Distinct influence of filter strips on acute and chronic pesticide aquatic environmental exposure assessments across U.S. EPA scenarios

Chemosphere 90 (2013) 195–202

Contents lists available at SciVerse ScienceDirect

Chemosphere journal homepage: www.elsevier.com/locate/chemosphere

Distinct influence of filter strips on acute and chronic pesticide aquatic environmental exposure assessments across U.S. EPA scenarios George J. Sabbagh a, Rafael Muñoz-Carpena b, Garey A. Fox c,⇑ a

Bayer CropScience, 2 T.W. Alexander Dr., Durham, NC 27709, United States Hydrology and Water Quality, Agricultural and Biological Engineering Department, University of Florida, 287 Frazier Rogers Hall, Gainesville, FL 32611-0570, United States c Orville L. and Helen L. Buchanan Endowed Chair, Department of Biosystems and Agricultural Engineering, Oklahoma State University, 120 Agricultural Hall, Stillwater, OK 74078-6016, United States b

h i g h l i g h t s " A pesticide exposure assessment framework can inform targeted filter strip design. " Pesticide application timing is an important input factor in exposure assessments. " Pesticide mass reduction is not equivalent to an exposure concentration reduction. " Acute and chronic exposure concentration buffer reductions are not equivalent. " Generic filter strip design consistent across EPA scenarios should be avoided.

a r t i c l e

i n f o

Article history: Received 27 January 2012 Received in revised form 18 June 2012 Accepted 26 June 2012 Available online 9 August 2012 Keywords: Exposure assessment Pesticides Sensitivity analysis Vegetative filter strips VFSMOD

a b s t r a c t Vegetative filter strips (VFS) are proposed for protection of receiving water bodies and aquatic organisms from pesticides in runoff, but there is debate regarding the efficiency and filter size requirements. This debate is largely due to the belief that no quantitative methodology exists for predicting runoff buffer efficiency when conducting acute and/or chronic environmental exposure assessments. Previous research has proposed a modeling approach that links the U.S. Environmental Protection Agency’s (EPA’s) PRZM/EXAMS with a well-tested process-based model for VFS (VFSMOD). In this research, we apply the modeling framework to determine (1) the most important input factors for quantifying mass reductions of pesticides by VFS in aquatic exposure assessments relative to three distinct U.S. EPA scenarios encompassing a wide range of conditions; (2) the expected range in percent reductions in acute and chronic estimated environmental concentrations (EECs); and (3) the differential influence of VFS when conducting acute versus chronic exposure assessments. This research utilized three, 30-yr U.S. EPA scenarios: Illinois corn, California tomato, and Oregon wheat. A global sensitivity analysis (GSA) method identified the most important input factors based on discrete uniform probability distributions for five input factors: VFS length (VL), organic-carbon sorption coefficient (Koc), half-lives in both water and soil phases, and application timing. For percent reductions in acute and chronic EECs, VL and application timing were consistently the most important input factors independent of EPA scenario. The potential ranges in acute and chronic EECs varied as a function of EPA scenario and application timing. Reductions in acute EECs were typically less than percent reductions in chronic EECs because acute exposure was driven primarily by large individual rainfall and runon events. Importantly, generic specification of VFS design characteristics equal across scenarios should be avoided. The revised pesticide assessment modeling framework offers the ability to elucidate the complex and non-linear relationships that can inform targeted VFS design specifications. Ó 2012 Elsevier Ltd. All rights reserved.

1. Introduction Vegetative filter strips (VFSs) reduce sediment and pesticide movement to receiving water bodies through infiltration and ⇑ Corresponding author. Tel.: +1 405 744 8423; fax: +1 405 744 6059. E-mail address: [email protected] (G.A. Fox). 0045-6535/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.chemosphere.2012.06.034

reduction of runoff flow volumes, by contact between the sediment and pesticide with vegetation and soil in the VFS, and by increasing hydraulic roughness to reduce the flow velocity and allow sediment-bound pesticide to settle out of the runoff (Muñoz-Carpena et al., 1999, 2010; Sabbagh et al., 2009; Fox et al., 2010). VFS pesticide trapping efficiency depends on their spatially and temporally dynamic hydrological and sedimentological conditions that result

