The effects of disproportional load contributions on quantifying vegetated filter strip sediment trapping efficiencies

Abstract

Vegetated filter strips (VFSs) are a best management practice (BMP) commonly implemented adjacent to row-cropped fields to trap overland transport of sediment and other constituents present in agricultural runoff. Although they have been widely adopted, insufficient data exist to understand their short and long-term effectiveness. High inter-event variability in performance has been observed, yet the majority of studies report average removal efficiencies over observed or simulated events, ignoring the disproportional effects of loads into and out of VFSs over longer periods of time. We argue that due to positively correlated sediment concentration-discharge relationships, disproportional contribution of runoff events transporting sediment over the course of a year (i.e., temporal inequality), decreased performance with increasing flow rates, and effects of antecedent moisture condition, VFS removal efficiencies over annual time scales may be significantly lower than reported per-event averages. By applying a stochastic approach, we investigated the extent of disparity between reporting average efficiencies from each runoff event over the course of 1 year versus the total annual load reduction. Additionally, we examined the effects of soil texture, concentration-discharge relationship, and VFS slope in contributing to this disparity, with the goal of revealing potential errors that may be incurred by ignoring the effects of temporal inequality in quantifying VFS performance. Simulation results suggest that ignoring temporal inequality can lead to overestimation of annual performance by as little as < 2% and to as much as > 20%, with the greatest disparities observed for soils with high clay content.

Introduction

The vegetated filter strip (VFS) is a widely adopted agricultural best management practice (BMP) due to its demonstrated ability to reduce surface runoff volumes and contaminant loads (e.g., sediment, nutrients, pesticides). Sediment removal efficiencies of VFSs have been studied for the past five decades, with reported values typically ranging from ~ 50% to greater than 90% (Abu-Zreig et al. 2001; Dorioz et al. 2006; Gharabaghi et al. 2002; Hayes et al. 1984; Liu et al. 2008; Magette et al. 1987, 1989; Wilson 1967). However, significant ranges in removal efficiencies have also been observed, with reported efficiencies less than 30% (Gumiere et al. 2011; Mekonnen et al. 2015; Yuan et al. 2009) and as low as 5% (Arora et al. 2010; Reichenberger et al. 2007).

Despite widespread promotion and adoption of VFSs as structural agricultural BMPs, current knowledge is insufficient to appropriately understand performance over both short and long term time scales (Liu et al. 2017). Current data in the literature report efficiencies for natural rainfall events for a short period of time, with no experimental durations lasting longer than 4 years and most often for less than 20 rainfall events over a two-year period (Arora et al. 2010; Helmers et al. 2011). Given that implementation of a VFS on a landscape is typically expected to be on the order of 25 years if a landowner has received federal aid (in the United States), these study periods do not represent the effectiveness of a VFS over the expected employment, as over time, factors such as the build-up of sediment and convergence of flow pathways (i.e., development of rills) can reduce their performance (Hénault-Ethier et al. 2017).

Furthermore, the disproportional contribution of runoff events in time (i.e., temporal inequality) to constituent loads entering and leaving a VFS is not typically considered when removal efficiencies for VFSs are reported. Instead, data from individual observed or simulated events are used to determine long-term average values (Arora et al. 2010; Liu et al. 2017). However, the sparsity of such data sets often misrepresents effects of temporal inequality. Various studies have shown that the vast majority of discharge and solutes are transported during short periods of time, typically during high-flow events. This temporal inequality has been widely recognized in contaminant transport from agricultural catchments. Richards et al. (2001) found that over a 20-year period (1975–1995), 79–92% of suspended sediment load exports from four dominantly agricultural catchments in the Lake Erie Basin (88–16,400 km²) occurred during storm runoff periods that accounted for ~ 1/3 of the time period. For dominantly agricultural catchments in Illinois, Royer et al. (2006) found that almost all nutrient loads were exported when flowrate exceeded median discharge. Similar patterns have been observed at the scale of the Chesapeake Bay watershed, with annual loads of nutrients and sediments discharged to the Bay correlated to the amount of precipitation and corresponding runoff (USGS 2011).

