Comparison of three pesticide fate models with respect to the leaching of two herbicides under field conditions in an irrigated maize cropping system

Science of the Total Environment 499 (2014) 533–545

Contents lists available at ScienceDirect

Science of the Total Environment journal homepage: www.elsevier.com/locate/scitotenv

Comparison of three pesticide fate models with respect to the leaching of two herbicides under field conditions in an irrigated maize cropping system J.M. Marín-Benito a,⁎, V. Pot a, L. Alletto b, L. Mamy c, C. Bedos a, E. Barriuso a, P. Benoit a a b c

Institut National de la Recherche Agronomique, INRA, AgroParisTech, UMR 1091 EGC, 1 Avenue Lucien Bretignières, 78850 Thiverval-Grignon, France Université de Toulouse — INPT-École d'ingénieurs de Purpan, UMR 1248 AGIR, 75 voie du TOEC BP 57611, 31076 Toulouse, France Institut National de la Recherche Agronomique, INRA, UR 251 PESSAC, Route de Saint Cyr, 78026 Versailles, France

H I G H L I G H T S • • • • •

The performances of PRZM, PEARL and MACRO are rigorously compared. Simulated evapotranspiration plays a major role in the prediction of water dynamic. The three models overestimated the total water volume leachate at 1 m depth. PEARL and MACRO predicted better water behavior than PRZM. PEARL simulated the observed herbicide concentrations better than MACRO and PRZM.

a r t i c l e

i n f o

Article history: Received 14 March 2014 Received in revised form 20 June 2014 Accepted 23 June 2014 Available online 15 August 2014 Keywords: Pesticide fate models Model comparison Field experiment Herbicide S-metolachlor Mesotrione

a b s t r a c t The ability of three models (PEARL, MACRO and PRZM) to describe the water transfer and leaching of the herbicides S-metolachlor and mesotrione as observed in an irrigated maize monoculture system in Toulouse area (France) was compared. The models were parameterized with field, laboratory and literature data, and pedotransfer functions using equivalent parameterization to better compare the results and the performance of the models. The models were evaluated and compared from soil water pressure, water content and temperature data monitored at 0.2, 0.5 and 1 m depth, together with water percolates and herbicide concentrations measured in a tension plate lysimeter at 1 m depth. Some hydraulic (n, θs) parameters and mesotrione DT50 needed calibration. After calibration, the comparison of the results obtained by the three models indicated that PRZM was not able to simulate properly the water dynamic in the soil profile. On the contrary, PEARL and MACRO simulated generally quite well the observed water pressure head and volumetric water content at the three different depths during wetting periods (e.g. irrigated cropping period) while a poorest performance was obtained for drying periods (fallow period with bare soil and beginning of crop period). Similar water flow dynamics were simulated by PEARL and MACRO in the soil profile although in general, and due to a higher evapotranspiration in MACRO, PEARL simulated a wetter soil than MACRO. For the whole simulated period, the performance of all models to simulate water leaching at 1 m depth was poor, with an overestimation of the total water volume measured in the lysimeter (ranging from 2.2 to 6.6 times). By contrast, soil temperature was properly reproduced by the three models. The models were able to simulate the leaching of herbicides at 1 m depth in similar appearance time and order of magnitude as field observations. Cumulative observed and simulated mesotrione losses by leaching were consistently higher than the observed and simulated losses of the less mobile herbicide, S-metolachlor. In general, PRZM predicted the highest concentrations for both herbicides in the leachates while PEARL simulated the observed herbicide concentrations better than MACRO and PRZM. © 2014 Elsevier B.V. All rights reserved.

1. Introduction

⁎ Corresponding author. Tel.: +33 1 30 81 59 91; fax: +33 1 30 81 53 96. E-mail address: [email protected] (J.M. Marín-Benito).

http://dx.doi.org/10.1016/j.scitotenv.2014.06.143 0048-9697/© 2014 Elsevier B.V. All rights reserved.

The use of pesticides in agriculture involves environmental risks such as the contamination of surface and ground waters. Indeed, monitoring studies of surface and groundwater bodies have shown the

534

J.M. Marín-Benito et al. / Science of the Total Environment 499 (2014) 533–545

presence of pesticides, and in some occasions at higher concentrations than the limit established by the EU for individual pesticide in drinking water (0.1 μg L− 1) (Belmonte et al., 2005; Blanchoud et al., 2004; Hildebrandt et al., 2008; Herrero-Hernández et al., 2013). Therefore, numerous models have been developed for a priori assessment of the transfer of pesticides from the soil surface to groundwater after their application to an agricultural field. These pesticide fate models are extensive, economic and efficient tools to predict the fate of pesticide in addition to field and laboratory data (Herbst et al., 2005; Mamy et al., 2008). Their ability to take into account most of the major processes involved in the environmental fate of pesticides (sorption, degradation, leaching, volatilization, plant uptake and wash-off, erosion and runoff) in varying degrees of complexity makes them to be also used for pesticide registration purposes (FOCUS, 2000; Regulation EC No 1107/2009, 2009). However, they are not always good enough (Jarvis and Larsbo, 2012) and in some cases, an inadequate parameterization of the models by the user, the quality and quantity of available experimental data and/or model limitations to describe some processes or pedological and agronomic conditions in a realistic way give unsatisfactory results relative to the field data (Garratt et al., 2002, 2007; Gottesbüren et al., 2000; Jarvis et al., 2000; Vanclooster et al., 2000). For this reason, European Food Safety Authority recommends that the modeling of pesticide transfer to water should be based on the use of at least two models, and on the comparison of their results (EFSA, 2004). Some comparative studies on the ability of different pesticide fate models to describe soil hydrology and pesticide transport under field conditions have been published in the literature (Armstrong et al., 2000; Garratt et al., 2002; Gottesbüren et al., 2000; Herbst et al., 2005; Vanclooster et al., 2000). Contrasting results were found: some studies showed an adequate description of the water and pesticide field data whereas others showed that models could not simulate correctly the observed data. Adequate or inadequate model predictions were not always linked to the complexity of the description of the different water and pesticide processes in the models. To the best of our knowledge, no study using equivalent parameterization and/or minimizing the differences in the parameterization for the processes specifically simulated by each model has been published to rigorously compare PEARL (Pesticide Emission Assessment at Regional and Local scales; Leistra et al., 2001), MACRO (Water and solute transport in macroporous soils; Larsbo and Jarvis, 2003), and PRZM (Pesticide Root Zone Model; Carsel et al., 1998), three of the four models used for pesticide registration in Europe (FOCUS, 2000). Indeed, it is often found in the literature that the discrepancies between the results predicted by different models were consequence of the different parameterizations used as authors did not try to minimize them. In addition, each user independently parameterizes each model with its own experience and/or subjectivity (Armstrong et al., 2000; Gottesbüren et al., 2000; Herbst et al., 2005; Vanclooster et al., 2000). Maize (Zea mays) is the most cultivated crop in the world with a sowing area about 175 million ha (Arvalis, 2013). It ranks the third leading crop after wheat and oilseed rape crops in France covering 1,532,000 ha with a mean grain yield of 9 t ha−1 in 2011. To maintain this yield, large amounts of water are needed and irrigation is frequently required along with an intensive use of pesticides, mainly herbicides: a mean of 3.2 kg active ingredients (a.i.) ha−1 are annually applied in the French maize cultures (UIPP, 2014). Because of the atrazine banning, S-metolachlor [2-chloro-N-(2ethyl-6-methylphenyl)-N-(methoxy-1-methylethyl) acetamide] and mesotrione [2-(4-mesyl-2-nitrobenzoyl)cyclohexane-1,3-dione] are now among the most used herbicides for maize. Both chemicals are used as selective pre- and post-emergence herbicides for the control of annual grasses and some broad-leaved weeds although S-metolachlor is more frequently used in agriculture than mesotrione because of its wider use in sorghum, cotton, potato, soybean, peanut or sunflower (Tomlin, 2003). Different behaviors in the soil sorption and degradation processes are observed for these herbicides, linked to