196

G.J. Sabbagh et al. / Chemosphere 90 (2013) 195–202

from unique combinations of climate, soil, vegetation and land uses. Because of this, generic and simple regression equations that attempt to relate reduction efficiency with VFS length, slope, and/ or physical characteristics are limited in applicability (Fox and Sabbagh, 2009). Two VFS with equivalent lengths, slopes, and vegetation may provide different sediment and pesticide trapping efficiencies, especially when comparing the responses under diffuse versus concentrated flow (Fox et al., 2010). This was demonstrated explicitly by Sabbagh et al. (2010) (additional details provided in the Supplementary information, see Fig. S-1), and is supported by extensive field data on pesticide reduction by VFS (see Poletika et al., 2009 and Sabbagh et al., 2009 for a summary of several field studies and Reichenberger et al., 2007 for a review on pesticide trapping efficiency). VFS are currently proposed for protection of threatened and endangered biological organisms from pesticide in runoff (National Marine Fisheries Service (NMFS) Biological Opinion (BiOp) issued November 18, 2008), but there is debate regarding the efficiency and filter size requirements. For example, in their response letter to the 2008 NMFS BiOp, EPA (2010) indicated that to date no quantitative methodology existed for predicting runoff buffer efficiency. The debate is created by the common intent to relate VFS efficiency to simple characteristics like their size or slope. The complex VFS time-variant behavior suggests the need for simulation models capable of accounting for hydrologic and sedimentological variability. Numerical models have been available for predicting runoff and sediment reductions by VFS such as the Vegetative Filter Strip Modeling System (VFSMOD), developed and evaluated by Muñoz-Carpena et al. (1999), Muñoz-Carpena and Parsons (2004), and tested by others (e.g., Abu-Zreig, 2001). VFSMOD is a finite-element, field-scale, storm-based model that routes an incoming surface flow hydrograph and sedigraph from an adjacent source area through a VFS and simulates the infiltration in the VFS using the Green-Ampt equation, sediment trapping based on GRASSF, and the resulting outflow. Sabbagh et al. (2009) proposed an empirical pesticide trapping function that was integrated with VFSMOD that considers sorption, pesticide reduction by infiltration of the dissolved phase, and sedimentation of the sorbed pesticide. The empirical function also depends on the percent clay content of the incoming sediment into the VFS and a phase distribution factor which represents the ratio between the mass of pesticide in the dissolved phase and the mass of pesticide sorbed to sediment. Degradation processes during runoff transport were not considered in the empirical component because of the small residence time during typical runoff events (min to h). Sabbagh et al. (2009) and Poletika et al. (2009) tested and evaluated the VFSMOD simulation tool linked with the pesticide trapping component. Muñoz-Carpena et al. (2010) and Fox et al. (2010) investigated the importance of various input factors on predicted sediment and pesticide reductions using this linked model. The EPA uses computer simulation models (PRZM/EXAMS) to evaluate pesticide estimated environmental concentrations (EECs) in surface water. The simulated exposure concentrations are then compared to toxicological endpoints for assessing potential risks to aquatic organisms. The simulation models are typically applied to benchmark scenarios for various crops (Lin et al., 2007; Lin, 2009). These conservative scenarios establish specific field, soil, and receiving water body (static pond) characteristics (additional details provided in the Supplementary information, see Fig. S-2), but do consider variations in weather and management practices through 30-yr simulations (1961–1990) that are conducted using daily weather data and maximum use rates and patterns. The current EPA PRZM/EXAMS assessment approach models pesticide transport from a 10 ha circular field flowing into a 1 ha, 2-m deep circular pond located in the center of the field. The volume of water and mass of sediment in the pond is assumed constant by EXAMS.

Risk is then quantified based on the upper 90th-percentile annual peak (acute risk) or the 60-d mean concentrations (chronic risk). Sabbagh et al. (2010) proposed a revised PRZM, VFSMOD, and EXAMS modeling framework that considers the presence of a VFS. Prior to this revised modeling framework, specification of the required VFS characteristics for reducing pesticide risk was largely subjective. The revised framework was applied to a single EPA scenario for four hypothetical pesticides. However, the importance of soil, hydrologic, pesticide and land use factors and their interactions need to be elucidated to advance their application in environmental exposure assessments for pesticides. Therefore, this research utilized the previously revised modeling framework (PRZM, VFSMOD, and EXAMS) to address three objectives: (1) to determine the most important input factors for quantifying mass reductions of pesticides by VFS in aquatic exposure assessments relative to EPA scenario, and in particular assess the relative importance of pesticide application timing among other variable factors; (2) to quantify the range in percent reductions in acute and chronic EECs for three prescribed EPA scenarios representing a wide range of conditions; and (3) to determine differences in the influence of VFS on aquatic exposure assessments when conducting acute versus chronic exposure assessments. The effect of the local environment (soil, precipitation, and runoff) was captured in three scenarios analyzed in this research, while application timing was considered a characteristic of the pesticide.