This temporal inequality is a direct result of positive concentration–discharge (C–Q) relationships typically exhibited by sediment. These relationships are generally given by the empirical relationship C = aQ^b, which has been used since as early as 1969 (e.g., Johnson et al. 1969; Haygarth et al. 2004; Vogel et al. 2005) to describe the relationship between concentration and flow. The parameter b can be interpreted as the strength of the dilution or accretion relationship between concentration and discharge (i.e., the slope of the linear log C–log Q relationship) and a is an empirical parameter (i.e., the y-intercept for the linear log C–log Q relationship). The higher the value of b, the higher the constituent concentrations when flow rates are high, and therefore the higher the temporal inequality (Gall et al. 2013; Jawitz and Mitchell 2011).

Additionally, during higher-flow events, runoff velocity increases and the mean residence time of water moving through the VFS shortens. However, as flow depth increases, suspended sediments have further to travel (in the vertical direction towards the ground) before they have settled, leading to decreased removal efficiencies of the VFS. Given the strong temporal inequality and sediment removal efficiencies decreasing with increasing flow rate (i.e., runoff depth), we hypothesize that VFS removal efficiency at an annual time scale may be significantly lower than the per-event averages typically reported in the literature. This disparity will become exaggerated when a data set does not sufficiently capture the multi-year, combined distribution of sediment concentration and runoff entering a VFS from an upgradient contributing area. Performance of a BMP at an annual time scale is critical for determining whether or not adoption of a suite of BMPs across a watershed is sufficient to meet long-term sediment and nutrient load reduction goals, such as those established for the Chesapeake Bay.

A literature review by Liu et al. (2017) of the effectiveness of BMPs found that most water quality models do not consider variability over time in removal efficiency, but instead assume a constant rate per BMP. Given that BMP effectiveness naturally fluctuates over time as a function of a wide variety of parameters including hydrologic forcing (i.e., intensity and duration of rainfall events), antecedent moisture conditions, season, maintenance activities, and vegetation conditions, the use of one value to represent effectiveness is misleading and may overestimate true performance over an extended period. However, insufficient data exist to appropriately incorporate these variables and all their interactions into model representations of BMPs. Existing complex VFS models such as VFSMOD (Munoz-Capena et al. 1992, 1993, 1999), APEX (Steglich and Williams 2013), PRZM-BUFF (Waterborne Environmental, Inc.), and REMM (Altier et al. 2002) are design-storm, field-scale models that produce removal efficiency results for single events of interest (e.g., 1 year-24 h design storm). Therefore, they cannot evaluate how temporal inequalities of contaminant loading throughout a year would impact overall performance, or how the effects of antecedent condition may influence simulated event results. These differences, if significant, could have important water quality implications for decision-making, given the multi-year time scales over which water quality and load reduction goals are often set.

Here, we use a two-component model that couples a field-scale water balance model to an existing process-based VFS sediment-trapping model (Haan et al. 1994) to test our hypothesis that the simple average of event-specific removal efficiencies is significantly different than a load-weighted average, particularly when evaluated at an annual scale. We expect this difference to be due to the combined effects of temporal inequality and lower performance of VFSs with greater storm event size and, thus, driven by soil texture, VFS slope, and the value of b in the C–Q relationship. Thus, we use a stochastic modeling approach to investigate how the annual load-weighted reduction differs from the mean event-based sediment removal efficiency as a function of (1) VFS slope; (2) strength of the C–Q relationship (i.e., b value); and (3) soil texture. The results have important implications for VFS representation in water quality models and the role that VFSs can play in achieving load reduction goals.

Methodology

The modeling approach consisted of coupling a field-scale water balance and sediment transport model to a previously published VFS model (Fig. 1). Hydro-climatic parameters (Table 1) representing typical farm conditions in Lancaster County, Pennsylvania, which is a hot spot for sediment and nutrient pollution to the Chesapeake Bay (USGS 2011), were used to create stochastic rainfall records. These stochastic rainfall records were input into the field-scale water balance and sediment transport component, which simulated daily soil moisture to drive surface runoff and suspended sediment loadings from a row-cropped field. Using particle size distribution (PSD) of the incoming suspended sediment (see Supplemental Material, Table S1) and physical characteristics of the VFS (Table 2), the VFS component then simulates daily sediment removal efficiencies.