their different chemical properties. S-metolachlor has a relatively high solubility in water (480 mg L−1) while a moderate solubility is shown for mesotrione (160 mg L−1) (PPDB, 2014). Mesotrione is a weak acid with a pKa of 3.12, it dissociates from the molecular to anionic form as pH increases. Such properties affect its adsorption and degradation rate in the soil (Dyson et al., 2002): the adsorption of mesotrione in the soil is significantly influenced by both pH and soil organic carbon (OC) content (Koc ranges from 390 L kg−1 at soil pH 4.6, to 15 L kg−1 at soil pH 7.7) while its degradation is mainly a function of soil pH (DT50 ranges from 32 days at pH 5.0 and 2.0% OC, to 4.5 days at pH 7.1 and 3.3% OC). In contrast, S-metolachlor is a non-ionic compound and its adsorption and degradation do not show pH sensitivity (PPDB, 2014). The OC content is the main soil property that controls the S-metolachlor adsorption process with values of Koc ranging from 62 to 372 L kg−1 (Alletto et al., 2013). The DT50 of S-metolachlor in the soil ranges from 7.6 to 37.6 days under laboratory conditions (PPDB, 2014). According to these properties and their GUS leaching index (GUSS-metolachlor = 1.94 and GUSMesotrione = 3.43), S-metolachlor and mesotrione are classified as herbicides with a moderate and high leachability, respectively. This explains that both herbicides have been frequently found in surface and ground waters (Alferness and Wiebe, 2002; Baran and Gourcy, 2013; Cerejeira et al., 2003; Hildebrandt et al., 2008). In particular, in the South-West region of France, water monitoring in 2012 showed that S-metolachlor was detected in 42% of water samples in surface water, with concentration reaching 25 μg L−1. In groundwater, metolachlor was also detected but less frequently (b10% of detection) (Agence de l'eau Adour-Garonne, 2012). Such a contamination of water resources is directly associated to the intensive irrigated maize production of this region. To better understand the relationships between agricultural practices and water contamination, ‘site-specific’ tools are needed to have a more accurate evaluation of the effects of agricultural practices on the risks of water pollution by pesticides. Numerical models are promising tools for risk assessment but, for now, no study has been published for the regional context of maize cropping systems. This work focused on the comparison of the ability of three models, PEARL, MACRO and PRZM, to describe the behavior of water and of the herbicides S-metolachlor and mesotrione as observed under field conditions in an irrigated conventional maize monoculture system representative of the main maize cropping system in France. In this work, we paid a special attention to reduce the sources of discrepancies by minimizing the differences in the parameterization of the three models that were used by a single operator.

2. Materials and methods 2.1. Field experiment Simulations were based on field experimentations set up in Toulouse area (France) over a 2-year monitored period (2011–2012). An irrigated maize monoculture system was designed in an experimental plot of 12 m × 60 m, under a clay loam soil (Stagnic Luvisol according to IUSS Working Group WRB, 2007). The main soil characteristics are given in Table 1. The studied maize cropping system, which is considered as a reference for maize cultivation in France, is based on a maximization of the growth margin. For this system, a late maize variety was sown after a conventional tillage with a mouldboard plow (25–28 cm depth). During the maize growing season, irrigation period generally starts in June and ends in September–October. The total amount of water applied by sprinkler irrigation was 560 mm (250 mm in summer 2011, and 310 mm in summer 2012) with 30 or 40 mm per event as shown on Fig. 1. After harvest, a bare soil is maintained during the fallow period. The development dates of maize for the two years are included in Table 2.

J.M. Marín-Benito et al. / Science of the Total Environment 499 (2014) 533–545

535

Table 1 Soil physicochemical and hydraulic characteristics. Characteristic/Depth (m)

0–0.1

0.1–0.3

0.3–0.6

0.6–1.0

1.0–2.0 a

Clay (%) Silt (%) Sand (%) OC (%) pH (0.01 M CaCl2) Bulk density (g cm−3) θr (m3 m−3)b θs (m3 m−3)b α (cm−1)b n (−) b Ksat (m d−1) CTEN (cm) Kb (m d−1) θb (m3 m−3)g ASCALE (m)f ZN (−)h θFC (m3 m−3)g θWP (m3 m−3)g

32.2 45.2 22.6 1.38 6.68 1.50 0.080 0.414 0.444c 0.010 1.464 1.391c 0.528 10 0.014 0.411 0.441c 0.006 3 0.352 0.383c 0.114 0.133c

34.6 42.8 22.6 1.07 6.40 1.50 0.083 0.419 0.449c 0.011 1.434 1.363c 1.560 10 0.024 0.415 0.445c 0.006 3 0.352 0.384c 0.120 0.141c

35.5 44.0 20.5 0.95 7.13 1.56 0.081 0.406 0.426c 0.011 1.409 1.338c 0.240d 10 0.034f 0.402 0.422c 0.030 2 0.343 0.359c 0.122 0.125 c

43.8 39.4 16.8 0.71 7.76 1.63 0.084 0.399 0.419c 0.014 1.307 1.242c 0.024e 10 0.007f 0.394 0.414c 0.030 2 0.338 0.365c 0.146 0.177c

33.9 22.1 44.0 0.24 7.87 1.63 0.071 0.381 0.020 1.258 0.240d 10 0.026f 0.373 0.020 3 0.312 0.142

Note: the parameters without exponent correspond to measured parameters. a Physicochemical and hydraulic characteristics measured or estimated for the horizon 1.0–1.6 m and also used to parameterize the horizon 1.6–2.0 m. b Estimated by Rosetta's pedotransfer functions. c Calibrated values. d From measurement done in a neighbor plot (Alletto, 2007). e From measurement done in a neighbor plot (Alletto, 2007) and considering the high clay content of the horizon. f Estimated using MACRO 5.0/5.1 pedotransfer functions. g Calculated from the water retention curves. h Estimated from Beulke et al. (2002).

S-metolachlor and mesotrione herbicides were applied to the whole surface of the plot. None of the two herbicides has been applied before. S-metolachlor was sprayed annually at 1.25 and 1.52 kg a.i./ha as Mercator Gold (Syngenta) formulation on May 5, 2011 (pre-emergence), and as Calibra (Syngenta) on May 3, 2012 (post-emergence), respectively. Post-emergence herbicide application involved an interception rate by maize, which was estimated as a function of the crop development stage. Two percent of the S-metolachlor applied rate was assumed to be intercepted by the crop in 2012 (expert judgment). Mesotrione was only applied in 2012. It was sprayed (post-emergence) twice at 0.152 and 0.150 kg a.i./ha as Calibra (Syngenta) on May 3, 2012, and as Callisto (Syngenta) on June 1, 2012, respectively. Two percent and 10% of the mesotrione applied rate were assumed to be intercepted by the plants in the first and second application, respectively. The climate was monitored with a meteorological station located on the site. Tensiometers, TDR probes and temperature sensors were installed at 0.2, 0.5 and 1 m depth to monitor soil pressure head, water content and temperature, respectively. One tension plate lysimeter of 32 cm-diameter with a fixed tension at −100 hPa was installed at 1 m depth in the middle of the plot to carry out water flow measurements, and the further quantification of herbicide leaching. A vacuum pump

was used to suck out periodically the water accumulated in the plate lysimeter. Leachate samples were kept at 4 °C until analysis (usually within 48 h). 2.2. Pesticide fate models Three models, used at the EU level for environmental risk assessment and registration purposes (EFSA, 2004; FOCUS, 2002; Vanclooster et al., 2003), were selected: PEARL 4.4.4. (Leistra et al., 2001), MACRO 5.2 (Larsbo and Jarvis, 2003), and PRZM 3.21 (Carsel et al., 1998). The models differ in their description of water and solute transport (Table 3). PRZM is a one-dimensional model used to simulate the movement of chemicals in unsaturated soil systems within and immediately below the plant root zone. Its description of soil hydrology is based on a ‘tipping-bucket’ approach where water will only percolate to the deeper soil layer if field capacity is exceeded. Solute transport is described by convection and numerical dispersion. PEARL and MACRO are also one-dimensional models but they show a more complex description of water transport than PRZM. PEARL implements Richards' equation and the convection–dispersion equation (CDE) to simulate the water flow and solute transport, respectively. MACRO is a dual-permeability model, which

Fig. 1. Measured rainfall and irrigation applied in the soil along the experimental period (10 July 2010 to 15 January 2013).