2. Materials and methods This research used the procedures developed by Sabbagh et al. (2010) for conducting aquatic exposure assessments with VFS based on PRZM, VFSMOD, and EXAMS simulation models. Three EPA scenarios were considered: Illinois corn, Oregon wheat, and California tomato. These scenarios were selected to provide a wide range of hydrological and sedimentological conditions (Table 1): Midwestern continental row-crop agriculture (Illinois Corn), wet maritime extensive agriculture (Oregon wheat), and dry Mediterranean irrigated horticulture (California tomato). It should be realized that results presented in this research are limited to the scenarios that the EPA developed for regulatory aquatic exposure assessments. In fact, the tools that are used in this research are applicable to other more realistic field and VFS conditions. Soils data as specified by the EPA scenario were used explicitly in the aquatic exposure assessments. Climate data from meteorological stations specified in the EPA scenarios were used for conducting 30-yr simulations. Pesticide application was assumed to occur at pre-emergence (10 d prior to emergence date), in-season (30 d after the emergence date), or post-harvest (10 d after the harvest date). The effect of the environment (location, rainfall, soil) was captured in three scenarios analyzed in this research, while application timing was considered a characteristic of the pesticide. Field and VFS slopes were assumed uniform as prescribed in the U.S. EPA scenario. Vegetation type in the VFS was assumed to be bluegrass and default parameters for VFS vegetation characteristics were used. A source of uncertainty when simulating actual VFS performance is the condition of the VFS relative to upkeep and maintenance (regular mowing to maintain design height and vegetation uniformity, resetting by leveling and reseeding every 5-yr). This research assumed a well-maintained VFS with shallow overland flow across the entire VFS width rather than concentrated flow (Fox et al., 2010; Muñoz-Carpena et al., 2010). Percent reduction in acute exposure was determined from the upper 90th-percentile of the annual peak concentrations during each year of the 30-yr simulations; percent reduction in chronic exposure was determined from the upper 90th-percentile of the

197

G.J. Sabbagh et al. / Chemosphere 90 (2013) 195–202

Table 1 Input parameters and boundary conditions of the three U.S. EPA scenarios (Illinois corn, Oregon wheat, and California tomato) for the PRZM/VFSMOD/EXAMS modeling package. Parameter

EPA scenario

Model input value

Model

Comment Prescribed by each EPA scenario

Precipitation/land use

All scenarios

30-yr time series (1961–1990)

PRZM, VFSMOD

Soil types

Illinois corn

Adair clay loam with 6% slope, 39% clay content, and 2.3% organic carbon Stockton clay with 0.25% slope, 48% clay content, and 1.0% organic carbon Bashow clay with 6% slope, 63% clay, and 4.6% organic carbon

PRZM, VFSMOD

California tomato Oregon wheat Soil hydraulic conductivity

Illinois corn California tomato Oregon wheat

3.2 cm h1 0.5 cm h1 1.5 cm h1

VFSMOD

SSURGO – Database

Soil bulk density

Illinois corn California tomato Oregon wheat

1.32 g cm3 1.32 g cm3 1.22 g cm3

PRZM, VFSMOD

SSURGO – Database

Pesticide molecular weight

All scenarios

500 g mol1

PRZM, EXAMS

Pesticide solubility at 20 °C

All scenarios

1 mg L1 (0.002 mol m3)

PRZM, EXAMS

Pesticide vapor pressure

All scenarios

1.3  1010 atm

PRZM, EXAMS

Henry’s law constant

All scenarios

6.5  108 atm m3 mol1

PRZM, EXAMS

Hydrolysis and aqueous photolysis

All scenarios

Stable

EXAMS

Biological degradation was assumed to be the only path for the product to dissipate in water.