Fig. 1

Schematic of the two-component model: a field-scale water balance and total suspended sediment (TSS) transport model coupled to a vegetated filter strip (VFS) model

Table 1 Hydro-climatic parameter values for Lancaster County, Pennsylvania, USA, used to create stochastic rainfall records

Table 2 Physical parameters used to model water and suspended sediment movement through the VFS

The model was run at a daily time step for one-year periods 1000 times for each of all 192 combinations of the following parameters: soil texture (all 12 textures), VFS slope (1, 2, 5, and 8%), and the value of b in the sediment C–Q relationship (0.5, 1, 1.5, and 2.5). Daily removal efficiencies were used to enable a comparison of simple-average event-based reductions versus annual load-weighted reductions.

Field-scale water balance and sediment transport model

The field-scale model was driven by stochastic daily rainfall with soil moisture storage (S) in the root zone calculated at a daily time step. Stochastic rainfall simulations, evapotranspiration calculations, and triggers for surface runoff are discussed below.

Rainfall

Daily rainfall, P [mm], was simulated as a marked Poisson process, in which depth (d, mm) and inter-arrival frequency (i.e., the time between rainfall events; λ, d⁻¹) were simulated randomly from independent exponential distributions (Rodriguez-Iturbe et al. 1999). Input parameters, d and λ, for the Poisson process were based on 60 years (1955–2015) of daily-scale data from the National Climatic Data Center (ncdc.noaa.gov) from the study region (Station Number USC00364763) (Table 1). All rain events are modeled as no longer than 1 day. Days with consecutive rainfall are considered individual, but consecutive, daily events.

Evapotranspiration

Potential evapotranspiration (PET) was calculated with a modified Blaney–Criddle model (Sammis et al. 1982). An advantage of the Blaney–Criddle model compared to other potential evapotranspiration models is that it only requires average monthly air temperature to be calibrated, and can therefore be easily applied to any location with observed air temperature data. We simulated daily mean temperatures, T_avg [°C], using the first harmonic of a Fourier function, as described by Grimenes and Nissen (2004):

$$T_{avg} (t) = A + Bsin\left( {\frac{2\pi }{365}\left( {t + 273.75 - D_{min} } \right)} \right)$$

(1)

where t is time [d], A is the mean annual temperature [°C], B is the half amplitude of the sine variation in air temperature [°C], and D_min is the day of the year with the minimum daily mean air temperature (see model parameter values in Table 1).

Seasonality was introduced by accounting for evapotranspirative changes throughout crop growth. The basal crop coefficient, used to adjust PET over the growing season based on crop growth stage, increases from 0.15 during early growth stages to 1.15 at the peak of growth, and then decreases to 0.60 at harvest (Basu et al. 2010). We further modified the Blaney–Criddle model to account for periods of soil moisture stress that would reduce actual evapotranspiration (AET). When soil moisture was insufficient to achieve PET, then AET was set equal to the amount of water in excess of the permanent wilting point (θ_wp). Change in soil moisture storage, ∆S [mm], was calculated as:

$$\Delta S(t) = P(t) - AET(t) - Q(t)$$

(2)

where Q(t) is the surface runoff depth (discussed below). Values for θ_wp vary by soil texture and are provided in the Supplemental Material (Table S1).

Surface runoff

Surface runoff for each event is driven by antecedent soil moisture and the event’s 2-hour rainfall intensity. This provides for representation of both Hortonian and Dunne runoff (i.e., intensity exceedance or storage capacity exceedance, respectively) as follows.