536

J.M. Marín-Benito et al. / Science of the Total Environment 499 (2014) 533–545

Table 2 Input parameters for maize development. Year

Date

Crop development

LAI (m2 m−2)a

Root depth (m)a

Root distributionb,c

Crop height (m)a

2011

9 May 20 July 14 October 27 April 12 July 8 October

Emergence Flowering Harvest Emergence Flowering Harvest

0.00 3.29 3.29 d 0.00 3.29 3.29 d

0.01 0.80 0.80 0.01 0.80 0.80

0.67

0.00 2.50 2.50 0.00 2.50 2.50

2012

0.67

a

Determined from field measurements in 2011 and re-used in 2012. Fraction of root density in the uppermost 25% of the root depth for PEARL and MACRO. c From Jarvis et al. (2007). d Green leaf area index at harvest (LAIHAR parameter) was set to 2 in MACRO, which is a default value for maize (FOCUS, 2000). b

includes a description of preferential flow processes by dividing the pore system into two flow domains, micropores and macropores. The boundary between the two domains is defined by a soil water pressure head close to saturation, and its associated water content and hydraulic conductivity. Water flow in micropores is calculated by Richards' equation while it is gravity driven in the macropore domain. Solute transport in micropores is given by the CDE but it is assumed to be solely convective in macropores. Exchange between the two domains is calculated according to approximate, physically based expressions using an effective aggregate half-width. Unlike PRZM, PEARL and MACRO are able to take into account the upward movement of water and solute. Linear, Freundlich and instantaneous adsorption can be simulated by the three models (Table 3). First-order degradation kinetics of pesticides is assumed by the three models, although PRZM also enables the use of a bi-phasic equation for this process. MACRO simulates degradation and sorption processes in both micro- and macropore domains. All three models are able to consider the effect of soil moisture content and soil temperature on the pesticide degradation rate. PRZM and PEARL can simulate the pesticide volatilization while MACRO does not include a comprehensive description of this process. PRZM is the only one of the three models, which simulates soil erosion and surface runoff using the Modified Universal Soil Loss Equation (MUSLE) and a

modified Soil Conservation Service curve number technique, respectively. These two subroutines were switched off in this study because the field slope was below 1% (negligible soil erosion) and because no surface runoff was observed in the field. A summary of the most important processes simulated by the three models is included in Table 3. 2.3. Model parameterization Models were parameterized with all available site-specific data, and the parameterization was completed using soil and pesticides properties from the literature and pedotransfer functions to determine the unavailable parameters. Different pedotransfer functions (PTF) (Rawls et al., 1982; HYPRES (Wösten et al., 1999); FOOTPRINT and MACRO 5.0/5.1 included in the MACRO 5.2 GUI; Rosetta's (Šimůnek et al., 2008)) were tested for the determination of some parameters. The PTF giving the best fit between the observed and simulated data were chosen. A 1-D soil profile of 2 m split into five horizons of different thickness was simulated by PEARL and MACRO with free drainage condition at the bottom of the soil profile because the groundwater level was below 2 m depth during the whole period of simulation. A soil profile of 1 m split into four horizons was considered for PRZM because it does not simulate the upward movement of water and solute. For MACRO and PEARL, van Genuchten's soil–water retention parameters (θr, θs, α and n) were obtained from soil texture and bulk density using Rosetta's pedotransfer functions of HYDRUS 1-D. To minimize the differences in parameterization of water-retention properties between models, the water retention curves were then used to calculate the soil water content at field capacity (θFC, pF = 2) and at wilting point (θWP, pF = 4.2) for PRZM. The parameterization of the description of macropores flows in MACRO was done as follows: default values were used for the parameter which defines the boundary soil water pressure head between micro and macropores (CTEN = 10 cm), and the water content corresponding to this boundary soil water pressure head (θb) was calculated from the water retention curves. The tortuosity/pore size distribution factor for macropores (ZN) was determined following the recommendations of Beulke et al. (2002). Other parameters like the boundary hydraulic conductivity (Kb) in the horizons 0.3–2 m, and the

Table 3 Summary of the most important processes simulated by PEARL, MACRO and PRZM models. Processes Water Hydrology

Pesticides Sorption

PEARL

MACRO

PRZM

Richards' equationa Capillarity

Micropores: Richards' equation Macropores: gravity flow Capillarity Preferential flows

‘Tipping-bucket’ approach

Linear or Freundlich Instantaneous or non-equilibrium

Linear or Freundlich Instantaneous or f(time) In micro- and macropores 1st order In micro- and macropores Empirical relationship or global coefficient (leaves) f(transpiration) Micropores: convection–dispersion Macropores: convection – –

Linear or Freundlich Instantaneous, reversible or f(time)

Degradation

1st order

Volatilization Absorption/plant uptake Solute transport

From soil and plant surface f(transpiration) Convection–dispersion

Erosion Surface runoff

– –

Crop Crop growth

Function of development stage

Bi-linear (height, root depth) and linear + two power-law phases (GLAI): annual crop Constant: perennial crop

Linear

Temperature Soil temperature

Heat conduction equation

Heat conduction equation

Heat conduction equation

a

PEARL uses the SWAP model (Kroes et al., 2008) as a sub-model for the simulation of water flow and soil temperature.

1st order or bi-phasic From soil and plant surface f(transpiration) Convection + numerical dispersion Modified Universal Soil Loss Equation Curve number approach

J.M. Marín-Benito et al. / Science of the Total Environment 499 (2014) 533–545

effective diffusion pathlength (ASCALE), which controls the exchange of both water and solute between the micropore and macropore flows were estimated with the MACRO 5.0/5.1 pedotransfer functions of the model. Saturated hydraulic conductivity values (Ksat) used in MACRO and PEARL were extrapolated from the K–h exponential relationship obtained for the hydraulic conductivity values measured at h = − 10, − 6, − 3 and − 1 hPa using a tension disc infiltrometer in the first 0.3 m depth. The hydraulic conductivities measured at h = − 10 hPa in the topsoil were directly used to parameterize the Kb parameter at this depth. The K–h relationship was calculated from the steady-state infiltration rates in accordance with the multipotential technique developed by Ankeny et al. (1991) and Reynolds and Elrick (1991), based on Wooding's solution for steady-state asymmetrical infiltration (Wooding, 1968). A value of 0.24 m d−1 was used for Ksat in the 0.3– 0.6 m and 1.0–2.0 m horizons. These values were based on measurement done in a neighbor plot (Alletto, 2007). The higher clay content of the 0.6–1.0 m horizon compared to the two other horizons justified the decrease of its Ksat value until 0.024 m d−1. Soil hydraulic parameters are shown in Table 1. Topsoil S-metolachlor sorption (Kd) and degradation (DT50) parameters came from the literature (Alletto et al., 2013) and from laboratory experiments carried out on a soil with similar characteristics than the studied soil, respectively (Table 4). The Kd of Alletto et al. (2013) was measured in a soil having the same OC content as the studied soil, and we assumed the OC was the main soil property which controls the adsorption of S-metolachlor (Alletto et al., 2013). The Kd and DT50 values for deeper soil layers were estimated assuming these coefficients are proportional to OC content as it is often done to model the fate of pesticides (Alavi et al., 2007; FOCUS, 2000; Jarvis, 1995) (Table 4). As no regional data were available for mesotrione, its Kd and DT50 were determined from the results of Dyson et al. (2002) who showed that they are related to soil pH (in CaCl2) and OC amount. Thus the mesotrione sorption coefficients for each soil horizon were determined with the following equation: log10 ðKoc Þ ¼ −0:376pH þ 4:05

ð1Þ

where Koc is the linear partition coefficient Kd (L kg−1) normalized by the soil OC as follows: Koc ¼ ðKd =%OCÞ  100:

ð2Þ

Then, the following equation was used to estimate the mesotrione half-life for each soil horizon (Dyson et al., 2002): log10 ðDT50 Þ ¼ −0:192pH þ 2:18:

k ¼ kref f W f T

ð4Þ

where k is the degradation rate coefficient (d−1), kref is the reference rate coefficient (d− 1), and fW and fT are factors accounting for the influence of soil moisture and temperature, respectively. In PRZM and PEARL, the effect of soil moisture is determined according to the Walker equation (Walker, 1974): fW ¼

θ θref

!β ð5Þ

where θ is the soil water content (m3 m−3), θref is the reference water content (m3 m−3) and β (−) is an empirical parameter which was set to 0.7 according to FOCUS (2000). In MACRO, fW is defined by a modified approach according to Schroll et al. (2006): fW ¼ 0   θ−0:5θW β fW ¼ θ100 −0:5θW fW ¼ 1

S-metolachlor 3

Kd (cm g

0–0.1 0.1–0.3 0.3–0.6 0.6–1.0 1.0–2.0 a

−1 a

)

Mesotrione DT50 (d)

b,c

Kd (cm3 g−1)d

DT50 (d)

Uncalibrated

Uncalibrated

Uncalibrated

Uncalibratede/ Calibrated

0.80 0.62 0.55 0.41 0.14

29 29 58 99 n.d.f

0.48 0.47 0.22 0.10 0.03

7.9/15.8 8.9/17.8 12.9/25.8 16.5/33.0 n.d.f/n.d.