Aerobic/anaerobic aquatic metabolism half-life (tw)

All scenarios

10, 100, or 1000 d

EXAMS

Aerobic and anaerobic degradations were assumed to occur at the same rate

Pesticide half-life in soil (ts)

All scenarios

10, 100, or 1000 d

EXAMS

Aerobic and anaerobic degradations were assumed to occur at the same rate

Aerobic soil metabolism

All scenarios

Stable

PRZM

Adsorption coefficient (Koc)

All scenarios

20, 200 or 2,000 L kg1 OC

PRZM, VFSMOD, EXAMS

Pesticide use pattern (Timing)

All scenarios

Pre-Emergence (10 d before), MidSeason (30 d after), or Post-Harvest Application (10 d after)

PRZM

Percent deposition

All scenarios

100%

PRZM

maximum 60-d mean concentrations during each year of the 30-yr simulations. Global Sensitivity Analysis (GSA) was used to identify important effects (underlying factors) in the context of all other effects (ranking based on direct or first order effects, and interactions). The screening GSA method proposed by Morris (1991) (herein ‘‘Morris method’’ or ‘‘Morris’’) and later modified by Campolongo et al. (2007), was used in this study because it is relatively easy to apply, requires few simulations, and its results are easily interpreted (Saltelli et al., 2004). The method is qualitative in nature and therefore can only be used to assess the relative importance of input factors. A brief summary of the method is given below with more details summarized by Muñoz-Carpena et al. (2007, 2010) and Fox et al. (2010). Morris (1991) proposed conducting individually randomized experiments that evaluate the elementary effects (relative output differences) of changing one parameter at a time. Each input may assume a discrete number of values called levels that are selected within an allocated range of variation for the parameter. For each parameter, two sensitivity measures are proposed: (1) the mean of the elementary effects, l, which estimates the overall direct (first-order) effect of the parameter on a given output; and (2) the standard deviation of the effects, r, which estimates the higher-order characteristics of the parameter (such as curvatures and interactions). Since sometimes the model output is non-monotonic, Campolongo et al. (2007) suggested considering the distribution of absolute values of the elementary effects, l⁄, to avoid the

Granular Application – 100% of applied material deposited in the field; no loss due to drift

canceling of effects of opposing signs. The number of simulations (N) to perform in the Morris analysis is given by:

N ¼ rðk þ 1Þ

ð1Þ

where r is the sampling size for the search trajectory (r = 10 produces satisfactory results) and k is the number of factors. Although elementary effects are local measures, the method is considered global because the final measure l⁄ is obtained by averaging the elementary effects along the k-dimensional search trajectories and this eliminates the need to consider the specific points at which they are computed (Saltelli et al., 2004). Morris (1991) recommended applying l (or l⁄ thereof) to rank parameters in order of importance and Saltelli et al. (2004) suggested applying the original Morris measure r when examining the effects due to interactions. To interpret the results in a manner that simultaneously informs about the parameter ranking and potential presence of interactions, Morris (1991) suggested plotting the points on a l (or l⁄)–r Cartesian plane. Input factors considered in the Morris analyses in this study (Table 1) included pesticide characteristics such as the organic carbon sorption coefficient (Koc), pesticide soil half-life (ts), and the pesticide half-life in the water phase (tw). Other input factors included the length of the vegetative filter strip in the direction of flow (VL) and the timing of pesticide application (Timing: preemergence, mid-season, and post-harvest application). All other factors were considered constant as defined by the scenario. In this study, three levels were considered for each input factor to

198

G.J. Sabbagh et al. / Chemosphere 90 (2013) 195–202

Fig. 1. Global sensitivity analysis results using the Morris method for the importance of input factors simulated by PRZM/VFSMOD/EXAMS on percent reductions in acute exposure due to vegetative filter strips (VL = length; Timing = application timing; Koc = organic carbon sorption coefficient; tw = half-life in water; and ts = soil half-life). Input factor importance is related to its separation distance from the l⁄–r plane along the l⁄ axis.

Fig. 2. Global sensitivity analysis results using the Morris method for the importance of input factors simulated by PRZM/VFSMOD/EXAMS on percent reductions in chronic exposure due to vegetative filter strips (VL = length; Timing = application timing; Koc = organic carbon sorption coefficient; tw = half-life in water; and ts = soil half-life). Input factor importance is related to its separation distance from the l⁄–r plane along the l⁄ axis.

encompass a wide range of pesticides and field conditions for each scenario (VL: 3, 5, 9 m; Koc: 20, 200, 2000 cm3 g1; ts and tw: 10, 100, 1000 d; and Timing: 1 = pre-emergent application, 2 = mid-

season application, 3 = post-harvest application). Based on this design a total of n = 243 long term (30-yr) simulations were conducted for each of the three EPA scenarios selected, and upper