When intensity of a simulated rainfall event exceeds saturated hydraulic conductivity (K_sat, mm d⁻¹), soil water is recharged to the lesser of saturation capacity or current soil moisture deficit. Any remaining rainfall becomes surface runoff and is calculated in depth, Q [mm] for time t [i.e., a given day] as:

$$Q(t) = P(t) - \hbox{min} \left( {K_{sat} \delta (t), \left( {\theta_{sat} - \theta (t)} \right)Z_{RZ} } \right)$$

(3)

where δ(t) is event duration [h] during time step t, θ_sat is soil moisture content at saturation, θ(t) is soil moisture content at time t, and Z_RZ is root zone depth [m].

When intensity of a simulated rainfall event does not exceed K_sat, surface runoff occurs whenever the current soil moisture deficit is satisfied:

$$Q(t) = P(t) - { \hbox{max} }\left( {0, \left( {\theta_{sat} - \theta (t)} \right)Z_{RZ} } \right)$$

(4)

An example time series of rainfall and resulting simulated surface runoff is shown in the Supplemental Material (Figure S1).

Sediment transport

Suspended sediment loads from the field were simulated via an empirical C–Q relationship: C = aQ^b (Haygarth et al. 2004; Vogel et al. 2005) with a held constant at 1 since it used to account for unit corrections. Higher values of b increase the temporal inequality of constituent loads (Gall et al. 2013). Thus, the value of b was adjusted, to explore how the strength of the C–Q relationship affected simulation results. Simulations were run with values of b of 0.5, 1, 1.5, and 2.5, as Forstner and Muller (1968) found that the C–Q relationship for sediment is typically positive, with values of b up to 2.5, and Walling and Webb (1982) reported that the values of b are typically between 1 and 2. The sediment load leaving the row-cropped field and entering the VFS, L_in (Fig. 1), was calculated as L_in = CQ = aQ^b+1.

Vegetated filter strip model

Water flow through VFS

Flow through the VFS on a given daily time through the VFS was calculated as:

$$q(t) = \frac{{d_{f}^{5/3} }}{n}\left( {\frac{{b_{g} }}{{2d_{f} + b_{g} }}} \right)^{2/3} S_{0}^{1/2}$$

(5)

where q(t) is the surface runoff rate [m²/s] calculated as the incoming runoff depth, Q, divided by the VFS length, L [m] (i.e., the length of the field), d_f is the depth of water flowing through the VFS [m], b_g, is the space between grass blades [mm], and S₀ is the VFS slope [m/m] (Haan et al. 1994). Kentucky Bluegrass, a common vegetation type implemented in agricultural VFSs, was used in all simulations. All VFS parameters are given in Table 2. Four slopes were simulated: 1, 2, 5, and 8%. The maximum allowable slope for VFSs in Pennsylvania is 8% (PA DEP 2006). A VFS width W of 10.7 m was used, as the minimum allowable width for receiving cost-share funds through the Natural Resources Conservation Service (NRCS) (NRCS 2011).

Sediment trapping efficiency

Sediment trapping efficiency within the VFS is a function of the particle size distribution (PSD) of each soil texture. Each particle size has a different trapping efficiency, with higher removal efficiencies for coarse particles (> 0.037 mm) and small removal efficiencies for fine particles (< 0.004 mm). A dimensionless number known as the fall number (N_f) is used to determine trapping efficiency of each particle size (Blanco-Canqui et al. 2004; Haan et al. 1994):

$$N_{f} = \frac{{V_{S} L}}{{V_{m} d_{f} }}$$

(6)

where V_S is the settling velocity of each particle size (V_C, V_M, and V_F for course, medium, and fine particles given in Table 2), V_m [m/s] is the Manning’s flow velocity through the VFS, d_f, is the depth of water flowing through the VFS (calculated using Eq. 5).

The trapping efficiency, F_d [–] for each sediment particle size is given as (Haan et al. 1994):

$$F_{d} = { \exp }\left( { - \,1.05*10^{ - 3} R_{e}^{0.82} N_{f}^{ - 0.91} } \right)$$

(7)

where R_e is Reynold’s number. Total sediment removal efficiency, F_t [–] is calculated as a weighted average of removal efficiency based on fractions of each particle size in the PSD (Haan et al. 1994):

$$F_{t} = f_{c} \cdot F_{d,C} + f_{M} \cdot F_{d,M} + f_{F} \cdot F_{d,F}$$

(8)

where f_C, f_M, and f_M are the fractions of coarse, medium, and fine particles in the PSD representative of each soil texture (Table S1).

