency (and cost) of future dredging projects and whether other options, such as engineered forebays or other stormwater best management practices (BMPs) upstream of the lake, would be preferable to address the source of sediments in the watershed.

Table 1 Land use in the Bill Moore Creek and Wise Branch subwatersheds
Full size table

Materials and methods

Stage and discharge

Stream and lake stages were measured via stilling wells in BMC, WB, and near the dam outflow in Enka Lake (Fig. 1). Pressure transducers (Onset HOBO U-20, Onset Corporation, Bourne, MA) were installed in slotted 5.08 cm-diameter PVC pipes to record water-level changes at 15-min increments; an additional pressure transducer was deployed above the water level to account for changes in atmospheric pressure. Stream discharge was manually calculated on several occasions at BMC (6) and WB (7) using measured cross sections, velocities (Global Water Flow Probe, Global Water Instrumentation Inc., Gold River, CA), and the USGS velocity-area method (Turnipseed and Sauer 2010) and used to construct stream rating curves. The outflow over the Enka Lake Dam was calculated according to the Francis formula for a rectangular weir with end contractions (USBR 1997):

$${\text{Q }} = { 3}.{33 }\left( {{\text{L }}{-} \, 0.{\text{2H}}} \right){\text{H}}^{{{1}.{5}}}$$

where Q is discharge (in cfs), L is the weir opening length (in ft), and H is the water level above the weir (in ft). A neighborhood resident measured 26.7 m (87.66 feet) for L, while H was calculated using water-level data from the stilling well in the lake.

Turbidity and suspended sediment

Biltmore Lake neighborhood residents were instrumental in monitoring streamflow conditions and collecting samples during stormflow events. Cleaned, empty 500-mL HDPE bottles were distributed to a team of volunteer samplers ahead of rainfall events. The sampling team communicated via group text ahead of predicted rainfall and in real time during “pop-up” thunderstorms to organize who could collect samples during and after the storm. Altogether, Biltmore Lake residents and UNC Asheville researchers collected nearly 500 grab samples in the Enka Lake watershed across all seasons and flow conditions (baseflow through stormflow) between October 2017 and October 2020, including 130 and 46 at the BMC and WB stream gauges, respectively.

Samples were refrigerated in the neighborhood clubhouse and/or at UNC Asheville until analysis. All samples were vigorously shaken to resuspend any settled sediments and analyzed for turbidity using a turbidimeter (Hach Company, USA), while a subset of 42 samples were randomly chosen across a range of turbidity values and analyzed for total suspended solids (TSS). These samples were filtered through 1.5-µm TSS-grade glass microfiber filters (I.W. Tremont Co., USA), which were then dried at 95 °C and weighed. The volume of filtered stormwater was recorded, the filtered sediment mass was calculated as the difference between the initial and final filter mass, and total suspended solids (TSS) were reported as mg/L (mg sediment/L water). Streamflow and TSS data from BMC (130 data pairs) and WB (80 data pairs) were then used to construct sediment rating curves. Since stream stage was measured (and stream discharge calculated) every 15 min, the sediment rating curves were then used to determine the sediment flux during each 15-min interval on record.

High-frequency stormflow sampling

The majority of stream sediment load is typically carried during short, discrete stormflow events (e.g., Coynel et al. 2004; Mano et al. 2009; Duvert et al. 2011). SRCs often underestimate high-flow sediment fluxes because the majority of sediment-discharge pairs used to construct the curves are from low-flow conditions (Efthimiou 2019). High-frequency sampling during these high-discharge, high-sediment conditions can help produce more representative SRCs and make better estimates of long-term sediment fluxes (Duvert et al. 2011). Duvert et al. (2011) suggest an hourly sampling interval for stormflow in mountainous watersheds ranging from 3 to 12 km2, which encompasses the size of the BMC and WB watersheds. Therefore, we manually collected samples approximately hourly for the duration of three storms (before, during, and after) on 2/6/2020, 5/20/2020, 9/17/2020. These dates were chosen ahead of predicted storms in three different seasons when Biltmore Lake and UNCA samplers would be available for the entire storm event.

Results and discussion

Discharge

Rating curves from BMC and WB (Fig. 2), along with stage data from the stilling wells (Supplementary Materials 1), were used to calculate discharge for the entire study period (Fig. 3); short data gaps exist when dataloggers were either moved between sites in the watershed or sent to the manufacturer for battery replacement (e.g. January 2020). Since manual discharge measurements were recorded during the third-largest stormflow event on April 19, 2019, extrapolations beyond the rating curve were only necessary for two slightly larger events on 10/29/2020 and 4/13/2020. Baseflow in BMC ranged from about 0.05 to 0.10 m3/s during late summer and fall to 0.30 m3/s in winter and spring, with seasonal variations likely due to changing evapotranspiration in the watershed. The greatest recorded stormflow through BMC was 6.14 m3/s (10/29/2020), with two additional peaks above 5.0 m3/s (5.47 m3/s on 4/13/2020 and 5.12 m3/s on 4/19/2019) and 35 peaks above 1.0 m3/s in just over 2 years. Baseflow in WB was typically between 0.05 and 0.10 m3/s, with a peak stormflow of 2.20 m3/s (10/29/2020). The median lake outflow over the dam was 0.31 m3/s, reaching as high as 13.4 m3/s (4/19/2019) and as low as 0.00 m3/s (no outflow) on summer days when lake evaporation exceeded stream inflow. The combined flow from BMC and WB provided the majority of streamflow into Enka Lake during baseflow conditions. However, direct rainfall, other minor tributaries, and ungauged direct stormwater runoff from the Biltmore Lake residential community provided a significant source of water during high-flow events (Fig. 4).