199

G.J. Sabbagh et al. / Chemosphere 90 (2013) 195–202

Table 2 Range and mean (in parentheses) estimated environmental concentrations (EECs) predicted in the receiving water body by the integrated PRZM/VFSMOD/EXAMS models relative to application timing and length of the VFS. Ranges are due to variability in values of Koc, tw, and ts for the Illinois corn (IL-Co), California tomato (CA-To), and Oregon wheat (ORWh) scenarios. Scenario

IL-Co

Application timing

Pre-emergent Mid-season Post-harvest

CA-To

Pre-emergent Mid-season Post-harvest

OR-Wh

Pre-emergent Mid-season Post-harvest

Acute (Peak) (lg L1)

Chronic (60-d) (lg L1)

EEC without VFS

EEC with 1-m VFS

EEC with 5-m VFS

EEC with 9-m VFS

EEC without VFS

EEC with 1-m VFS

EEC with 5-m VFS

EEC with 9-m VFS

3.7–76.2 (27.0) 3.4–74.0 (20.9) 1.5–87.6 (27.5) 0.7–17.3 (5.7) 0.8–12.5 (5.0) 0.3–32.5 (7.2) 0.6–43.9 (12.2) 0.4–38.3 (9.2) 0.2–43.9 (10.3)

2.7–44.9 (16.0) 2.2–48.3 (14.2) 0.5–33.0 (11.2) 0.4–10.2 (2.9)

1.6–24.8 (9.0)

0.3–7.9 (2.1)

0.1–7.9 (1.6)

0.3–14.4 (4.0)

0.3–31.1 (7.8)

0.1–14.2 (3.5)

0.2–14.4 (5.4)

0.0–4.4 (1.0)

0.1–14.3 (4.6)

0.0–3.9 (0.7)

0.1–5.8 (1.2)

0.0–3.3 (0.6)

0.8–44.2 (13.3) 0.6–47.7 (12.1) 0.3–32.8 (10.0) 0.1–10.1 (2.5)

0.5–24.1 (7.2)

1.3–31.7 (9.2)

0.0–5.8 (1.1)

0.0–3.2 (0.5)

0.4–8.4 (2.9)

0.0–4.8 (1.1)

0.0–3.2 (0.6)

0.1–8.4 (2.3)

0.0–4.8 (1.0)

0.0–3.2 (0.6)

0.2–17.9 (4.0)

0.0–8.3 (1.9)

0.0–3.8 (1.1)

0.0–16.8 (3.3)

0.0–7.5 (1.4)

0.0–3.4 (0.8)

0.5–37.5 (10.2) 0.3–29.8 (7.3)

0.4–31.2 (8.4)

0.2–25.4 (6.5)

0.1–37.1 (9.0)

0.1–30.8 (7.3)

0.1–25.3 (5.7)

0.2–22.6 (5.7)

0.1–16.2 (4.3)

0.1–30.0 (6.6)

0.1–21.8 (5.1)

0.1–15.5 (3.8)

0.1–38.4 (8.9)

0.1–32.7 (7.5)

0.0–25.9 (6.1)

1.2–75.1 (23.1) 0.9–73.1 (18.1) 0.8–85.9 (25.2) 0.2–17.2 (4.8) 0.3–12.4 (4.0) 0.1–31.6 (6.1) 0.2–43.5 (10.8) 0.2–37.3 (8.3) 0.0–43.6 (9.2)

0.0–38.2 (7.9)

0.0–32.5 (6.7)

0.0–25.7 (5.4)

90th-percentile EEC reduction efficiencies were calculated for either acute or chronic exposure. Note that the pesticide was applied at the same application time each year of the 30-yr simulations. For the Morris analysis, discrete uniform (DU) probability distributions (equal probability for the factor’s levels) were selected for each of the five input factors. For k = 5 factors a total of 60 input sample sets (and simulations) were drawn from the complete data set of upper 90th-percentile reduction efficiencies (n = 243 for each scenario and for either acute or chronic exposure). Parameter importance was assessed based on the value of l for either percent acute or chronic EEC reductions. Relationships were derived between the percent reductions in acute (peak) and chronic (60-d) EECs relative to the most important input factors from the Morris results. Ranges in predicted individual event’s VFS percent runoff (dQ) and sediment (dE) reductions were also analyzed. Acute versus chronic EECs and EEC reductions by the VFS were compared within and between scenarios, providing a first estimate of potential values for different EPA scenarios.