Topsoil data (0–0.1 m) taken from Alletto et al. (2013). Values for deeper soil layers were estimated assuming sorption coefficients are proportional to organic carbon content. b Topsoil data (0–0.1 m) obtained from laboratory experiments on a soil with similar characteristics than the studied soil. c Variation of degradation rate k (k (d−1) = ln (2)/DT50) with depth: k for 0–0.3 m, k × 0.5 for 0.3–0.6 m, k × 0.3 for 0.6–1.0 m, and k = 0 for N 1.0 m (FOCUS, 2000). d Calculated from Eqs. (1) and (2) (Dyson et al., 2002). e Calculated from Eq. (3) (Dyson et al., 2002) and considering the variation of k with depth. f n.d. = no degradation.

if θ b 0:5θW if 0:5θW ≤ θ ≤ θ100

ð6Þ

if θ N θ100

where θW is the water content at wilting point (m3 m−3) and θ100 is the water content corresponding to a pressure head of −100 hPa. The value of β was set to 0.49 to be consistent with PRZM and PEARL. In PEARL, the effect of temperature on degradation rate is considered using the Arrhenius equation: "

Depth (m)

ð3Þ

Mesotrione sorption coefficients and half-lives are shown in Table 4. For both S-metolachlor and mesotrione, the recommendations of FOCUS (2000) were followed to take into account the influence of depth on the degradation rates, and the Freundlich exponent, 1/n, was set to 1 in order to represent linearity of sorption (PPDB, 2014; Tomlin, 2003). The effects of soil moisture and temperature on the degradation of both herbicides are calculated from the following general equation (the formalism is different in the three models, see below):

f T ¼ exp − Table 4 Sorption coefficients (Kd) and half-lives (DT50) of S-metolachlor and mesotrione in the soil profile.

537

Ea R

1 1 − T T ref

!# for 273K ≤ T ≤ 308K

ð7Þ

where Ea is the molar activation energy (kJ mol−1), R is the molar gas constant (kJ mol−1 K−1), T is the soil temperature (K), and Tref is the reference temperature (K). The value of Ea was set to 65.4 kJ mol−1 (EFSA, 2007). In PRZM, the effect of the temperature is taken into account according to the following equation (similar to an Arrhenius equation) (Carsel et al., 1998): Q10 ¼ QFAC

ðT−T ref Þ

. 10

ð8Þ

where Q10 is the correction factor for degradation based on soil temperature, QFAC is the factor for degradation rate increase when temperature increases by 10 °C. The default value of 2.58 as proposed by EFSA (2007) was used for Q10. Finally, in MACRO, the influence of temperature on the degradation rates is determined according to the approach of Boesten and van der Linden (1991): f T ¼ exp ½α ðT−T ref Þ for T N 278K

ð9Þ

538

J.M. Marín-Benito et al. / Science of the Total Environment 499 (2014) 533–545

where α (K− 1) is an empirical parameter which can be determined from the following equations: 10α

while an overestimation of the observed values takes place if negative CRM values are determined. A CRM value of 0 indicates the absence of trends.

ð10Þ

Q 10 ¼ e

3. Results and discussion α ¼ Ea =686; 270 with Ea in J mol

−1

:

ð11Þ

A value of 0.095 K−1 was obtained. Required climatic data (solar radiation, air temperature, relative humidity, wind speed and rainfall) came from the meteorological station aforementioned. Potential evapotranspiration was calculated outside the models using the Penman equation (Penman, 1948) and inputted into the three models in such a way that all models have the same ETP. PRZM usually uses pan evaporation data and a crop factor (PFAC) to estimate the ETP, therefore PFAC was set to 1. Finally, the cropping parameters were determined from field measurements in 2011 and re-used in 2012 or taken from literature (Table 2). Root density profiles were defined as similar as possible in all models. MACRO assumes that the root density varies logarithmically with depth and allows the user to set a value for the fraction of root density in the uppermost 25% of the root depth. PEARL allows a free definition of the root density profile and thus it was parameterized similarly than in MACRO (Table 2). On the contrary, PRZM assumes a triangular root distribution from the soil surface to the maximum rooting depth, with the maximum root density being near the surface (this cannot be modified by the user). Each simulation started with a warm up period of 9 months in order to decrease the impact of the initial conditions onto the results because temperature and soil moisture measurements were not available at the beginning of the experiment along the soil profile. Hydrostatic pressure equilibrium was thus initially considered in the soil profile. The simulation period lasted from 10 July 2010 to 15 January 2013. 2.4. Evaluation of model performance Two statistical indices, the modeling efficiency (EF) (Nash and Sutcliffe, 1970), and the coefficient of residual mass (CRM) (Smith et al., 1996) were calculated to evaluate the performance of the models. EF indicates whether the simulated values correspond closely to observed values and it is determined with the following equation: 2 n X ðSi −Oi Þ

EF ¼ 1− ni¼1 X 2 ðOi −Om Þ

ð12Þ

i¼1

CRM ¼

Oi

i¼1

n X

Model calibration was necessary because important discrepancies between measured and simulated data were first observed, especially at the end of the simulation period (since 20 May 2012) when the studied pesticides were detected in the leachates (EF b 0.23, results not shown). The calibration was done in two steps: the first step involved the calibration of the soil hydraulic dynamic, and the second one the herbicide behavior. In the first step, we calibrated the hydraulic parameters of PEARL and MACRO that were determined by pedotransfer functions and that are known to have a large influence on the prediction of soil moisture and/or percolation (Dubus et al., 2003; Kolupaeva et al., 2006; Scorza Júnior and Boesten, 2005). The best results were obtained with the Rosetta's pedotransfer function and the MACRO 5.0/5.1 for the additional Kb and ASCALE parameters of MACRO. We further improved the results by decreasing by 5–7% the n values, and by increasing by 5% the θs values in the calibration process for the first four soil horizons. The model results were not found sensitive to a further variation of the alpha values. Calibrated values are shown in Table 1. The calibration allowed a better description of the experimental data, especially at the end of the simulation period (since 20 May 2012) (Figs. 2, 3, Table 5). After the hydraulic parameter calibration, mesotrione DT50 was calibrated to minimize the marked underestimation of its leaching as simulated by the three models at the lysimeter depth (1 m). The initial simulations using the uncalibrated DT50 values gave mean concentrations of mesotrione from 2 to 3 orders of magnitude lower than the measured concentrations in the leachates (results not shown). Using the Kd values estimated by Eqs. (1) and (2), the simulated mobility of mesotrione agreed with the field data. However, a good description of mesotrione field persistence could not be obtained with the initial half-lives that were estimated according to Eq. (3) (Dyson et al., 2002). For this reason, the initial DT50 as calculated with Eq. (3) was multiplied by a factor of two according to Dyson et al. (2002) who observed that the DT50 values fitted by this equation were within a factor of two of the measured DT50 that were used to determinate Eq. (3). These new half-lives are within the range of values reported in the literature (PPDB, 2014). Finally, the recommendations of FOCUS (2000) were followed to take into account the variation of degradation with depth. Initial and modified half-lives are shown in Table 4. 3.2. Water balance

where Si and Oi are the simulated and observed values, respectively, Om is the mean of observed values and n is the number of sampling dates. This coefficient ranges from −∞ to 1. EF = 1 indicates a perfect fit while values lower than 0 result from a worse fit than the average of observations. CRM gives an indication of the consistent errors in the distribution of all simulated values across all measurements with no consideration of the order of the measurements. It was calculated with the following expression: n X

3.1. Calibration

n X i¼1

Si

The total water balance simulated by the models is shown in Table 6. Some differences are observed despite the same potential evapotranspiration was used in the three models. The actual evapotranspiration (ETR) as predicted by PEARL was lower than the ETR simulated by MACRO and PRZM by about 7%. As a result, total water drainage at the bottom of soil profile was smaller for MACRO and PRZM compared to PEARL. MACRO and PRZM showed a similar ETR although MACRO simulated a smaller drainage than the capacity model which showed the lowest change in soil water storage in the soil profile. 3.3. Soil water content and water pressure head

:

ð13Þ

Oi

i¼1

This coefficient ranges from −∞ to +∞. Positive CRM values indicate an underestimation of observed values with respect to simulated data

Figs. 2 and 3 show the observed and simulated soil water content and soil water pressure head at 0.2, 0.5 and 1 m depth after the calibration aforementioned, respectively. PEARL and MACRO simulations showed similar water flow dynamics for the whole period at the three different depths although PEARL generally predicted a higher degree

J.M. Marín-Benito et al. / Science of the Total Environment 499 (2014) 533–545

Soil water content (m3 m-3)

0.5

crop

0.2 m depth

539

crop

0.4 0.3 0.2 0.1

Observed 0.0 02/05/10

30/08/10

28/12/10

27/04/11

PEARL

25/08/11

23/12/11

MACRO 21/04/12

PRZM

19/08/12

17/12/12

Soil water content (m3 m-3)

Date 0.5

crop

0.5 m depth

crop

0.4 0.3 0.2 0.1

Observed 0.0 02/05/10

30/08/10

28/12/10

27/04/11

PEARL

25/08/11

23/12/11

MACRO 21/04/12

PRZM

19/08/12

17/12/12

Date

Soil water content (m3 m-3)

0.5

crop

1 m depth

crop

0.4 0.3 0.2 0.1

Observed 0.0 02/05/10

30/08/10

28/12/10

27/04/11

PEARL

25/08/11

23/12/11

MACRO 21/04/12

PRZM

19/08/12

17/12/12

Date Fig. 2. Observed and simulated (after calibration) soil water content at 0.2, 0.5 and 1 m depth.