Fig. 2
figure 2

Stream rating curves for Bill Moore Creek (BMC) and Wise Branch (WB)

Full size image
Fig. 3
figure 3

Discharge through Bill Moore Creek (BMC), Wise Branch (WB), and over the Enka Lake dam, 2018–2020

Full size image
Fig. 4
figure 4

Discharge through Bill Moore Creek (BMC), Wise Branch (WB) and over the Enka Lake dam, beginning at 00:00 on a 4/19/2019, b 2/4/2020, and c 5/17/2020

Full size image

Sediment rating curves (SRCs)

Water flowing into Enka Lake during baseflow conditions was typically very clear, with turbidity ranging from 1 to 20 NTU. During stormflow conditions, however, turbidity frequently reached several hundred NTU (with the highest readings well above 1000 NTU). The 42 paired samples where both turbidity and TSS were measured (Fig. 5) showed no distinguishable differences between samples collected in the BMC, WB, or other Enka Lake tributaries; the best-fit relationship was then applied to remaining samples where only turbidity measurements were recorded (see Supplementary Materials 2).

Fig. 5
figure 5

Relationship between total suspended solids (TSS) and turbidity in the Enka Lake watershed

Full size image

SRCs were constructed using paired stream discharge and TSS data from BMC (130 samples) and WB (80 samples) (Fig. 6). The log-transformed linear slopes in this study (1.11 and 0.77 for BMC and WB, respectively) were similar to those found near Atlanta, Georgia (1.1–1.2, Horowitz et al. 2008); in the Mississippi River (0.9–1.3, Horowitz 2003); and across five different streams in the Netherlands (0.32–1.21, mean = 0.93, Asselman 2000). Asselman (2000) noted that although they don’t have any physical meaning, greater sediment rating curve slopes (BMC) correspond with higher erosive river power, and higher intercepts (WB) indicate more easily eroded material in the watershed.

Fig. 6
figure 6

Sediment rating curves for a Bill Moore Creek and b Wise Branch

Full size image

There was considerable scatter above and below both SRC regression lines, which may be attributed to uncertainty in the turbidity-TSS relationship, variations in sediment supply, recent previous storms, or seasonal factors such as vegetation and ground cover (Asselman 2000). Nevertheless, the log-transformed r2 found in this study (0.41–0.42) were similar to those found in the previously-mentioned Dutch study, which presented r2 values ranging from 0.11 to 0.49 with a mean of 0.37 (Asselman 2000). Horowitz (2003) noted that the r2 regression statistic is a poor indicator of whether a sediment rating curve will produce a reasonable estimate of annual sediment fluxes. For example, two sediment rating curves in their study both produced excellent estimates of sediment load (within 5% of measured values) despite very different r2 values of 0.23 and 0.83 (Horowitz 2003).

Storm sediment fluxes

The sediment load-discharge relationships displayed clockwise hysteresis for all storm events (Fig. 7), which occurs when sediment concentration peaks with or before peak discharge and leaves the stream relatively sediment-free on the falling limb of the hydrograph (Dominic et al. 2015). This can be explained by readily available streambed or streambank sediment being quickly transported during the rising limb of the hydrograph (Fang et al. 2011; Keesstra et al. 2019). Reworked streambed sediment (both erosion and deposition) were observed following several large storm events.

Fig. 7
figure 7

The sediment-discharge relationship for all storm events exhibited clockwise hysteresis loops

Full size image

Predicted sediment fluxes for the three high-frequency sampling events were mostly consistent with measured values (Fig. 8), but they underestimated large peak storm fluxes on 2/6/2020 and overestimated sediment fluxes on the falling limbs of all storm hydrographs. Both results—underestimating high fluxes and overestimating low fluxes—are common when using SRCs (e.g., Asselman 2000; Efthimiou 2019) but have been shown in many studies to balance out over longer periods of time. For example, Horowitz (2003) reported that sediment load estimates over quarterly time frames or longer were generally within 15–20% of measured values, while Asselman (2000) reported that long-term sediment flux rates differed by less than 20% from measured values at most locations.