3. Results and discussion The Morris method established the importance of each input factor relative to each of the scenarios and demonstrated that input factor importance also depended on whether the focus was on acute versus chronic exposure assessment (Figs. 1 and 2). In this analysis technique, input factor importance is related to its separation distance from the origin of the l⁄–r plane along the l⁄ axis. For percent reductions in acute exposure, VL was consistently the most important and, interestingly, application timing the second most important input factor regardless of EPA scenario (Fig. 1). Fox et al. (2010) and Muñoz-Carpena et al. (2010) reported the saturated, hydraulic conductivity as the most important input factor when applying the Morris method to VFS edge of the field observations. However, this research did not consider variability in the hydraulic conductivity as the EPA assessment scenarios prescribed the soil properties. The importance of application timing was higher in the Illinois corn scenario than the other two scenarios. For percent reductions in chronic exposure, VL and application timing

Fig. 3. Edge of the field percent reductions in (a) event runoff and (b) sediment due to VFS of various lengths (1, 5, and 9 m, data separation at a specific length is presentation purposes only) for the three U.S. EPA scenarios: Oregon Wheat (ORWh), California tomato (CA-To), and Illinois corn (IL-Co). Symbols are average values and error bars represent the event data range (minimum to maximum) for the 30 yr scenarios. Note that n is the number of rainfall events.

200

G.J. Sabbagh et al. / Chemosphere 90 (2013) 195–202

were still the most important input factors, but the importance of Koc increased (ranked 3rd versus 4th for acute exposure) when conducting these assessments (Fig. 2). The ts was consistently the least important input factor for all scenarios whether conducting acute or chronic assessments. Therefore, it was less important which value for ts within the simulated range was utilized in the simulations compared to the effects of other input factors. In Figs. 1 and 2, the standard deviation of the elementary effects, r, is used as a statistic indicating interactions of the input factor with other factors. For the scenarios tested and for both types of exposures, interactions are limited (as denoted by the low r values in Figs. 1 and 2), with only tw and timing introducing some effects in all three scenarios. Simulations across the three scenarios and for the various input factors demonstrated that a VFS percent pesticide mass reduction (dP) entering the receiving water body did not always correlate to an equivalent EEC percent reduction. Also, acute and chronic EECs and reductions vary based on EPA scenario and therefore cannot be assumed equivalent between scenarios (Table 2). EECs without VFS were typically the greatest for the Illinois corn scenario, followed by the Oregon wheat and California tomato scenarios. Percent pesticide mass reductions and therefore reductions in EECs are driven by dQ and dE. The scenario averaged dQ and dE across all variables were at the upper limit of their ranges, suggesting the majority of the data lied in the upper quartile of the distribution (Fig. 3). In fact, only a few outliers controlled the extent of the lower range in dQ and dE. Although the ranges in dE were greater for the Illinois corn scenario, there was not much difference in the average dE between scenarios (Fig. 3). These results are correlated

to the total runoff simulated in the 30-yr simulation of each scenario. The Oregon wheat scenario had the lowest average dQ (Fig. 3a) leading to lower acute and chronic EEC reductions (Fig. 4). In fact, the Oregon wheat scenario even had cases with a negative dQ (Fig. 3a), i.e., more runoff water out from the VFS than the amount that entered the VFS due to the precipitation rate exceeding the infiltration rate for some large events. A 9 m VFS length in the Illinois corn and California tomato scenarios typically reduced EECs by greater than 80% on average, but only reduced EECs by approximately 50% on average in the Oregon wheat scenario. Small VFS lengths (1 m) resulted in greater percent reductions in the California tomato scenario (typically greater than 40–60%) as compared to the Illinois corn scenario (30–60%) and the Oregon wheat scenario (15–20%). Note that EECs without VFS were lower in the California tomato scenario. The Illinois corn scenario had the larger range of dE (the lowest minima in Fig. 3b), so it contributed to extend the range of dP for those pesticides with high Koc (Fig. 4a), and in some way counteract the higher observed dQ, especially for larger VL. The California tomato scenario exhibited the largest overall dP reductions due to the dry Mediterranean controlled irrigation (drip) conditions where runoff was limited to a few isolated storm events (number of storm events, n = 334 versus n = 1326 and 1247 for the OR-wheat and IL-corn scenarios, respectively). All of these results suggest inherent limitations associated with using a standard ‘‘one-size-fits-all’’ approach to VFS design for predicting EECs of different EPA scenarios. As noted earlier, the largest ranges in reduction for acute and chronic exposure were typically associated with post-harvest