of saturation of the soil than MACRO because MACRO simulated higher ETR than PEARL during an important part of the simulation period (data not shown). However, a close degree of saturation of the soil was predicted by both models in periods with similar ETR predictions as for example in the cropping periods (ETRMACRO = 521 mm and ETRPEARL = 523 mm in 2011; ETRMACRO = 551 mm and ETRPEARL = 542 mm in 2012). In particular, both models displayed generally similar predictions at 0.2 m depth during wet conditions on bare soil periods (autumn–winter periods characterized by low ETR, data not shown), and during the irrigated cropping period or after important events of rainfall (10 October 2010, 20 May 2012 and 21 October 2012). The simulated values from the irrigated cropping period of 2012 agree closely with available observed data as shown by the good EF values obtained for water pressure head at these periods (generally higher or equal to 0.48), and CRM values close to 0 (Table 5). However, simulated soil moisture and water pressure were overestimated during the period before the rainfall event of 20 May 2012 compared to the observed data. These discrepancies explained the lower EF and higher CRM values obtained when considering the whole periods of available data (Table 5). MACRO generally showed better fitting coefficients than PEARL, especially at 0.5 m depth. Jarvis (1995) also observed a

decrease in the agreement between simulated and measured soil water contents with depth. Compared to MACRO and PEARL, PRZM did not simulate properly water dynamics (Fig. 2, Table 5). Gottesbüren et al. (2000) also observed the difficulties of other capacity models to describe the water dynamics in soil. PRZM predicted that the soil was at field capacity during extended periods of the simulation at the three depths. The duration of these periods increased with depth because evaporation and transpiration decrease. Thereby at 1 m depth, the water content kept fixed to the water content value of field capacity. It means that microporosity was always saturated at this depth, probably because the maximum root depth was 0.8 m. Similar results were observed by Garratt et al. (2002). By contrast, PRZM predicted similar fluctuations of the soil moisture than PEARL and MACRO during the cropping periods at 0.2 m depth while for the remaining time it described the soil at field capacity. This behavior is explained by the plant transpiration process in the cropping periods while the absence of fluctuations in the bare soil periods is due to the lack of simulation of evapotranspiration. Without crop, the evapotranspiration is controlled by the “maximum depth for evaporation” (ANETD) parameter, which was set to 10 cm. Thus, there is no evaporation below this depth and consequently the

J.M. Marín-Benito et al. / Science of the Total Environment 499 (2014) 533–545

Pressure head (hPa)

540

02/05/10 200

30/08/10

28/12/10

27/04/11

25/08/11

23/12/11

21/04/12

crop

19/08/12

17/12/12

crop

0 -200 -400 -600 -800

Observed

0.2 m depth

PEARL

MACRO

Date

Pressure head (hPa)

02/05/10 200

30/08/10

28/12/10

27/04/11

25/08/11

23/12/11

21/04/12

crop

19/08/12

17/12/12

crop

0 -200 -400 -600

Observed

0.5 m depth

PEARL

MACRO

-800

Pressure head (hPa)

Date 02/05/10 0

30/08/10

28/12/10

27/04/11

25/08/11

23/12/11

21/04/12

crop

19/08/12

17/12/12

crop

-200 -400 -600

Observed

1 m depth

PEARL

MACRO

-800

Date Fig. 3. Observed and simulated (after calibration) soil water pressure head at 0.2, 0.5 and 1 m depth.

soil water content is field capacity. A marked decrease of soil water content was simulated by PRZM at 0.2 m depth between 23 May and 22 June 2012, when it decreased close to the wilting point. This indicates that the root water uptake was overestimated in this period, and it was only predicted by PRZM probably due to its root distribution

assumption, which is different from that of PEARL and MACRO (see Section 2.3). Differences in the herbicide leaching can be expected as a consequence of the different soil water contents as simulated by the three models, and taking into account that they consider the effect of soil moisture content on the pesticide degradation rate.

Table 5 Modeling efficiency (EF) and coefficient of residual mass (CRM) of PEARL, MACRO and PRZM for the soil water content (θ), pressure head (h) and temperature (T) at 0.2, 0.5 and 1 m depth, and for the percolation at 1 m depth (after calibration). Parameter

PEARL

MACRO

PRZM

0.2 m

0.5 m

1m

0.2 m

0.5 m

1m

0.2 m

0.5 m

1m

EF θ (whole period) θ (since 20/05/2012) h (whole period) h (since 20/05/2012) T (whole period) Percolation

0.34 0.65 0.09 0.66 0.92 –

−0.46 0.23 −0.04 −0.25 0.78 –

−0.82 0.39 −0.52 0.59 0.57 −13.80

0.47 0.38 0.49 0.55 0.92 –

0.19 0.48 0.52 0.55 0.89 –

−0.65 0.50 −0.16 0.65 0.94 −14.07

−1.79 −3.00 – – 0.92 –

−1.31 −1.53 – – 0.94 –

−2.75 −1.29 – – 0.94 −55.09

CRM θ (whole period) θ (since 20/05/2012) h (whole period) h (since 20/05/2012) T (whole period) Percolation

−0.04 0.00 −0.30 −0.13 −0.05 –

−0.10 −0.04 −0.29 −0.24 −0.06 –

−0.07 −0.03 −0.21 −0.05 −0.05 −2.80

0.01 0.05 −0.07 0.13 −0.04 –

−0.05 0.01 −0.03 0.06 −0.04 –

−0.06 −0.02 −0.13 0.05 0.00 −3.67

0.05 0.12 – – −0.01 –

−0.02 0.09 – – −0.02 –

−0.10 −0.06 – – 0.01 −8.40

J.M. Marín-Benito et al. / Science of the Total Environment 499 (2014) 533–545

(PEARL), 2.2 (MACRO) and 6.6 (PRZM) for the period of available data (18 April 2011–17 December 2012) although this overestimation is generally concentrated at the end of the simulation period (30 October–17 December 2012). The highest percolation predicted by PRZM compared to MACRO and PEARL was mainly due to the approximation carried out on the percolation results predicted by the models based on Richard's equation to mimic the lysimeter condition (h = − 100 hPa). While in PEARL and MACRO vertical water flow occurs throughout all water potential range according to the K–h relationship, it occurs only in PRZM if saturation condition of the soil is above field capacity (h = −100 hPa). Applying the lysimeter condition on PEARL and MACRO leaching results allowed a better description of the experimental data suggesting that a non-negligible part of water flow was in the range of water potential inferior to h = −100 hPa. Indeed, the water flow simulated by PEARL and MACRO at h b −100 hPa potential, and at lysimeter depth, represented 66% and 64% of the total leached water, respectively. The lysimeter condition allowed the decrease in the overestimation of total water volume leachate at 1 m depth from 7.3 times to 2.5 times in PEARL, and from 6.0 times to 2.2 times in MACRO. In previous comparative studies, Garratt et al. (2002) also simulated a higher percolation with capacity models than with models based on Richard's equation but they assumed it was due to the absence of upward movement of water in the capacity models. In our study, PEARL and MACRO simulated a small upward movement of water by capillary rise (b 0.27 mm) in the whole period of simulation suggesting it was not an important process to explain the percolation discrepancies between both models and PRZM. MACRO did not simulate any leachate volume at 1 m depth until the end of the simulation period (30 October 2012–17 December 2012). From 23 May to 16 July 2012, two important events of leaching were observed in the field but only the percolation observed from 23 May to 12 June 2012 was predicted by PEARL and PRZM although with some differences. While PEARL simulated quite well the volume of leached water and the period in which it took place, PRZM strongly overestimated the volume, and predicted that the percolation happened in the prior period (26 April–22 May 2012) when water leachate was actually measured in the plate lysimeter. This can be explained by the fact that PRZM predicted that the whole soil profile was saturated after the precipitation events of 19–21 May 2012. Consequently the water fallen on the soil surface percolated quickly until 1 m depth while the movement of water predicted in PEARL was slower. Contrary to PEARL, MACRO did not simulate any percolation in this period due to its high predicted ETR (ETRMACRO = 75 mm and ETRPEARL = 64 mm). The second important event of percolation observed in the field, from 13 June to 16 July 2012 at the beginning of the irrigation period, was not predicted by any model. It could be explained by lateral flows above the lysimeter, which are not simulated by these onedimensional models. For the rest of the simulation period, PRZM markedly overestimated the water drainage in several periods before

Table 6 Soil water balance simulated by PEARL, MACRO and PRZM from 10 July 2010 to 15 January 2013. Process

PEARL

MACRO

PRZM

Precipitation + Irrigation (mm) Actual evapotranspiration (mm) Seepage at bottom of profile (mm) a Surface runoff (mm) Change in soil water storage in the soil profile (mm)