Fig. 8
figure 8

Measured sediment loads in Bill Moore Creek (orange) and Wise Branch (blue) compared to model estimates (gray) for three storms

Full size image

Total sediment loads

Sediment rating curves were used to estimate total annual sediment loads of about 540 tonnes/year, or 300–420 m3/year (based on a bulk density of 1.3–1.8 tonnes/m3) into Enka Lake from BMC and WB. The actual sediment flux is most likely higher, though, as converting log-transformed data can result in a bias that underestimates total sediment loads (Duan 1983). In a previous study, Horowitz et al. (2001) applied a bias correction factor only when it was necessary to match predicted and actual flux estimates. Given both the general agreement between (uncorrected) predicted and measured sediment loads for the three storm events (Fig. 8) and the mathematical basis for applying a correction factor, we include both estimates here (Table 2). To quantify potential bias, we used the Rating Curve Analysis tool in HEC-RAS (HEC 2021) to calculate Duan Factors of 1.80 for BMC and 1.44 for WB. The resulting sediment load estimates after bias correction were about 900 tonnes/year, or 500–700 m3/year. To put this into perspective, the 2016 dredging project removed approximately 920 m3 (1200 cubic yards) from the inlet where BMC and WB flow into Enka Lake from the south, or around just 1–3 years of sediment according to our estimation. Given that the southern inlet to the lake has not filled in every few years (or sooner), the vast majority of sediment must be flowing through the inlet and further out into the lake. This conclusion would not be surprising to area residents, who have witnessed cloudy sediment fronts migrating across the lake after large storm events.

Table 2 Total sediment load estimates to Enka Lake from Bill Moore Creek and Wise Branch
Full size table

Annual sediment load estimates were greatly impacted by the total amount of rainfall/streamflow in a given year, and also whether that rainfall occurred primarily as short, intense storms. For example, we estimated much greater sediment loads in BMC in 2020 than 2019 (824 tonnes vs 540 tonnes, bias corrected), correlating with greater annual precipitation—1.68 m vs. 1.37 m, respectively (Thornton et al. 2022)—and one additional peak flow above 5m3/s (4/13 and 10/29 in 2020 vs. 4/19 in 2019). Several previous studies noted that a large portion (e.g., 35–80%) of annual suspended sediment loads can be transported in relatively short, discrete periods of time corresponding to extreme storm events. (e.g., Asselman 2000; Meybeck et al. 2003; Mano et al. 2009). In this study, we estimated roughly 13% of the total sediment flux from BMC and WB during the study period occurred during just three storm events (see Fig. 3): 4/19/2019 (5.0%), 4/13/2020 (4.8%), and 10/29/2020 (3.7%). So while the annual sediment loads presented here can be used to help evaluate sediment management strategies, it should be done with the understanding that a particularly wet year (or one with a couple particularly powerful storms) could yield a much greater sediment load than an average or drier year.

Sediment management implications

The overall motivation for monitoring streamflow and sediment fluxes into Enka Lake was to evaluate potential management options: dredging, engineered forebays, and BMPs in the watershed. Based on observations made during major storm events in December 2018 and April 2019, streamflow sediment primarily originates in low-density rural residential areas: dirt roads, gravel driveways, and poorly-vegetated yards. As these are privately-owned, individual residences located upstream from Enka Lake, it is unlikely that the Biltmore Lake neighborhood association can enact BMPs to address the introduction of these sediments into BMC or WB. The annual sediment load estimates presented here suggest future mechanical dredging would be required every few years to maintain inlet depth and surface area where BMC and WB flow into Enka Lake. This not only represents a significant recurring financial cost to the homeowners’ association, but also affects the recreational and aesthetic value of the lake while water levels are lowered. Beyond the cost and inconvenience of dredging, undesirable sediment surges would still affect water clarity throughout the lake following large stormflows. Another option considered by the Biltmore Lake Association was installing sediment forebays in neighborhood-owned property upstream from Enka Lake. These constructed reservoirs would allow sedimentation within the forebay, where accumulated sediment could be removed (dredged) with easier access and without affecting lake levels. Estimates of stormflows (at least 6.7 m3/s combined from BMC and WB) and sediment loads (at least 540 tonnes/year) will be useful for sizing sediment forebays and determining cleanout frequencies.

Conclusions

Reasonable estimates of stormflow and sediment loads are essential for planning dredging frequencies or designing sediment retention structures. In this study, we measured streamflow and estimated sediment fluxes over a 2-year period in two tributaries feeding a Southern Appalachian mountain reservoir. Most sediment transport occurred during large storm events, when the sediment rating curve tended to somewhat underestimate peak sediment fluxes but overestimate fluxes on the falling limb of the hydrograph. Nevertheless, high-frequency sampling during three storm events demonstrated general agreement between predicted and actual total storm sediment loads. Installing sediment forebays or other retention structures is likely the preferred sediment management strategy moving forward (over a regular dredging schedule) since they can be maintained more easily without affecting lake levels. The monitoring protocols used in the study—including the help of community-based volunteer samplers—provide a model that could be used at other sites in the Southern Appalachians or beyond.