Analysis methods

To quantify the effects of temporal inequality on the disparity between reporting sediment removal efficiency as simple event-specific means versus annual load-weighted reductions, we calculated sediment trapping efficiency using two different calculation methods. The first method represents the way that VFS efficiencies are typically reported in the literature, as the average efficiency of all events observed or simulated over a study period. We refer to this calculation method as the annual per-event average (APEA), which equally weights the sediment trapping efficiency, F_t, of each event, i, over the course of a year, as:

$$APEA = \frac{{\mathop \sum \nolimits_{i = 1}^{n} F_{t,i} }}{n}$$

(9)

where n is the total number of surface runoff events in a stochastically simulated year.

The second method represents the actual overall performance of the VFS on an annual time scale and includes the effects of temporal inequality. We refer to this calculation method as the annual load reduction (ALR), which was calculated as the total volume of sediment that the VFS removes over the course of a stochastically simulated year, as:

$$ALR = \frac{{\mathop \sum \nolimits_{i = 1}^{n} F_{t,i} L_{in, i} }}{{\mathop \sum \nolimits_{i = 1}^{n} L_{in,i} }}$$

(10)

An example to highlight the potential differences between the VFS performance using Eqs. 9 and 10 is provided in the Supporting Material. Statistical significance between results from Eqs. 9 and 10, for each one-year simulation, was determined using One-Way Analysis of Variance (α = 0.01).

Results and discussion

The model was run at a daily time step for one-year periods 1000 times for each parameter set (192 different combinations of soil texture, VFS slope, and b value). The stochastic simulations were intended to examine variability in the performance of a VFS at an annual time scale rather than to investigate the performance of a VFS over a 1000-year period.

Event-specific trapping efficiencies

The simulation results suggested that the ranges of trapping efficiencies vary by soil texture (Table 3), with values ranging as high as > 95% for sand to as low as ~ 20% for clay. These results are in general agreement with the values reported in the literature. Additionally, simulation results suggest that the ranges of F_t values and the importance of antecedent moisture condition varied (Fig. 2), since for the same rainfall intensity, multiple values of F_t may be observed for a given soil texture (i.e., field). Given that the hydro-climatic drivers were the same (i.e., representative of a humid climate in Southeastern, PA, USA, see Table 1), this is likely driven by variability in surface runoff due to antecedent moisture conditions in the soil profile. A rainfall event that occurred when soil moisture was high generated a higher surface runoff depth, and therefore had a lower removal efficiency relative to a rainfall event of a similar size that generated less surface runoff due to drier soil conditions at the time of the event. This held true for rainfall intensities that were less than K_sat; however, rainfall intensities that were greater than K_sat triggered Horton-type runoff and therefore were independent of antecedent moisture condition.

Table 3 Averages and standard deviations of the annual per-event average (APEA) and annual load reduction (ALR) VFS trapping efficiency distributions for each soil texture, VFS slope, and b value in the C–Q relationship; n = 1000

Fig. 2

Event-specific trapping efficiencies, F_t, for each soil texture as a function of rainfall intensity and surface runoff rates. Intensities of storm events with return periods of 1, 2, 5, 10, 25, 50, and 100 years for the area of interest (Lancaster, PA) are indicated with red asterisks. The black lines indicate the saturated hydraulic conductivity (K_sat, Table S1) for each soil texture. Simulation results shown here were run with a VFS slope, S₀, of 2% and average values of F_t across b values

The range of trapping efficiencies simulated by the model was largest for clay and smallest for sand. The course particles in the sand led to relatively high F_t values (> 90%) for even the highest rainfall intensities and runoff rates, since the settling time for course-sized particles is fast relative to the retention time in the VFS. The soil textures with higher percentages of medium and fine particles exhibited the largest ranges of simulated trapping efficiencies. For larger events, the removal of medium-sized particles decreased to less than 50% relative to > 90% for smaller events (see Figure S2), thereby contributing to the high variability in the performance of soils with high percentages of silt. For soil textures that were dominated by clay content, simulations suggested that the removal efficiencies were high for small events, but decreased to less than 50% and as low was ~ 20% for large runoff events (Fig. 2).