2148 1531 551 0 66

2148 1635 450 0 63

2148 1627 496 0b 25

a

Bottom of soil profile is at 1 m depth in PRZM, and at 2 m depth in PEARL and MACRO. Surface runoff was not considered in PRZM during the simulations because it was not observed in the field. b

Considering the whole simulated period, the performance of all models to simulate the water behavior was generally poor, even for the models based on Richards' equation: EF ranges from − 2.75 (capacity model) to 0.52 (Richards' model), and CRM ranges from − 0.30 to 0.05 (Table 5). Since 20 May 2012, EF values range from − 3.00 (capacity model) to 0.66 (Richards' model), and CRM ranges from −0.24 to 0.13. Simulating soil moisture with Richards' and capacity models, Vanclooster and Boesten (2000) obtained, after calibration, EF values ranging from − 4.77 (capacity model) to 0.94 (Richards' model). The global poorer performance of PEARL and MACRO on the whole data set as compared to the second period (after May 20, 2012) (Table 5) suggested that the parameterization was very sensitive to the available data. The water pressure head data in the first period were very scarce (Fig. 3) as compared to the water content data (Fig. 2) so that these data were not sufficient to achieve a good parameterization of the water retention curves of PEARL and MACRO models in the whole range of water saturation. 3.4. Percolation

PEARL

Observed

PRZM

MACRO

2.0 1.5 1.0 0.5

Period Fig. 4. Observed and simulated (after calibration) percolation at 1 m depth.

17/12/12

05/11/12

29/10/12

07/08/12

16/07/12

12/06/12

22/05/12

26/04/12

06/04/12

07/03/12

01/02/12

16/01/12

14/11/11

17/10/11

19/09/11

25/07/11

21/06/11

13/06/11

0.0

18/04/11

Leachate volume (mm)

Taking into account, on the one hand, that leachate samples were collected only in the plate lysimeter at soil pressure head higher than − 100 hPa and, on the other hand, that none of the three models can impose this condition, an approximation had to be applied to compare observed and simulated leaching data. Daily water percolation as simulated by PEARL and MACRO at 1 m depth was considered or not taking into account the daily pressure head at the same depth along the following condition: water percolation = 0 if soil pressure head b − 100 hPa. This approximation is not used for the capacity model PRZM because it was parameterized with a field capacity at pF = 2 (h =−100 hPa) (see Section 2.3). Observed and simulated leachates at 1 m depth after calibration are shown in Fig. 4. Neither the dynamic nor the total amount of water leachate was correctly simulated by any model as shown by the poor EF and the high CRM values (Table 5). The three models overestimated the total water volume leachate at 1 m depth (40 mm) by factors of 2.5

100.0 80.0 60.0 40.0 20.0

541

542

J.M. Marín-Benito et al. / Science of the Total Environment 499 (2014) 533–545

29 October 2012 in contrast to PEARL and MACRO that did not predict any event of leaching in agreement with the measured drainage data which were generally zero or close to zero. The soil, as simulated by PRZM at 1 m depth, was always at field capacity after 5 August 2010 which could explain the high simulated leaching in periods with important rainfall or irrigation events and low ETR. These two conditions were generally observed (data not shown) in each period where PRZM overestimated the leachate volume. 3.5. Soil temperature Pesticide degradation depends on temperature and consequently a good simulation of soil temperature along the soil profile by the models is important. Fig. 5 shows the observed and simulated soil temperature at 0.2, 0.5 and 1 m depth. PRZM overestimated the observed soil temperatures by as much as 7 °C at 0.2 m depth in February because PRZM cannot simulate negative soil temperatures. Apart from this specific case, the soil temperature was correctly reproduced by the three models at the three depths, as evidenced by the high EF and small absolute CRM values (Table 5). PRZM and MACRO showed the best predictions with similar values along the whole soil profile. The same EF values were obtained for the three models at 0.2 m depth

probably because the atmospheric conditions manage the soil heat flux at this depth. Herbst et al. (2005) obtained similar EF values for superficial soil temperature simulations with different leaching models. Contrary to PRZM and MACRO, the efficiency of PEARL to reproduce the soil temperature decreases with depth according to its high overestimation (from July to October) and underestimation (especially in February) of the measured values at deeper horizons (Fig. 5). Indeed, a detailed parameterization of initial soil temperature profile can be done in MACRO (one value of temperature per layer of soil) and in PRZM (one value of temperature per horizon of soil). Moreover, we used a constant value of temperature at the bottom of the soil profile in PEARL while PRZM is parameterized with one monthly value, and MACRO with the annual average air temperature and the annual amplitude in air temperature on a monthly basis that influence the bottom boundary condition in the model. Models generally simulated soil temperatures within ±3 °C of the observed values although this rate of variation increased until ±5 °C in PEARL predictions at 0.5 and 1 m depth. Due to these differences, associated differences in herbicide concentrations can be expected in the leachate because the herbicide degradation rates depend on the soil temperature. In any case, some authors indicate that this order of error in the temperature prediction are acceptable taking into account that there are other sources of

Soil temperature (°C)

30 25

0.2 m depth

20 15

Observed PEARL MACRO PRZM

10 5 0 07/04/11 -5

26/06/11

14/09/11

03/12/11

21/02/12

11/05/12

30/07/12

18/10/12

11/05/12

30/07/12

18/10/12

11/05/12

30/07/12

18/10/12

-10

Date 30

Soil temperature (°C)

0.5 m depth 25 20 15

Observed PEARL MACRO PRZM

10 5 0 07/04/11 -5

26/06/11

14/09/11

03/12/11

21/02/12

Date 30

Soil temperature (°C)

1 m depth 25 20 15 10 5 0 07/04/11

Observed PEARL MACRO PRZM 26/06/11

14/09/11

03/12/11

21/02/12

Date Fig. 5. Observed and simulated soil temperature at 0.2, 0.5 and 1 m depth.

J.M. Marín-Benito et al. / Science of the Total Environment 499 (2014) 533–545

December 2012, and by a factor of 2.2 for the period from 6 to 17 December 2012. In general, PRZM predicted the highest concentrations for both herbicides possibly due to a greater transport of these compounds related to high PRZM overestimation of water flow. Armstrong et al. (2000) and Beulke et al. (2001) also simulated the largest pesticide concentrations in drainage water with a capacity model (PLM) compared to models based on Richards' equation. Moreover, PRZM predicted the appearance of both herbicides in the lysimeter sooner than PEARL and MACRO because it predicted the event of percolation measured between 23 May and 12 June 2012 earlier than observed, as it was aforementioned. PEARL generally simulated the observed herbicide concentrations better than MACRO and PRZM. Considering that the models simulate the herbicide degradation rates as a function of soil moisture and soil temperature, the discrepancies in the results of simulation of these processes may have an important influence on the simulated herbicide concentrations, as has been found by other authors (Garratt et al., 2002). However, Gottesbüren et al. (2000) observed that a good prediction of soil water content does not always involve a right simulation of pesticide transport and vice versa. The comparison of pesticide concentrations can give misleading information because of the dilution effect due to the volumes of percolated water therefore it is better to compare pesticide leaching flows. The comparison of the observed and simulated herbicide flows (Fig. 7a and b) showed that in general the models underestimated the measured flows of both compounds in the periods ranging from 23 May to 16 July 2012, while they overestimated the flows for the periods ranging from 30 October to 17 December 2012. PRZM showed the highest overestimation of the herbicide flows. Cumulative observed mesotrione losses by leaching reached almost 0.033% of the applied dose, but the models predicted losses of 0.047% (MACRO) b 0.060% (PEARL) b 0.164% (PRZM). As expected, the less mobile herbicide, S-metolachlor, showed a smaller percentage of losses by leaching than

uncertainty that are much more influential in the simulation of pesticide fate, such as the spatial variability of soil properties (Jarvis, 1995).