Fig. 4. Percent reductions in peak (acute) and 60-d (chronic) estimated environmental concentrations (EECs) due to VFS of various lengths (1, 5, and 9 m, data separation at a specific length is for presentation purposes only) and pesticide application timing for the three U.S. EPA scenarios: (a) Oregon Wheat (OR-Wh), (b) California tomato (CA-To), and (c) Illinois corn (IL-Co). Symbols are average values and error bars represent the data range (minimum to maximum) in percent reductions (black error bars for chronic and gray error bars for acute).

G.J. Sabbagh et al. / Chemosphere 90 (2013) 195–202

Fig. 5. Average cumulative runoff after pesticide application for a post-harvest application over the 30-yr simulation period for three U.S. EPA scenarios: Oregon Wheat (OR-Wh), California tomato (CA-To), and Illinois corn (IL-Co).

applications. Verifying the Morris results (Figs. 1 and 2), application timing was more important in the Illinois corn scenario (Fig. 4a) especially for post-harvest pesticide applications, which typically resulted in 10–20% greater percent reductions in EECs compared to pre-emergence and in-season applications, especially for the smaller VFS lengths (VL = 1 and 5 m). Ranges in acute and chronic exposure were similar except for the post-harvest scenarios due to differences in runoff amounts after harvest (Fig. 5). Postharvest scenarios assume that all crop residue is removed from the field leaving bare soil conditions susceptible for erosion and sediment mobilization. Higher runoff amounts correspond to greater mobilization potential and therefore less opportunity for dE and dP by the VFS. The Illinois corn scenario experienced the greatest average and maximum cumulative runoff amounts compared to the other two scenarios (Fig. 5). Note that the California tomato scenario was influenced primarily by irrigation events and therefore experienced the lowest average cumulative runoff. Percent reductions in acute and chronic EECs were not equivalent in most cases (Table 2 and Fig. 4), especially when considering the percent change between acute and chronic EECs (changes between 11% and 22%). Reductions in acute EECs were typically less than percent reductions in chronic EECs because acute exposure was driven primarily by large rainfall and runon events. Chronic exposure was driven more by cumulative loadings to the receiving water body in EXAMs and therefore less dependent on single events, as discussed by Sabbagh et al. (2010). This result was observed in specific scenarios when comparing events with smaller and larger sediment transport rates. With greater sediment loading typically associated with large storm and runon events, there is less opportunity for sediment trapping by the VFS. For example, sediment transport rates predicted by PRZM in the Oregon wheat scenario were typically smaller and therefore smaller differences in percent reduction in acute and chronic EECs were observed. Larger differences in acute versus chronic EEC percent reductions were observed for the Illinois corn scenario where ranges in dE were much greater than the corresponding ranges in the other scenarios (Fig. 3b).

4. Conclusions The revised pesticide assessment modeling framework (PRZM, VFSMOD, and EXAMS) offers the ability to elucidate the complex and non-linear relationships between field land use, climate, soil, application timing, and vegetation that can inform targeted VFS design specifications. When conducting long-term exposure