By simulating soil moisture storage between rainfall events to predict surface runoff from the field, the coupled model enables rainfall depths of a given storm size to produce varying, condition-dependent depths and sediment loads into the VFS. Simulation results suggest that capturing this variability is particularly important for storms with average return periods of 1, 2, and 5 years (i.e., design storms). As shown in Fig. 2, the trapping efficiency can vary by as much as 15–20% for 1, 2, and 5-year events for silty loam, silt, sandy loam, silty clay loam, and loam. However, the simulation results suggest that for humid hydroclimatic conditions, VFS performance for sediment trapping during larger storms (i.e., 10-year return periods and higher) can be considered to be independent of antecedent moisture conditions (Fig. 2), with less than a 5% variability in F_t for these larger storms for all soil textures.

Annual load removal efficiencies

Simulation results revealed that reporting the average of the trapping efficiency for each runoff event over the course of a year (APEA, Eq. 9) overestimates the performance of the VFS relative to calculating the annual-scale load reduction (ALR, Eq. 10). However, the extent of the disparity between the APEA and ALR values varied by soil texture (Table 3). The differences were generally greater for soils with a higher fraction of fine particles, as the removal efficiency is higher for coarse and medium-sized particles (Figure S2). Based on the results of the one-way Analysis of Variance tests, the distributions of ALR values were statistically significantly different (p < 0.01) than the distributions of APEA values for each soil type and for each VFS slope (Table 3). This was true even for the soil textures with relatively small differences in the mean values, such as sand and loamy sand (Table 3), due to the ALR distributions generally having a higher coefficient of variation and a higher probability of years with lower annual performance.

These results highlight the importance of considering the effects of temporal inequality in understanding and quantifying the performance of VFSs at the field scale. Deviations from the 1:1 line, where ALR and APEA would be equal, are shown in Fig. 3. The differences in the two performance calculation methods (Eqs. 9 and 10) were less than 5% for sand and loamy sand, between 5 and 10% for sandy loam and sandy clay loam, between 10 and 15% for loam, silt loam, and clay loam, and between 15 and 20% for sandy clay, silt, silty clay loam, silty clay, and clay. Disparity in the results is shown in the soil textural triangle to aid in visual clarity of the results (Fig. 4).

Fig. 3

Deviation from the 1:1 line for the annual load reduction (ALR) versus per event average (APEA) method of assessing annual VFS performance as a function of soil texture. Error bars show standard deviations; some bars are < 2% and are imperceptible on the figure

Fig. 4

Visual representation of the average differences between annual per-event average (APEA) values and annual load reduction (ALR) values as a function of soil texture. Particle size distributions used to represent each soil texture are given in the Supplemental Material

Other parameters explored in this analysis were the value of b in the sediment C–Q relationship and the slope of the VFS, S₀. Since APEA does not capture the temporal distribution of loads over a year, the value of b does not affect APEA. Overall, increasing values of b led to lower overall annual performance (see ALR values in Table 3). However, the effects were generally minor, with an average of < 4% difference between the ALR values calculated for b = 0.5 and b = 2.5 (Table 3). The largest difference was ~ 8% for silt, where the average ALR value for b = 0.5 was 76 and 68% for b = 2.5. The effects of VFS slope were also minor, with average APEA values approximately 2% lower when the VFS slope was 8% versus when it was 1%. Therefore, it is clear that the PSD in the surface runoff (here, simulated by changing the soil texture) was the dominant factor in determining the overall performance of the VFS, with VFS slope and strength of the sediment C-Q relationship playing minor roles in the overall performance of the VFS.