3.6. Herbicide leaching

Mean concentration of S-metolachlor (µg L-1)

Because the simulated water leaching dynamics and absolute values of leachate water by period were very different from the observed data in some cases, we mainly compared from a qualitative point of view the simulated and observed time of appearance of the herbicides at 1 m depth. S-metolachlor was detected in the water leachate at 1 m depth 404 days after the first application date (May 5, 2011) at a concentration level of 0.230 μg L−1 (Fig. 6a). Then, S-metolachlor concentration decreased until zero in the last event of observed leachate. Simulated S-metolachlor concentrations at 1 m depth are also shown in Fig. 6a. The simulated concentrations increased with time contrary to the observed data. In general, the simulated concentrations were lower than the observed data except in the period 6–17 December 2012 when the three models overestimated the concentration of S-metolachlor. For mesotrione, the increase in DT50 values following calibration allowed the improvement of the leaching results (Fig. 6b). This herbicide was detected and quantified much sooner than S-metolachlor in the water leachate at 1 m depth, just 19 days after its first application date (May 3, 2012), and at a similar concentration level (0.240 μg L−1, Fig. 6b) despite the dose of mesotrione applied in the field was 10 times lower than that of S-metolachlor. Measured mesotrione concentrations do not show a clear trend with alternate increases and decreases whereas the simulated concentrations always increased as for S-metolachlor. For a majority of events, the three models underestimated the mesotrione concentrations as it happened for S-metolachlor. However, PRZM markedly overestimated the mesotrione concentration by a factor of 3.5 for the period from 30 October to 5 1.0

PEARL

Observed

543

a)

PRZM

MACRO

0.8 0.6 0.4 0.2 0.0 2

/12

/12

2/0

/12

/04

-1

/12

/12

/08

07

/12

/12

/12

17

/12

30

17

/12

/12

05

/10

/07

13

23

/12

/12

/07

16

/06

/05

26

/12

6/1

/05

22

06

Mean concentration of mesotrione (µg L-1)

Period 1.0

PEARL

Observed

PRZM

MACRO

b)

0.8 0.6 0.4 0.2 0.0 2

2 /12

/05

23

/1

/06

13

0 2-

7/1

0 2-

/1

/1

/07

17

2/1

5/1

7/0

1 2-

2

2/1

8/1

6/0

-1

2

2

7/1

2/0

/1

/04

26

2

6/1

5/1

2/0

2 2-

/10

30

1 2-

/1

/12

06

Period Fig. 6. Observed and simulated (after calibration) mean concentrations of S-metolachlor (a) and mesotrione (b) in the leachates for each period of measurement. The arrows mean that there was no observation (volumes of water in the lysimeter not sufficient for herbicide analyses).

J.M. Marín-Benito et al. / Science of the Total Environment 499 (2014) 533–545

Cumulative flows of S-metolachlor in the leachates (µg m-2)

544

40 36 32

PRZM

MACRO

a)

16 14 12 10 8 6 4 2 0 2

/12

/12

2/0

/12

/04

/12

6/1

/05

22

-1

/12

/12

/12

/08

07

/07

13

23

/12

/07

16

/06

/05

26

Cumulative flows of mesotrione in the leachates (µg m-2)

PEARL

Observed

/12

/10

17

/12

/12

05

/12

/12

17

/12

30

06

Period 40 36 32

PEARL

Observed

PRZM

MACRO

b)

16 14 12 10 8 6 4 2 0 2

2 /12

/05

23

/1

/06

13

0 2-

7/1

0 2-

/1

/1

/07

/10

30

17

2/1

5/1

7/0

1 2-

2

2/1

8/1

6/0

-1

2

2

7/1

2/0

/1

/04

26

2

6/1

5/1

2/0

2 2-

1 2-

/1

/12

06

Period Fig. 7. Observed and simulated (after calibration) cumulative flows of S-metolachlor (a) and mesotrione (b) in the leachates for each period of measurement. The arrows mean that there was no observation (volumes of water in the lysimeter not sufficient for herbicides analyses).

mesotrione: 0.001% (observed) b 0.004% (MACRO) b 0.007% (PEARL) b 0.025% (PRZM). In most cases, the herbicide concentrations quantified in the leachate were higher than the EU drinking water limit of 0.1 μg L−1 (Fig. 6a and b). Considering some unexplained inconsistencies between measured and simulated leached water volumes and the fact that the 1-D models 1) do not capture potential lateral flows due to the 3-D geometry of the velocity fields above the lysimeters and 2) were not able to apply a seepage face at a water potential of − 100 hPa, we did not go further into the calibration of the herbicide leaching. Moreover, the limited number of herbicide concentration data in the lysimeter may be insufficient to characterize correctly the processes and to better parameterize the models. Nevertheless, the qualitative comparison of time occurrence of the herbicides at 1 m depth helped improving the parameterization of herbicide behavior. 4. Conclusions For the first time, the PEARL, MACRO and PRZM models were rigorously compared (equivalent parameterization) and evaluated in terms of their ability to predict soil, hydrology and thermic balance as basic properties involved in their ability to predict the fate of pesticide. Field experimental data on leaching of two herbicides, S-metolachlor and mesotrione, were used to conclude this evaluation in an irrigated maize monoculture system. Discrepancies in the ETR as simulated by the three models played an important role in the different water dynamics as predicted by each model in the soil profile. Neither the dynamic nor the total amount of percolated water was correctly simulated by any model after the calibration of hydraulic parameters although PEARL and MACRO

predicted the water behavior better than the capacity model PRZM. Model's performance was better when distinguishing the bare soil periods from the cropping periods than when considering the whole period. For the whole simulated period, the performance of all models to simulate the water flow dynamics was poor, even for the models based on Richards' equation. The models were able to simulate the leaching of herbicides at 1 m depth in similar appearance time and order of magnitude as field observations even if some discrepancies can still be noticed, probably linked to the difficulties in describing water flow dynamics. Contrary to PRZM and MACRO, PEARL efficiency to simulate the soil temperature decreased with the depth because the parameterization of this module was simpler than that of PRZM and MACRO. However, PEARL generally simulated the observed herbicide concentrations better than MACRO and PRZM so that other factors as the soil moisture overridden or at least compensated the temperature effect on the herbicide degradation rate. The results obtained in this work showed the complexity of parameterizing the PEARL, MACRO and PRZM 1-D models on field datasets whose conditions could not be completely reproduced by the models, like the lysimeter condition. It also illustrates that in situations with a deficit of information, the use of data from the literature or of pedotransfer functions to determine some of the model parameters that are not often measured in situ or in laboratory is very useful although it does not always work. Finally, an equivalent parameterization of all models allows a better comparison of their performance. Acknowledgments JM Marín-Benito thanks the ANR Systerra (ANR-09-STRA-06: the research project Mic-Mac Design) for the financial support of his first

J.M. Marín-Benito et al. / Science of the Total Environment 499 (2014) 533–545

post-doctoral contract, and the financial support of ONEMA through the ECoPESt project for his second post-doctoral contract. The ECoPest project is supported by the research program “Assessing and reducing environmental risks from plant protection products” funded by the French Ministries in charge of Ecology and Agriculture. The authors thank Erik van den Berg for his collaboration in PEARL modeling, and Simon Giuliano for his help with the field data. References Agence de l'eau Adour-Garonne. Qualité des eaux et produits phytosanitaires sur le bassin Adour-Garonne – Situation en 2012; 2012 [11 pp.]. Alavi G, Dusek J, Vogel T, Green RE, Ray C. Evaluation of dual-permeability models for chemical leaching assessment to assist pesticide regulation in Hawaii. Vadose Zone J 2007;6:735–45. Alferness P, Wiebe L. Determination of mesotrione residues and metabolites in crops, soil, and water by liquid chromatography with fluorescence detection. J Agric Food Chem 2002;50:3926–34. Alletto L. Dynamique de l'eau et dissipation de l'isoxaflutole et du dicétonitrile en monoculture de maïs irrigué: effects du mode de travail du sol et de gestión de l'interculture (PhD Thesis) AgroParisTech; 2007 [318 pp.]. Alletto L, Benoit P, Bolognési B, Couffignal M, Bergheaud V, Dumény V, et al. Sorption and mineralisation of S-metolachlor in soils from fields cultivated with different conservation tillage systems. Soil Tillage Res 2013;128:97–103. Ankeny MD, Ahmed M, Kaspar TC, Horton R. Simple field method for determining unsaturated hydraulic conductivity. Soil Sci Soc Am J 1991;55:467–70. Arvalis. Institut du végétal website. http://www.arvalisinstitutduvegetal.fr/rapports-dactivites-@/_plugins/WMS_BO_Gallery/page/getElementStream.html?id= 17511&prop=file, 2013. (accessed 12.03.14). Armstrong A, Aden K, Amraoui N, Diekkrüger B, Jarvis N, Mouvet C, et al. Comparison of the performance of pesticide-leaching models on a cracking clay soil: results using the Brimstone Farm dataset. Agric Water Manag 2000;44:85–104. Baran N, Gourcy L. Sorption and mineralization of S-metolachlor and its ionic metabolites in soils and vadose zone solids: consequences on groundwater quality in an alluvial aquifer (Ain Plain, France). J Contam Hydrol 2013;154:20–8. Belmonte A, Garrido A, Martínez JL. Monitoring of pesticides in agricultural water and soil samples from Andalusia by liquid chromatography coupled to mass spectrometry. Anal Chim Acta 2005;538:117–27. Beulke S, Brown CD, Dubus IG, Harris G. Evaluation of uncalibrated referential flow models against data for isoproturon movement to drains through a heavy clay soil. Pest Manag Sci 2001;57:537–47. Beulke S, Renaud F, Brown CD. Development of guidance on parameter estimation for the preferential flow model MACRO 4.2. Final Report of the DEFRA project PL0538. Cranfield Centre for EcoChemistryUK: University of Cranfield; 2002. [contract No. JA3749E, 68 pp.]. Blanchoud H, Farrugia F, Mouchel JM. Pesticide uses and transfer in urbanised catchments. Chemosphere 2004;55:905–13. Boesten JJTI, van der Linden AMA. Modeling the influence of sorption and transformation on pesticide leaching and persistence. J Environ Qual 1991;20:425–35. Carsel RF, Imhoff JC, Hummel PR, Cheplick JM, Donigian Jr AS. PRZM-3: a model for predicting pesticide and nitrogen fate in the crop root and unsaturated soil zones: user's manual for release 3.12. Athens, GA: National Exposure Research Laboratory, Office of Research and Development, US Environmental Protection Agency; 1998. Cerejeira MJ, Viana P, Batista S, Pereira T, Silva E, Ferreira M, et al. Pesticides in Portuguese surface and ground waters. Water Res 2003;37:1055–63. Dubus IG, Brown CD, Beulke S. Sensitivity analyses for four pesticide leaching models. Pest Manag Sci 2003;59:962–82. Dyson JS, Beulke S, Brown CD, Lane MCG. Adsorption and degradation of the weak acid mesotrione in soil and environmental fate implications. J Environ Qual 2002;31:613–8. EFSA. Opinion of the scientific panel on plant health, plant protection products and their residues on a request of EFSA related to FOCUS groundwater models. EFSA J 2004;93: 1–20. EFSA. Scientific opinion of the panel on plant protection products and their residues on a request from EFSA related to the default Q10 value used to describe the temperature effect on transformation rates of pesticides in soil. EFSA J 2007;622:1–32. FOCUS. Groundwater scenarios in the EU review of active substances. Report of the FOCUS Groundwater Scenarios Workgroup, EC document reference Sanco/321/ 2000 rev.2; 2000. [202 pp.]. FOCUS. Generic guidance for focus groundwater scenarios. Report of the FOCUS Groundwater Scenarios Workgroup. Version 1.1; 2002. [61 pp.]. Garratt JA, Capri E, Trevisan M, Errera G, Wilkins RM. Parameterisation, evaluation and comparison of pesticide leaching models to data from a Bologna field site, Italy. Pest Manag Sci 2002;58:3–20. Garratt JA, Kennedy A, Wilkins RM, Ureña-Amate MD, González-Pradas E, FloresCéspedes F, et al. Modeling pesticide leaching and dissipation in a Mediterranean littoral greenhouse. J Agric Food Chem 2007;55:7052–61. Gottesbüren B, Aden K, Bärlund I, Brown C, Dust M, Görlitz G, et al. Comparison of pesticide leaching models: results using the Weiherbach data set. Agric Water Manag 2000;44:153–81.