201

assessments, application timing (pre-emergence, mid-season, or postharvest) is an important input factor that should be explicitly considered. A percent mass reduction in pesticide entering the receiving water body does not correlate simply to an equivalent EEC percent reduction, which depends on several important factors such as the pesticide’s mobility and persistence and whether one is calculating an acute or chronic EEC. Relationships obtained between percent reduction relative to the VFS length and application timing provide a broad-level estimate of potential reductions in acute and chronic EECs, although acute and chronic exposure reductions vary based on EPA scenario. Generic specification of VFS design characteristics consistent across all assessment scenarios should be avoided. For example, a 9 m long VFS in the Illinois corn and California tomato scenario typically reduced EECs by greater than 80%, but only reduced EECs by approximately 50% in the Oregon wheat scenario. Across all VFS lengths, percent reductions in acute and chronic EECs were greater in the California tomato and Illinois corn scenarios, due to differences in event runoff and sediment reductions for each of the agroecological scenarios. Reductions in acute EECs were typically less than percent reductions in chronic EECs because acute exposure was driven primarily by large events. A quantitative methodology does now exist for predicting runoff buffer efficiency and eliminates the need to relate VFS efficiency to simple characteristics like their size or slope. It should be realized that these results are limited to the scenarios that the EPA developed for regulatory aquatic exposure assessments. The philosophy of the EPA is to be conservative in estimating exposure and the framework that EPA follows is based on certain assumptions that may or may not reflect reality in all cases. Future work should apply the integrated modeling framework to other EPA scenarios. Acknowledgements The USDA CSREES Regional Project S1042 supported this work. Dr. Muñoz-Carpena acknowledges support by University of Florida Foundation for support during his academic sabbatical year. Dr. Garey Fox acknowledges support from the Buchanan Foundation as part of the Orville L. and Helen L. Buchanan Endowed Chair in Biosystems and Agricultural Engineering. Appendix A. Supplementary material Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.chemosphere. 2012.06.034. References Abu-Zreig, M., 2001. Factors affecting sediment trapping in vegetated filter strips: simulation study using VFSMOD. Hydrol. Proc. 15 (8), 1477–1488. Campolongo, F., Cariboni, J., Saltelli, A., 2007. An effective screening design for sensitivity analysis of large models. Environ. Modell. Softw. 22 (10), 1509–1518. EPA (U.S. Environmental Protection Agency, 2010. EPA’s comments on the National Marine Fisheries Service (NMFS), June 16, 2010, Draft Biological Opinion (BiOp); . Fox, G.A., Sabbagh, G.J., 2009. Comment on major factors influencing the efficacy of vegetated buffers on sediment trapping: a review and analysis. J. Environ. Qual. 38 (1), 1–3. Fox, G.A., Muñoz-Carpena, R., Sabbagh, G.J., 2010. Influence of flow concentration on parameter importance and prediction uncertainty of pesticide trapping by vegetative filter strips. J. Hydrol. 384 (1–2), 164–173. Lin, J., 2009. A progress report for aquatic exposure assessment in the U.S. EPA Office of Pesticide Programs. U.S. EPA Office of Pesticide Programs, Washington, DC, 2009. . Lin, J., Young, D., Kennedy, I., 2007. The Tier II Modeling approach for aquatic exposure assessment in the U.S. EPA Office of Pesticide Programs. U.S. EPA Office of Pesticide Programs, Washington, DC. .

202

G.J. Sabbagh et al. / Chemosphere 90 (2013) 195–202

Morris, M.D., 1991. Factorial sampling plans for preliminary computational experiments. Technometric 33 (2), 161–174. Muñoz-Carpena, R., Fox, G.A., Sabbagh, G.J., 2010. Parameter importance and uncertainty in predicting runoff pesticide reduction with filter strips. J. Environ. Qual. 39 (2), 630–641. Muñoz-Carpena, R., Parsons, J.E., 2004. A design procedure for vegetative filter strips using VFSMOD-W. T. ASAE 47 (6), 1933–1941. Muñoz-Carpena, R., Parsons, J.E., Gilliam, J.W., 1999. Modeling hydrology and sediment transport in vegetative filter strips. J. Hydrol. 214, 111–129. Poletika, N.N., Coody, P.N., Fox, G.A., Sabbagh, G.J., Dolder, S.C., White, J., 2009. Chlorpyrifos and atrazine removal from runoff by vegetated filter strips: experiments and predictive modeling. J. Environ. Qual. 38 (3), 1042–1052.

Reichenberger, S., Bach, M., Skitschak, A., Frede, H.-G., 2007. Mitigation strategies to reduce pesticide inputs into ground- and surface water and their effectiveness: a review. Sci. Total Environ. 384, 1–35. Sabbagh, G.J., Fox, G.A., Kamanzi, A., Roepke, B., Tang, J.-Z., 2009. Effectiveness of vegetative filter strips in reducing pesticide loading: quantifying pesticide trapping efficiency. J. Environ. Qual. 38 (2), 762–771. Sabbagh, G.J., Fox, G.A., Muñoz-Carpena, R., Lenz, M., 2010. Revised framework for pesticide aquatic environmental exposure assessment that accounts for vegetative filter strips. Environ. Sci. Technol. 44 (10), 3839–3845. Saltelli, A., Tarantola, S., Campolongo, F., Ratto, M., 2004. Sensitivity Analysis in Practice. A Guide to Assessing Scientific Models. John Wiley and Sons, Chichester, U.K.