Implications for VFS adoption

Simulation results suggested that VFS treating soils with the highest percentages of coarse-sized particles have the highest performance and are least impacted by effects of antecedent moisture condition. However, sandy soils may produce less erosion than soils with greater fractions of medium and fine particles. Indeed, sediment loads from clay soils were simulated to be significantly higher relative to sediment loads exported from sandy soils (Fig. 5). Therefore, a VFS with a lower trapping efficiency, due to slope, width, or vegetation characteristics, could still to reduce larger sediment loads than a VFS with a higher trapping efficiency, depending on the location of its implementation. Overall, the simulation results suggest that VFSs could be most effective when implemented on or adjacent to fields with high clay content (i.e., > 30%) and low sand content (i.e., < 50%). Using these results to identify locations that likely provide the most sediment into a VFS (i.e., soils with high clay content) combined with VFS designs that are most effective in trapping those particular soils and an understanding of typical location rainfall patterns, we can maximize our sediment trapping efforts.

Fig. 5

Annual sediment loads exported from a field with and without a vegetated filter strip. S, sand; LS, loamy sand; SL, sandy loam; Si, silt; L, loam; SiL, silt loam; SCL, sandy clay loam; CL, clay loam; SC, sandy clay; SiCL, silty clay loam; C, clay; SiC, silty clay. Error bars show standard deviation

Overall, simulation results reveal that temporal inequality plays an important role in the potential effectiveness of VFSs. The stronger the temporal inequality, the greater the potential for achieving higher annual removal efficiencies with “temporal targeting”. By identifying “hot moments” in contaminant transport and the corresponding flowrates leading to these “hot events”, the “window of opportunity” for load reduction can be targeted such that the annual load reduction goals can be met by targeting a few large events. This could reduce maintenance costs associated with BMPs, as the BMP would only need to perform for shorter periods of times (presumably when flow is high) for the load reduction goals to be met. For example, if vegetation is to be mowed or cut back, this maintenance could be done during a period of time (e.g., in the summer) when runoff to the VFS is anticipated to be low. Furthermore, sediment that has built up in the VFS over time should be removed prior to when high flows into the VFS are expected to provide conditions in the VFS that are more amenable to incoming sediment being captured by the VFS. Additionally, re-grading may be helpful if flow through the VFS has started to create rills, as sheet flow rather than shallow concentrated flow through the VFS is desirable.

Conclusions

A field-scale water balance and sediment transport model was coupled with a physically-based VFS model to better understand the effectiveness of VFSs at event-based and annual-based time scales. One thousand 1-yr stochastic simulations were run at a daily time scale using a parsimonious modeling approach. This approach was novel in that no other VFS modeling approach has simulated the non-stationarity of storm-event based removal efficiencies resulting from antecedent soil moisture of the contributing area or produced results that enable comparison of simple-average, storm-event based VFS removal efficiencies to the long-term, load-weighted, annual removal efficiency. The straight-forward modeling approach provides valuable insight into the use of VFSs as an agricultural BMP for reducing sediment loads and for achieving water quality goals.

Simulation results revealed the importance of antecedent conditions on VFS performance and suggest that water quality models need to explicitly consider variability in the relationship between the performance of structural BMPs, such as VFSs, and rainfall events. VFSs cannot be expected to have the same removal efficiency each time a storm with a specific return period (e.g., 1, 2, 5-year storm) occurs. Antecedent conditions in the field and seasonal differences in soil moisture deficit due to crop growth and corresponding differences in evapotranspiration play important roles in determining the amount of effective rainfall that each storm event produces, therefore leading to variability in the values of F_t for similar size storm events. Additionally, simply averaging the removal efficiencies of each observed (or simulated) event is not necessarily an accurate indicator of the overall removal efficiency of VFSs due to the strong temporal inequality exhibited by many contaminants of interest, such as sediments, as well as phosphorus and pesticides (Gall et al. 2013). Therefore, temporal inequality and non-stationarity of F_t must explicitly be considered when VFSs and other structural BMPs are simulated in water quality models.