545

Herbst M, Fialkiewicz W, Chen T, Pütz T, Thiéry D, Mouvet C, et al. Intercomparison of flow and transport models applied to vertical drainage in cropped lysimeters. Vadose Zone J 2005;4:240–54. Herrero-Hernández E, Andrades MS, Álvarez-Martín A, Pose-Juan E, Rodríguez-Cruz MS, Sánchez-Martín MJ. Occurrence of pesticides and some of their degradation products in waters in a Spanish wine region. J Hydrol 2013;486:234–45. Hildebrandt A, Guillamon M, Lacorte S, Tauler R, Barceló D. Impact of pesticides used in agriculture and vineyards to surface and groundwater quality (North Spain). Water Res 2008;42:3315–26. IUSS Working Group WRB. World Reference Base for Soil Resources 2006, first update 2007. World Soil Resources Reports No. 103. Rome: FAO; 2007. Jarvis NJ. Simulation of soil water dynamics and herbicide persistence in a silt loam soil using the MACRO model. Ecol Model 1995;81:97–109. Jarvis N, Larsbo M. MACRO (v5.2): model use, calibration, and validation. Trans ASABE 2012;55:1413–23. Jarvis NJ, Brown CD, Granitza E. Sources of error in model predictions of pesticide leaching: a case study using the MACRO model. Agric Water Manag 2000;44:247–62. Jarvis N, Lindahl A, Messing I, Stenemo F, Hollis J, Reichenberger S, et al. Algorithm to completely parameterize MACRO from basic soil property data and soil and crop management options. Report DL21 of the FP6 EU-funded FOOTPRINT project; 2007. [18 pp.]. Kolupaeva VN, Gorbatov VS, Shein EV, Leonova AA. The use of the PEARL model for assessing the migration of metribuzin in soil. Eurasian Soil Sci 2006;39:597–603. Kroes JG, van Dam JC, Groenendijk P, Hendriks RFA, Jacobs CMJ. SWAP version 3.2. Theory description and user manual. Alterra Rep 1649. The Netherlands: Alterra, Wageningen; 2008. [ISSN 1566-7197, 262 pp.]. Larsbo M, Jarvis NJMACRO 5.0. A model of water flow and solute transport in macroporous soil. Technical description. Rep EmergoUppsala, Sweden: Swedish University of Agricultural Sciences; 2003. [49 pp.]. Leistra M, van der Linden AMA, Boesten JJTI, Tiktak A, van den Berg F. PEARL model for pesticide behaviour and emissions in soil-plant systems: description of the processes. Alterra Rep 13. The Netherlands: Wageningen University and Research Centre, Wageningen; 2001. [115 pp.]. Mamy L, Gabrielle B, Barriuso E. Measurement and modelling of glyphosate fate compared with that of herbicides replaced as a result of the introduction of glyphosateresistant oilseed rape. Pest Manag Sci 2008;64:262–75. Nash JE, Sutcliffe JV. River flow forecasting through conceptual models, part I: a discussion of principles. J Hydrol 1970;10:282–90. Penman HL. Natural evaporation from open water, bare soil and grass. Proc R Soc Lond 1948;193:120–46. PPDB. The FOOTPRINT pesticide properties database. UK: University of Hertfordshire; 2014 [http://sitem.herts.ac.uk/aeru/footprint/es/index2.htm (accessed 12.03.14)]. Rawls WJ, Brakensiek DL, Saxton KE. Estimation of soil water properties. Trans ASAE 1982;1316–20. [Paper No. 81–2510]. Regulation EC No 1107/2009. Regulation of the European parliament and of the council of 21 October 2009 concerning the placing of plant protection products on the market and repealing Council Directives 79/117/EEC and 91/414/EEC. Off J Eur Union, L 309; 2009. p. 1–50. Reynolds WD, Elrick DE. Determination of hydraulic conductivity using a tension infiltrometer. Soil Sci Soc Am J 1991;55:633–9. Schroll R, Becher HH, Dörfler U, Gayler S, Grundmann S, Hartmann HP, et al. Quantifying the effect of soil moisture on the aerobic microbial mineralization of selected pesticides in different soils. Environ Sci Technol 2006;40:3305–12. Scorza Júnior RP, Boesten JJTI. Simulation of pesticide leaching in a cracking clay soil with the PEARL model. Pest Manag Sci 2005;61:432–48. Šimůnek J, van Genuchten MTh, Šejna M. Development and applications of HYDRUS and STANMOD software packages, and related codes. Vadose Zone J 2008;7:587–600. Smith J, Smith P, Addiscott T. Quantitative methods to evaluate and compare soil organic matter (SOM) models. NATO ASI Series, 38. ; 1996. p. 181–99. Tomlin CDS. The pesticide manual. Cambridge, UK: British Crop Protection Council; 2003. UIPP. Union des Industries de la Protection des Plantes website. http://www.uipp.org/Services-pro/Chiffres-cles/Production-vegetale-en-France-en-2011, 2014. (accessed 12. 03.14). Vanclooster M, Boesten JJTI. Application of pesticide simulation models to the Vredepeel dataset I. Water, solute and heat transport. Agric Water Manag 2000;44:105–17. Vanclooster M, Boesten JJTI, Trevisan M, Brown CD, Capri E, Eklo OM, et al. A European test of pesticide-leaching models: methodology and major recommendations. Agric Water Manag 2000;44:1–19. Vanclooster M, Pineros Garcet J, Boesten J, van den Berg F, Leistra M, Smelt J, et al. Effective approaches for assessing the predicted environmental concentrations of pesticides — final report. APECOP; 2003 [QLK4-CT-1999-01238, 158 pp.]. Walker A. A simulation model for prediction of herbicide persistence. J Environ Qual 1974;3:396–401. Wooding RA. Steady infiltration from a shallow circular pond. Water Resour Res 1968;4: 1259–73. Wösten JHM, Lilly A, Nemes A, Le Bas C. Development and use of a database of hydraulic properties of European soils. Geoderma 1999;90:169–85.